Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | #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 Motohiro | 10ed273 | 2008-03-04 14:28:32 -0800 | [diff] [blame] | 22 | #include <linux/jiffies.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | #include <linux/bootmem.h> |
| 24 | #include <linux/compiler.h> |
Randy Dunlap | 9f15833 | 2005-09-13 01:25:16 -0700 | [diff] [blame] | 25 | #include <linux/kernel.h> |
Vegard Nossum | b1eeab6 | 2008-11-25 16:55:53 +0100 | [diff] [blame] | 26 | #include <linux/kmemcheck.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | #include <linux/module.h> |
| 28 | #include <linux/suspend.h> |
| 29 | #include <linux/pagevec.h> |
| 30 | #include <linux/blkdev.h> |
| 31 | #include <linux/slab.h> |
David Rientjes | 5a3135c2 | 2007-10-16 23:25:53 -0700 | [diff] [blame] | 32 | #include <linux/oom.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | #include <linux/notifier.h> |
| 34 | #include <linux/topology.h> |
| 35 | #include <linux/sysctl.h> |
| 36 | #include <linux/cpu.h> |
| 37 | #include <linux/cpuset.h> |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 38 | #include <linux/memory_hotplug.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | #include <linux/nodemask.h> |
| 40 | #include <linux/vmalloc.h> |
Christoph Lameter | 4be38e3 | 2006-01-06 00:11:17 -0800 | [diff] [blame] | 41 | #include <linux/mempolicy.h> |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 42 | #include <linux/stop_machine.h> |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 43 | #include <linux/sort.h> |
| 44 | #include <linux/pfn.h> |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 45 | #include <linux/backing-dev.h> |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 46 | #include <linux/fault-inject.h> |
KAMEZAWA Hiroyuki | a5d76b54 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 47 | #include <linux/page-isolation.h> |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 48 | #include <linux/page_cgroup.h> |
Thomas Gleixner | 3ac7fe5 | 2008-04-30 00:55:01 -0700 | [diff] [blame] | 49 | #include <linux/debugobjects.h> |
Catalin Marinas | dbb1f81 | 2009-06-11 13:23:19 +0100 | [diff] [blame] | 50 | #include <linux/kmemleak.h> |
Robert Jennings | 925cc71 | 2009-12-17 14:44:38 +0000 | [diff] [blame] | 51 | #include <linux/memory.h> |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 52 | #include <linux/compaction.h> |
Mel Gorman | 0d3d062 | 2009-09-21 17:02:44 -0700 | [diff] [blame] | 53 | #include <trace/events/kmem.h> |
Wu Fengguang | 718a382 | 2010-03-10 15:20:43 -0800 | [diff] [blame] | 54 | #include <linux/ftrace_event.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | |
| 56 | #include <asm/tlbflush.h> |
Andrew Morton | ac924c6 | 2006-05-15 09:43:59 -0700 | [diff] [blame] | 57 | #include <asm/div64.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | #include "internal.h" |
| 59 | |
Lee Schermerhorn | 7281201 | 2010-05-26 14:44:56 -0700 | [diff] [blame] | 60 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
| 61 | DEFINE_PER_CPU(int, numa_node); |
| 62 | EXPORT_PER_CPU_SYMBOL(numa_node); |
| 63 | #endif |
| 64 | |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 65 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
| 66 | /* |
| 67 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. |
| 68 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. |
| 69 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() |
| 70 | * defined in <linux/topology.h>. |
| 71 | */ |
| 72 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ |
| 73 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); |
| 74 | #endif |
| 75 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 76 | /* |
Christoph Lameter | 1380891 | 2007-10-16 01:25:27 -0700 | [diff] [blame] | 77 | * Array of node states. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 78 | */ |
Christoph Lameter | 1380891 | 2007-10-16 01:25:27 -0700 | [diff] [blame] | 79 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
| 80 | [N_POSSIBLE] = NODE_MASK_ALL, |
| 81 | [N_ONLINE] = { { [0] = 1UL } }, |
| 82 | #ifndef CONFIG_NUMA |
| 83 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, |
| 84 | #ifdef CONFIG_HIGHMEM |
| 85 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, |
| 86 | #endif |
| 87 | [N_CPU] = { { [0] = 1UL } }, |
| 88 | #endif /* NUMA */ |
| 89 | }; |
| 90 | EXPORT_SYMBOL(node_states); |
| 91 | |
Ravikiran G Thirumalai | 6c231b7 | 2005-09-06 15:17:45 -0700 | [diff] [blame] | 92 | unsigned long totalram_pages __read_mostly; |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 93 | unsigned long totalreserve_pages __read_mostly; |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 94 | int percpu_pagelist_fraction; |
Benjamin Herrenschmidt | dcce284 | 2009-06-18 13:24:12 +1000 | [diff] [blame] | 95 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 97 | #ifdef CONFIG_PM_SLEEP |
| 98 | /* |
| 99 | * The following functions are used by the suspend/hibernate code to temporarily |
| 100 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations |
| 101 | * while devices are suspended. To avoid races with the suspend/hibernate code, |
| 102 | * they should always be called with pm_mutex held (gfp_allowed_mask also should |
| 103 | * only be modified with pm_mutex held, unless the suspend/hibernate code is |
| 104 | * guaranteed not to run in parallel with that modification). |
| 105 | */ |
| 106 | void set_gfp_allowed_mask(gfp_t mask) |
| 107 | { |
| 108 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
| 109 | gfp_allowed_mask = mask; |
| 110 | } |
| 111 | |
| 112 | gfp_t clear_gfp_allowed_mask(gfp_t mask) |
| 113 | { |
| 114 | gfp_t ret = gfp_allowed_mask; |
| 115 | |
| 116 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
| 117 | gfp_allowed_mask &= ~mask; |
| 118 | return ret; |
| 119 | } |
| 120 | #endif /* CONFIG_PM_SLEEP */ |
| 121 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 122 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
| 123 | int pageblock_order __read_mostly; |
| 124 | #endif |
| 125 | |
