blob: 07a654486f75cfdfe13035f7bf501e2d2dbf0a4a [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>
David Rientjes5a3135c22007-10-16 23:25:53 -070033#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/notifier.h>
35#include <linux/topology.h>
36#include <linux/sysctl.h>
37#include <linux/cpu.h>
38#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070039#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <linux/nodemask.h>
41#include <linux/vmalloc.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080042#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070043#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070044#include <linux/sort.h>
45#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070046#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080047#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -070048#include <linux/page-isolation.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070049#include <linux/page_cgroup.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070050#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010051#include <linux/kmemleak.h>
Robert Jennings925cc712009-12-17 14:44:38 +000052#include <linux/memory.h>
Mel Gorman56de7262010-05-24 14:32:30 -070053#include <linux/compaction.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070054#include <trace/events/kmem.h>
Wu Fengguang718a3822010-03-10 15:20:43 -080055#include <linux/ftrace_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
57#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070058#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070059#include "internal.h"
60
Lee Schermerhorn72812012010-05-26 14:44:56 -070061#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
62DEFINE_PER_CPU(int, numa_node);
63EXPORT_PER_CPU_SYMBOL(numa_node);
64#endif
65
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070066#ifdef CONFIG_HAVE_MEMORYLESS_NODES
67/*
68 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
69 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
70 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
71 * defined in <linux/topology.h>.
72 */
73DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
74EXPORT_PER_CPU_SYMBOL(_numa_mem_);
75#endif
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077/*
Christoph Lameter13808912007-10-16 01:25:27 -070078 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 */
Christoph Lameter13808912007-10-16 01:25:27 -070080nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
81 [N_POSSIBLE] = NODE_MASK_ALL,
82 [N_ONLINE] = { { [0] = 1UL } },
83#ifndef CONFIG_NUMA
84 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
85#ifdef CONFIG_HIGHMEM
86 [N_HIGH_MEMORY] = { { [0] = 1UL } },
87#endif
88 [N_CPU] = { { [0] = 1UL } },
89#endif /* NUMA */
90};
91EXPORT_SYMBOL(node_states);
92
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070093unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070094unsigned long totalreserve_pages __read_mostly;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080095int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +100096gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Rafael J. Wysocki452aa692010-03-05 13:42:13 -080098#ifdef CONFIG_PM_SLEEP
99/*
100 * The following functions are used by the suspend/hibernate code to temporarily
101 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
102 * while devices are suspended. To avoid races with the suspend/hibernate code,
103 * they should always be called with pm_mutex held (gfp_allowed_mask also should
104 * only be modified with pm_mutex held, unless the suspend/hibernate code is
105 * guaranteed not to run in parallel with that modification).
106 */
107void set_gfp_allowed_mask(gfp_t mask)
108{
109 WARN_ON(!mutex_is_locked(&pm_mutex));
110 gfp_allowed_mask = mask;
111}
112
113gfp_t clear_gfp_allowed_mask(gfp_t mask)
114{
115 gfp_t ret = gfp_allowed_mask;
116
117 WARN_ON(!mutex_is_locked(&pm_mutex));
118 gfp_allowed_mask &= ~mask;
119 return ret;
120}
121#endif /* CONFIG_PM_SLEEP */
122
Mel Gormand9c23402007-10-16 01:26:01 -0700123#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
124int pageblock_order __read_mostly;
125#endif
126
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800127static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800128
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129/*
130 * results with 256, 32 in the lowmem_reserve sysctl:
131 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
132 * 1G machine -> (16M dma, 784M normal, 224M high)
133 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
134 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
135 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100136 *
137 * TBD: should special case ZONE_DMA32 machines here - in those we normally
138 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700140int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800141#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700142 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800143#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700144#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700145 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700146#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700147#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700148 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700149#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700150 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700151};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
153EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
Helge Deller15ad7cd2006-12-06 20:40:36 -0800155static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800156#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700157 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800158#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700159#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700160 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700161#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700162 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700163#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700164 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700165#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700166 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700167};
168
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169int min_free_kbytes = 1024;
170
Jan Beulich2c85f512009-09-21 17:03:07 -0700171static unsigned long __meminitdata nr_kernel_pages;
172static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700173static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Mel Gormanc7132162006-09-27 01:49:43 -0700175#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
176 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200177 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
Mel Gormanc7132162006-09-27 01:49:43 -0700178 * ranges of memory (RAM) that may be registered with add_active_range().
179 * Ranges passed to add_active_range() will be merged if possible
180 * so the number of times add_active_range() can be called is
181 * related to the number of nodes and the number of holes
182 */
183 #ifdef CONFIG_MAX_ACTIVE_REGIONS
184 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
185 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
186 #else
187 #if MAX_NUMNODES >= 32
188 /* If there can be many nodes, allow up to 50 holes per node */
189 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
190 #else
191 /* By default, allow up to 256 distinct regions */
192 #define MAX_ACTIVE_REGIONS 256
193 #endif
194 #endif
195
Jan Beulich98011f52007-07-15 23:38:17 -0700196 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
197 static int __meminitdata nr_nodemap_entries;
198 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
199 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Adrian Bunkb69a7282008-07-23 21:28:12 -0700200 static unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700201 static unsigned long __initdata required_movablecore;
Adrian Bunkb69a7282008-07-23 21:28:12 -0700202 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gorman2a1e2742007-07-17 04:03:12 -0700203
204 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
205 int movable_zone;
206 EXPORT_SYMBOL(movable_zone);
Mel Gormanc7132162006-09-27 01:49:43 -0700207#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
208
Miklos Szeredi418508c2007-05-23 13:57:55 -0700209#if MAX_NUMNODES > 1
210int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700211int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700212EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700213EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700214#endif
215
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700216int page_group_by_mobility_disabled __read_mostly;
217
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700218static void set_pageblock_migratetype(struct page *page, int migratetype)
219{
Mel Gorman49255c62009-06-16 15:31:58 -0700220
221 if (unlikely(page_group_by_mobility_disabled))
222 migratetype = MIGRATE_UNMOVABLE;
223
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700224 set_pageblock_flags_group(page, (unsigned long)migratetype,
225 PB_migrate, PB_migrate_end);
226}
227
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700228bool oom_killer_disabled __read_mostly;
229
Nick Piggin13e74442006-01-06 00:10:58 -0800230#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700231static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700233 int ret = 0;
234 unsigned seq;
235 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700236
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700237 do {
238 seq = zone_span_seqbegin(zone);
239 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
240 ret = 1;
241 else if (pfn < zone->zone_start_pfn)
242 ret = 1;
243 } while (zone_span_seqretry(zone, seq));
244
245 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700246}
247
248static int page_is_consistent(struct zone *zone, struct page *page)
249{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700250 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700251 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700253 return 0;
254
255 return 1;
256}
257/*
258 * Temporary debugging check for pages not lying within a given zone.
259 */
260static int bad_range(struct zone *zone, struct page *page)
261{
262 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700264 if (!page_is_consistent(zone, page))
265 return 1;
266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 return 0;
268}
Nick Piggin13e74442006-01-06 00:10:58 -0800269#else
270static inline int bad_range(struct zone *zone, struct page *page)
271{
272 return 0;
273}
274#endif
275
Nick Piggin224abf92006-01-06 00:11:11 -0800276static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800278 static unsigned long resume;
279 static unsigned long nr_shown;
280 static unsigned long nr_unshown;
281
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200282 /* Don't complain about poisoned pages */
283 if (PageHWPoison(page)) {
284 __ClearPageBuddy(page);
285 return;
286 }
287
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800288 /*
289 * Allow a burst of 60 reports, then keep quiet for that minute;
290 * or allow a steady drip of one report per second.
291 */
292 if (nr_shown == 60) {
293 if (time_before(jiffies, resume)) {
294 nr_unshown++;
295 goto out;
296 }
297 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800298 printk(KERN_ALERT
299 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800300 nr_unshown);
301 nr_unshown = 0;
302 }
303 nr_shown = 0;
304 }
305 if (nr_shown++ == 0)
306 resume = jiffies + 60 * HZ;
307
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800308 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800309 current->comm, page_to_pfn(page));
Wu Fengguang718a3822010-03-10 15:20:43 -0800310 dump_page(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700311
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800313out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800314 /* Leave bad fields for debug, except PageBuddy could make trouble */
315 __ClearPageBuddy(page);
Randy Dunlap9f158332005-09-13 01:25:16 -0700316 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317}
318
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319/*
320 * Higher-order pages are called "compound pages". They are structured thusly:
321 *
322 * The first PAGE_SIZE page is called the "head page".
323 *
324 * The remaining PAGE_SIZE pages are called "tail pages".
325 *
326 * All pages have PG_compound set. All pages have their ->private pointing at
327 * the head page (even the head page has this).
328 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800329 * The first tail page's ->lru.next holds the address of the compound page's
330 * put_page() function. Its ->lru.prev holds the order of allocation.
331 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800333
334static void free_compound_page(struct page *page)
335{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700336 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800337}
338
Andi Kleen01ad1c02008-07-23 21:27:46 -0700339void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340{
341 int i;
342 int nr_pages = 1 << order;
343
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800344 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700345 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700346 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800347 for (i = 1; i < nr_pages; i++) {
348 struct page *p = page + i;
349
Christoph Lameterd85f3382007-05-06 14:49:39 -0700350 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700351 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 }
353}
354
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800355static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800356{
357 int i;
358 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800359 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800360
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800361 if (unlikely(compound_order(page) != order) ||
362 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800363 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800364 bad++;
365 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
Christoph Lameter6d777952007-05-06 14:49:40 -0700367 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800368
Andy Whitcroft18229df2008-11-06 12:53:27 -0800369 for (i = 1; i < nr_pages; i++) {
370 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
Alexey Zaytseve713a212009-01-10 02:47:57 +0300372 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800373 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800374 bad++;
375 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700376 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800378
379 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
Nick Piggin17cf4402006-03-22 00:08:41 -0800382static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
383{
384 int i;
385
Andrew Morton6626c5d2006-03-22 00:08:42 -0800386 /*
387 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
388 * and __GFP_HIGHMEM from hard or soft interrupt context.
389 */
Nick Piggin725d7042006-09-25 23:30:55 -0700390 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800391 for (i = 0; i < (1 << order); i++)
392 clear_highpage(page + i);
393}
394
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700395static inline void set_page_order(struct page *page, int order)
396{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700397 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000398 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399}
400
401static inline void rmv_page_order(struct page *page)
402{
Nick Piggin676165a2006-04-10 11:21:48 +1000403 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700404 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405}
406
407/*
408 * Locate the struct page for both the matching buddy in our
409 * pair (buddy1) and the combined O(n+1) page they form (page).
410 *
411 * 1) Any buddy B1 will have an order O twin B2 which satisfies
412 * the following equation:
413 * B2 = B1 ^ (1 << O)
414 * For example, if the starting buddy (buddy2) is #8 its order
415 * 1 buddy is #10:
416 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
417 *
418 * 2) Any buddy B will have an order O+1 parent P which
419 * satisfies the following equation:
420 * P = B & ~(1 << O)
421 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200422 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 */
424static inline struct page *
425__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
426{
427 unsigned long buddy_idx = page_idx ^ (1 << order);
428
429 return page + (buddy_idx - page_idx);
430}
431
432static inline unsigned long
433__find_combined_index(unsigned long page_idx, unsigned int order)
434{
435 return (page_idx & ~(1 << order));
436}
437
438/*
439 * This function checks whether a page is free && is the buddy
440 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800441 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000442 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700443 * (c) a page and its buddy have the same order &&
444 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 *
Nick Piggin676165a2006-04-10 11:21:48 +1000446 * For recording whether a page is in the buddy system, we use PG_buddy.
447 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
448 *
449 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700451static inline int page_is_buddy(struct page *page, struct page *buddy,
452 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700454 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800455 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800456
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700457 if (page_zone_id(page) != page_zone_id(buddy))
458 return 0;
459
460 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700461 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700462 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000463 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700464 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465}
466
467/*
468 * Freeing function for a buddy system allocator.
469 *
470 * The concept of a buddy system is to maintain direct-mapped table
471 * (containing bit values) for memory blocks of various "orders".
472 * The bottom level table contains the map for the smallest allocatable
473 * units of memory (here, pages), and each level above it describes
474 * pairs of units from the levels below, hence, "buddies".
475 * At a high level, all that happens here is marking the table entry
476 * at the bottom level available, and propagating the changes upward
477 * as necessary, plus some accounting needed to play nicely with other
478 * parts of the VM system.
479 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000480 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700481 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 * So when we are allocating or freeing one, we can derive the state of the
483 * other. That is, if we allocate a small block, and both were
484 * free, the remainder of the region must be split into blocks.
485 * If a block is freed, and its buddy is also free, then this
486 * triggers coalescing into a block of larger size.
487 *
488 * -- wli
489 */
490
Nick Piggin48db57f2006-01-08 01:00:42 -0800491static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700492 struct zone *zone, unsigned int order,
493 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494{
495 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700496 unsigned long combined_idx;
497 struct page *buddy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
Nick Piggin224abf92006-01-06 00:11:11 -0800499 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800500 if (unlikely(destroy_compound_page(page, order)))
501 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
Mel Gormaned0ae212009-06-16 15:32:07 -0700503 VM_BUG_ON(migratetype == -1);
504
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
506
Mel Gormanf2260e62009-06-16 15:32:13 -0700507 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700508 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 while (order < MAX_ORDER-1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700512 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700513 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800514
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700515 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700517 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800519 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 page = page + (combined_idx - page_idx);
521 page_idx = combined_idx;
522 order++;
523 }
524 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700525
526 /*
527 * If this is not the largest possible page, check if the buddy
528 * of the next-highest order is free. If it is, it's possible
529 * that pages are being freed that will coalesce soon. In case,
530 * that is happening, add the free page to the tail of the list
531 * so it's less likely to be used soon and more likely to be merged
532 * as a higher order page
533 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700534 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700535 struct page *higher_page, *higher_buddy;
536 combined_idx = __find_combined_index(page_idx, order);
537 higher_page = page + combined_idx - page_idx;
538 higher_buddy = __page_find_buddy(higher_page, combined_idx, order + 1);
539 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
540 list_add_tail(&page->lru,
541 &zone->free_area[order].free_list[migratetype]);
542 goto out;
543 }
544 }
545
546 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
547out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 zone->free_area[order].nr_free++;
549}
550
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700551/*
552 * free_page_mlock() -- clean up attempts to free and mlocked() page.
553 * Page should not be on lru, so no need to fix that up.
554 * free_pages_check() will verify...
555 */
556static inline void free_page_mlock(struct page *page)
557{
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700558 __dec_zone_page_state(page, NR_MLOCK);
559 __count_vm_event(UNEVICTABLE_MLOCKFREED);
560}
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700561
Nick Piggin224abf92006-01-06 00:11:11 -0800562static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563{
Nick Piggin92be2e332006-01-06 00:10:57 -0800564 if (unlikely(page_mapcount(page) |
565 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700566 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800567 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800568 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800569 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800570 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800571 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
572 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
573 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574}
575
576/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700577 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700579 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 *
581 * If the zone was previously in an "all pages pinned" state then look to
582 * see if this freeing clears that state.
583 *
584 * And clear the zone's pages_scanned counter, to hold off the "all pages are
585 * pinned" detection logic.
586 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700587static void free_pcppages_bulk(struct zone *zone, int count,
588 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700590 int migratetype = 0;
Mel Gormana6f9edd62009-09-21 17:03:20 -0700591 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700592 int to_free = count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700593
Nick Pigginc54ad302006-01-06 00:10:56 -0800594 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800595 zone->all_unreclaimable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700597
Mel Gorman72853e22010-09-09 16:38:16 -0700598 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800599 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700600 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800601
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700602 /*
Mel Gormana6f9edd62009-09-21 17:03:20 -0700603 * Remove pages from lists in a round-robin fashion. A
604 * batch_free count is maintained that is incremented when an
605 * empty list is encountered. This is so more pages are freed
606 * off fuller lists instead of spinning excessively around empty
607 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700608 */
609 do {
Mel Gormana6f9edd62009-09-21 17:03:20 -0700610 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700611 if (++migratetype == MIGRATE_PCPTYPES)
612 migratetype = 0;
613 list = &pcp->lists[migratetype];
614 } while (list_empty(list));
615
Mel Gormana6f9edd62009-09-21 17:03:20 -0700616 do {
617 page = list_entry(list->prev, struct page, lru);
618 /* must delete as __free_one_page list manipulates */
619 list_del(&page->lru);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000620 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
621 __free_one_page(page, zone, 0, page_private(page));
622 trace_mm_page_pcpu_drain(page, 0, page_private(page));
Mel Gorman72853e22010-09-09 16:38:16 -0700623 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 }
Mel Gorman72853e22010-09-09 16:38:16 -0700625 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
Nick Pigginc54ad302006-01-06 00:10:56 -0800626 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627}
628
Mel Gormaned0ae212009-06-16 15:32:07 -0700629static void free_one_page(struct zone *zone, struct page *page, int order,
630 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800631{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700632 spin_lock(&zone->lock);
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -0800633 zone->all_unreclaimable = 0;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700634 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700635
Mel Gormaned0ae212009-06-16 15:32:07 -0700636 __free_one_page(page, zone, order, migratetype);
Mel Gorman72853e22010-09-09 16:38:16 -0700637 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700638 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800639}
640
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700641static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800644 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
Li Hongf6503162010-03-05 13:41:52 -0800646 trace_mm_page_free_direct(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100647 kmemcheck_free_shadow(page, order);
648
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700649 for (i = 0; i < (1 << order); i++) {
650 struct page *pg = page + i;
651
652 if (PageAnon(pg))
653 pg->mapping = NULL;
654 bad += free_pages_check(pg);
655 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800656 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700657 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800658
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700659 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700660 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700661 debug_check_no_obj_freed(page_address(page),
662 PAGE_SIZE << order);
663 }
Nick Piggindafb1362006-10-11 01:21:30 -0700664 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800665 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700666
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700667 return true;
668}
669
670static void __free_pages_ok(struct page *page, unsigned int order)
671{
672 unsigned long flags;
673 int wasMlocked = __TestClearPageMlocked(page);
674
675 if (!free_pages_prepare(page, order))
676 return;
677
Nick Pigginc54ad302006-01-06 00:10:56 -0800678 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200679 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700680 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700681 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700682 free_one_page(page_zone(page), page, order,
683 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800684 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685}
686
David Howellsa226f6c2006-01-06 00:11:08 -0800687/*
688 * permit the bootmem allocator to evade page validation on high-order frees
689 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700690void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800691{
692 if (order == 0) {
693 __ClearPageReserved(page);
694 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800695 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800696 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800697 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800698 int loop;
699
Nick Piggin545b1ea2006-03-22 00:08:07 -0800700 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800701 for (loop = 0; loop < BITS_PER_LONG; loop++) {
702 struct page *p = &page[loop];
703
Nick Piggin545b1ea2006-03-22 00:08:07 -0800704 if (loop + 1 < BITS_PER_LONG)
705 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800706 __ClearPageReserved(p);
707 set_page_count(p, 0);
708 }
709
Nick Piggin7835e982006-03-22 00:08:40 -0800710 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800711 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800712 }
713}
714
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715
716/*
717 * The order of subdivision here is critical for the IO subsystem.
718 * Please do not alter this order without good reasons and regression
719 * testing. Specifically, as large blocks of memory are subdivided,
720 * the order in which smaller blocks are delivered depends on the order
721 * they're subdivided in this function. This is the primary factor
722 * influencing the order in which pages are delivered to the IO
723 * subsystem according to empirical testing, and this is also justified
724 * by considering the behavior of a buddy system containing a single
725 * large block of memory acted on by a series of small allocations.
726 * This behavior is a critical factor in sglist merging's success.
727 *
728 * -- wli
729 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800730static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700731 int low, int high, struct free_area *area,
732 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733{
734 unsigned long size = 1 << high;
735
736 while (high > low) {
737 area--;
738 high--;
739 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700740 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700741 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 area->nr_free++;
743 set_page_order(&page[size], high);
744 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745}
746
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747/*
748 * This page is about to be returned from the page allocator
749 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200750static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751{
Nick Piggin92be2e332006-01-06 00:10:57 -0800752 if (unlikely(page_mapcount(page) |
753 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700754 (atomic_read(&page->_count) != 0) |
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800755 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800756 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800757 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800758 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200759 return 0;
760}
761
762static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
763{
764 int i;
765
766 for (i = 0; i < (1 << order); i++) {
767 struct page *p = page + i;
768 if (unlikely(check_new_page(p)))
769 return 1;
770 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800771
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700772 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800773 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800774
775 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800777
778 if (gfp_flags & __GFP_ZERO)
779 prep_zero_page(page, order, gfp_flags);
780
781 if (order && (gfp_flags & __GFP_COMP))
782 prep_compound_page(page, order);
783
Hugh Dickins689bceb2005-11-21 21:32:20 -0800784 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785}
786
Mel Gorman56fd56b2007-10-16 01:25:58 -0700787/*
788 * Go through the free lists for the given migratetype and remove
789 * the smallest available page from the freelists
790 */
Mel Gorman728ec982009-06-16 15:32:04 -0700791static inline
792struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700793 int migratetype)
794{
795 unsigned int current_order;
796 struct free_area * area;
797 struct page *page;
798
799 /* Find a page of the appropriate size in the preferred list */
800 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
801 area = &(zone->free_area[current_order]);
802 if (list_empty(&area->free_list[migratetype]))
803 continue;
804
805 page = list_entry(area->free_list[migratetype].next,
806 struct page, lru);
807 list_del(&page->lru);
808 rmv_page_order(page);
809 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700810 expand(zone, page, order, current_order, area, migratetype);
811 return page;
812 }
813
814 return NULL;
815}
816
817
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700818/*
819 * This array describes the order lists are fallen back to when
820 * the free lists for the desirable migrate type are depleted
821 */
822static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700823 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
824 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
825 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
826 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700827};
828
Mel Gormanc361be52007-10-16 01:25:51 -0700829/*
830 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700831 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700832 * boundary. If alignment is required, use move_freepages_block()
833 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700834static int move_freepages(struct zone *zone,
835 struct page *start_page, struct page *end_page,
836 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700837{
838 struct page *page;
839 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700840 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700841
842#ifndef CONFIG_HOLES_IN_ZONE
843 /*
844 * page_zone is not safe to call in this context when
845 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
846 * anyway as we check zone boundaries in move_freepages_block().
847 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700848 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700849 */
850 BUG_ON(page_zone(start_page) != page_zone(end_page));
851#endif
852
853 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700854 /* Make sure we are not inadvertently changing nodes */
855 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
856
Mel Gormanc361be52007-10-16 01:25:51 -0700857 if (!pfn_valid_within(page_to_pfn(page))) {
858 page++;
859 continue;
860 }
861
862 if (!PageBuddy(page)) {
863 page++;
864 continue;
865 }
866
867 order = page_order(page);
868 list_del(&page->lru);
869 list_add(&page->lru,
870 &zone->free_area[order].free_list[migratetype]);
871 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700872 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700873 }
874
Mel Gormand1003132007-10-16 01:26:00 -0700875 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700876}
877
Adrian Bunkb69a7282008-07-23 21:28:12 -0700878static int move_freepages_block(struct zone *zone, struct page *page,
879 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700880{
881 unsigned long start_pfn, end_pfn;
882 struct page *start_page, *end_page;
883
884 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700885 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700886 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700887 end_page = start_page + pageblock_nr_pages - 1;
888 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700889
890 /* Do not cross zone boundaries */
891 if (start_pfn < zone->zone_start_pfn)
892 start_page = page;
893 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
894 return 0;
895
896 return move_freepages(zone, start_page, end_page, migratetype);
897}
898
Mel Gorman2f66a682009-09-21 17:02:31 -0700899static void change_pageblock_range(struct page *pageblock_page,
900 int start_order, int migratetype)
901{
902 int nr_pageblocks = 1 << (start_order - pageblock_order);
903
904 while (nr_pageblocks--) {
905 set_pageblock_migratetype(pageblock_page, migratetype);
906 pageblock_page += pageblock_nr_pages;
907 }
908}
909
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700910/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -0700911static inline struct page *
912__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700913{
914 struct free_area * area;
915 int current_order;
916 struct page *page;
917 int migratetype, i;
918
919 /* Find the largest possible block of pages in the other list */
920 for (current_order = MAX_ORDER-1; current_order >= order;
921 --current_order) {
922 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
923 migratetype = fallbacks[start_migratetype][i];
924
Mel Gorman56fd56b2007-10-16 01:25:58 -0700925 /* MIGRATE_RESERVE handled later if necessary */
926 if (migratetype == MIGRATE_RESERVE)
927 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700928
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700929 area = &(zone->free_area[current_order]);
930 if (list_empty(&area->free_list[migratetype]))
931 continue;
932
933 page = list_entry(area->free_list[migratetype].next,
934 struct page, lru);
935 area->nr_free--;
936
937 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700938 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700939 * pages to the preferred allocation list. If falling
940 * back for a reclaimable kernel allocation, be more
941 * agressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700942 */
Mel Gormand9c23402007-10-16 01:26:01 -0700943 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gormandd5d2412009-09-05 11:17:11 -0700944 start_migratetype == MIGRATE_RECLAIMABLE ||
945 page_group_by_mobility_disabled) {
Mel Gorman46dafbc2007-10-16 01:25:55 -0700946 unsigned long pages;
947 pages = move_freepages_block(zone, page,
948 start_migratetype);
949
950 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -0700951 if (pages >= (1 << (pageblock_order-1)) ||
952 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -0700953 set_pageblock_migratetype(page,
954 start_migratetype);
955
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700956 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700957 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700958
959 /* Remove the page from the freelists */
960 list_del(&page->lru);
961 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700962
Mel Gorman2f66a682009-09-21 17:02:31 -0700963 /* Take ownership for orders >= pageblock_order */
964 if (current_order >= pageblock_order)
965 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700966 start_migratetype);
967
968 expand(zone, page, order, current_order, area, migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -0700969
970 trace_mm_page_alloc_extfrag(page, order, current_order,
971 start_migratetype, migratetype);
972
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700973 return page;
974 }
975 }
976
Mel Gorman728ec982009-06-16 15:32:04 -0700977 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700978}
979
Mel Gorman56fd56b2007-10-16 01:25:58 -0700980/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 * Do the hard work of removing an element from the buddy allocator.
982 * Call me with the zone->lock already held.
983 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700984static struct page *__rmqueue(struct zone *zone, unsigned int order,
985 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 struct page *page;
988
Mel Gorman728ec982009-06-16 15:32:04 -0700989retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -0700990 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
Mel Gorman728ec982009-06-16 15:32:04 -0700992 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Mel Gorman56fd56b2007-10-16 01:25:58 -0700993 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700994
Mel Gorman728ec982009-06-16 15:32:04 -0700995 /*
996 * Use MIGRATE_RESERVE rather than fail an allocation. goto
997 * is used because __rmqueue_smallest is an inline function
998 * and we want just one call site
999 */
1000 if (!page) {
1001 migratetype = MIGRATE_RESERVE;
1002 goto retry_reserve;
1003 }
1004 }
1005
Mel Gorman0d3d0622009-09-21 17:02:44 -07001006 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001007 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008}
1009
1010/*
1011 * Obtain a specified number of elements from the buddy allocator, all under
1012 * a single hold of the lock, for efficiency. Add them to the supplied list.
1013 * Returns the number of new pages which were placed at *list.
1014 */
1015static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001016 unsigned long count, struct list_head *list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001017 int migratetype, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
Nick Pigginc54ad302006-01-06 00:10:56 -08001021 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001023 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001024 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001026
1027 /*
1028 * Split buddy pages returned by expand() are received here
1029 * in physical page order. The page is added to the callers and
1030 * list and the list head then moves forward. From the callers
1031 * perspective, the linked list is ordered by page number in
1032 * some conditions. This is useful for IO devices that can
1033 * merge IO requests if the physical pages are ordered
1034 * properly.
1035 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001036 if (likely(cold == 0))
1037 list_add(&page->lru, list);
1038 else
1039 list_add_tail(&page->lru, list);
Mel Gorman535131e62007-10-16 01:25:49 -07001040 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001041 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001043 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001044 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001045 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046}
1047
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001048#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001049/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001050 * Called from the vmstat counter updater to drain pagesets of this
1051 * currently executing processor on remote nodes after they have
1052 * expired.
1053 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001054 * Note that this function must be called with the thread pinned to
1055 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001056 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001057void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001058{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001059 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001060 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001061
Christoph Lameter4037d452007-05-09 02:35:14 -07001062 local_irq_save(flags);
1063 if (pcp->count >= pcp->batch)
1064 to_drain = pcp->batch;
1065 else
1066 to_drain = pcp->count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001067 free_pcppages_bulk(zone, to_drain, pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -07001068 pcp->count -= to_drain;
1069 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001070}
1071#endif
1072
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001073/*
1074 * Drain pages of the indicated processor.
1075 *
1076 * The processor must either be the current processor and the
1077 * thread pinned to the current processor or a processor that
1078 * is not online.
1079 */
1080static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081{
Nick Pigginc54ad302006-01-06 00:10:56 -08001082 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001085 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001087 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001089 local_irq_save(flags);
1090 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001092 pcp = &pset->pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001093 free_pcppages_bulk(zone, pcp->count, pcp);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001094 pcp->count = 0;
1095 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 }
1097}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001099/*
1100 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1101 */
1102void drain_local_pages(void *arg)
1103{
1104 drain_pages(smp_processor_id());
1105}
1106
1107/*
1108 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1109 */
1110void drain_all_pages(void)
1111{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001112 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001113}
1114
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001115#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
1117void mark_free_pages(struct zone *zone)
1118{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001119 unsigned long pfn, max_zone_pfn;
1120 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001121 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 struct list_head *curr;
1123
1124 if (!zone->spanned_pages)
1125 return;
1126
1127 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001128
1129 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1130 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1131 if (pfn_valid(pfn)) {
1132 struct page *page = pfn_to_page(pfn);
1133
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001134 if (!swsusp_page_is_forbidden(page))
1135 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001136 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001138 for_each_migratetype_order(order, t) {
1139 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001140 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001142 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1143 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001144 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001145 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 spin_unlock_irqrestore(&zone->lock, flags);
1148}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001149#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150
1151/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001153 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 */
Li Hongfc916682010-03-05 13:41:54 -08001155void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156{
1157 struct zone *zone = page_zone(page);
1158 struct per_cpu_pages *pcp;
1159 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001160 int migratetype;
Johannes Weiner451ea252009-09-21 17:01:48 -07001161 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001163 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001164 return;
1165
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001166 migratetype = get_pageblock_migratetype(page);
1167 set_page_private(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001169 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001170 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001171 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001172
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001173 /*
1174 * We only track unmovable, reclaimable and movable on pcp lists.
1175 * Free ISOLATE pages back to the allocator because they are being
1176 * offlined but treat RESERVE as movable pages so we can get those
1177 * areas back if necessary. Otherwise, we may have to free
1178 * excessively into the page allocator
1179 */
1180 if (migratetype >= MIGRATE_PCPTYPES) {
1181 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1182 free_one_page(zone, page, 0, migratetype);
1183 goto out;
1184 }
1185 migratetype = MIGRATE_MOVABLE;
1186 }
1187
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001188 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001189 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001190 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001191 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001192 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001194 if (pcp->count >= pcp->high) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001195 free_pcppages_bulk(zone, pcp->batch, pcp);
Nick Piggin48db57f2006-01-08 01:00:42 -08001196 pcp->count -= pcp->batch;
1197 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001198
1199out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201}
1202
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001203/*
1204 * split_page takes a non-compound higher-order page, and splits it into
1205 * n (1<<order) sub-pages: page[0..n]
1206 * Each sub-page must be freed individually.
1207 *
1208 * Note: this is probably too low level an operation for use in drivers.
1209 * Please consult with lkml before using this in your driver.
1210 */
1211void split_page(struct page *page, unsigned int order)
1212{
1213 int i;
1214
Nick Piggin725d7042006-09-25 23:30:55 -07001215 VM_BUG_ON(PageCompound(page));
1216 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001217
1218#ifdef CONFIG_KMEMCHECK
1219 /*
1220 * Split shadow pages too, because free(page[0]) would
1221 * otherwise free the whole shadow.
1222 */
1223 if (kmemcheck_page_is_tracked(page))
1224 split_page(virt_to_page(page[0].shadow), order);
1225#endif
1226
Nick Piggin7835e982006-03-22 00:08:40 -08001227 for (i = 1; i < (1 << order); i++)
1228 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001229}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001230
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231/*
Mel Gorman748446b2010-05-24 14:32:27 -07001232 * Similar to split_page except the page is already free. As this is only
1233 * being used for migration, the migratetype of the block also changes.
1234 * As this is called with interrupts disabled, the caller is responsible
1235 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1236 * are enabled.
1237 *
1238 * Note: this is probably too low level an operation for use in drivers.
1239 * Please consult with lkml before using this in your driver.
1240 */
1241int split_free_page(struct page *page)
1242{
1243 unsigned int order;
1244 unsigned long watermark;
1245 struct zone *zone;
1246
1247 BUG_ON(!PageBuddy(page));
1248
1249 zone = page_zone(page);
1250 order = page_order(page);
1251
1252 /* Obey watermarks as if the page was being allocated */
1253 watermark = low_wmark_pages(zone) + (1 << order);
1254 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1255 return 0;
1256
1257 /* Remove page from free list */
1258 list_del(&page->lru);
1259 zone->free_area[order].nr_free--;
1260 rmv_page_order(page);
1261 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1262
1263 /* Split into individual pages */
1264 set_page_refcounted(page);
1265 split_page(page, order);
1266
1267 if (order >= pageblock_order - 1) {
1268 struct page *endpage = page + (1 << order) - 1;
1269 for (; page < endpage; page += pageblock_nr_pages)
1270 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1271 }
1272
1273 return 1 << order;
1274}
1275
1276/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1278 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1279 * or two.
1280 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001281static inline
1282struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001283 struct zone *zone, int order, gfp_t gfp_flags,
1284 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285{
1286 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001287 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 int cold = !!(gfp_flags & __GFP_COLD);
1289
Hugh Dickins689bceb2005-11-21 21:32:20 -08001290again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001291 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001293 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001296 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1297 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001298 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001299 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001300 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07001301 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001302 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001303 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07001304 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001305
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001306 if (cold)
1307 page = list_entry(list->prev, struct page, lru);
1308 else
1309 page = list_entry(list->next, struct page, lru);
1310
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001311 list_del(&page->lru);
1312 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001313 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001314 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1315 /*
1316 * __GFP_NOFAIL is not to be used in new code.
1317 *
1318 * All __GFP_NOFAIL callers should be fixed so that they
1319 * properly detect and handle allocation failures.
1320 *
1321 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001322 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001323 * __GFP_NOFAIL.
1324 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001325 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001326 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001328 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001329 spin_unlock(&zone->lock);
1330 if (!page)
1331 goto failed;
KOSAKI Motohiro6ccf80e2010-01-15 17:01:18 -08001332 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 }
1334
Christoph Lameterf8891e52006-06-30 01:55:45 -07001335 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Mel Gorman18ea7e72008-04-28 02:12:14 -07001336 zone_statistics(preferred_zone, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001337 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Nick Piggin725d7042006-09-25 23:30:55 -07001339 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001340 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001341 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001343
1344failed:
1345 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001346 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347}
1348
Mel Gorman41858962009-06-16 15:32:12 -07001349/* The ALLOC_WMARK bits are used as an index to zone->watermark */
1350#define ALLOC_WMARK_MIN WMARK_MIN
1351#define ALLOC_WMARK_LOW WMARK_LOW
1352#define ALLOC_WMARK_HIGH WMARK_HIGH
1353#define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1354
1355/* Mask to get the watermark bits */
1356#define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1357
Nick Piggin31488902005-11-28 13:44:03 -08001358#define ALLOC_HARDER 0x10 /* try to alloc harder */
1359#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1360#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001361
Akinobu Mita933e3122006-12-08 02:39:45 -08001362#ifdef CONFIG_FAIL_PAGE_ALLOC
1363
1364static struct fail_page_alloc_attr {
1365 struct fault_attr attr;
1366
1367 u32 ignore_gfp_highmem;
1368 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001369 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001370
1371#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1372
1373 struct dentry *ignore_gfp_highmem_file;
1374 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001375 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001376
1377#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1378
1379} fail_page_alloc = {
1380 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001381 .ignore_gfp_wait = 1,
1382 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001383 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001384};
1385
1386static int __init setup_fail_page_alloc(char *str)
1387{
1388 return setup_fault_attr(&fail_page_alloc.attr, str);
1389}
1390__setup("fail_page_alloc=", setup_fail_page_alloc);
1391
1392static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1393{
Akinobu Mita54114992007-07-15 23:40:23 -07001394 if (order < fail_page_alloc.min_order)
1395 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001396 if (gfp_mask & __GFP_NOFAIL)
1397 return 0;
1398 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1399 return 0;
1400 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1401 return 0;
1402
1403 return should_fail(&fail_page_alloc.attr, 1 << order);
1404}
1405
1406#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1407
1408static int __init fail_page_alloc_debugfs(void)
1409{
1410 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1411 struct dentry *dir;
1412 int err;
1413
1414 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1415 "fail_page_alloc");
1416 if (err)
1417 return err;
1418 dir = fail_page_alloc.attr.dentries.dir;
1419
1420 fail_page_alloc.ignore_gfp_wait_file =
1421 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1422 &fail_page_alloc.ignore_gfp_wait);
1423
1424 fail_page_alloc.ignore_gfp_highmem_file =
1425 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1426 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001427 fail_page_alloc.min_order_file =
1428 debugfs_create_u32("min-order", mode, dir,
1429 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001430
1431 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001432 !fail_page_alloc.ignore_gfp_highmem_file ||
1433 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001434 err = -ENOMEM;
1435 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1436 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001437 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001438 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1439 }
1440
1441 return err;
1442}
1443
1444late_initcall(fail_page_alloc_debugfs);
1445
1446#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1447
1448#else /* CONFIG_FAIL_PAGE_ALLOC */
1449
1450static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1451{
1452 return 0;
1453}
1454
1455#endif /* CONFIG_FAIL_PAGE_ALLOC */
1456
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457/*
1458 * Return 1 if free pages are above 'mark'. This takes into account the order
1459 * of the allocation.
1460 */
1461int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001462 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463{
1464 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001465 long min = mark;
Christoph Lameteraa454842010-09-09 16:38:17 -07001466 long free_pages = zone_nr_free_pages(z) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 int o;
1468
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001469 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001471 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 min -= min / 4;
1473
1474 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1475 return 0;
1476 for (o = 0; o < order; o++) {
1477 /* At the next order, this order's pages become unavailable */
1478 free_pages -= z->free_area[o].nr_free << o;
1479
1480 /* Require fewer higher order pages to be free */
1481 min >>= 1;
1482
1483 if (free_pages <= min)
1484 return 0;
1485 }
1486 return 1;
1487}
1488
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001489#ifdef CONFIG_NUMA
1490/*
1491 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1492 * skip over zones that are not allowed by the cpuset, or that have
1493 * been recently (in last second) found to be nearly full. See further
1494 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001495 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001496 *
1497 * If the zonelist cache is present in the passed in zonelist, then
1498 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001499 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001500 *
1501 * If the zonelist cache is not available for this zonelist, does
1502 * nothing and returns NULL.
1503 *
1504 * If the fullzones BITMAP in the zonelist cache is stale (more than
1505 * a second since last zap'd) then we zap it out (clear its bits.)
1506 *
1507 * We hold off even calling zlc_setup, until after we've checked the
1508 * first zone in the zonelist, on the theory that most allocations will
1509 * be satisfied from that first zone, so best to examine that zone as
1510 * quickly as we can.
1511 */
1512static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1513{
1514 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1515 nodemask_t *allowednodes; /* zonelist_cache approximation */
1516
1517 zlc = zonelist->zlcache_ptr;
1518 if (!zlc)
1519 return NULL;
1520
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001521 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001522 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1523 zlc->last_full_zap = jiffies;
1524 }
1525
1526 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1527 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001528 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001529 return allowednodes;
1530}
1531
1532/*
1533 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1534 * if it is worth looking at further for free memory:
1535 * 1) Check that the zone isn't thought to be full (doesn't have its
1536 * bit set in the zonelist_cache fullzones BITMAP).
1537 * 2) Check that the zones node (obtained from the zonelist_cache
1538 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1539 * Return true (non-zero) if zone is worth looking at further, or
1540 * else return false (zero) if it is not.
1541 *
1542 * This check -ignores- the distinction between various watermarks,
1543 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1544 * found to be full for any variation of these watermarks, it will
1545 * be considered full for up to one second by all requests, unless
1546 * we are so low on memory on all allowed nodes that we are forced
1547 * into the second scan of the zonelist.
1548 *
1549 * In the second scan we ignore this zonelist cache and exactly
1550 * apply the watermarks to all zones, even it is slower to do so.