Hugh Dickins | d98c7a0 | 2006-02-14 13:52:59 -0800 | [diff] [blame] | 126 | static void __free_pages_ok(struct page *page, unsigned int order); |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 127 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 128 | /* |
| 129 | * results with 256, 32 in the lowmem_reserve sysctl: |
| 130 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) |
| 131 | * 1G machine -> (16M dma, 784M normal, 224M high) |
| 132 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA |
| 133 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL |
| 134 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA |
Andi Kleen | a2f1b42 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 135 | * |
| 136 | * TBD: should special case ZONE_DMA32 machines here - in those we normally |
| 137 | * don't need any ZONE_NORMAL reservation |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 138 | */ |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 139 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 140 | #ifdef CONFIG_ZONE_DMA |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 141 | 256, |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 142 | #endif |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 143 | #ifdef CONFIG_ZONE_DMA32 |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 144 | 256, |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 145 | #endif |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 146 | #ifdef CONFIG_HIGHMEM |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 147 | 32, |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 148 | #endif |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 149 | 32, |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 150 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 151 | |
| 152 | EXPORT_SYMBOL(totalram_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 153 | |
Helge Deller | 15ad7cd | 2006-12-06 20:40:36 -0800 | [diff] [blame] | 154 | static char * const zone_names[MAX_NR_ZONES] = { |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 155 | #ifdef CONFIG_ZONE_DMA |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 156 | "DMA", |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 157 | #endif |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 158 | #ifdef CONFIG_ZONE_DMA32 |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 159 | "DMA32", |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 160 | #endif |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 161 | "Normal", |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 162 | #ifdef CONFIG_HIGHMEM |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 163 | "HighMem", |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 164 | #endif |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 165 | "Movable", |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 166 | }; |
| 167 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 168 | int min_free_kbytes = 1024; |
| 169 | |
Jan Beulich | 2c85f51 | 2009-09-21 17:03:07 -0700 | [diff] [blame] | 170 | static unsigned long __meminitdata nr_kernel_pages; |
| 171 | static unsigned long __meminitdata nr_all_pages; |
Yasunori Goto | a3142c8 | 2007-05-08 00:23:07 -0700 | [diff] [blame] | 172 | static unsigned long __meminitdata dma_reserve; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 174 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
| 175 | /* |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 176 | * MAX_ACTIVE_REGIONS determines the maximum number of distinct |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 177 | * ranges of memory (RAM) that may be registered with add_active_range(). |
| 178 | * Ranges passed to add_active_range() will be merged if possible |
| 179 | * so the number of times add_active_range() can be called is |
| 180 | * related to the number of nodes and the number of holes |
| 181 | */ |
| 182 | #ifdef CONFIG_MAX_ACTIVE_REGIONS |
| 183 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ |
| 184 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS |
| 185 | #else |
| 186 | #if MAX_NUMNODES >= 32 |
| 187 | /* If there can be many nodes, allow up to 50 holes per node */ |
| 188 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) |
| 189 | #else |
| 190 | /* By default, allow up to 256 distinct regions */ |
| 191 | #define MAX_ACTIVE_REGIONS 256 |
| 192 | #endif |
| 193 | #endif |
| 194 | |
Jan Beulich | 98011f5 | 2007-07-15 23:38:17 -0700 | [diff] [blame] | 195 | static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS]; |
| 196 | static int __meminitdata nr_nodemap_entries; |
| 197 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; |
| 198 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 199 | static unsigned long __initdata required_kernelcore; |
Adrian Bunk | 484f51f | 2007-10-16 01:26:03 -0700 | [diff] [blame] | 200 | static unsigned long __initdata required_movablecore; |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 201 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 202 | |
| 203 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ |
| 204 | int movable_zone; |
| 205 | EXPORT_SYMBOL(movable_zone); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 206 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
| 207 | |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 208 | #if MAX_NUMNODES > 1 |
| 209 | int nr_node_ids __read_mostly = MAX_NUMNODES; |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 210 | int nr_online_nodes __read_mostly = 1; |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 211 | EXPORT_SYMBOL(nr_node_ids); |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 212 | EXPORT_SYMBOL(nr_online_nodes); |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 213 | #endif |
| 214 | |
Mel Gorman | 9ef9acb | 2007-10-16 01:25:54 -0700 | [diff] [blame] | 215 | int page_group_by_mobility_disabled __read_mostly; |
| 216 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 217 | static void set_pageblock_migratetype(struct page *page, int migratetype) |
| 218 | { |
Mel Gorman | 49255c6 | 2009-06-16 15:31:58 -0700 | [diff] [blame] | 219 | |
| 220 | if (unlikely(page_group_by_mobility_disabled)) |