1551 * We are low on memory in the second scan, and should leave no stone
1552 * unturned looking for a free page.
1553 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001554static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001555 nodemask_t *allowednodes)
1556{
1557 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1558 int i; /* index of *z in zonelist zones */
1559 int n; /* node that zone *z is on */
1560
1561 zlc = zonelist->zlcache_ptr;
1562 if (!zlc)
1563 return 1;
1564
Mel Gormandd1a2392008-04-28 02:12:17 -07001565 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001566 n = zlc->z_to_n[i];
1567
1568 /* This zone is worth trying if it is allowed but not full */
1569 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1570}
1571
1572/*
1573 * Given 'z' scanning a zonelist, set the corresponding bit in
1574 * zlc->fullzones, so that subsequent attempts to allocate a page
1575 * from that zone don't waste time re-examining it.
1576 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001577static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001578{
1579 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1580 int i; /* index of *z in zonelist zones */
1581
1582 zlc = zonelist->zlcache_ptr;
1583 if (!zlc)
1584 return;
1585
Mel Gormandd1a2392008-04-28 02:12:17 -07001586 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001587
1588 set_bit(i, zlc->fullzones);
1589}
1590
1591#else /* CONFIG_NUMA */
1592
1593static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1594{
1595 return NULL;
1596}
1597
Mel Gormandd1a2392008-04-28 02:12:17 -07001598static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001599 nodemask_t *allowednodes)
1600{
1601 return 1;
1602}
1603
Mel Gormandd1a2392008-04-28 02:12:17 -07001604static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001605{
1606}
1607#endif /* CONFIG_NUMA */
1608
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001609/*
Paul Jackson0798e512006-12-06 20:31:38 -08001610 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001611 * a page.
1612 */
1613static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001614get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001615 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001616 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001617{
Mel Gormandd1a2392008-04-28 02:12:17 -07001618 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001619 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001620 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001621 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001622 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1623 int zlc_active = 0; /* set if using zonelist_cache */
1624 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001625
Mel Gorman19770b32008-04-28 02:12:18 -07001626 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001627zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001628 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001629 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001630 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1631 */
Mel Gorman19770b32008-04-28 02:12:18 -07001632 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1633 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001634 if (NUMA_BUILD && zlc_active &&
1635 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1636 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001637 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001638 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001639 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001640
Mel Gorman41858962009-06-16 15:32:12 -07001641 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001642 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001643 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001644 int ret;
1645
Mel Gorman41858962009-06-16 15:32:12 -07001646 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001647 if (zone_watermark_ok(zone, order, mark,
1648 classzone_idx, alloc_flags))
1649 goto try_this_zone;
1650
1651 if (zone_reclaim_mode == 0)
1652 goto this_zone_full;
1653
1654 ret = zone_reclaim(zone, gfp_mask, order);
1655 switch (ret) {
1656 case ZONE_RECLAIM_NOSCAN:
1657 /* did not scan */
1658 goto try_next_zone;
1659 case ZONE_RECLAIM_FULL:
1660 /* scanned but unreclaimable */
1661 goto this_zone_full;
1662 default:
1663 /* did we reclaim enough */
1664 if (!zone_watermark_ok(zone, order, mark,
1665 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001666 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001667 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001668 }
1669
Mel Gormanfa5e0842009-06-16 15:33:22 -07001670try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001671 page = buffered_rmqueue(preferred_zone, zone, order,
1672 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001673 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001674 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001675this_zone_full:
1676 if (NUMA_BUILD)
1677 zlc_mark_zone_full(zonelist, z);
1678try_next_zone:
Christoph Lameter62bc62a2009-06-16 15:32:15 -07001679 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormand395b732009-06-16 15:32:09 -07001680 /*
1681 * we do zlc_setup after the first zone is tried but only
1682 * if there are multiple nodes make it worthwhile
1683 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001684 allowednodes = zlc_setup(zonelist, alloc_flags);
1685 zlc_active = 1;
1686 did_zlc_setup = 1;
1687 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001688 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001689
1690 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1691 /* Disable zlc cache for second zonelist scan */
1692 zlc_active = 0;
1693 goto zonelist_scan;
1694 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001695 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001696}
1697
Mel Gorman11e33f62009-06-16 15:31:57 -07001698static inline int
1699should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1700 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701{
Mel Gorman11e33f62009-06-16 15:31:57 -07001702 /* Do not loop if specifically requested */
1703 if (gfp_mask & __GFP_NORETRY)
1704 return 0;
1705
1706 /*
1707 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1708 * means __GFP_NOFAIL, but that may not be true in other
1709 * implementations.
1710 */
1711 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1712 return 1;
1713
1714 /*
1715 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1716 * specified, then we retry until we no longer reclaim any pages
1717 * (above), or we've reclaimed an order of pages at least as
1718 * large as the allocation's order. In both cases, if the
1719 * allocation still fails, we stop retrying.
1720 */
1721 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1722 return 1;
1723
1724 /*
1725 * Don't let big-order allocations loop unless the caller
1726 * explicitly requests that.
1727 */
1728 if (gfp_mask & __GFP_NOFAIL)
1729 return 1;
1730
1731 return 0;
1732}
1733
1734static inline struct page *
1735__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1736 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001737 nodemask_t *nodemask, struct zone *preferred_zone,
1738 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001739{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741
Mel Gorman11e33f62009-06-16 15:31:57 -07001742 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07001743 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07001744 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 return NULL;
1746 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001747
Mel Gorman11e33f62009-06-16 15:31:57 -07001748 /*
1749 * Go through the zonelist yet one more time, keep very high watermark
1750 * here, this is only to catch a parallel oom killing, we must fail if
1751 * we're still under heavy pressure.
1752 */
1753 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1754 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001755 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07001756 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001757 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07001758 goto out;
1759
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001760 if (!(gfp_mask & __GFP_NOFAIL)) {
1761 /* The OOM killer will not help higher order allocs */
1762 if (order > PAGE_ALLOC_COSTLY_ORDER)
1763 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07001764 /* The OOM killer does not needlessly kill tasks for lowmem */
1765 if (high_zoneidx < ZONE_NORMAL)
1766 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001767 /*
1768 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1769 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1770 * The caller should handle page allocation failure by itself if
1771 * it specifies __GFP_THISNODE.
1772 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1773 */
1774 if (gfp_mask & __GFP_THISNODE)
1775 goto out;
1776 }
Mel Gorman11e33f62009-06-16 15:31:57 -07001777 /* Exhausted what can be done so it's blamo time */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08001778 out_of_memory(zonelist, gfp_mask, order, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07001779
1780out:
1781 clear_zonelist_oom(zonelist, gfp_mask);
1782 return page;
1783}
1784
Mel Gorman56de7262010-05-24 14:32:30 -07001785#ifdef CONFIG_COMPACTION
1786/* Try memory compaction for high-order allocations before reclaim */
1787static struct page *
1788__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1789 struct zonelist *zonelist, enum zone_type high_zoneidx,
1790 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1791 int migratetype, unsigned long *did_some_progress)
1792{
1793 struct page *page;
1794
Mel Gorman4f92e252010-05-24 14:32:32 -07001795 if (!order || compaction_deferred(preferred_zone))
Mel Gorman56de7262010-05-24 14:32:30 -07001796 return NULL;
1797
1798 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
1799 nodemask);
1800 if (*did_some_progress != COMPACT_SKIPPED) {
1801
1802 /* Page migration frees to the PCP lists but we want merging */
1803 drain_pages(get_cpu());
1804 put_cpu();
1805
1806 page = get_page_from_freelist(gfp_mask, nodemask,
1807 order, zonelist, high_zoneidx,
1808 alloc_flags, preferred_zone,
1809 migratetype);
1810 if (page) {
Mel Gorman4f92e252010-05-24 14:32:32 -07001811 preferred_zone->compact_considered = 0;
1812 preferred_zone->compact_defer_shift = 0;
Mel Gorman56de7262010-05-24 14:32:30 -07001813 count_vm_event(COMPACTSUCCESS);
1814 return page;
1815 }
1816
1817 /*
1818 * It's bad if compaction run occurs and fails.
1819 * The most likely reason is that pages exist,
1820 * but not enough to satisfy watermarks.
1821 */
1822 count_vm_event(COMPACTFAIL);
Mel Gorman4f92e252010-05-24 14:32:32 -07001823 defer_compaction(preferred_zone);
Mel Gorman56de7262010-05-24 14:32:30 -07001824
1825 cond_resched();
1826 }
1827
1828 return NULL;
1829}
1830#else
1831static inline struct page *
1832__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1833 struct zonelist *zonelist, enum zone_type high_zoneidx,
1834 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1835 int migratetype, unsigned long *did_some_progress)
1836{
1837 return NULL;
1838}
1839#endif /* CONFIG_COMPACTION */
1840
Mel Gorman11e33f62009-06-16 15:31:57 -07001841/* The really slow allocator path where we enter direct reclaim */
1842static inline struct page *
1843__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1844 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07001845 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001846 int migratetype, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07001847{
1848 struct page *page = NULL;
1849 struct reclaim_state reclaim_state;
1850 struct task_struct *p = current;
Mel Gorman9ee493c2010-09-09 16:38:18 -07001851 bool drained = false;
Mel Gorman11e33f62009-06-16 15:31:57 -07001852
1853 cond_resched();
1854
1855 /* We now go into synchronous reclaim */
1856 cpuset_memory_pressure_bump();
Mel Gorman11e33f62009-06-16 15:31:57 -07001857 p->flags |= PF_MEMALLOC;
1858 lockdep_set_current_reclaim_state(gfp_mask);
1859 reclaim_state.reclaimed_slab = 0;
1860 p->reclaim_state = &reclaim_state;
1861
1862 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1863
1864 p->reclaim_state = NULL;
1865 lockdep_clear_current_reclaim_state();
1866 p->flags &= ~PF_MEMALLOC;
1867
1868 cond_resched();
1869
Mel Gorman9ee493c2010-09-09 16:38:18 -07001870 if (unlikely(!(*did_some_progress)))
1871 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07001872
Mel Gorman9ee493c2010-09-09 16:38:18 -07001873retry:
1874 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001875 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001876 alloc_flags, preferred_zone,
1877 migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07001878
1879 /*
1880 * If an allocation failed after direct reclaim, it could be because
1881 * pages are pinned on the per-cpu lists. Drain them and try again
1882 */
1883 if (!page && !drained) {
1884 drain_all_pages();
1885 drained = true;
1886 goto retry;
1887 }
1888
Mel Gorman11e33f62009-06-16 15:31:57 -07001889 return page;
1890}
1891
Mel Gorman11e33f62009-06-16 15:31:57 -07001892/*
1893 * This is called in the allocator slow-path if the allocation request is of
1894 * sufficient urgency to ignore watermarks and take other desperate measures
1895 */
1896static inline struct page *
1897__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1898 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001899 nodemask_t *nodemask, struct zone *preferred_zone,
1900 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001901{
1902 struct page *page;
1903
1904 do {
1905 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07001906 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07001907 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07001908
1909 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07001910 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07001911 } while (!page && (gfp_mask & __GFP_NOFAIL));
1912
1913 return page;
1914}
1915
1916static inline
1917void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1918 enum zone_type high_zoneidx)
1919{
1920 struct zoneref *z;
1921 struct zone *zone;
1922
1923 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1924 wakeup_kswapd(zone, order);
1925}
1926
Peter Zijlstra341ce062009-06-16 15:32:02 -07001927static inline int
1928gfp_to_alloc_flags(gfp_t gfp_mask)
1929{
1930 struct task_struct *p = current;
1931 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1932 const gfp_t wait = gfp_mask & __GFP_WAIT;
1933
Mel Gormana56f57f2009-06-16 15:32:02 -07001934 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07001935 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07001936
Peter Zijlstra341ce062009-06-16 15:32:02 -07001937 /*
1938 * The caller may dip into page reserves a bit more if the caller
1939 * cannot run direct reclaim, or if the caller has realtime scheduling
1940 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1941 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1942 */
Namhyung Kime6223a32010-10-26 14:21:59 -07001943 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07001944
1945 if (!wait) {
1946 alloc_flags |= ALLOC_HARDER;
1947 /*
1948 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1949 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1950 */
1951 alloc_flags &= ~ALLOC_CPUSET;
Mel Gorman9d0ed602009-11-11 14:26:17 -08001952 } else if (unlikely(rt_task(p)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07001953 alloc_flags |= ALLOC_HARDER;
1954
1955 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1956 if (!in_interrupt() &&
1957 ((p->flags & PF_MEMALLOC) ||
1958 unlikely(test_thread_flag(TIF_MEMDIE))))
1959 alloc_flags |= ALLOC_NO_WATERMARKS;
1960 }
1961
1962 return alloc_flags;
1963}
1964
Mel Gorman11e33f62009-06-16 15:31:57 -07001965static inline struct page *
1966__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1967 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07001968 nodemask_t *nodemask, struct zone *preferred_zone,
1969 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07001970{
1971 const gfp_t wait = gfp_mask & __GFP_WAIT;
1972 struct page *page = NULL;
1973 int alloc_flags;
1974 unsigned long pages_reclaimed = 0;
1975 unsigned long did_some_progress;
1976 struct task_struct *p = current;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001977
Christoph Lameter952f3b52006-12-06 20:33:26 -08001978 /*
Mel Gorman72807a72009-06-16 15:32:18 -07001979 * In the slowpath, we sanity check order to avoid ever trying to
1980 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
1981 * be using allocators in order of preference for an area that is
1982 * too large.
1983 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07001984 if (order >= MAX_ORDER) {
1985 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07001986 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07001987 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001988
Christoph Lameter952f3b52006-12-06 20:33:26 -08001989 /*
1990 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1991 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1992 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1993 * using a larger set of nodes after it has established that the
1994 * allowed per node queues are empty and that nodes are
1995 * over allocated.
1996 */
1997 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1998 goto nopage;
1999
Mel Gormancc4a6852009-11-11 14:26:14 -08002000restart:
Mel Gorman11e33f62009-06-16 15:31:57 -07002001 wake_all_kswapd(order, zonelist, high_zoneidx);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002002
Paul Jackson9bf22292005-09-06 15:18:12 -07002003 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002004 * OK, we're below the kswapd watermark and have kicked background
2005 * reclaim. Now things get more complex, so set up alloc_flags according
2006 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002007 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002008 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009
Peter Zijlstra341ce062009-06-16 15:32:02 -07002010 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002011 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002012 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2013 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002014 if (page)
2015 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08002017rebalance:
Mel Gorman11e33f62009-06-16 15:31:57 -07002018 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002019 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2020 page = __alloc_pages_high_priority(gfp_mask, order,
2021 zonelist, high_zoneidx, nodemask,
2022 preferred_zone, migratetype);
2023 if (page)
2024 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 }
2026
2027 /* Atomic allocations - we can't balance anything */
2028 if (!wait)
2029 goto nopage;
2030
Peter Zijlstra341ce062009-06-16 15:32:02 -07002031 /* Avoid recursion of direct reclaim */
2032 if (p->flags & PF_MEMALLOC)
2033 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034
David Rientjes6583bb62009-07-29 15:02:06 -07002035 /* Avoid allocations with no watermarks from looping endlessly */
2036 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2037 goto nopage;
2038
Mel Gorman56de7262010-05-24 14:32:30 -07002039 /* Try direct compaction */
2040 page = __alloc_pages_direct_compact(gfp_mask, order,
2041 zonelist, high_zoneidx,
2042 nodemask,
2043 alloc_flags, preferred_zone,
2044 migratetype, &did_some_progress);
2045 if (page)
2046 goto got_pg;
2047
Mel Gorman11e33f62009-06-16 15:31:57 -07002048 /* Try direct reclaim and then allocating */
2049 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2050 zonelist, high_zoneidx,
2051 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002052 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002053 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002054 if (page)
2055 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002058 * If we failed to make any progress reclaiming, then we are
2059 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002061 if (!did_some_progress) {
2062 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002063 if (oom_killer_disabled)
2064 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002065 page = __alloc_pages_may_oom(gfp_mask, order,
2066 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002067 nodemask, preferred_zone,
2068 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002069 if (page)
2070 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071
David Rientjes03668b32010-08-09 17:18:54 -07002072 if (!(gfp_mask & __GFP_NOFAIL)) {
2073 /*
2074 * The oom killer is not called for high-order
2075 * allocations that may fail, so if no progress
2076 * is being made, there are no other options and
2077 * retrying is unlikely to help.
2078 */
2079 if (order > PAGE_ALLOC_COSTLY_ORDER)
2080 goto nopage;
2081 /*
2082 * The oom killer is not called for lowmem
2083 * allocations to prevent needlessly killing
2084 * innocent tasks.
2085 */
2086 if (high_zoneidx < ZONE_NORMAL)
2087 goto nopage;
2088 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 goto restart;
2091 }
2092 }
2093
Mel Gorman11e33f62009-06-16 15:31:57 -07002094 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002095 pages_reclaimed += did_some_progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002096 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
2097 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002098 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 goto rebalance;
2100 }
2101
2102nopage:
2103 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
2104 printk(KERN_WARNING "%s: page allocation failure."
2105 " order:%d, mode:0x%x\n",
2106 p->comm, order, gfp_mask);
2107 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07002108 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 }
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002110 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002112 if (kmemcheck_enabled)
2113 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07002115
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116}
Mel Gorman11e33f62009-06-16 15:31:57 -07002117
2118/*
2119 * This is the 'heart' of the zoned buddy allocator.
2120 */
2121struct page *
2122__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2123 struct zonelist *zonelist, nodemask_t *nodemask)
2124{
2125 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002126 struct zone *preferred_zone;
Mel Gorman11e33f62009-06-16 15:31:57 -07002127 struct page *page;
Mel Gorman3dd28262009-06-16 15:32:00 -07002128 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002129
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002130 gfp_mask &= gfp_allowed_mask;
2131
Mel Gorman11e33f62009-06-16 15:31:57 -07002132 lockdep_trace_alloc(gfp_mask);
2133
2134 might_sleep_if(gfp_mask & __GFP_WAIT);
2135
2136 if (should_fail_alloc_page(gfp_mask, order))
2137 return NULL;
2138
2139 /*
2140 * Check the zones suitable for the gfp_mask contain at least one
2141 * valid zone. It's possible to have an empty zonelist as a result
2142 * of GFP_THISNODE and a memoryless node
2143 */
2144 if (unlikely(!zonelist->_zonerefs->zone))
2145 return NULL;
2146
Miao Xiec0ff7452010-05-24 14:32:08 -07002147 get_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002148 /* The preferred zone is used for statistics later */
2149 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
Miao Xiec0ff7452010-05-24 14:32:08 -07002150 if (!preferred_zone) {
2151 put_mems_allowed();
Mel Gorman5117f452009-06-16 15:31:59 -07002152 return NULL;
Miao Xiec0ff7452010-05-24 14:32:08 -07002153 }
Mel Gorman5117f452009-06-16 15:31:59 -07002154
2155 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002156 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002157 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002158 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002159 if (unlikely(!page))
2160 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002161 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002162 preferred_zone, migratetype);
Miao Xiec0ff7452010-05-24 14:32:08 -07002163 put_mems_allowed();
Mel Gorman11e33f62009-06-16 15:31:57 -07002164
Mel Gorman4b4f2782009-09-21 17:02:41 -07002165 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002166 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167}
Mel Gormand2391712009-06-16 15:31:52 -07002168EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169
2170/*
2171 * Common helper functions.
2172 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002173unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174{
Akinobu Mita945a1112009-09-21 17:01:47 -07002175 struct page *page;
2176
2177 /*
2178 * __get_free_pages() returns a 32-bit address, which cannot represent
2179 * a highmem page
2180 */
2181 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2182
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 page = alloc_pages(gfp_mask, order);
2184 if (!page)
2185 return 0;
2186 return (unsigned long) page_address(page);
2187}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188EXPORT_SYMBOL(__get_free_pages);
2189
Harvey Harrison920c7a52008-02-04 22:29:26 -08002190unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191{
Akinobu Mita945a1112009-09-21 17:01:47 -07002192 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194EXPORT_SYMBOL(get_zeroed_page);
2195
2196void __pagevec_free(struct pagevec *pvec)
2197{
2198 int i = pagevec_count(pvec);
2199
Mel Gorman4b4f2782009-09-21 17:02:41 -07002200 while (--i >= 0) {
2201 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 free_hot_cold_page(pvec->pages[i], pvec->cold);
Mel Gorman4b4f2782009-09-21 17:02:41 -07002203 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204}
2205
Harvey Harrison920c7a52008-02-04 22:29:26 -08002206void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207{
Nick Pigginb5810032005-10-29 18:16:12 -07002208 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002210 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 else
2212 __free_pages_ok(page, order);
2213 }
2214}
2215
2216EXPORT_SYMBOL(__free_pages);
2217
Harvey Harrison920c7a52008-02-04 22:29:26 -08002218void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219{
2220 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002221 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 __free_pages(virt_to_page((void *)addr), order);
2223 }
2224}
2225
2226EXPORT_SYMBOL(free_pages);
2227
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002228/**
2229 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2230 * @size: the number of bytes to allocate
2231 * @gfp_mask: GFP flags for the allocation
2232 *
2233 * This function is similar to alloc_pages(), except that it allocates the
2234 * minimum number of pages to satisfy the request. alloc_pages() can only
2235 * allocate memory in power-of-two pages.
2236 *
2237 * This function is also limited by MAX_ORDER.
2238 *
2239 * Memory allocated by this function must be released by free_pages_exact().
2240 */
2241void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2242{
2243 unsigned int order = get_order(size);
2244 unsigned long addr;
2245
2246 addr = __get_free_pages(gfp_mask, order);
2247 if (addr) {
2248 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2249 unsigned long used = addr + PAGE_ALIGN(size);
2250
Kevin Cernekee5bfd7562009-07-05 12:08:19 -07002251 split_page(virt_to_page((void *)addr), order);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002252 while (used < alloc_end) {
2253 free_page(used);
2254 used += PAGE_SIZE;
2255 }
2256 }
2257
2258 return (void *)addr;
2259}
2260EXPORT_SYMBOL(alloc_pages_exact);
2261
2262/**
2263 * free_pages_exact - release memory allocated via alloc_pages_exact()
2264 * @virt: the value returned by alloc_pages_exact.
2265 * @size: size of allocation, same value as passed to alloc_pages_exact().
2266 *
2267 * Release the memory allocated by a previous call to alloc_pages_exact.
2268 */
2269void free_pages_exact(void *virt, size_t size)
2270{
2271 unsigned long addr = (unsigned long)virt;
2272 unsigned long end = addr + PAGE_ALIGN(size);
2273
2274 while (addr < end) {
2275 free_page(addr);
2276 addr += PAGE_SIZE;
2277 }
2278}
2279EXPORT_SYMBOL(free_pages_exact);
2280
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281static unsigned int nr_free_zone_pages(int offset)
2282{
Mel Gormandd1a2392008-04-28 02:12:17 -07002283 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002284 struct zone *zone;
2285
Martin J. Blighe310fd42005-07-29 22:59:18 -07002286 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 unsigned int sum = 0;
2288
Mel Gorman0e884602008-04-28 02:12:14 -07002289 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290
Mel Gorman54a6eb52008-04-28 02:12:16 -07002291 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002292 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002293 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002294 if (size > high)
2295 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 }
2297
2298 return sum;
2299}
2300
2301/*
2302 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2303 */
2304unsigned int nr_free_buffer_pages(void)
2305{
Al Viroaf4ca452005-10-21 02:55:38 -04002306 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002308EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309
2310/*
2311 * Amount of free RAM allocatable within all zones
2312 */
2313unsigned int nr_free_pagecache_pages(void)
2314{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002315 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002317
2318static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002320 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002321 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324void si_meminfo(struct sysinfo *val)
2325{
2326 val->totalram = totalram_pages;
2327 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002328 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 val->totalhigh = totalhigh_pages;
2331 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 val->mem_unit = PAGE_SIZE;
2333}
2334
2335EXPORT_SYMBOL(si_meminfo);
2336
2337#ifdef CONFIG_NUMA
2338void si_meminfo_node(struct sysinfo *val, int nid)
2339{
2340 pg_data_t *pgdat = NODE_DATA(nid);
2341
2342 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002343 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002344#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002346 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2347 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002348#else
2349 val->totalhigh = 0;
2350 val->freehigh = 0;
2351#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 val->mem_unit = PAGE_SIZE;
2353}
2354#endif
2355
2356#define K(x) ((x) << (PAGE_SHIFT-10))
2357
2358/*
2359 * Show free area list (used inside shift_scroll-lock stuff)
2360 * We also calculate the percentage fragmentation. We do this by counting the
2361 * memory on each free list with the exception of the first item on the list.
2362 */
2363void show_free_areas(void)
2364{
Jes Sorensenc7241912006-09-27 01:50:05 -07002365 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 struct zone *zone;
2367
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002368 for_each_populated_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07002369 show_node(zone);
2370 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371
Dave Jones6b482c62005-11-10 15:45:56 -05002372 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 struct per_cpu_pageset *pageset;
2374
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002375 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002377 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2378 cpu, pageset->pcp.high,
2379 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 }
2381 }
2382
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002383 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2384 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002385 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07002386 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002387 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002388 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002389 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002390 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002391 global_page_state(NR_ISOLATED_ANON),
2392 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002393 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002394 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002395 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002396 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002397 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002398 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002399 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002400 global_page_state(NR_SLAB_RECLAIMABLE),
2401 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002402 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002403 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08002404 global_page_state(NR_PAGETABLE),
2405 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002407 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 int i;
2409
2410 show_node(zone);
2411 printk("%s"
2412 " free:%lukB"
2413 " min:%lukB"
2414 " low:%lukB"
2415 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002416 " active_anon:%lukB"
2417 " inactive_anon:%lukB"
2418 " active_file:%lukB"
2419 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002420 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002421 " isolated(anon):%lukB"
2422 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 " present:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002424 " mlocked:%lukB"
2425 " dirty:%lukB"
2426 " writeback:%lukB"
2427 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002428 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002429 " slab_reclaimable:%lukB"
2430 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002431 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002432 " pagetables:%lukB"
2433 " unstable:%lukB"
2434 " bounce:%lukB"
2435 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 " pages_scanned:%lu"
2437 " all_unreclaimable? %s"
2438 "\n",
2439 zone->name,
Christoph Lameteraa454842010-09-09 16:38:17 -07002440 K(zone_nr_free_pages(zone)),
Mel Gorman41858962009-06-16 15:32:12 -07002441 K(min_wmark_pages(zone)),
2442 K(low_wmark_pages(zone)),
2443 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002444 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2445 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2446 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2447 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002448 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002449 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2450 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 K(zone->present_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002452 K(zone_page_state(zone, NR_MLOCK)),
2453 K(zone_page_state(zone, NR_FILE_DIRTY)),
2454 K(zone_page_state(zone, NR_WRITEBACK)),
2455 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002456 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002457 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2458 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002459 zone_page_state(zone, NR_KERNEL_STACK) *
2460 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002461 K(zone_page_state(zone, NR_PAGETABLE)),
2462 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2463 K(zone_page_state(zone, NR_BOUNCE)),
2464 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 zone->pages_scanned,
KOSAKI Motohiro93e4a892010-03-05 13:41:55 -08002466 (zone->all_unreclaimable ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 );
2468 printk("lowmem_reserve[]:");
2469 for (i = 0; i < MAX_NR_ZONES; i++)
2470 printk(" %lu", zone->lowmem_reserve[i]);
2471 printk("\n");
2472 }
2473
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002474 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002475 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476
2477 show_node(zone);
2478 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479
2480 spin_lock_irqsave(&zone->lock, flags);
2481 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002482 nr[order] = zone->free_area[order].nr_free;
2483 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 }
2485 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07002486 for (order = 0; order < MAX_ORDER; order++)
2487 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 printk("= %lukB\n", K(total));
2489 }
2490
Larry Woodmane6f36022008-02-04 22:29:30 -08002491 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2492
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 show_swap_cache_info();
2494}
2495
Mel Gorman19770b32008-04-28 02:12:18 -07002496static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2497{
2498 zoneref->zone = zone;
2499 zoneref->zone_idx = zone_idx(zone);
2500}
2501
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502/*
2503 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08002504 *
2505 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002507static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2508 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509{
Christoph Lameter1a932052006-01-06 00:11:16 -08002510 struct zone *zone;
2511
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002512 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002513 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08002514
2515 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002516 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08002517 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08002518 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002519 zoneref_set_zone(zone,
2520 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08002521 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08002523
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002524 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08002525 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526}
2527
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002528
2529/*
2530 * zonelist_order:
2531 * 0 = automatic detection of better ordering.
2532 * 1 = order by ([node] distance, -zonetype)
2533 * 2 = order by (-zonetype, [node] distance)
2534 *
2535 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2536 * the same zonelist. So only NUMA can configure this param.
2537 */
2538#define ZONELIST_ORDER_DEFAULT 0
2539#define ZONELIST_ORDER_NODE 1
2540#define ZONELIST_ORDER_ZONE 2
2541
2542/* zonelist order in the kernel.
2543 * set_zonelist_order() will set this to NODE or ZONE.
2544 */
2545static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2546static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2547
2548
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002550/* The value user specified ....changed by config */
2551static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2552/* string for sysctl */
2553#define NUMA_ZONELIST_ORDER_LEN 16
2554char numa_zonelist_order[16] = "default";
2555
2556/*
2557 * interface for configure zonelist ordering.
2558 * command line option "numa_zonelist_order"
2559 * = "[dD]efault - default, automatic configuration.
2560 * = "[nN]ode - order by node locality, then by zone within node
2561 * = "[zZ]one - order by zone, then by locality within zone
2562 */
2563
2564static int __parse_numa_zonelist_order(char *s)
2565{
2566 if (*s == 'd' || *s == 'D') {
2567 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2568 } else if (*s == 'n' || *s == 'N') {
2569 user_zonelist_order = ZONELIST_ORDER_NODE;
2570 } else if (*s == 'z' || *s == 'Z') {
2571 user_zonelist_order = ZONELIST_ORDER_ZONE;
2572 } else {
2573 printk(KERN_WARNING
2574 "Ignoring invalid numa_zonelist_order value: "
2575 "%s\n", s);
2576 return -EINVAL;
2577 }
2578 return 0;
2579}
2580
2581static __init int setup_numa_zonelist_order(char *s)
2582{
2583 if (s)
2584 return __parse_numa_zonelist_order(s);
2585 return 0;
2586}
2587early_param("numa_zonelist_order", setup_numa_zonelist_order);
2588
2589/*
2590 * sysctl handler for numa_zonelist_order
2591 */
2592int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002593 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002594 loff_t *ppos)
2595{
2596 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2597 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01002598 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002599
Andi Kleen443c6f12009-12-23 21:00:47 +01002600 mutex_lock(&zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002601 if (write)
Andi Kleen443c6f12009-12-23 21:00:47 +01002602 strcpy(saved_string, (char*)table->data);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07002603 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002604 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01002605 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002606 if (write) {
2607 int oldval = user_zonelist_order;
2608 if (__parse_numa_zonelist_order((char*)table->data)) {
2609 /*
2610 * bogus value. restore saved string
2611 */
2612 strncpy((char*)table->data, saved_string,
2613 NUMA_ZONELIST_ORDER_LEN);
2614 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002615 } else if (oldval != user_zonelist_order) {
2616 mutex_lock(&zonelists_mutex);
Haicheng Li1f522502010-05-24 14:32:51 -07002617 build_all_zonelists(NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002618 mutex_unlock(&zonelists_mutex);
2619 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002620 }
Andi Kleen443c6f12009-12-23 21:00:47 +01002621out:
2622 mutex_unlock(&zl_order_mutex);
2623 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002624}
2625
2626
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002627#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002628static int node_load[MAX_NUMNODES];
2629
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002631 * 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 -07002632 * @node: node whose fallback list we're appending
2633 * @used_node_mask: nodemask_t of already used nodes
2634 *
2635 * We use a number of factors to determine which is the next node that should
2636 * appear on a given node's fallback list. The node should not have appeared
2637 * already in @node's fallback list, and it should be the next closest node
2638 * according to the distance array (which contains arbitrary distance values
2639 * from each node to each node in the system), and should also prefer nodes
2640 * with no CPUs, since presumably they'll have very little allocation pressure
2641 * on them otherwise.
2642 * It returns -1 if no node is found.
2643 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002644static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645{
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01002646 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 int min_val = INT_MAX;
2648 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10302649 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01002651 /* Use the local node if we haven't already */
2652 if (!node_isset(node, *used_node_mask)) {
2653 node_set(node, *used_node_mask);
2654 return node;
2655 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002657 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658
2659 /* Don't want a node to appear more than once */
2660 if (node_isset(n, *used_node_mask))
2661 continue;
2662
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 /* Use the distance array to find the distance */
2664 val = node_distance(node, n);
2665
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01002666 /* Penalize nodes under us ("prefer the next node") */
2667 val += (n < node);
2668
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10302670 tmp = cpumask_of_node(n);
2671 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 val += PENALTY_FOR_NODE_WITH_CPUS;
2673
2674 /* Slight preference for less loaded node */
2675 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2676 val += node_load[n];
2677
2678 if (val < min_val) {
2679 min_val = val;
2680 best_node = n;
2681 }
2682 }
2683
2684 if (best_node >= 0)
2685 node_set(best_node, *used_node_mask);
2686
2687 return best_node;
2688}
2689
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002690
2691/*
2692 * Build zonelists ordered by node and zones within node.
2693 * This results in maximum locality--normal zone overflows into local
2694 * DMA zone, if any--but risks exhausting DMA zone.
2695 */
2696static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002698 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002700
Mel Gorman54a6eb52008-04-28 02:12:16 -07002701 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002702 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002703 ;
2704 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2705 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002706 zonelist->_zonerefs[j].zone = NULL;
2707 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002708}
2709
2710/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002711 * Build gfp_thisnode zonelists
2712 */
2713static void build_thisnode_zonelists(pg_data_t *pgdat)
2714{
Christoph Lameter523b9452007-10-16 01:25:37 -07002715 int j;
2716 struct zonelist *zonelist;
2717
Mel Gorman54a6eb52008-04-28 02:12:16 -07002718 zonelist = &pgdat->node_zonelists[1];
2719 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002720 zonelist->_zonerefs[j].zone = NULL;
2721 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002722}
2723
2724/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002725 * Build zonelists ordered by zone and nodes within zones.
2726 * This results in conserving DMA zone[s] until all Normal memory is
2727 * exhausted, but results in overflowing to remote node while memory
2728 * may still exist in local DMA zone.
2729 */
2730static int node_order[MAX_NUMNODES];
2731
2732static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2733{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002734 int pos, j, node;
2735 int zone_type; /* needs to be signed */
2736 struct zone *z;
2737 struct zonelist *zonelist;
2738
Mel Gorman54a6eb52008-04-28 02:12:16 -07002739 zonelist = &pgdat->node_zonelists[0];
2740 pos = 0;
2741 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2742 for (j = 0; j < nr_nodes; j++) {
2743 node = node_order[j];
2744 z = &NODE_DATA(node)->node_zones[zone_type];
2745 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002746 zoneref_set_zone(z,
2747 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002748 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002749 }
2750 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002751 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002752 zonelist->_zonerefs[pos].zone = NULL;
2753 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002754}
2755
2756static int default_zonelist_order(void)
2757{
2758 int nid, zone_type;
2759 unsigned long low_kmem_size,total_size;
2760 struct zone *z;
2761 int average_size;
2762 /*
Thomas Weber88393162010-03-16 11:47:56 +01002763 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002764 * If they are really small and used heavily, the system can fall
2765 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07002766 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002767 */
2768 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2769 low_kmem_size = 0;
2770 total_size = 0;
2771 for_each_online_node(nid) {
2772 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2773 z = &NODE_DATA(nid)->node_zones[zone_type];
2774 if (populated_zone(z)) {
2775 if (zone_type < ZONE_NORMAL)
2776 low_kmem_size += z->present_pages;
2777 total_size += z->present_pages;
David Rientjese325c902010-05-24 14:32:13 -07002778 } else if (zone_type == ZONE_NORMAL) {
2779 /*
2780 * If any node has only lowmem, then node order
2781 * is preferred to allow kernel allocations
2782 * locally; otherwise, they can easily infringe
2783 * on other nodes when there is an abundance of
2784 * lowmem available to allocate from.
2785 */
2786 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002787 }
2788 }
2789 }
2790 if (!low_kmem_size || /* there are no DMA area. */
2791 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2792 return ZONELIST_ORDER_NODE;
2793 /*
2794 * look into each node's config.
2795 * If there is a node whose DMA/DMA32 memory is very big area on
2796 * local memory, NODE_ORDER may be suitable.
2797 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002798 average_size = total_size /
2799 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002800 for_each_online_node(nid) {
2801 low_kmem_size = 0;
2802 total_size = 0;
2803 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2804 z = &NODE_DATA(nid)->node_zones[zone_type];
2805 if (populated_zone(z)) {
2806 if (zone_type < ZONE_NORMAL)
2807 low_kmem_size += z->present_pages;
2808 total_size += z->present_pages;
2809 }
2810 }
2811 if (low_kmem_size &&
2812 total_size > average_size && /* ignore small node */
2813 low_kmem_size > total_size * 70/100)
2814 return ZONELIST_ORDER_NODE;
2815 }
2816 return ZONELIST_ORDER_ZONE;
2817}
2818
2819static void set_zonelist_order(void)
2820{
2821 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2822 current_zonelist_order = default_zonelist_order();
2823 else
2824 current_zonelist_order = user_zonelist_order;
2825}
2826
2827static void build_zonelists(pg_data_t *pgdat)
2828{
2829 int j, node, load;
2830 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002832 int local_node, prev_node;
2833 struct zonelist *zonelist;
2834 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835
2836 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002837 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07002839 zonelist->_zonerefs[0].zone = NULL;
2840 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 }
2842
2843 /* NUMA-aware ordering of nodes */
2844 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07002845 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 prev_node = local_node;
2847 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002848
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002849 memset(node_order, 0, sizeof(node_order));
2850 j = 0;
2851
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002853 int distance = node_distance(local_node, node);
2854
2855 /*
2856 * If another node is sufficiently far away then it is better
2857 * to reclaim pages in a zone before going off node.
2858 */
2859 if (distance > RECLAIM_DISTANCE)
2860 zone_reclaim_mode = 1;
2861
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 /*
2863 * We don't want to pressure a particular node.
2864 * So adding penalty to the first node in same
2865 * distance group to make it round-robin.
2866 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002867 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002868 node_load[node] = load;
2869
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 prev_node = node;
2871 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002872 if (order == ZONELIST_ORDER_NODE)
2873 build_zonelists_in_node_order(pgdat, node);
2874 else
2875 node_order[j++] = node; /* remember order */
2876 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002878 if (order == ZONELIST_ORDER_ZONE) {
2879 /* calculate node order -- i.e., DMA last! */
2880 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002882
2883 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884}
2885
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002886/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002887static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002888{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002889 struct zonelist *zonelist;
2890 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07002891 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002892
Mel Gorman54a6eb52008-04-28 02:12:16 -07002893 zonelist = &pgdat->node_zonelists[0];
2894 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2895 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07002896 for (z = zonelist->_zonerefs; z->zone; z++)
2897 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002898}
2899
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07002900#ifdef CONFIG_HAVE_MEMORYLESS_NODES
2901/*
2902 * Return node id of node used for "local" allocations.
2903 * I.e., first node id of first zone in arg node's generic zonelist.
2904 * Used for initializing percpu 'numa_mem', which is used primarily
2905 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
2906 */
2907int local_memory_node(int node)
2908{
2909 struct zone *zone;
2910
2911 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
2912 gfp_zone(GFP_KERNEL),
2913 NULL,
2914 &zone);
2915 return zone->node;
2916}
2917#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002918
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919#else /* CONFIG_NUMA */
2920
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002921static void set_zonelist_order(void)
2922{
2923 current_zonelist_order = ZONELIST_ORDER_ZONE;
2924}
2925
2926static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927{
Christoph Lameter19655d32006-09-25 23:31:19 -07002928 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002929 enum zone_type j;
2930 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931
2932 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933
Mel Gorman54a6eb52008-04-28 02:12:16 -07002934 zonelist = &pgdat->node_zonelists[0];
2935 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936
Mel Gorman54a6eb52008-04-28 02:12:16 -07002937 /*
2938 * Now we build the zonelist so that it contains the zones
2939 * of all the other nodes.
2940 * We don't want to pressure a particular node, so when
2941 * building the zones for node N, we make sure that the
2942 * zones coming right after the local ones are those from
2943 * node N+1 (modulo N)
2944 */
2945 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2946 if (!node_online(node))
2947 continue;
2948 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2949 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002951 for (node = 0; node < local_node; node++) {
2952 if (!node_online(node))
2953 continue;
2954 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2955 MAX_NR_ZONES - 1);
2956 }
2957
Mel Gormandd1a2392008-04-28 02:12:17 -07002958 zonelist->_zonerefs[j].zone = NULL;
2959 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960}
2961
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002962/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002963static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002964{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002965 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002966}
2967
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968#endif /* CONFIG_NUMA */
2969
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002970/*
2971 * Boot pageset table. One per cpu which is going to be used for all
2972 * zones and all nodes. The parameters will be set in such a way
2973 * that an item put on a list will immediately be handed over to
2974 * the buddy list. This is safe since pageset manipulation is done
2975 * with interrupts disabled.