| 221 | migratetype = MIGRATE_UNMOVABLE; |
| 222 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 223 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
| 224 | PB_migrate, PB_migrate_end); |
| 225 | } |
| 226 | |
Rafael J. Wysocki | 7f33d49 | 2009-06-16 15:32:41 -0700 | [diff] [blame] | 227 | bool oom_killer_disabled __read_mostly; |
| 228 | |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 229 | #ifdef CONFIG_DEBUG_VM |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 230 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 231 | { |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 232 | int ret = 0; |
| 233 | unsigned seq; |
| 234 | unsigned long pfn = page_to_pfn(page); |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 235 | |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 236 | do { |
| 237 | seq = zone_span_seqbegin(zone); |
| 238 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) |
| 239 | ret = 1; |
| 240 | else if (pfn < zone->zone_start_pfn) |
| 241 | ret = 1; |
| 242 | } while (zone_span_seqretry(zone, seq)); |
| 243 | |
| 244 | return ret; |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 245 | } |
| 246 | |
| 247 | static int page_is_consistent(struct zone *zone, struct page *page) |
| 248 | { |
Andy Whitcroft | 14e0729 | 2007-05-06 14:49:14 -0700 | [diff] [blame] | 249 | if (!pfn_valid_within(page_to_pfn(page))) |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 250 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 251 | if (zone != page_zone(page)) |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 252 | return 0; |
| 253 | |
| 254 | return 1; |
| 255 | } |
| 256 | /* |
| 257 | * Temporary debugging check for pages not lying within a given zone. |
| 258 | */ |
| 259 | static int bad_range(struct zone *zone, struct page *page) |
| 260 | { |
| 261 | if (page_outside_zone_boundaries(zone, page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 262 | return 1; |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 263 | if (!page_is_consistent(zone, page)) |
| 264 | return 1; |
| 265 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 266 | return 0; |
| 267 | } |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 268 | #else |
| 269 | static inline int bad_range(struct zone *zone, struct page *page) |
| 270 | { |
| 271 | return 0; |
| 272 | } |
| 273 | #endif |
| 274 | |
Nick Piggin | 224abf9 | 2006-01-06 00:11:11 -0800 | [diff] [blame] | 275 | static void bad_page(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 276 | { |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 277 | static unsigned long resume; |
| 278 | static unsigned long nr_shown; |
| 279 | static unsigned long nr_unshown; |
| 280 | |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 281 | /* Don't complain about poisoned pages */ |
| 282 | if (PageHWPoison(page)) { |
| 283 | __ClearPageBuddy(page); |
| 284 | return; |
| 285 | } |
| 286 | |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 287 | /* |
| 288 | * Allow a burst of 60 reports, then keep quiet for that minute; |
| 289 | * or allow a steady drip of one report per second. |
| 290 | */ |
| 291 | if (nr_shown == 60) { |
| 292 | if (time_before(jiffies, resume)) { |
| 293 | nr_unshown++; |
| 294 | goto out; |
| 295 | } |
| 296 | if (nr_unshown) { |
Hugh Dickins | 1e9e636 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 297 | printk(KERN_ALERT |
| 298 | "BUG: Bad page state: %lu messages suppressed\n", |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 299 | nr_unshown); |
| 300 | nr_unshown = 0; |
| 301 | } |
| 302 | nr_shown = 0; |
| 303 | } |
| 304 | if (nr_shown++ == 0) |
| 305 | resume = jiffies + 60 * HZ; |
| 306 | |
Hugh Dickins | 1e9e636 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 307 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 308 | current->comm, page_to_pfn(page)); |
Wu Fengguang | 718a382 | 2010-03-10 15:20:43 -0800 | [diff] [blame] | 309 | dump_page(page); |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 310 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 311 | dump_stack(); |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 312 | out: |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 313 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
| 314 | __ClearPageBuddy(page); |
Randy Dunlap | 9f15833 | 2005-09-13 01:25:16 -0700 | [diff] [blame] | 315 | add_taint(TAINT_BAD_PAGE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 316 | } |
| 317 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 318 | /* |
| 319 | * Higher-order pages are called "compound pages". They are structured thusly: |
| 320 | * |
| 321 | * The first PAGE_SIZE page is called the "head page". |
| 322 | * |
| 323 | * The remaining PAGE_SIZE pages are called "tail pages". |
| 324 | * |
| 325 | * All pages have PG_compound set. All pages have their ->private pointing at |
| 326 | * the head page (even the head page has this). |
| 327 | * |
Hugh Dickins | 41d78ba | 2006-02-14 13:52:58 -0800 | [diff] [blame] | 328 | * The first tail page's ->lru.next holds the address of the compound page's |
| 329 | * put_page() function. Its ->lru.prev holds the order of allocation. |
| 330 | * This usage means that zero-order pages may not be compound. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 331 | */ |
Hugh Dickins | d98c7a0 | 2006-02-14 13:52:59 -0800 | [diff] [blame] | 332 | |
| 333 | static void free_compound_page(struct page *page) |
| 334 | { |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 335 | __free_pages_ok(page, compound_order(page)); |
Hugh Dickins | d98c7a0 | 2006-02-14 13:52:59 -0800 | [diff] [blame] | 336 | } |
| 337 | |
Andi Kleen | 01ad1c0 | 2008-07-23 21:27:46 -0700 | [diff] [blame] | 338 | void prep_compound_page(struct page *page, unsigned long order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 339 | { |
| 340 | int i; |
| 341 | int nr_pages = 1 << order; |
| 342 | |
Andy Whitcroft | 33f2ef8 | 2006-12-06 20:33:32 -0800 | [diff] [blame] | 343 | set_compound_page_dtor(page, free_compound_page); |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 344 | set_compound_order(page, order); |