2976 *
2977 * The boot_pagesets must be kept even after bootup is complete for
2978 * unused processors and/or zones. They do play a role for bootstrapping
2979 * hotplugged processors.
2980 *
2981 * zoneinfo_show() and maybe other functions do
2982 * not check if the processor is online before following the pageset pointer.
2983 * Other parts of the kernel may not check if the zone is available.
2984 */
2985static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
2986static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07002987static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002988
Haicheng Li4eaf3f62010-05-24 14:32:52 -07002989/*
2990 * Global mutex to protect against size modification of zonelists
2991 * as well as to serialize pageset setup for the new populated zone.
2992 */
2993DEFINE_MUTEX(zonelists_mutex);
2994
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002995/* return values int ....just for stop_machine() */
Haicheng Li1f522502010-05-24 14:32:51 -07002996static __init_refok int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997{
Yasunori Goto68113782006-06-23 02:03:11 -07002998 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002999 int cpu;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003000
Bo Liu7f9cfb32009-08-18 14:11:19 -07003001#ifdef CONFIG_NUMA
3002 memset(node_load, 0, sizeof(node_load));
3003#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003004 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003005 pg_data_t *pgdat = NODE_DATA(nid);
3006
3007 build_zonelists(pgdat);
3008 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003009 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003010
Haicheng Li1f522502010-05-24 14:32:51 -07003011#ifdef CONFIG_MEMORY_HOTPLUG
3012 /* Setup real pagesets for the new zone */
3013 if (data) {
3014 struct zone *zone = data;
3015 setup_zone_pageset(zone);
3016 }
3017#endif
3018
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003019 /*
3020 * Initialize the boot_pagesets that are going to be used
3021 * for bootstrapping processors. The real pagesets for
3022 * each zone will be allocated later when the per cpu
3023 * allocator is available.
3024 *
3025 * boot_pagesets are used also for bootstrapping offline
3026 * cpus if the system is already booted because the pagesets
3027 * are needed to initialize allocators on a specific cpu too.
3028 * F.e. the percpu allocator needs the page allocator which
3029 * needs the percpu allocator in order to allocate its pagesets
3030 * (a chicken-egg dilemma).
3031 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003032 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003033 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3034
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003035#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3036 /*
3037 * We now know the "local memory node" for each node--
3038 * i.e., the node of the first zone in the generic zonelist.
3039 * Set up numa_mem percpu variable for on-line cpus. During
3040 * boot, only the boot cpu should be on-line; we'll init the
3041 * secondary cpus' numa_mem as they come on-line. During
3042 * node/memory hotplug, we'll fixup all on-line cpus.
3043 */
3044 if (cpu_online(cpu))
3045 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3046#endif
3047 }
3048
Yasunori Goto68113782006-06-23 02:03:11 -07003049 return 0;
3050}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003052/*
3053 * Called with zonelists_mutex held always
3054 * unless system_state == SYSTEM_BOOTING.
3055 */
Haicheng Li1f522502010-05-24 14:32:51 -07003056void build_all_zonelists(void *data)
Yasunori Goto68113782006-06-23 02:03:11 -07003057{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003058 set_zonelist_order();
3059
Yasunori Goto68113782006-06-23 02:03:11 -07003060 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003061 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003062 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003063 cpuset_init_current_mems_allowed();
3064 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02003065 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07003066 of zonelist */
Haicheng Li1f522502010-05-24 14:32:51 -07003067 stop_machine(__build_all_zonelists, data, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003068 /* cpuset refresh routine should be here */
3069 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003070 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003071 /*
3072 * Disable grouping by mobility if the number of pages in the
3073 * system is too low to allow the mechanism to work. It would be
3074 * more accurate, but expensive to check per-zone. This check is
3075 * made on memory-hotadd so a system can start with mobility
3076 * disabled and enable it later
3077 */
Mel Gormand9c23402007-10-16 01:26:01 -07003078 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003079 page_group_by_mobility_disabled = 1;
3080 else
3081 page_group_by_mobility_disabled = 0;
3082
3083 printk("Built %i zonelists in %s order, mobility grouping %s. "
3084 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003085 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003086 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003087 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003088 vm_total_pages);
3089#ifdef CONFIG_NUMA
3090 printk("Policy zone: %s\n", zone_names[policy_zone]);
3091#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092}
3093
3094/*
3095 * Helper functions to size the waitqueue hash table.
3096 * Essentially these want to choose hash table sizes sufficiently
3097 * large so that collisions trying to wait on pages are rare.
3098 * But in fact, the number of active page waitqueues on typical
3099 * systems is ridiculously low, less than 200. So this is even
3100 * conservative, even though it seems large.
3101 *
3102 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3103 * waitqueues, i.e. the size of the waitq table given the number of pages.
3104 */
3105#define PAGES_PER_WAITQUEUE 256
3106
Yasunori Gotocca448f2006-06-23 02:03:10 -07003107#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003108static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109{
3110 unsigned long size = 1;
3111
3112 pages /= PAGES_PER_WAITQUEUE;
3113
3114 while (size < pages)
3115 size <<= 1;
3116
3117 /*
3118 * Once we have dozens or even hundreds of threads sleeping
3119 * on IO we've got bigger problems than wait queue collision.
3120 * Limit the size of the wait table to a reasonable size.
3121 */
3122 size = min(size, 4096UL);
3123
3124 return max(size, 4UL);
3125}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003126#else
3127/*
3128 * A zone's size might be changed by hot-add, so it is not possible to determine
3129 * a suitable size for its wait_table. So we use the maximum size now.
3130 *
3131 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3132 *
3133 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3134 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3135 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3136 *
3137 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3138 * or more by the traditional way. (See above). It equals:
3139 *
3140 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3141 * ia64(16K page size) : = ( 8G + 4M)byte.
3142 * powerpc (64K page size) : = (32G +16M)byte.
3143 */
3144static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3145{
3146 return 4096UL;
3147}
3148#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149
3150/*
3151 * This is an integer logarithm so that shifts can be used later
3152 * to extract the more random high bits from the multiplicative
3153 * hash function before the remainder is taken.
3154 */
3155static inline unsigned long wait_table_bits(unsigned long size)
3156{
3157 return ffz(~size);
3158}
3159
3160#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3161
Mel Gorman56fd56b2007-10-16 01:25:58 -07003162/*
Mel Gormand9c23402007-10-16 01:26:01 -07003163 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003164 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3165 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003166 * higher will lead to a bigger reserve which will get freed as contiguous
3167 * blocks as reclaim kicks in
3168 */
3169static void setup_zone_migrate_reserve(struct zone *zone)
3170{
3171 unsigned long start_pfn, pfn, end_pfn;
3172 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003173 unsigned long block_migratetype;
3174 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003175
3176 /* Get the start pfn, end pfn and the number of blocks to reserve */
3177 start_pfn = zone->zone_start_pfn;
3178 end_pfn = start_pfn + zone->spanned_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003179 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003180 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003181
Mel Gorman78986a62009-09-21 17:03:02 -07003182 /*
3183 * Reserve blocks are generally in place to help high-order atomic
3184 * allocations that are short-lived. A min_free_kbytes value that
3185 * would result in more than 2 reserve blocks for atomic allocations
3186 * is assumed to be in place to help anti-fragmentation for the
3187 * future allocation of hugepages at runtime.
3188 */
3189 reserve = min(2, reserve);
3190
Mel Gormand9c23402007-10-16 01:26:01 -07003191 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003192 if (!pfn_valid(pfn))
3193 continue;
3194 page = pfn_to_page(pfn);
3195
Adam Litke344c7902008-09-02 14:35:38 -07003196 /* Watch out for overlapping nodes */
3197 if (page_to_nid(page) != zone_to_nid(zone))
3198 continue;
3199
Mel Gorman56fd56b2007-10-16 01:25:58 -07003200 /* Blocks with reserved pages will never free, skip them. */
3201 if (PageReserved(page))
3202 continue;
3203
3204 block_migratetype = get_pageblock_migratetype(page);
3205
3206 /* If this block is reserved, account for it */
3207 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
3208 reserve--;
3209 continue;
3210 }
3211
3212 /* Suitable for reserving if this block is movable */
3213 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
3214 set_pageblock_migratetype(page, MIGRATE_RESERVE);
3215 move_freepages_block(zone, page, MIGRATE_RESERVE);
3216 reserve--;
3217 continue;
3218 }
3219
3220 /*
3221 * If the reserve is met and this is a previous reserved block,
3222 * take it back
3223 */
3224 if (block_migratetype == MIGRATE_RESERVE) {
3225 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3226 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3227 }
3228 }
3229}
Mel Gormanac0e5b72007-10-16 01:25:58 -07003230
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231/*
3232 * Initially all pages are reserved - free ones are freed
3233 * up by free_all_bootmem() once the early boot process is
3234 * done. Non-atomic initialization, single-pass.
3235 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003236void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003237 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07003240 unsigned long end_pfn = start_pfn + size;
3241 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003242 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003243
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003244 if (highest_memmap_pfn < end_pfn - 1)
3245 highest_memmap_pfn = end_pfn - 1;
3246
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003247 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08003248 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08003249 /*
3250 * There can be holes in boot-time mem_map[]s
3251 * handed to this function. They do not
3252 * exist on hotplugged memory.
3253 */
3254 if (context == MEMMAP_EARLY) {
3255 if (!early_pfn_valid(pfn))
3256 continue;
3257 if (!early_pfn_in_nid(pfn, nid))
3258 continue;
3259 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003260 page = pfn_to_page(pfn);
3261 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07003262 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08003263 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264 reset_page_mapcount(page);
3265 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003266 /*
3267 * Mark the block movable so that blocks are reserved for
3268 * movable at startup. This will force kernel allocations
3269 * to reserve their blocks rather than leaking throughout
3270 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07003271 * kernel allocations are made. Later some blocks near
3272 * the start are marked MIGRATE_RESERVE by
3273 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003274 *
3275 * bitmap is created for zone's valid pfn range. but memmap
3276 * can be created for invalid pages (for alignment)
3277 * check here not to call set_pageblock_migratetype() against
3278 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003279 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003280 if ((z->zone_start_pfn <= pfn)
3281 && (pfn < z->zone_start_pfn + z->spanned_pages)
3282 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003283 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003284
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285 INIT_LIST_HEAD(&page->lru);
3286#ifdef WANT_PAGE_VIRTUAL
3287 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3288 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07003289 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003290#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291 }
3292}
3293
Andi Kleen1e548de2008-02-04 22:29:26 -08003294static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003296 int order, t;
3297 for_each_migratetype_order(order, t) {
3298 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299 zone->free_area[order].nr_free = 0;
3300 }
3301}
3302
3303#ifndef __HAVE_ARCH_MEMMAP_INIT
3304#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08003305 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306#endif
3307
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08003308static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003309{
David Howells3a6be872009-05-06 16:03:03 -07003310#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003311 int batch;
3312
3313 /*
3314 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003315 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003316 *
3317 * OK, so we don't know how big the cache is. So guess.
3318 */
3319 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07003320 if (batch * PAGE_SIZE > 512 * 1024)
3321 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003322 batch /= 4; /* We effectively *= 4 below */
3323 if (batch < 1)
3324 batch = 1;
3325
3326 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003327 * Clamp the batch to a 2^n - 1 value. Having a power
3328 * of 2 value was found to be more likely to have
3329 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003330 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003331 * For example if 2 tasks are alternately allocating
3332 * batches of pages, one task can end up with a lot
3333 * of pages of one half of the possible page colors
3334 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003335 */
David Howells91552032009-05-06 16:03:02 -07003336 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07003337
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003338 return batch;
David Howells3a6be872009-05-06 16:03:03 -07003339
3340#else
3341 /* The deferral and batching of frees should be suppressed under NOMMU
3342 * conditions.
3343 *
3344 * The problem is that NOMMU needs to be able to allocate large chunks
3345 * of contiguous memory as there's no hardware page translation to
3346 * assemble apparent contiguous memory from discontiguous pages.
3347 *
3348 * Queueing large contiguous runs of pages for batching, however,
3349 * causes the pages to actually be freed in smaller chunks. As there
3350 * can be a significant delay between the individual batches being
3351 * recycled, this leads to the once large chunks of space being
3352 * fragmented and becoming unavailable for high-order allocations.
3353 */
3354 return 0;
3355#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003356}
3357
Adrian Bunkb69a7282008-07-23 21:28:12 -07003358static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07003359{
3360 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003361 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003362
Magnus Damm1c6fe942005-10-26 01:58:59 -07003363 memset(p, 0, sizeof(*p));
3364
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003365 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003366 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003367 pcp->high = 6 * batch;
3368 pcp->batch = max(1UL, 1 * batch);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003369 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3370 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07003371}
3372
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003373/*
3374 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3375 * to the value high for the pageset p.
3376 */
3377
3378static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3379 unsigned long high)
3380{
3381 struct per_cpu_pages *pcp;
3382
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003383 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003384 pcp->high = high;
3385 pcp->batch = max(1UL, high/4);
3386 if ((high/4) > (PAGE_SHIFT * 8))
3387 pcp->batch = PAGE_SHIFT * 8;
3388}
3389
Wu Fengguang319774e2010-05-24 14:32:49 -07003390static __meminit void setup_zone_pageset(struct zone *zone)
3391{
3392 int cpu;
3393
3394 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3395
3396 for_each_possible_cpu(cpu) {
3397 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3398
3399 setup_pageset(pcp, zone_batchsize(zone));
3400
3401 if (percpu_pagelist_fraction)
3402 setup_pagelist_highmark(pcp,
3403 (zone->present_pages /
3404 percpu_pagelist_fraction));
3405 }
3406}
3407
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003408/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003409 * Allocate per cpu pagesets and initialize them.
3410 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07003411 */
Al Viro78d99552005-12-15 09:18:25 +00003412void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003413{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003414 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003415
Wu Fengguang319774e2010-05-24 14:32:49 -07003416 for_each_populated_zone(zone)
3417 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003418}
3419
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003420static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07003421int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07003422{
3423 int i;
3424 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003425 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07003426
3427 /*
3428 * The per-page waitqueue mechanism uses hashed waitqueues
3429 * per zone.
3430 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07003431 zone->wait_table_hash_nr_entries =
3432 wait_table_hash_nr_entries(zone_size_pages);
3433 zone->wait_table_bits =
3434 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003435 alloc_size = zone->wait_table_hash_nr_entries
3436 * sizeof(wait_queue_head_t);
3437
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07003438 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07003439 zone->wait_table = (wait_queue_head_t *)
3440 alloc_bootmem_node(pgdat, alloc_size);
3441 } else {
3442 /*
3443 * This case means that a zone whose size was 0 gets new memory
3444 * via memory hot-add.
3445 * But it may be the case that a new node was hot-added. In
3446 * this case vmalloc() will not be able to use this new node's
3447 * memory - this wait_table must be initialized to use this new
3448 * node itself as well.
3449 * To use this new node's memory, further consideration will be
3450 * necessary.
3451 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07003452 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003453 }
3454 if (!zone->wait_table)
3455 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07003456
Yasunori Goto02b694d2006-06-23 02:03:08 -07003457 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07003458 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07003459
3460 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003461}
3462
Shaohua Li112067f2009-09-21 17:01:16 -07003463static int __zone_pcp_update(void *data)
3464{
3465 struct zone *zone = data;
3466 int cpu;
3467 unsigned long batch = zone_batchsize(zone), flags;
3468
Thomas Gleixner2d30a1f2010-03-10 15:20:40 -08003469 for_each_possible_cpu(cpu) {
Shaohua Li112067f2009-09-21 17:01:16 -07003470 struct per_cpu_pageset *pset;
3471 struct per_cpu_pages *pcp;
3472
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003473 pset = per_cpu_ptr(zone->pageset, cpu);
Shaohua Li112067f2009-09-21 17:01:16 -07003474 pcp = &pset->pcp;
3475
3476 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003477 free_pcppages_bulk(zone, pcp->count, pcp);
Shaohua Li112067f2009-09-21 17:01:16 -07003478 setup_pageset(pset, batch);
3479 local_irq_restore(flags);
3480 }
3481 return 0;
3482}
3483
3484void zone_pcp_update(struct zone *zone)
3485{
3486 stop_machine(__zone_pcp_update, zone, NULL);
3487}
3488
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003489static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07003490{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003491 /*
3492 * per cpu subsystem is not up at this point. The following code
3493 * relies on the ability of the linker to provide the
3494 * offset of a (static) per cpu variable into the per cpu area.
3495 */
3496 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07003497
Anton Blanchardf5335c02006-03-25 03:06:49 -08003498 if (zone->present_pages)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003499 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3500 zone->name, zone->present_pages,
3501 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07003502}
3503
Yasunori Goto718127c2006-06-23 02:03:10 -07003504__meminit int init_currently_empty_zone(struct zone *zone,
3505 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08003506 unsigned long size,
3507 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07003508{
3509 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07003510 int ret;
3511 ret = zone_wait_table_init(zone, size);
3512 if (ret)
3513 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07003514 pgdat->nr_zones = zone_idx(zone) + 1;
3515
Dave Hansened8ece22005-10-29 18:16:50 -07003516 zone->zone_start_pfn = zone_start_pfn;
3517
Mel Gorman708614e2008-07-23 21:26:51 -07003518 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3519 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3520 pgdat->node_id,
3521 (unsigned long)zone_idx(zone),
3522 zone_start_pfn, (zone_start_pfn + size));
3523
Andi Kleen1e548de2008-02-04 22:29:26 -08003524 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07003525
3526 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07003527}
3528
Mel Gormanc7132162006-09-27 01:49:43 -07003529#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3530/*
3531 * Basic iterator support. Return the first range of PFNs for a node
3532 * Note: nid == MAX_NUMNODES returns first region regardless of node
3533 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003534static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003535{
3536 int i;
3537
3538 for (i = 0; i < nr_nodemap_entries; i++)
3539 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3540 return i;
3541
3542 return -1;
3543}
3544
3545/*
3546 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02003547 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07003548 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003549static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003550{
3551 for (index = index + 1; index < nr_nodemap_entries; index++)
3552 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3553 return index;
3554
3555 return -1;
3556}
3557
3558#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3559/*
3560 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3561 * Architectures may implement their own version but if add_active_range()
3562 * was used and there are no special requirements, this is a convenient
3563 * alternative
3564 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003565int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003566{
3567 int i;
3568
3569 for (i = 0; i < nr_nodemap_entries; i++) {
3570 unsigned long start_pfn = early_node_map[i].start_pfn;
3571 unsigned long end_pfn = early_node_map[i].end_pfn;
3572
3573 if (start_pfn <= pfn && pfn < end_pfn)
3574 return early_node_map[i].nid;
3575 }
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003576 /* This is a memory hole */
3577 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07003578}
3579#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3580
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003581int __meminit early_pfn_to_nid(unsigned long pfn)
3582{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003583 int nid;
3584
3585 nid = __early_pfn_to_nid(pfn);
3586 if (nid >= 0)
3587 return nid;
3588 /* just returns 0 */
3589 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003590}
3591
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08003592#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3593bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3594{
3595 int nid;
3596
3597 nid = __early_pfn_to_nid(pfn);
3598 if (nid >= 0 && nid != node)
3599 return false;
3600 return true;
3601}
3602#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08003603
Mel Gormanc7132162006-09-27 01:49:43 -07003604/* Basic iterator support to walk early_node_map[] */
3605#define for_each_active_range_index_in_nid(i, nid) \
3606 for (i = first_active_region_index_in_nid(nid); i != -1; \
3607 i = next_active_region_index_in_nid(i, nid))
3608
3609/**
3610 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003611 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3612 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07003613 *
3614 * If an architecture guarantees that all ranges registered with
3615 * add_active_ranges() contain no holes and may be freed, this
3616 * this function may be used instead of calling free_bootmem() manually.