Christoph Lameter | 6d77795 | 2007-05-06 14:49:40 -0700 | [diff] [blame] | 345 | __SetPageHead(page); |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 346 | for (i = 1; i < nr_pages; i++) { |
| 347 | struct page *p = page + i; |
| 348 | |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 349 | __SetPageTail(p); |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 350 | p->first_page = page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 351 | } |
| 352 | } |
| 353 | |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 354 | static int destroy_compound_page(struct page *page, unsigned long order) |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 355 | { |
| 356 | int i; |
| 357 | int nr_pages = 1 << order; |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 358 | int bad = 0; |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 359 | |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 360 | if (unlikely(compound_order(page) != order) || |
| 361 | unlikely(!PageHead(page))) { |
Nick Piggin | 224abf9 | 2006-01-06 00:11:11 -0800 | [diff] [blame] | 362 | bad_page(page); |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 363 | bad++; |
| 364 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | |
Christoph Lameter | 6d77795 | 2007-05-06 14:49:40 -0700 | [diff] [blame] | 366 | __ClearPageHead(page); |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 367 | |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 368 | for (i = 1; i < nr_pages; i++) { |
| 369 | struct page *p = page + i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 370 | |
Alexey Zaytsev | e713a21 | 2009-01-10 02:47:57 +0300 | [diff] [blame] | 371 | if (unlikely(!PageTail(p) || (p->first_page != page))) { |
Nick Piggin | 224abf9 | 2006-01-06 00:11:11 -0800 | [diff] [blame] | 372 | bad_page(page); |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 373 | bad++; |
| 374 | } |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 375 | __ClearPageTail(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 376 | } |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 377 | |
| 378 | return bad; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 379 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 380 | |
Nick Piggin | 17cf440 | 2006-03-22 00:08:41 -0800 | [diff] [blame] | 381 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
| 382 | { |
| 383 | int i; |
| 384 | |
Andrew Morton | 6626c5d | 2006-03-22 00:08:42 -0800 | [diff] [blame] | 385 | /* |
| 386 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO |
| 387 | * and __GFP_HIGHMEM from hard or soft interrupt context. |
| 388 | */ |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 389 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
Nick Piggin | 17cf440 | 2006-03-22 00:08:41 -0800 | [diff] [blame] | 390 | for (i = 0; i < (1 << order); i++) |
| 391 | clear_highpage(page + i); |
| 392 | } |
| 393 | |
Andrew Morton | 6aa3001b2 | 2006-04-18 22:20:52 -0700 | [diff] [blame] | 394 | static inline void set_page_order(struct page *page, int order) |
| 395 | { |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 396 | set_page_private(page, order); |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 397 | __SetPageBuddy(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 398 | } |
| 399 | |
| 400 | static inline void rmv_page_order(struct page *page) |
| 401 | { |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 402 | __ClearPageBuddy(page); |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 403 | set_page_private(page, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 404 | } |
| 405 | |
| 406 | /* |
| 407 | * Locate the struct page for both the matching buddy in our |
| 408 | * pair (buddy1) and the combined O(n+1) page they form (page). |
| 409 | * |
| 410 | * 1) Any buddy B1 will have an order O twin B2 which satisfies |
| 411 | * the following equation: |
| 412 | * B2 = B1 ^ (1 << O) |
| 413 | * For example, if the starting buddy (buddy2) is #8 its order |
| 414 | * 1 buddy is #10: |
| 415 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 |
| 416 | * |
| 417 | * 2) Any buddy B will have an order O+1 parent P which |
| 418 | * satisfies the following equation: |
| 419 | * P = B & ~(1 << O) |
| 420 | * |
Andreas Mohr | d6e05ed | 2006-06-26 18:35:02 +0200 | [diff] [blame] | 421 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 422 | */ |
| 423 | static inline struct page * |
| 424 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) |
| 425 | { |
| 426 | unsigned long buddy_idx = page_idx ^ (1 << order); |
| 427 | |
| 428 | return page + (buddy_idx - page_idx); |
| 429 | } |
| 430 | |
| 431 | static inline unsigned long |
| 432 | __find_combined_index(unsigned long page_idx, unsigned int order) |
| 433 | { |
| 434 | return (page_idx & ~(1 << order)); |
| 435 | } |
| 436 | |
| 437 | /* |
| 438 | * This function checks whether a page is free && is the buddy |
| 439 | * we can do coalesce a page and its buddy if |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 440 | * (a) the buddy is not in a hole && |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 441 | * (b) the buddy is in the buddy system && |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 442 | * (c) a page and its buddy have the same order && |
| 443 | * (d) a page and its buddy are in the same zone. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 444 | * |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 445 | * For recording whether a page is in the buddy system, we use PG_buddy. |
| 446 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. |
| 447 | * |
| 448 | * For recording page's order, we use page_private(page). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 449 | */ |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 450 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
| 451 | int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 452 | { |
Andy Whitcroft | 14e0729 | 2007-05-06 14:49:14 -0700 | [diff] [blame] | 453 | if (!pfn_valid_within(page_to_pfn(buddy))) |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 454 | return 0; |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 455 | |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 456 | if (page_zone_id(page) != page_zone_id(buddy)) |
| 457 | return 0; |
| 458 | |
| 459 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
Nick Piggin | a3af9c3 | 2009-06-16 15:32:10 -0700 | [diff] [blame] | 460 | VM_BUG_ON(page_count(buddy) != 0); |
Andrew Morton | 6aa3001b2 | 2006-04-18 22:20:52 -0700 | [diff] [blame] | 461 | return 1; |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 462 | } |
Andrew Morton | 6aa3001b2 | 2006-04-18 22:20:52 -0700 | [diff] [blame] | 463 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 464 | } |
| 465 | |
| 466 | /* |
| 467 | * Freeing function for a buddy system allocator. |
| 468 | * |
| 469 | * The concept of a buddy system is to maintain direct-mapped table |
| 470 | * (containing bit values) for memory blocks of various "orders". |
| 471 | * The bottom level table contains the map for the smallest allocatable |
| 472 | * units of memory (here, pages), and each level above it describes |
| 473 | * pairs of units from the levels below, hence, "buddies". |
| 474 | * At a high level, all that happens here is marking the table entry |
| 475 | * at the bottom level available, and propagating the changes upward |
| 476 | * as necessary, plus some accounting needed to play nicely with other |
| 477 | * parts of the VM system. |
| 478 | * At each level, we keep a list of pages, which are heads of continuous |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 479 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 480 | * order is recorded in page_private(page) field. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 481 | * So when we are allocating or freeing one, we can derive the state of the |
| 482 | * other. That is, if we allocate a small block, and both were |
| 483 | * free, the remainder of the region must be split into blocks. |
| 484 | * If a block is freed, and its buddy is also free, then this |
| 485 | * triggers coalescing into a block of larger size. |
| 486 | * |
| 487 | * -- wli |
| 488 | */ |
| 489 | |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 490 | static inline void __free_one_page(struct page *page, |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 491 | struct zone *zone, unsigned int order, |
| 492 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 493 | { |
| 494 | unsigned long page_idx; |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 495 | unsigned long combined_idx; |
| 496 | struct page *buddy; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 497 | |
Nick Piggin | 224abf9 | 2006-01-06 00:11:11 -0800 | [diff] [blame] | 498 | if (unlikely(PageCompound(page))) |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 499 | if (unlikely(destroy_compound_page(page, order))) |
| 500 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 501 | |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 502 | VM_BUG_ON(migratetype == -1); |
| 503 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 504 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); |
| 505 | |
Mel Gorman | f2260e6 | 2009-06-16 15:32:13 -0700 | [diff] [blame] | 506 | VM_BUG_ON(page_idx & ((1 << order) - 1)); |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 507 | VM_BUG_ON(bad_range(zone, page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 508 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 509 | while (order < MAX_ORDER-1) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 510 | buddy = __page_find_buddy(page, page_idx, order); |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 511 | if (!page_is_buddy(page, buddy, order)) |
Andy Whitcroft | 3c82d0c | 2008-07-23 21:27:11 -0700 | [diff] [blame] | 512 | break; |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 513 | |
Andy Whitcroft | 3c82d0c | 2008-07-23 21:27:11 -0700 | [diff] [blame] | 514 | /* Our buddy is free, merge with it and move up one order. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 515 | list_del(&buddy->lru); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 516 | zone->free_area[order].nr_free--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 517 | rmv_page_order(buddy); |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 518 | combined_idx = __find_combined_index(page_idx, order); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 519 | page = page + (combined_idx - page_idx); |
| 520 | page_idx = combined_idx; |
| 521 | order++; |
| 522 | } |
| 523 | set_page_order(page, order); |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 524 | |
| 525 | /* |
| 526 | * If this is not the largest possible page, check if the buddy |
| 527 | * of the next-highest order is free. If it is, it's possible |
| 528 | * that pages are being freed that will coalesce soon. In case, |
| 529 | * that is happening, add the free page to the tail of the list |
| 530 | * so it's less likely to be used soon and more likely to be merged |
| 531 | * as a higher order page |
| 532 | */ |
| 533 | if ((order < MAX_ORDER-1) && pfn_valid_within(page_to_pfn(buddy))) { |
| 534 | struct page *higher_page, *higher_buddy; |
| 535 | combined_idx = __find_combined_index(page_idx, order); |
| 536 | higher_page = page + combined_idx - page_idx; |
| 537 | higher_buddy = __page_find_buddy(higher_page, combined_idx, order + 1); |
| 538 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
| 539 | list_add_tail(&page->lru, |
| 540 | &zone->free_area[order].free_list[migratetype]); |
| 541 | goto out; |
| 542 | } |
| 543 | } |
| 544 | |
| 545 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); |
| 546 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 547 | zone->free_area[order].nr_free++; |
| 548 | } |
| 549 | |
KOSAKI Motohiro | 092cead | 2009-06-16 15:32:17 -0700 | [diff] [blame] | 550 | /* |
| 551 | * free_page_mlock() -- clean up attempts to free and mlocked() page. |
| 552 | * Page should not be on lru, so no need to fix that up. |
| 553 | * free_pages_check() will verify... |
| 554 | */ |
| 555 | static inline void free_page_mlock(struct page *page) |
| 556 | { |