3617 */
3618void __init free_bootmem_with_active_regions(int nid,
3619 unsigned long max_low_pfn)
3620{
3621 int i;
3622
3623 for_each_active_range_index_in_nid(i, nid) {
3624 unsigned long size_pages = 0;
3625 unsigned long end_pfn = early_node_map[i].end_pfn;
3626
3627 if (early_node_map[i].start_pfn >= max_low_pfn)
3628 continue;
3629
3630 if (end_pfn > max_low_pfn)
3631 end_pfn = max_low_pfn;
3632
3633 size_pages = end_pfn - early_node_map[i].start_pfn;
3634 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3635 PFN_PHYS(early_node_map[i].start_pfn),
3636 size_pages << PAGE_SHIFT);
3637 }
3638}
3639
Yinghai Luedbe7d22010-08-25 13:39:16 -07003640#ifdef CONFIG_HAVE_MEMBLOCK
3641u64 __init find_memory_core_early(int nid, u64 size, u64 align,
3642 u64 goal, u64 limit)
3643{
3644 int i;
3645
3646 /* Need to go over early_node_map to find out good range for node */
3647 for_each_active_range_index_in_nid(i, nid) {
3648 u64 addr;
3649 u64 ei_start, ei_last;
3650 u64 final_start, final_end;
3651
3652 ei_last = early_node_map[i].end_pfn;
3653 ei_last <<= PAGE_SHIFT;
3654 ei_start = early_node_map[i].start_pfn;
3655 ei_start <<= PAGE_SHIFT;
3656
3657 final_start = max(ei_start, goal);
3658 final_end = min(ei_last, limit);
3659
3660 if (final_start >= final_end)
3661 continue;
3662
3663 addr = memblock_find_in_range(final_start, final_end, size, align);
3664
3665 if (addr == MEMBLOCK_ERROR)
3666 continue;
3667
3668 return addr;
3669 }
3670
3671 return MEMBLOCK_ERROR;
3672}
3673#endif
3674
Yinghai Lu08677212010-02-10 01:20:20 -08003675int __init add_from_early_node_map(struct range *range, int az,
3676 int nr_range, int nid)
3677{
3678 int i;
3679 u64 start, end;
3680
3681 /* need to go over early_node_map to find out good range for node */
3682 for_each_active_range_index_in_nid(i, nid) {
3683 start = early_node_map[i].start_pfn;
3684 end = early_node_map[i].end_pfn;
3685 nr_range = add_range(range, az, nr_range, start, end);
3686 }
3687 return nr_range;
3688}
3689
Yinghai Lu2ee78f72010-02-17 11:29:49 -08003690#ifdef CONFIG_NO_BOOTMEM
Yinghai Lu08677212010-02-10 01:20:20 -08003691void * __init __alloc_memory_core_early(int nid, u64 size, u64 align,
3692 u64 goal, u64 limit)
3693{
Yinghai Lu08677212010-02-10 01:20:20 -08003694 void *ptr;
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003695 u64 addr;
Yinghai Lu08677212010-02-10 01:20:20 -08003696
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003697 if (limit > memblock.current_limit)
3698 limit = memblock.current_limit;
Yinghai Lub8ab9f82010-07-20 13:24:31 -07003699
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003700 addr = find_memory_core_early(nid, size, align, goal, limit);
Yinghai Lu08677212010-02-10 01:20:20 -08003701
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003702 if (addr == MEMBLOCK_ERROR)
3703 return NULL;
Yinghai Lu08677212010-02-10 01:20:20 -08003704
Yinghai Lu72d7c3b2010-08-25 13:39:17 -07003705 ptr = phys_to_virt(addr);
3706 memset(ptr, 0, size);
3707 memblock_x86_reserve_range(addr, addr + size, "BOOTMEM");
3708 /*
3709 * The min_count is set to 0 so that bootmem allocated blocks
3710 * are never reported as leaks.
3711 */
3712 kmemleak_alloc(ptr, size, 0, 0);
3713 return ptr;
Yinghai Lu08677212010-02-10 01:20:20 -08003714}
Yinghai Lu2ee78f72010-02-17 11:29:49 -08003715#endif
Yinghai Lu08677212010-02-10 01:20:20 -08003716
3717
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003718void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3719{
3720 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07003721 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003722
Yinghai Lud52d53b2008-06-16 20:10:55 -07003723 for_each_active_range_index_in_nid(i, nid) {
3724 ret = work_fn(early_node_map[i].start_pfn,
3725 early_node_map[i].end_pfn, data);
3726 if (ret)
3727 break;
3728 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07003729}
Mel Gormanc7132162006-09-27 01:49:43 -07003730/**
3731 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003732 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07003733 *
3734 * If an architecture guarantees that all ranges registered with
3735 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003736 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07003737 */
3738void __init sparse_memory_present_with_active_regions(int nid)
3739{
3740 int i;
3741
3742 for_each_active_range_index_in_nid(i, nid)
3743 memory_present(early_node_map[i].nid,
3744 early_node_map[i].start_pfn,
3745 early_node_map[i].end_pfn);
3746}
3747
3748/**
3749 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003750 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3751 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3752 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003753 *
3754 * It returns the start and end page frame of a node based on information
3755 * provided by an arch calling add_active_range(). If called for a node
3756 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003757 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003758 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003759void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003760 unsigned long *start_pfn, unsigned long *end_pfn)
3761{
3762 int i;
3763 *start_pfn = -1UL;
3764 *end_pfn = 0;
3765
3766 for_each_active_range_index_in_nid(i, nid) {
3767 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3768 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3769 }
3770
Christoph Lameter633c0662007-10-16 01:25:37 -07003771 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003772 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003773}
3774
3775/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003776 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3777 * assumption is made that zones within a node are ordered in monotonic
3778 * increasing memory addresses so that the "highest" populated zone is used
3779 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003780static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003781{
3782 int zone_index;
3783 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3784 if (zone_index == ZONE_MOVABLE)
3785 continue;
3786
3787 if (arch_zone_highest_possible_pfn[zone_index] >
3788 arch_zone_lowest_possible_pfn[zone_index])
3789 break;
3790 }
3791
3792 VM_BUG_ON(zone_index == -1);
3793 movable_zone = zone_index;
3794}
3795
3796/*
3797 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3798 * because it is sized independant of architecture. Unlike the other zones,
3799 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3800 * in each node depending on the size of each node and how evenly kernelcore
3801 * is distributed. This helper function adjusts the zone ranges
3802 * provided by the architecture for a given node by using the end of the
3803 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3804 * zones within a node are in order of monotonic increases memory addresses
3805 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003806static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003807 unsigned long zone_type,
3808 unsigned long node_start_pfn,
3809 unsigned long node_end_pfn,
3810 unsigned long *zone_start_pfn,
3811 unsigned long *zone_end_pfn)
3812{
3813 /* Only adjust if ZONE_MOVABLE is on this node */
3814 if (zone_movable_pfn[nid]) {
3815 /* Size ZONE_MOVABLE */
3816 if (zone_type == ZONE_MOVABLE) {
3817 *zone_start_pfn = zone_movable_pfn[nid];
3818 *zone_end_pfn = min(node_end_pfn,
3819 arch_zone_highest_possible_pfn[movable_zone]);
3820
3821 /* Adjust for ZONE_MOVABLE starting within this range */
3822 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3823 *zone_end_pfn > zone_movable_pfn[nid]) {
3824 *zone_end_pfn = zone_movable_pfn[nid];
3825
3826 /* Check if this whole range is within ZONE_MOVABLE */
3827 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3828 *zone_start_pfn = *zone_end_pfn;
3829 }
3830}
3831
3832/*
Mel Gormanc7132162006-09-27 01:49:43 -07003833 * Return the number of pages a zone spans in a node, including holes
3834 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3835 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003836static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003837 unsigned long zone_type,
3838 unsigned long *ignored)
3839{
3840 unsigned long node_start_pfn, node_end_pfn;
3841 unsigned long zone_start_pfn, zone_end_pfn;
3842
3843 /* Get the start and end of the node and zone */
3844 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3845 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3846 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003847 adjust_zone_range_for_zone_movable(nid, zone_type,
3848 node_start_pfn, node_end_pfn,
3849 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003850
3851 /* Check that this node has pages within the zone's required range */
3852 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3853 return 0;
3854
3855 /* Move the zone boundaries inside the node if necessary */
3856 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3857 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3858
3859 /* Return the spanned pages */
3860 return zone_end_pfn - zone_start_pfn;
3861}
3862
3863/*
3864 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003865 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003866 */
Yinghai Lu32996252009-12-15 17:59:02 -08003867unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003868 unsigned long range_start_pfn,
3869 unsigned long range_end_pfn)
3870{
3871 int i = 0;
3872 unsigned long prev_end_pfn = 0, hole_pages = 0;
3873 unsigned long start_pfn;
3874
3875 /* Find the end_pfn of the first active range of pfns in the node */
3876 i = first_active_region_index_in_nid(nid);
3877 if (i == -1)
3878 return 0;
3879
Mel Gormanb5445f92007-07-26 10:41:18 -07003880 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3881
Mel Gorman9c7cd682006-09-27 01:49:58 -07003882 /* Account for ranges before physical memory on this node */
3883 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003884 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003885
3886 /* Find all holes for the zone within the node */
3887 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3888
3889 /* No need to continue if prev_end_pfn is outside the zone */
3890 if (prev_end_pfn >= range_end_pfn)
3891 break;
3892
3893 /* Make sure the end of the zone is not within the hole */
3894 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3895 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3896
3897 /* Update the hole size cound and move on */
3898 if (start_pfn > range_start_pfn) {
3899 BUG_ON(prev_end_pfn > start_pfn);
3900 hole_pages += start_pfn - prev_end_pfn;
3901 }
3902 prev_end_pfn = early_node_map[i].end_pfn;
3903 }
3904
Mel Gorman9c7cd682006-09-27 01:49:58 -07003905 /* Account for ranges past physical memory on this node */
3906 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003907 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003908 max(range_start_pfn, prev_end_pfn);
3909
Mel Gormanc7132162006-09-27 01:49:43 -07003910 return hole_pages;
3911}
3912
3913/**
3914 * absent_pages_in_range - Return number of page frames in holes within a range
3915 * @start_pfn: The start PFN to start searching for holes
3916 * @end_pfn: The end PFN to stop searching for holes
3917 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003918 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003919 */
3920unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3921 unsigned long end_pfn)
3922{
3923 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3924}
3925
3926/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003927static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003928 unsigned long zone_type,
3929 unsigned long *ignored)
3930{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003931 unsigned long node_start_pfn, node_end_pfn;
3932 unsigned long zone_start_pfn, zone_end_pfn;
3933
3934 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3935 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3936 node_start_pfn);
3937 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3938 node_end_pfn);
3939
Mel Gorman2a1e2742007-07-17 04:03:12 -07003940 adjust_zone_range_for_zone_movable(nid, zone_type,
3941 node_start_pfn, node_end_pfn,
3942 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07003943 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003944}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003945
Mel Gormanc7132162006-09-27 01:49:43 -07003946#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07003947static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003948 unsigned long zone_type,
3949 unsigned long *zones_size)
3950{
3951 return zones_size[zone_type];
3952}
3953
Paul Mundt6ea6e682007-07-15 23:38:20 -07003954static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003955 unsigned long zone_type,
3956 unsigned long *zholes_size)
3957{
3958 if (!zholes_size)
3959 return 0;
3960
3961 return zholes_size[zone_type];
3962}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003963
Mel Gormanc7132162006-09-27 01:49:43 -07003964#endif
3965
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003966static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07003967 unsigned long *zones_size, unsigned long *zholes_size)
3968{
3969 unsigned long realtotalpages, totalpages = 0;
3970 enum zone_type i;
3971
3972 for (i = 0; i < MAX_NR_ZONES; i++)
3973 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3974 zones_size);
3975 pgdat->node_spanned_pages = totalpages;
3976
3977 realtotalpages = totalpages;
3978 for (i = 0; i < MAX_NR_ZONES; i++)
3979 realtotalpages -=
3980 zone_absent_pages_in_node(pgdat->node_id, i,
3981 zholes_size);
3982 pgdat->node_present_pages = realtotalpages;
3983 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3984 realtotalpages);
3985}
3986
Mel Gorman835c1342007-10-16 01:25:47 -07003987#ifndef CONFIG_SPARSEMEM
3988/*
3989 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07003990 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3991 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07003992 * round what is now in bits to nearest long in bits, then return it in
3993 * bytes.
3994 */
3995static unsigned long __init usemap_size(unsigned long zonesize)
3996{
3997 unsigned long usemapsize;
3998
Mel Gormand9c23402007-10-16 01:26:01 -07003999 usemapsize = roundup(zonesize, pageblock_nr_pages);
4000 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004001 usemapsize *= NR_PAGEBLOCK_BITS;
4002 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4003
4004 return usemapsize / 8;
4005}
4006
4007static void __init setup_usemap(struct pglist_data *pgdat,
4008 struct zone *zone, unsigned long zonesize)
4009{
4010 unsigned long usemapsize = usemap_size(zonesize);
4011 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004012 if (usemapsize)
Mel Gorman835c1342007-10-16 01:25:47 -07004013 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07004014}
4015#else
4016static void inline setup_usemap(struct pglist_data *pgdat,
4017 struct zone *zone, unsigned long zonesize) {}
4018#endif /* CONFIG_SPARSEMEM */
4019
Mel Gormand9c23402007-10-16 01:26:01 -07004020#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004021
4022/* Return a sensible default order for the pageblock size. */
4023static inline int pageblock_default_order(void)
4024{
4025 if (HPAGE_SHIFT > PAGE_SHIFT)
4026 return HUGETLB_PAGE_ORDER;
4027
4028 return MAX_ORDER-1;
4029}
4030
Mel Gormand9c23402007-10-16 01:26:01 -07004031/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4032static inline void __init set_pageblock_order(unsigned int order)
4033{
4034 /* Check that pageblock_nr_pages has not already been setup */
4035 if (pageblock_order)
4036 return;
4037
4038 /*
4039 * Assume the largest contiguous order of interest is a huge page.
4040 * This value may be variable depending on boot parameters on IA64
4041 */
4042 pageblock_order = order;
4043}
4044#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4045
Mel Gormanba72cb82007-11-28 16:21:13 -08004046/*
4047 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4048 * and pageblock_default_order() are unused as pageblock_order is set
4049 * at compile-time. See include/linux/pageblock-flags.h for the values of
4050 * pageblock_order based on the kernel config
4051 */
4052static inline int pageblock_default_order(unsigned int order)
4053{
4054 return MAX_ORDER-1;
4055}
Mel Gormand9c23402007-10-16 01:26:01 -07004056#define set_pageblock_order(x) do {} while (0)
4057
4058#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4059
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060/*
4061 * Set up the zone data structures:
4062 * - mark all pages reserved
4063 * - mark all memory queues empty
4064 * - clear the memory bitmaps
4065 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004066static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067 unsigned long *zones_size, unsigned long *zholes_size)
4068{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004069 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004070 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004072 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004073
Dave Hansen208d54e2005-10-29 18:16:52 -07004074 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075 pgdat->nr_zones = 0;
4076 init_waitqueue_head(&pgdat->kswapd_wait);
4077 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004078 pgdat_page_cgroup_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004079
4080 for (j = 0; j < MAX_NR_ZONES; j++) {
4081 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004082 unsigned long size, realsize, memmap_pages;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004083 enum lru_list l;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004084
Mel Gormanc7132162006-09-27 01:49:43 -07004085 size = zone_spanned_pages_in_node(nid, j, zones_size);
4086 realsize = size - zone_absent_pages_in_node(nid, j,
4087 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004088
Mel Gorman0e0b8642006-09-27 01:49:56 -07004089 /*
4090 * Adjust realsize so that it accounts for how much memory
4091 * is used by this zone for memmap. This affects the watermark
4092 * and per-cpu initialisations
4093 */
Johannes Weinerf7232152008-05-23 13:04:21 -07004094 memmap_pages =
4095 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004096 if (realsize >= memmap_pages) {
4097 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004098 if (memmap_pages)
4099 printk(KERN_DEBUG
4100 " %s zone: %lu pages used for memmap\n",
4101 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004102 } else
4103 printk(KERN_WARNING
4104 " %s zone: %lu pages exceeds realsize %lu\n",
4105 zone_names[j], memmap_pages, realsize);
4106
Christoph Lameter62672762007-02-10 01:43:07 -08004107 /* Account for reserved pages */
4108 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07004109 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004110 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004111 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004112 }
4113
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004114 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004115 nr_kernel_pages += realsize;
4116 nr_all_pages += realsize;
4117
4118 zone->spanned_pages = size;
4119 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07004120#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004121 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07004122 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004123 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07004124 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004125#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004126 zone->name = zone_names[j];
4127 spin_lock_init(&zone->lock);
4128 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004129 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004130 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004131
Dave Hansened8ece22005-10-29 18:16:50 -07004132 zone_pcp_init(zone);
Christoph Lameterb69408e2008-10-18 20:26:14 -07004133 for_each_lru(l) {
4134 INIT_LIST_HEAD(&zone->lru[l].list);
Wu Fengguangf8629632009-09-21 17:03:11 -07004135 zone->reclaim_stat.nr_saved_scan[l] = 0;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004136 }
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08004137 zone->reclaim_stat.recent_rotated[0] = 0;
4138 zone->reclaim_stat.recent_rotated[1] = 0;
4139 zone->reclaim_stat.recent_scanned[0] = 0;
4140 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07004141 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07004142 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004143 if (!size)
4144 continue;
4145
Mel Gormanba72cb82007-11-28 16:21:13 -08004146 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07004147 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08004148 ret = init_currently_empty_zone(zone, zone_start_pfn,
4149 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004150 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004151 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004152 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004153 }
4154}
4155
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004156static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004157{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158 /* Skip empty nodes */
4159 if (!pgdat->node_spanned_pages)
4160 return;
4161
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004162#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004163 /* ia64 gets its own node_mem_map, before this, without bootmem */
4164 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004165 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004166 struct page *map;
4167
Bob Piccoe984bb42006-05-20 15:00:31 -07004168 /*
4169 * The zone's endpoints aren't required to be MAX_ORDER
4170 * aligned but the node_mem_map endpoints must be in order
4171 * for the buddy allocator to function correctly.
4172 */
4173 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4174 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4175 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4176 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004177 map = alloc_remap(pgdat->node_id, size);
4178 if (!map)
4179 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07004180 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004181 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004182#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183 /*
4184 * With no DISCONTIG, the global mem_map is just set as node 0's
4185 */
Mel Gormanc7132162006-09-27 01:49:43 -07004186 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004187 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07004188#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4189 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004190 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07004191#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004193#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004194#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004195}
4196
Johannes Weiner9109fb72008-07-23 21:27:20 -07004197void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4198 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004199{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004200 pg_data_t *pgdat = NODE_DATA(nid);
4201
Linus Torvalds1da177e2005-04-16 15:20:36 -07004202 pgdat->node_id = nid;
4203 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004204 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004205
4206 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004207#ifdef CONFIG_FLAT_NODE_MEM_MAP
4208 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4209 nid, (unsigned long)pgdat,
4210 (unsigned long)pgdat->node_mem_map);
4211#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004212
4213 free_area_init_core(pgdat, zones_size, zholes_size);
4214}
4215
Mel Gormanc7132162006-09-27 01:49:43 -07004216#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004217
4218#if MAX_NUMNODES > 1
4219/*
4220 * Figure out the number of possible node ids.
4221 */
4222static void __init setup_nr_node_ids(void)
4223{
4224 unsigned int node;
4225 unsigned int highest = 0;
4226
4227 for_each_node_mask(node, node_possible_map)
4228 highest = node;
4229 nr_node_ids = highest + 1;
4230}
4231#else
4232static inline void setup_nr_node_ids(void)
4233{
4234}
4235#endif
4236
Mel Gormanc7132162006-09-27 01:49:43 -07004237/**
4238 * add_active_range - Register a range of PFNs backed by physical memory
4239 * @nid: The node ID the range resides on
4240 * @start_pfn: The start PFN of the available physical memory
4241 * @end_pfn: The end PFN of the available physical memory
4242 *
4243 * These ranges are stored in an early_node_map[] and later used by
4244 * free_area_init_nodes() to calculate zone sizes and holes. If the
4245 * range spans a memory hole, it is up to the architecture to ensure
4246 * the memory is not freed by the bootmem allocator. If possible
4247 * the range being registered will be merged with existing ranges.