KOSAKI Motohiro | 092cead | 2009-06-16 15:32:17 -0700 | [diff] [blame] | 557 | __dec_zone_page_state(page, NR_MLOCK); |
| 558 | __count_vm_event(UNEVICTABLE_MLOCKFREED); |
| 559 | } |
KOSAKI Motohiro | 092cead | 2009-06-16 15:32:17 -0700 | [diff] [blame] | 560 | |
Nick Piggin | 224abf9 | 2006-01-06 00:11:11 -0800 | [diff] [blame] | 561 | static inline int free_pages_check(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 562 | { |
Nick Piggin | 92be2e33 | 2006-01-06 00:10:57 -0800 | [diff] [blame] | 563 | if (unlikely(page_mapcount(page) | |
| 564 | (page->mapping != NULL) | |
Nick Piggin | a3af9c3 | 2009-06-16 15:32:10 -0700 | [diff] [blame] | 565 | (atomic_read(&page->_count) != 0) | |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 566 | (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) { |
Nick Piggin | 224abf9 | 2006-01-06 00:11:11 -0800 | [diff] [blame] | 567 | bad_page(page); |
Hugh Dickins | 79f4b7b | 2009-01-06 14:40:05 -0800 | [diff] [blame] | 568 | return 1; |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 569 | } |
Hugh Dickins | 79f4b7b | 2009-01-06 14:40:05 -0800 | [diff] [blame] | 570 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
| 571 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; |
| 572 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 573 | } |
| 574 | |
| 575 | /* |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 576 | * Frees a number of pages from the PCP lists |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 577 | * Assumes all pages on list are in same zone, and of same order. |
Renaud Lienhart | 207f36e | 2005-09-10 00:26:59 -0700 | [diff] [blame] | 578 | * count is the number of pages to free. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | * |
| 580 | * If the zone was previously in an "all pages pinned" state then look to |
| 581 | * see if this freeing clears that state. |
| 582 | * |
| 583 | * And clear the zone's pages_scanned counter, to hold off the "all pages are |
| 584 | * pinned" detection logic. |
| 585 | */ |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 586 | static void free_pcppages_bulk(struct zone *zone, int count, |
| 587 | struct per_cpu_pages *pcp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 588 | { |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 589 | int migratetype = 0; |
Mel Gorman | a6f9edd6 | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 590 | int batch_free = 0; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 591 | |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 592 | spin_lock(&zone->lock); |
KOSAKI Motohiro | 93e4a89a | 2010-03-05 13:41:55 -0800 | [diff] [blame] | 593 | zone->all_unreclaimable = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 594 | zone->pages_scanned = 0; |
Mel Gorman | f2260e6 | 2009-06-16 15:32:13 -0700 | [diff] [blame] | 595 | |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 596 | __mod_zone_page_state(zone, NR_FREE_PAGES, count); |
Mel Gorman | a6f9edd6 | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 597 | while (count) { |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 598 | struct page *page; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 599 | struct list_head *list; |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 600 | |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 601 | /* |
Mel Gorman | a6f9edd6 | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 602 | * Remove pages from lists in a round-robin fashion. A |
| 603 | * batch_free count is maintained that is incremented when an |
| 604 | * empty list is encountered. This is so more pages are freed |
| 605 | * off fuller lists instead of spinning excessively around empty |
| 606 | * lists |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 607 | */ |
| 608 | do { |
Mel Gorman | a6f9edd6 | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 609 | batch_free++; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 610 | if (++migratetype == MIGRATE_PCPTYPES) |
| 611 | migratetype = 0; |
| 612 | list = &pcp->lists[migratetype]; |
| 613 | } while (list_empty(list)); |
| 614 | |
Mel Gorman | a6f9edd6 | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 615 | do { |
| 616 | page = list_entry(list->prev, struct page, lru); |
| 617 | /* must delete as __free_one_page list manipulates */ |
| 618 | list_del(&page->lru); |
Hugh Dickins | a701623 | 2010-01-29 17:46:34 +0000 | [diff] [blame] | 619 | /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */ |
| 620 | __free_one_page(page, zone, 0, page_private(page)); |
| 621 | trace_mm_page_pcpu_drain(page, 0, page_private(page)); |
Mel Gorman | a6f9edd6 | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 622 | } while (--count && --batch_free && !list_empty(list)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 623 | } |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 624 | spin_unlock(&zone->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 625 | } |
| 626 | |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 627 | static void free_one_page(struct zone *zone, struct page *page, int order, |
| 628 | int migratetype) |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 629 | { |
Christoph Lameter | 006d22d | 2006-09-25 23:31:48 -0700 | [diff] [blame] | 630 | spin_lock(&zone->lock); |
KOSAKI Motohiro | 93e4a89a | 2010-03-05 13:41:55 -0800 | [diff] [blame] | 631 | zone->all_unreclaimable = 0; |
Christoph Lameter | 006d22d | 2006-09-25 23:31:48 -0700 | [diff] [blame] | 632 | zone->pages_scanned = 0; |
Mel Gorman | f2260e6 | 2009-06-16 15:32:13 -0700 | [diff] [blame] | 633 | |
| 634 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order); |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 635 | __free_one_page(page, zone, order, migratetype); |
Christoph Lameter | 006d22d | 2006-09-25 23:31:48 -0700 | [diff] [blame] | 636 | spin_unlock(&zone->lock); |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 637 | } |
| 638 | |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 639 | static bool free_pages_prepare(struct page *page, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 640 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 641 | int i; |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 642 | int bad = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 643 | |
Li Hong | f650316 | 2010-03-05 13:41:52 -0800 | [diff] [blame] | 644 | trace_mm_page_free_direct(page, order); |
Vegard Nossum | b1eeab6 | 2008-11-25 16:55:53 +0100 | [diff] [blame] | 645 | kmemcheck_free_shadow(page, order); |
| 646 | |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 647 | for (i = 0; i < (1 << order); i++) { |
| 648 | struct page *pg = page + i; |
| 649 | |
| 650 | if (PageAnon(pg)) |
| 651 | pg->mapping = NULL; |
| 652 | bad += free_pages_check(pg); |
| 653 | } |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 654 | if (bad) |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 655 | return false; |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 656 | |
Thomas Gleixner | 3ac7fe5 | 2008-04-30 00:55:01 -0700 | [diff] [blame] | 657 | if (!PageHighMem(page)) { |
Nick Piggin | 9858db5 | 2006-10-11 01:21:30 -0700 | [diff] [blame] | 658 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); |
Thomas Gleixner | 3ac7fe5 | 2008-04-30 00:55:01 -0700 | [diff] [blame] | 659 | debug_check_no_obj_freed(page_address(page), |
| 660 | PAGE_SIZE << order); |
| 661 | } |
Nick Piggin | dafb1367 | 2006-10-11 01:21:30 -0700 | [diff] [blame] | 662 | arch_free_page(page, order); |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 663 | kernel_map_pages(page, 1 << order, 0); |
Nick Piggin | dafb1367 | 2006-10-11 01:21:30 -0700 | [diff] [blame] | 664 | |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 665 | return true; |
| 666 | } |
| 667 | |
| 668 | static void __free_pages_ok(struct page *page, unsigned int order) |
| 669 | { |
| 670 | unsigned long flags; |
| 671 | int wasMlocked = __TestClearPageMlocked(page); |
| 672 | |
| 673 | if (!free_pages_prepare(page, order)) |
| 674 | return; |
| 675 | |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 676 | local_irq_save(flags); |
Johannes Weiner | c277331 | 2009-06-19 19:30:56 +0200 | [diff] [blame] | 677 | if (unlikely(wasMlocked)) |
Mel Gorman | da456f1 | 2009-06-16 15:32:08 -0700 | [diff] [blame] | 678 | free_page_mlock(page); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 679 | __count_vm_events(PGFREE, 1 << order); |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 680 | free_one_page(page_zone(page), page, order, |
| 681 | get_pageblock_migratetype(page)); |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 682 | local_irq_restore(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 683 | } |
| 684 | |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 685 | /* |
| 686 | * permit the bootmem allocator to evade page validation on high-order frees |
| 687 | */ |
Yasunori Goto | af370fb | 2008-07-23 21:28:17 -0700 | [diff] [blame] | 688 | void __meminit __free_pages_bootmem(struct page *page, unsigned int order) |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 689 | { |
| 690 | if (order == 0) { |
| 691 | __ClearPageReserved(page); |
| 692 | set_page_count(page, 0); |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 693 | set_page_refcounted(page); |
Nick Piggin | 545b1ea | 2006-03-22 00:08:07 -0800 | [diff] [blame] | 694 | __free_page(page); |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 695 | } else { |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 696 | int loop; |
| 697 | |
Nick Piggin | 545b1ea | 2006-03-22 00:08:07 -0800 | [diff] [blame] | 698 | prefetchw(page); |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 699 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
| 700 | struct page *p = &page[loop]; |
| 701 | |
Nick Piggin | 545b1ea | 2006-03-22 00:08:07 -0800 | [diff] [blame] | 702 | if (loop + 1 < BITS_PER_LONG) |
| 703 | prefetchw(p + 1); |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 704 | __ClearPageReserved(p); |
| 705 | set_page_count(p, 0); |
| 706 | } |
| 707 | |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 708 | set_page_refcounted(page); |
Nick Piggin | 545b1ea | 2006-03-22 00:08:07 -0800 | [diff] [blame] | 709 | __free_pages(page, order); |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 710 | } |
| 711 | } |
| 712 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 713 | |
| 714 | /* |
| 715 | * The order of subdivision here is critical for the IO subsystem. |
| 716 | * Please do not alter this order without good reasons and regression |
| 717 | * testing. Specifically, as large blocks of memory are subdivided, |
| 718 | * the order in which smaller blocks are delivered depends on the order |
| 719 | * they're subdivided in this function. This is the primary factor |
| 720 | * influencing the order in which pages are delivered to the IO |
| 721 | * subsystem according to empirical testing, and this is also justified |
| 722 | * by considering the behavior of a buddy system containing a single |
| 723 | * large block of memory acted on by a series of small allocations. |
| 724 | * This behavior is a critical factor in sglist merging's success. |
| 725 | * |
| 726 | * -- wli |
| 727 | */ |
Nick Piggin | 085cc7d5 | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 728 | static inline void expand(struct zone *zone, struct page *page, |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 729 | int low, int high, struct free_area *area, |
| 730 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 731 | { |
| 732 | unsigned long size = 1 << high; |
| 733 | |
| 734 | while (high > low) { |
| 735 | area--; |
| 736 | high--; |
| 737 | size >>= 1; |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 738 | VM_BUG_ON(bad_range(zone, &page[size])); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 739 | list_add(&page[size].lru, &area->free_list[migratetype]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 740 | area->nr_free++; |
| 741 | set_page_order(&page[size], high); |
| 742 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 743 | } |
| 744 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 745 | /* |
| 746 | * This page is about to be returned from the page allocator |