4248 */
4249void __init add_active_range(unsigned int nid, unsigned long start_pfn,
4250 unsigned long end_pfn)
4251{
4252 int i;
4253
Mel Gorman6b74ab92008-07-23 21:26:49 -07004254 mminit_dprintk(MMINIT_TRACE, "memory_register",
4255 "Entering add_active_range(%d, %#lx, %#lx) "
4256 "%d entries of %d used\n",
4257 nid, start_pfn, end_pfn,
4258 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07004259
Mel Gorman2dbb51c2008-07-23 21:26:52 -07004260 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
4261
Mel Gormanc7132162006-09-27 01:49:43 -07004262 /* Merge with existing active regions if possible */
4263 for (i = 0; i < nr_nodemap_entries; i++) {
4264 if (early_node_map[i].nid != nid)
4265 continue;
4266
4267 /* Skip if an existing region covers this new one */
4268 if (start_pfn >= early_node_map[i].start_pfn &&
4269 end_pfn <= early_node_map[i].end_pfn)
4270 return;
4271
4272 /* Merge forward if suitable */
4273 if (start_pfn <= early_node_map[i].end_pfn &&
4274 end_pfn > early_node_map[i].end_pfn) {
4275 early_node_map[i].end_pfn = end_pfn;
4276 return;
4277 }
4278
4279 /* Merge backward if suitable */
Kazuhisa Ichikawad2dbe082010-01-15 17:01:20 -08004280 if (start_pfn < early_node_map[i].start_pfn &&
Mel Gormanc7132162006-09-27 01:49:43 -07004281 end_pfn >= early_node_map[i].start_pfn) {
4282 early_node_map[i].start_pfn = start_pfn;
4283 return;
4284 }
4285 }
4286
4287 /* Check that early_node_map is large enough */
4288 if (i >= MAX_ACTIVE_REGIONS) {
4289 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4290 MAX_ACTIVE_REGIONS);
4291 return;
4292 }
4293
4294 early_node_map[i].nid = nid;
4295 early_node_map[i].start_pfn = start_pfn;
4296 early_node_map[i].end_pfn = end_pfn;
4297 nr_nodemap_entries = i + 1;
4298}
4299
4300/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07004301 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07004302 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07004303 * @start_pfn: The new PFN of the range
4304 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07004305 *
4306 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004307 * The map is kept near the end physical page range that has already been
4308 * registered. This function allows an arch to shrink an existing registered
4309 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07004310 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004311void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4312 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004313{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004314 int i, j;
4315 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004316
Yinghai Lucc1050b2008-06-13 19:08:52 -07004317 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4318 nid, start_pfn, end_pfn);
4319
Mel Gormanc7132162006-09-27 01:49:43 -07004320 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004321 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07004322 if (early_node_map[i].start_pfn >= start_pfn &&
4323 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004324 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07004325 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004326 early_node_map[i].end_pfn = 0;
4327 removed = 1;
4328 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004329 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07004330 if (early_node_map[i].start_pfn < start_pfn &&
4331 early_node_map[i].end_pfn > start_pfn) {
4332 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4333 early_node_map[i].end_pfn = start_pfn;
4334 if (temp_end_pfn > end_pfn)
4335 add_active_range(nid, end_pfn, temp_end_pfn);
4336 continue;
4337 }
4338 if (early_node_map[i].start_pfn >= start_pfn &&
4339 early_node_map[i].end_pfn > end_pfn &&
4340 early_node_map[i].start_pfn < end_pfn) {
4341 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07004342 continue;
4343 }
4344 }
4345
4346 if (!removed)
4347 return;
4348
4349 /* remove the blank ones */
4350 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4351 if (early_node_map[i].nid != nid)
4352 continue;
4353 if (early_node_map[i].end_pfn)
4354 continue;
4355 /* we found it, get rid of it */
4356 for (j = i; j < nr_nodemap_entries - 1; j++)
4357 memcpy(&early_node_map[j], &early_node_map[j+1],
4358 sizeof(early_node_map[j]));
4359 j = nr_nodemap_entries - 1;
4360 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4361 nr_nodemap_entries--;
4362 }
Mel Gormanc7132162006-09-27 01:49:43 -07004363}
4364
4365/**
4366 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004367 *
Mel Gormanc7132162006-09-27 01:49:43 -07004368 * During discovery, it may be found that a table like SRAT is invalid
4369 * and an alternative discovery method must be used. This function removes
4370 * all currently registered regions.
4371 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004372void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004373{
4374 memset(early_node_map, 0, sizeof(early_node_map));
4375 nr_nodemap_entries = 0;
4376}
4377
4378/* Compare two active node_active_regions */
4379static int __init cmp_node_active_region(const void *a, const void *b)
4380{
4381 struct node_active_region *arange = (struct node_active_region *)a;
4382 struct node_active_region *brange = (struct node_active_region *)b;
4383
4384 /* Done this way to avoid overflows */
4385 if (arange->start_pfn > brange->start_pfn)
4386 return 1;
4387 if (arange->start_pfn < brange->start_pfn)
4388 return -1;
4389
4390 return 0;
4391}
4392
4393/* sort the node_map by start_pfn */
Yinghai Lu32996252009-12-15 17:59:02 -08004394void __init sort_node_map(void)
Mel Gormanc7132162006-09-27 01:49:43 -07004395{
4396 sort(early_node_map, (size_t)nr_nodemap_entries,
4397 sizeof(struct node_active_region),
4398 cmp_node_active_region, NULL);
4399}
4400
Mel Gormana6af2bc2007-02-10 01:42:57 -08004401/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004402static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004403{
4404 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08004405 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004406
Mel Gormanc7132162006-09-27 01:49:43 -07004407 /* Assuming a sorted map, the first range found has the starting pfn */
4408 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08004409 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004410
Mel Gormana6af2bc2007-02-10 01:42:57 -08004411 if (min_pfn == ULONG_MAX) {
4412 printk(KERN_WARNING
Paul Jackson2bc0d2612008-06-22 07:22:17 -07004413 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004414 return 0;
4415 }
4416
4417 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004418}
4419
4420/**
4421 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4422 *
4423 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004424 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004425 */
4426unsigned long __init find_min_pfn_with_active_regions(void)
4427{
4428 return find_min_pfn_for_node(MAX_NUMNODES);
4429}
4430
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004431/*
4432 * early_calculate_totalpages()
4433 * Sum pages in active regions for movable zone.
4434 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4435 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004436static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004437{
4438 int i;
4439 unsigned long totalpages = 0;
4440
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004441 for (i = 0; i < nr_nodemap_entries; i++) {
4442 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07004443 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004444 totalpages += pages;
4445 if (pages)
4446 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4447 }
4448 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004449}
4450
Mel Gorman2a1e2742007-07-17 04:03:12 -07004451/*
4452 * Find the PFN the Movable zone begins in each node. Kernel memory
4453 * is spread evenly between nodes as long as the nodes have enough
4454 * memory. When they don't, some nodes will have more kernelcore than
4455 * others
4456 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004457static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004458{
4459 int i, nid;
4460 unsigned long usable_startpfn;
4461 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004462 /* save the state before borrow the nodemask */
4463 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004464 unsigned long totalpages = early_calculate_totalpages();
4465 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004466
Mel Gorman7e63efe2007-07-17 04:03:15 -07004467 /*
4468 * If movablecore was specified, calculate what size of
4469 * kernelcore that corresponds so that memory usable for
4470 * any allocation type is evenly spread. If both kernelcore
4471 * and movablecore are specified, then the value of kernelcore
4472 * will be used for required_kernelcore if it's greater than
4473 * what movablecore would have allowed.
4474 */
4475 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004476 unsigned long corepages;
4477
4478 /*
4479 * Round-up so that ZONE_MOVABLE is at least as large as what
4480 * was requested by the user
4481 */
4482 required_movablecore =
4483 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4484 corepages = totalpages - required_movablecore;
4485
4486 required_kernelcore = max(required_kernelcore, corepages);
4487 }
4488
Mel Gorman2a1e2742007-07-17 04:03:12 -07004489 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4490 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004491 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004492
4493 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4494 find_usable_zone_for_movable();
4495 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4496
4497restart:
4498 /* Spread kernelcore memory as evenly as possible throughout nodes */
4499 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004500 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004501 /*
4502 * Recalculate kernelcore_node if the division per node
4503 * now exceeds what is necessary to satisfy the requested
4504 * amount of memory for the kernel
4505 */
4506 if (required_kernelcore < kernelcore_node)
4507 kernelcore_node = required_kernelcore / usable_nodes;
4508
4509 /*
4510 * As the map is walked, we track how much memory is usable
4511 * by the kernel using kernelcore_remaining. When it is
4512 * 0, the rest of the node is usable by ZONE_MOVABLE
4513 */
4514 kernelcore_remaining = kernelcore_node;
4515
4516 /* Go through each range of PFNs within this node */
4517 for_each_active_range_index_in_nid(i, nid) {
4518 unsigned long start_pfn, end_pfn;
4519 unsigned long size_pages;
4520
4521 start_pfn = max(early_node_map[i].start_pfn,
4522 zone_movable_pfn[nid]);
4523 end_pfn = early_node_map[i].end_pfn;
4524 if (start_pfn >= end_pfn)
4525 continue;
4526
4527 /* Account for what is only usable for kernelcore */
4528 if (start_pfn < usable_startpfn) {
4529 unsigned long kernel_pages;
4530 kernel_pages = min(end_pfn, usable_startpfn)
4531 - start_pfn;
4532
4533 kernelcore_remaining -= min(kernel_pages,
4534 kernelcore_remaining);
4535 required_kernelcore -= min(kernel_pages,
4536 required_kernelcore);
4537
4538 /* Continue if range is now fully accounted */
4539 if (end_pfn <= usable_startpfn) {
4540
4541 /*
4542 * Push zone_movable_pfn to the end so
4543 * that if we have to rebalance
4544 * kernelcore across nodes, we will
4545 * not double account here
4546 */
4547 zone_movable_pfn[nid] = end_pfn;
4548 continue;
4549 }
4550 start_pfn = usable_startpfn;
4551 }
4552
4553 /*
4554 * The usable PFN range for ZONE_MOVABLE is from
4555 * start_pfn->end_pfn. Calculate size_pages as the
4556 * number of pages used as kernelcore
4557 */
4558 size_pages = end_pfn - start_pfn;
4559 if (size_pages > kernelcore_remaining)
4560 size_pages = kernelcore_remaining;
4561 zone_movable_pfn[nid] = start_pfn + size_pages;
4562
4563 /*
4564 * Some kernelcore has been met, update counts and
4565 * break if the kernelcore for this node has been
4566 * satisified
4567 */
4568 required_kernelcore -= min(required_kernelcore,
4569 size_pages);
4570 kernelcore_remaining -= size_pages;
4571 if (!kernelcore_remaining)
4572 break;
4573 }
4574 }
4575
4576 /*
4577 * If there is still required_kernelcore, we do another pass with one
4578 * less node in the count. This will push zone_movable_pfn[nid] further
4579 * along on the nodes that still have memory until kernelcore is
4580 * satisified
4581 */
4582 usable_nodes--;
4583 if (usable_nodes && required_kernelcore > usable_nodes)
4584 goto restart;
4585
4586 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4587 for (nid = 0; nid < MAX_NUMNODES; nid++)
4588 zone_movable_pfn[nid] =
4589 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004590
4591out:
4592 /* restore the node_state */
4593 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004594}
4595
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004596/* Any regular memory on that node ? */
4597static void check_for_regular_memory(pg_data_t *pgdat)
4598{
4599#ifdef CONFIG_HIGHMEM
4600 enum zone_type zone_type;
4601
4602 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4603 struct zone *zone = &pgdat->node_zones[zone_type];
4604 if (zone->present_pages)
4605 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4606 }
4607#endif
4608}
4609
Mel Gormanc7132162006-09-27 01:49:43 -07004610/**
4611 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004612 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004613 *
4614 * This will call free_area_init_node() for each active node in the system.
4615 * Using the page ranges provided by add_active_range(), the size of each
4616 * zone in each node and their holes is calculated. If the maximum PFN
4617 * between two adjacent zones match, it is assumed that the zone is empty.
4618 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4619 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4620 * starts where the previous one ended. For example, ZONE_DMA32 starts
4621 * at arch_max_dma_pfn.
4622 */
4623void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4624{
4625 unsigned long nid;
Andrew Mortondb991002008-10-15 22:01:07 -07004626 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004627
Mel Gormana6af2bc2007-02-10 01:42:57 -08004628 /* Sort early_node_map as initialisation assumes it is sorted */
4629 sort_node_map();
4630
Mel Gormanc7132162006-09-27 01:49:43 -07004631 /* Record where the zone boundaries are */
4632 memset(arch_zone_lowest_possible_pfn, 0,
4633 sizeof(arch_zone_lowest_possible_pfn));
4634 memset(arch_zone_highest_possible_pfn, 0,
4635 sizeof(arch_zone_highest_possible_pfn));
4636 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4637 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4638 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004639 if (i == ZONE_MOVABLE)
4640 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004641 arch_zone_lowest_possible_pfn[i] =
4642 arch_zone_highest_possible_pfn[i-1];
4643 arch_zone_highest_possible_pfn[i] =
4644 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4645 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004646 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4647 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4648
4649 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4650 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4651 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004652
Mel Gormanc7132162006-09-27 01:49:43 -07004653 /* Print out the zone ranges */
4654 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004655 for (i = 0; i < MAX_NR_ZONES; i++) {
4656 if (i == ZONE_MOVABLE)
4657 continue;
David Rientjes72f0ba02010-03-05 13:42:14 -08004658 printk(" %-8s ", zone_names[i]);
4659 if (arch_zone_lowest_possible_pfn[i] ==
4660 arch_zone_highest_possible_pfn[i])
4661 printk("empty\n");
4662 else
4663 printk("%0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004664 arch_zone_lowest_possible_pfn[i],
4665 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004666 }
4667
4668 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4669 printk("Movable zone start PFN for each node\n");
4670 for (i = 0; i < MAX_NUMNODES; i++) {
4671 if (zone_movable_pfn[i])
4672 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4673 }
Mel Gormanc7132162006-09-27 01:49:43 -07004674
4675 /* Print out the early_node_map[] */
4676 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4677 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07004678 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004679 early_node_map[i].start_pfn,
4680 early_node_map[i].end_pfn);
4681
4682 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07004683 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08004684 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004685 for_each_online_node(nid) {
4686 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004687 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004688 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004689
4690 /* Any memory on that node */
4691 if (pgdat->node_present_pages)
4692 node_set_state(nid, N_HIGH_MEMORY);
4693 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004694 }
4695}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004696
Mel Gorman7e63efe2007-07-17 04:03:15 -07004697static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004698{
4699 unsigned long long coremem;
4700 if (!p)
4701 return -EINVAL;
4702
4703 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004704 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004705
Mel Gorman7e63efe2007-07-17 04:03:15 -07004706 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004707 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4708
4709 return 0;
4710}
Mel Gormaned7ed362007-07-17 04:03:14 -07004711
Mel Gorman7e63efe2007-07-17 04:03:15 -07004712/*
4713 * kernelcore=size sets the amount of memory for use for allocations that
4714 * cannot be reclaimed or migrated.
4715 */
4716static int __init cmdline_parse_kernelcore(char *p)
4717{
4718 return cmdline_parse_core(p, &required_kernelcore);
4719}
4720
4721/*
4722 * movablecore=size sets the amount of memory for use for allocations that
4723 * can be reclaimed or migrated.
4724 */
4725static int __init cmdline_parse_movablecore(char *p)
4726{
4727 return cmdline_parse_core(p, &required_movablecore);
4728}
4729
Mel Gormaned7ed362007-07-17 04:03:14 -07004730early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004731early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004732
Mel Gormanc7132162006-09-27 01:49:43 -07004733#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4734
Mel Gorman0e0b8642006-09-27 01:49:56 -07004735/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004736 * set_dma_reserve - set the specified number of pages reserved in the first zone
4737 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004738 *
4739 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4740 * In the DMA zone, a significant percentage may be consumed by kernel image
4741 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004742 * function may optionally be used to account for unfreeable pages in the
4743 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4744 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004745 */
4746void __init set_dma_reserve(unsigned long new_dma_reserve)
4747{
4748 dma_reserve = new_dma_reserve;
4749}
4750
Dave Hansen93b75042005-06-23 00:07:47 -07004751#ifndef CONFIG_NEED_MULTIPLE_NODES
Yinghai Lu08677212010-02-10 01:20:20 -08004752struct pglist_data __refdata contig_page_data = {
4753#ifndef CONFIG_NO_BOOTMEM
4754 .bdata = &bootmem_node_data[0]
4755#endif
4756 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07004757EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07004758#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004759
4760void __init free_area_init(unsigned long *zones_size)
4761{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004762 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004763 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4764}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004765
Linus Torvalds1da177e2005-04-16 15:20:36 -07004766static int page_alloc_cpu_notify(struct notifier_block *self,
4767 unsigned long action, void *hcpu)
4768{
4769 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004770
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004771 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004772 drain_pages(cpu);
4773
4774 /*
4775 * Spill the event counters of the dead processor
4776 * into the current processors event counters.
4777 * This artificially elevates the count of the current
4778 * processor.
4779 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004780 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004781
4782 /*
4783 * Zero the differential counters of the dead processor
4784 * so that the vm statistics are consistent.
4785 *
4786 * This is only okay since the processor is dead and cannot
4787 * race with what we are doing.
4788 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004789 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004790 }
4791 return NOTIFY_OK;
4792}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004793
4794void __init page_alloc_init(void)
4795{
4796 hotcpu_notifier(page_alloc_cpu_notify, 0);
4797}
4798
4799/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004800 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4801 * or min_free_kbytes changes.
4802 */
4803static void calculate_totalreserve_pages(void)
4804{
4805 struct pglist_data *pgdat;
4806 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004807 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004808
4809 for_each_online_pgdat(pgdat) {
4810 for (i = 0; i < MAX_NR_ZONES; i++) {
4811 struct zone *zone = pgdat->node_zones + i;
4812 unsigned long max = 0;
4813
4814 /* Find valid and maximum lowmem_reserve in the zone */
4815 for (j = i; j < MAX_NR_ZONES; j++) {
4816 if (zone->lowmem_reserve[j] > max)
4817 max = zone->lowmem_reserve[j];
4818 }
4819
Mel Gorman41858962009-06-16 15:32:12 -07004820 /* we treat the high watermark as reserved pages. */
4821 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004822
4823 if (max > zone->present_pages)
4824 max = zone->present_pages;
4825 reserve_pages += max;
4826 }
4827 }
4828 totalreserve_pages = reserve_pages;
4829}
4830
4831/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004832 * setup_per_zone_lowmem_reserve - called whenever
4833 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4834 * has a correct pages reserved value, so an adequate number of
4835 * pages are left in the zone after a successful __alloc_pages().
4836 */
4837static void setup_per_zone_lowmem_reserve(void)
4838{
4839 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004840 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004841
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004842 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004843 for (j = 0; j < MAX_NR_ZONES; j++) {
4844 struct zone *zone = pgdat->node_zones + j;
4845 unsigned long present_pages = zone->present_pages;
4846
4847 zone->lowmem_reserve[j] = 0;
4848
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004849 idx = j;
4850 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004851 struct zone *lower_zone;
4852
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004853 idx--;
4854
Linus Torvalds1da177e2005-04-16 15:20:36 -07004855 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4856 sysctl_lowmem_reserve_ratio[idx] = 1;
4857
4858 lower_zone = pgdat->node_zones + idx;
4859 lower_zone->lowmem_reserve[j] = present_pages /
4860 sysctl_lowmem_reserve_ratio[idx];
4861 present_pages += lower_zone->present_pages;
4862 }
4863 }
4864 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004865
4866 /* update totalreserve_pages */
4867 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004868}
4869
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004870/**
Minchan Kimbc75d332009-06-16 15:32:48 -07004871 * setup_per_zone_wmarks - called when min_free_kbytes changes
Minchan Kimbce73942009-06-16 15:32:50 -07004872 * or when memory is hot-{added|removed}
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004873 *
Minchan Kimbc75d332009-06-16 15:32:48 -07004874 * Ensures that the watermark[min,low,high] values for each zone are set
4875 * correctly with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004876 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004877void setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004878{
4879 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4880 unsigned long lowmem_pages = 0;
4881 struct zone *zone;
4882 unsigned long flags;
4883
4884 /* Calculate total number of !ZONE_HIGHMEM pages */
4885 for_each_zone(zone) {
4886 if (!is_highmem(zone))
4887 lowmem_pages += zone->present_pages;
4888 }
4889
4890 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004891 u64 tmp;
4892
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004893 spin_lock_irqsave(&zone->lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004894 tmp = (u64)pages_min * zone->present_pages;
4895 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004896 if (is_highmem(zone)) {
4897 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004898 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4899 * need highmem pages, so cap pages_min to a small
4900 * value here.
4901 *
Mel Gorman41858962009-06-16 15:32:12 -07004902 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08004903 * deltas controls asynch page reclaim, and so should
4904 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004905 */
4906 int min_pages;
4907
4908 min_pages = zone->present_pages / 1024;
4909 if (min_pages < SWAP_CLUSTER_MAX)
4910 min_pages = SWAP_CLUSTER_MAX;
4911 if (min_pages > 128)
4912 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07004913 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004914 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004915 /*
4916 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004917 * proportionate to the zone's size.
4918 */
Mel Gorman41858962009-06-16 15:32:12 -07004919 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004920 }
4921
Mel Gorman41858962009-06-16 15:32:12 -07004922 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
4923 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004924 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07004925 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004926 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004927
4928 /* update totalreserve_pages */
4929 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004930}
4931
Randy Dunlap55a44622009-09-21 17:01:20 -07004932/*
Rik van Riel556adec2008-10-18 20:26:34 -07004933 * The inactive anon list should be small enough that the VM never has to
4934 * do too much work, but large enough that each inactive page has a chance
4935 * to be referenced again before it is swapped out.
4936 *
4937 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4938 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4939 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4940 * the anonymous pages are kept on the inactive list.
4941 *
4942 * total target max
4943 * memory ratio inactive anon
4944 * -------------------------------------
4945 * 10MB 1 5MB
4946 * 100MB 1 50MB
4947 * 1GB 3 250MB
4948 * 10GB 10 0.9GB
4949 * 100GB 31 3GB
4950 * 1TB 101 10GB
4951 * 10TB 320 32GB
4952 */
Minchan Kim96cb4df2009-06-16 15:32:49 -07004953void calculate_zone_inactive_ratio(struct zone *zone)
4954{
4955 unsigned int gb, ratio;
4956
4957 /* Zone size in gigabytes */
4958 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4959 if (gb)
4960 ratio = int_sqrt(10 * gb);
4961 else
4962 ratio = 1;
4963
4964 zone->inactive_ratio = ratio;
4965}
4966
Cyrill Gorcunove9bb35d2009-06-16 15:32:32 -07004967static void __init setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07004968{
4969 struct zone *zone;
4970
Minchan Kim96cb4df2009-06-16 15:32:49 -07004971 for_each_zone(zone)
4972 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07004973}
4974
Linus Torvalds1da177e2005-04-16 15:20:36 -07004975/*
4976 * Initialise min_free_kbytes.
4977 *
4978 * For small machines we want it small (128k min). For large machines
4979 * we want it large (64MB max). But it is not linear, because network
4980 * bandwidth does not increase linearly with machine size. We use
4981 *
4982 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4983 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4984 *
4985 * which yields
4986 *
4987 * 16MB: 512k
4988 * 32MB: 724k
4989 * 64MB: 1024k
4990 * 128MB: 1448k
4991 * 256MB: 2048k
4992 * 512MB: 2896k
4993 * 1024MB: 4096k
4994 * 2048MB: 5792k
4995 * 4096MB: 8192k
4996 * 8192MB: 11584k
4997 * 16384MB: 16384k
4998 */
Minchan Kimbc75d332009-06-16 15:32:48 -07004999static int __init init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005000{
5001 unsigned long lowmem_kbytes;
5002
5003 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5004
5005 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5006 if (min_free_kbytes < 128)
5007 min_free_kbytes = 128;
5008 if (min_free_kbytes > 65536)
5009 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07005010 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005011 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005012 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005013 return 0;
5014}
Minchan Kimbc75d332009-06-16 15:32:48 -07005015module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005016
5017/*
5018 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5019 * that we can call two helper functions whenever min_free_kbytes
5020 * changes.
5021 */
5022int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005023 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005024{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005025 proc_dointvec(table, write, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07005026 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07005027 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005028 return 0;
5029}
5030
Christoph Lameter96146342006-07-03 00:24:13 -07005031#ifdef CONFIG_NUMA
5032int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005033 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005034{
5035 struct zone *zone;
5036 int rc;
5037
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005038 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005039 if (rc)
5040 return rc;
5041
5042 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07005043 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005044 sysctl_min_unmapped_ratio) / 100;
5045 return 0;
5046}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005047
5048int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005049 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005050{
5051 struct zone *zone;
5052 int rc;
5053
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005054 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005055 if (rc)
5056 return rc;
5057
5058 for_each_zone(zone)
5059 zone->min_slab_pages = (zone->present_pages *
5060 sysctl_min_slab_ratio) / 100;
5061 return 0;
5062}
Christoph Lameter96146342006-07-03 00:24:13 -07005063#endif
5064
Linus Torvalds1da177e2005-04-16 15:20:36 -07005065/*
5066 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5067 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5068 * whenever sysctl_lowmem_reserve_ratio changes.
5069 *
5070 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005071 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005072 * if in function of the boot time zone sizes.
5073 */
5074int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005075 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005076{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005077 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005078 setup_per_zone_lowmem_reserve();
5079 return 0;
5080}
5081
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005082/*
5083 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5084 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5085 * can have before it gets flushed back to buddy allocator.
5086 */
5087
5088int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005089 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005090{
5091 struct zone *zone;
5092 unsigned int cpu;
5093 int ret;
5094
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005095 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005096 if (!write || (ret == -EINVAL))
5097 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005098 for_each_populated_zone(zone) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005099 for_each_possible_cpu(cpu) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005100 unsigned long high;
5101 high = zone->present_pages / percpu_pagelist_fraction;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005102 setup_pagelist_highmark(
5103 per_cpu_ptr(zone->pageset, cpu), high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005104 }
5105 }
5106 return 0;
5107}
5108
David S. Millerf034b5d2006-08-24 03:08:07 -07005109int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005110
5111#ifdef CONFIG_NUMA
5112static int __init set_hashdist(char *str)
5113{
5114 if (!str)
5115 return 0;
5116 hashdist = simple_strtoul(str, &str, 0);
5117 return 1;
5118}
5119__setup("hashdist=", set_hashdist);
5120#endif
5121
5122/*
5123 * allocate a large system hash table from bootmem
5124 * - it is assumed that the hash table must contain an exact power-of-2
5125 * quantity of entries
5126 * - limit is the number of hash buckets, not the total allocation size
5127 */
5128void *__init alloc_large_system_hash(const char *tablename,
5129 unsigned long bucketsize,
5130 unsigned long numentries,
5131 int scale,
5132 int flags,
5133 unsigned int *_hash_shift,
5134 unsigned int *_hash_mask,
5135 unsigned long limit)
5136{
5137 unsigned long long max = limit;
5138 unsigned long log2qty, size;
5139 void *table = NULL;
5140
5141 /* allow the kernel cmdline to have a say */
5142 if (!numentries) {
5143 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005144 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005145 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5146 numentries >>= 20 - PAGE_SHIFT;
5147 numentries <<= 20 - PAGE_SHIFT;
5148
5149 /* limit to 1 bucket per 2^scale bytes of low memory */
5150 if (scale > PAGE_SHIFT)
5151 numentries >>= (scale - PAGE_SHIFT);
5152 else
5153 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005154
5155 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005156 if (unlikely(flags & HASH_SMALL)) {
5157 /* Makes no sense without HASH_EARLY */
5158 WARN_ON(!(flags & HASH_EARLY));
5159 if (!(numentries >> *_hash_shift)) {
5160 numentries = 1UL << *_hash_shift;
5161 BUG_ON(!numentries);
5162 }
5163 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005164 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005165 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005166 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005167
5168 /* limit allocation size to 1/16 total memory by default */
5169 if (max == 0) {
5170 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5171 do_div(max, bucketsize);
5172 }
5173
5174 if (numentries > max)
5175 numentries = max;
5176
David Howellsf0d1b0b2006-12-08 02:37:49 -08005177 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005178
5179 do {
5180 size = bucketsize << log2qty;
5181 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07005182 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005183 else if (hashdist)
5184 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5185 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005186 /*
5187 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005188 * some pages at the end of hash table which
5189 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005190 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005191 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005192 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005193 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5194 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005195 }
5196 } while (!table && size > PAGE_SIZE && --log2qty);
5197
5198 if (!table)
5199 panic("Failed to allocate %s hash table\n", tablename);
5200
Robin Holtf241e6602010-10-07 12:59:26 -07005201 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005202 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07005203 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005204 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005205 size);
5206
5207 if (_hash_shift)
5208 *_hash_shift = log2qty;
5209 if (_hash_mask)
5210 *_hash_mask = (1 << log2qty) - 1;
5211
5212 return table;
5213}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005214
Mel Gorman835c1342007-10-16 01:25:47 -07005215/* Return a pointer to the bitmap storing bits affecting a block of pages */
5216static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5217 unsigned long pfn)
5218{
5219#ifdef CONFIG_SPARSEMEM
5220 return __pfn_to_section(pfn)->pageblock_flags;
5221#else
5222 return zone->pageblock_flags;
5223#endif /* CONFIG_SPARSEMEM */
5224}
Andrew Morton6220ec72006-10-19 23:29:05 -07005225
Mel Gorman835c1342007-10-16 01:25:47 -07005226static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5227{
5228#ifdef CONFIG_SPARSEMEM
5229 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005230 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005231#else
5232 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07005233 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005234#endif /* CONFIG_SPARSEMEM */
5235}
5236
5237/**
Mel Gormand9c23402007-10-16 01:26:01 -07005238 * 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 -07005239 * @page: The page within the block of interest
5240 * @start_bitidx: The first bit of interest to retrieve
5241 * @end_bitidx: The last bit of interest
5242 * returns pageblock_bits flags
5243 */
5244unsigned long get_pageblock_flags_group(struct page *page,
5245 int start_bitidx, int end_bitidx)
5246{
5247 struct zone *zone;
5248 unsigned long *bitmap;
5249 unsigned long pfn, bitidx;
5250 unsigned long flags = 0;
5251 unsigned long value = 1;
5252
5253 zone = page_zone(page);
5254 pfn = page_to_pfn(page);
5255 bitmap = get_pageblock_bitmap(zone, pfn);
5256 bitidx = pfn_to_bitidx(zone, pfn);
5257
5258 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5259 if (test_bit(bitidx + start_bitidx, bitmap))
5260 flags |= value;
5261
5262 return flags;
5263}
5264
5265/**
Mel Gormand9c23402007-10-16 01:26:01 -07005266 * 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 -07005267 * @page: The page within the block of interest
5268 * @start_bitidx: The first bit of interest
5269 * @end_bitidx: The last bit of interest
5270 * @flags: The flags to set
5271 */
5272void set_pageblock_flags_group(struct page *page, unsigned long flags,
5273 int start_bitidx, int end_bitidx)
5274{
5275 struct zone *zone;
5276 unsigned long *bitmap;
5277 unsigned long pfn, bitidx;
5278 unsigned long value = 1;
5279
5280 zone = page_zone(page);
5281 pfn = page_to_pfn(page);
5282 bitmap = get_pageblock_bitmap(zone, pfn);
5283 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07005284 VM_BUG_ON(pfn < zone->zone_start_pfn);
5285 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07005286
5287 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5288 if (flags & value)
5289 __set_bit(bitidx + start_bitidx, bitmap);
5290 else
5291 __clear_bit(bitidx + start_bitidx, bitmap);
5292}
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005293
5294/*
5295 * This is designed as sub function...plz see page_isolation.c also.
5296 * set/clear page block's type to be ISOLATE.
5297 * page allocater never alloc memory from ISOLATE block.
5298 */
5299
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005300static int
5301__count_immobile_pages(struct zone *zone, struct page *page, int count)
5302{
5303 unsigned long pfn, iter, found;
5304 /*
5305 * For avoiding noise data, lru_add_drain_all() should be called
5306 * If ZONE_MOVABLE, the zone never contains immobile pages
5307 */
5308 if (zone_idx(zone) == ZONE_MOVABLE)
5309 return true;
5310
5311 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5312 return true;
5313
5314 pfn = page_to_pfn(page);
5315 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5316 unsigned long check = pfn + iter;
5317
5318 if (!pfn_valid_within(check)) {
5319 iter++;
5320 continue;
5321 }
5322 page = pfn_to_page(check);
5323 if (!page_count(page)) {
5324 if (PageBuddy(page))
5325 iter += (1 << page_order(page)) - 1;
5326 continue;
5327 }
5328 if (!PageLRU(page))
5329 found++;
5330 /*
5331 * If there are RECLAIMABLE pages, we need to check it.
5332 * But now, memory offline itself doesn't call shrink_slab()
5333 * and it still to be fixed.
5334 */
5335 /*
5336 * If the page is not RAM, page_count()should be 0.
5337 * we don't need more check. This is an _used_ not-movable page.
5338 *
5339 * The problematic thing here is PG_reserved pages. PG_reserved
5340 * is set to both of a memory hole page and a _used_ kernel
5341 * page at boot.
5342 */
5343 if (found > count)
5344 return false;
5345 }
5346 return true;
5347}
5348
5349bool is_pageblock_removable_nolock(struct page *page)
5350{
5351 struct zone *zone = page_zone(page);
5352 return __count_immobile_pages(zone, page, 0);
5353}
5354
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005355int set_migratetype_isolate(struct page *page)
5356{
5357 struct zone *zone;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005358 unsigned long flags, pfn;
Robert Jennings925cc712009-12-17 14:44:38 +00005359 struct memory_isolate_notify arg;
5360 int notifier_ret;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005361 int ret = -EBUSY;
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005362 int zone_idx;
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005363
5364 zone = page_zone(page);
Shaohua Li8e7e40d2009-09-21 17:01:18 -07005365 zone_idx = zone_idx(zone);
Robert Jennings925cc712009-12-17 14:44:38 +00005366
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005367 spin_lock_irqsave(&zone->lock, flags);
Robert Jennings925cc712009-12-17 14:44:38 +00005368
5369 pfn = page_to_pfn(page);
5370 arg.start_pfn = pfn;
5371 arg.nr_pages = pageblock_nr_pages;
5372 arg.pages_found = 0;
5373
5374 /*
5375 * It may be possible to isolate a pageblock even if the
5376 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5377 * notifier chain is used by balloon drivers to return the
5378 * number of pages in a range that are held by the balloon
5379 * driver to shrink memory. If all the pages are accounted for
5380 * by balloons, are free, or on the LRU, isolation can continue.
5381 * Later, for example, when memory hotplug notifier runs, these
5382 * pages reported as "can be isolated" should be isolated(freed)
5383 * by the balloon driver through the memory notifier chain.
5384 */
5385 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5386 notifier_ret = notifier_to_errno(notifier_ret);
KAMEZAWA Hiroyuki4b204772010-10-26 14:21:29 -07005387 if (notifier_ret)
Robert Jennings925cc712009-12-17 14:44:38 +00005388 goto out;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005389 /*
5390 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5391 * We just check MOVABLE pages.
5392 */
5393 if (__count_immobile_pages(zone, page, arg.pages_found))
Robert Jennings925cc712009-12-17 14:44:38 +00005394 ret = 0;
5395
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005396 /*
5397 * immobile means "not-on-lru" paes. If immobile is larger than
5398 * removable-by-driver pages reported by notifier, we'll fail.
5399 */
5400
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005401out:
Robert Jennings925cc712009-12-17 14:44:38 +00005402 if (!ret) {
5403 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5404 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5405 }
5406
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005407 spin_unlock_irqrestore(&zone->lock, flags);
5408 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005409 drain_all_pages();
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07005410 return ret;
5411}
5412
5413void unset_migratetype_isolate(struct page *page)
5414{
5415 struct zone *zone;
5416 unsigned long flags;
5417 zone = page_zone(page);
5418 spin_lock_irqsave(&zone->lock, flags);
5419 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5420 goto out;
5421 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5422 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5423out:
5424 spin_unlock_irqrestore(&zone->lock, flags);
5425}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005426
5427#ifdef CONFIG_MEMORY_HOTREMOVE
5428/*
5429 * All pages in the range must be isolated before calling this.
5430 */
5431void
5432__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5433{
5434 struct page *page;
5435 struct zone *zone;
5436 int order, i;
5437 unsigned long pfn;
5438 unsigned long flags;
5439 /* find the first valid pfn */
5440 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5441 if (pfn_valid(pfn))
5442 break;
5443 if (pfn == end_pfn)
5444 return;
5445 zone = page_zone(pfn_to_page(pfn));
5446 spin_lock_irqsave(&zone->lock, flags);
5447 pfn = start_pfn;
5448 while (pfn < end_pfn) {
5449 if (!pfn_valid(pfn)) {
5450 pfn++;
5451 continue;
5452 }
5453 page = pfn_to_page(pfn);
5454 BUG_ON(page_count(page));
5455 BUG_ON(!PageBuddy(page));
5456 order = page_order(page);
5457#ifdef CONFIG_DEBUG_VM
5458 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5459 pfn, 1 << order, end_pfn);
5460#endif
5461 list_del(&page->lru);
5462 rmv_page_order(page);
5463 zone->free_area[order].nr_free--;
5464 __mod_zone_page_state(zone, NR_FREE_PAGES,
5465 - (1UL << order));
5466 for (i = 0; i < (1 << order); i++)
5467 SetPageReserved((page+i));
5468 pfn += (1 << order);
5469 }
5470 spin_unlock_irqrestore(&zone->lock, flags);
5471}
5472#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01005473
5474#ifdef CONFIG_MEMORY_FAILURE
5475bool is_free_buddy_page(struct page *page)
5476{
5477 struct zone *zone = page_zone(page);
5478 unsigned long pfn = page_to_pfn(page);
5479 unsigned long flags;
5480 int order;
5481
5482 spin_lock_irqsave(&zone->lock, flags);
5483 for (order = 0; order < MAX_ORDER; order++) {
5484 struct page *page_head = page - (pfn & ((1 << order) - 1));
5485
5486 if (PageBuddy(page_head) && page_order(page_head) >= order)
5487 break;
5488 }
5489 spin_unlock_irqrestore(&zone->lock, flags);
5490
5491 return order < MAX_ORDER;
5492}
5493#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08005494
5495static struct trace_print_flags pageflag_names[] = {
5496 {1UL << PG_locked, "locked" },
5497 {1UL << PG_error, "error" },
5498 {1UL << PG_referenced, "referenced" },
5499 {1UL << PG_uptodate, "uptodate" },
5500 {1UL << PG_dirty, "dirty" },
5501 {1UL << PG_lru, "lru" },
5502 {1UL << PG_active, "active" },
5503 {1UL << PG_slab, "slab" },
5504 {1UL << PG_owner_priv_1, "owner_priv_1" },
5505 {1UL << PG_arch_1, "arch_1" },
5506 {1UL << PG_reserved, "reserved" },
5507 {1UL << PG_private, "private" },
5508 {1UL << PG_private_2, "private_2" },
5509 {1UL << PG_writeback, "writeback" },
5510#ifdef CONFIG_PAGEFLAGS_EXTENDED
5511 {1UL << PG_head, "head" },
5512 {1UL << PG_tail, "tail" },
5513#else
5514 {1UL << PG_compound, "compound" },
5515#endif
5516 {1UL << PG_swapcache, "swapcache" },
5517 {1UL << PG_mappedtodisk, "mappedtodisk" },
5518 {1UL << PG_reclaim, "reclaim" },
5519 {1UL << PG_buddy, "buddy" },
5520 {1UL << PG_swapbacked, "swapbacked" },
5521 {1UL << PG_unevictable, "unevictable" },
5522#ifdef CONFIG_MMU
5523 {1UL << PG_mlocked, "mlocked" },
5524#endif
5525#ifdef CONFIG_ARCH_USES_PG_UNCACHED
5526 {1UL << PG_uncached, "uncached" },
5527#endif
5528#ifdef CONFIG_MEMORY_FAILURE
5529 {1UL << PG_hwpoison, "hwpoison" },
5530#endif
5531 {-1UL, NULL },
5532};
5533
5534static void dump_page_flags(unsigned long flags)
5535{
5536 const char *delim = "";
5537 unsigned long mask;
5538 int i;
5539
5540 printk(KERN_ALERT "page flags: %#lx(", flags);
5541
5542 /* remove zone id */
5543 flags &= (1UL << NR_PAGEFLAGS) - 1;
5544
5545 for (i = 0; pageflag_names[i].name && flags; i++) {
5546
5547 mask = pageflag_names[i].mask;
5548 if ((flags & mask) != mask)
5549 continue;
5550
5551 flags &= ~mask;
5552 printk("%s%s", delim, pageflag_names[i].name);
5553 delim = "|";
5554 }
5555
5556 /* check for left over flags */
5557 if (flags)
5558 printk("%s%#lx", delim, flags);
5559
5560 printk(")\n");
5561}
5562
5563void dump_page(struct page *page)
5564{
5565 printk(KERN_ALERT
5566 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
5567 page, page_count(page), page_mapcount(page),
5568 page->mapping, page->index);
5569 dump_page_flags(page->flags);
5570}