| 747 | */ |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 748 | static inline int check_new_page(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 749 | { |
Nick Piggin | 92be2e33 | 2006-01-06 00:10:57 -0800 | [diff] [blame] | 750 | if (unlikely(page_mapcount(page) | |
| 751 | (page->mapping != NULL) | |
Nick Piggin | a3af9c3 | 2009-06-16 15:32:10 -0700 | [diff] [blame] | 752 | (atomic_read(&page->_count) != 0) | |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 753 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) { |
Nick Piggin | 224abf9 | 2006-01-06 00:11:11 -0800 | [diff] [blame] | 754 | bad_page(page); |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 755 | return 1; |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 756 | } |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 757 | return 0; |
| 758 | } |
| 759 | |
| 760 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
| 761 | { |
| 762 | int i; |
| 763 | |
| 764 | for (i = 0; i < (1 << order); i++) { |
| 765 | struct page *p = page + i; |
| 766 | if (unlikely(check_new_page(p))) |
| 767 | return 1; |
| 768 | } |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 769 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 770 | set_page_private(page, 0); |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 771 | set_page_refcounted(page); |
Nick Piggin | cc102509 | 2006-12-06 20:32:00 -0800 | [diff] [blame] | 772 | |
| 773 | arch_alloc_page(page, order); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 774 | kernel_map_pages(page, 1 << order, 1); |
Nick Piggin | 17cf440 | 2006-03-22 00:08:41 -0800 | [diff] [blame] | 775 | |
| 776 | if (gfp_flags & __GFP_ZERO) |
| 777 | prep_zero_page(page, order, gfp_flags); |
| 778 | |
| 779 | if (order && (gfp_flags & __GFP_COMP)) |
| 780 | prep_compound_page(page, order); |
| 781 | |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 782 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 783 | } |
| 784 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 785 | /* |
| 786 | * Go through the free lists for the given migratetype and remove |
| 787 | * the smallest available page from the freelists |
| 788 | */ |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 789 | static inline |
| 790 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 791 | int migratetype) |
| 792 | { |
| 793 | unsigned int current_order; |
| 794 | struct free_area * area; |
| 795 | struct page *page; |
| 796 | |
| 797 | /* Find a page of the appropriate size in the preferred list */ |
| 798 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { |
| 799 | area = &(zone->free_area[current_order]); |
| 800 | if (list_empty(&area->free_list[migratetype])) |
| 801 | continue; |
| 802 | |
| 803 | page = list_entry(area->free_list[migratetype].next, |
| 804 | struct page, lru); |
| 805 | list_del(&page->lru); |
| 806 | rmv_page_order(page); |
| 807 | area->nr_free--; |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 808 | expand(zone, page, order, current_order, area, migratetype); |
| 809 | return page; |
| 810 | } |
| 811 | |
| 812 | return NULL; |
| 813 | } |
| 814 | |
| 815 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 816 | /* |
| 817 | * This array describes the order lists are fallen back to when |
| 818 | * the free lists for the desirable migrate type are depleted |
| 819 | */ |
| 820 | static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = { |
Mel Gorman | 64c5e13 | 2007-10-16 01:25:59 -0700 | [diff] [blame] | 821 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, |
| 822 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, |
| 823 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, |
| 824 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */ |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 825 | }; |
| 826 | |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 827 | /* |
| 828 | * Move the free pages in a range to the free lists of the requested type. |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 829 | * Note that start_page and end_pages are not aligned on a pageblock |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 830 | * boundary. If alignment is required, use move_freepages_block() |
| 831 | */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 832 | static int move_freepages(struct zone *zone, |
| 833 | struct page *start_page, struct page *end_page, |
| 834 | int migratetype) |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 835 | { |
| 836 | struct page *page; |
| 837 | unsigned long order; |
Mel Gorman | d100313 | 2007-10-16 01:26:00 -0700 | [diff] [blame] | 838 | int pages_moved = 0; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 839 | |
| 840 | #ifndef CONFIG_HOLES_IN_ZONE |
| 841 | /* |
| 842 | * page_zone is not safe to call in this context when |
| 843 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant |
| 844 | * anyway as we check zone boundaries in move_freepages_block(). |
| 845 | * Remove at a later date when no bug reports exist related to |
Mel Gorman | ac0e5b7 | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 846 | * grouping pages by mobility |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 847 | */ |
| 848 | BUG_ON(page_zone(start_page) != page_zone(end_page)); |
| 849 | #endif |
| 850 | |
| 851 | for (page = start_page; page <= end_page;) { |
Adam Litke | 344c790 | 2008-09-02 14:35:38 -0700 | [diff] [blame] | 852 | /* Make sure we are not inadvertently changing nodes */ |
| 853 | VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone)); |
| 854 | |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 855 | if (!pfn_valid_within(page_to_pfn(page))) { |
| 856 | page++; |
| 857 | continue; |
| 858 | } |
| 859 | |
| 860 | if (!PageBuddy(page)) { |
| 861 | page++; |
| 862 | continue; |
| 863 | } |
| 864 | |
| 865 | order = page_order(page); |
| 866 | list_del(&page->lru); |
| 867 | list_add(&page->lru, |
| 868 | &zone->free_area[order].free_list[migratetype]); |
| 869 | page += 1 << order; |
Mel Gorman | d100313 | 2007-10-16 01:26:00 -0700 | [diff] [blame] | 870 | pages_moved += 1 << order; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 871 | } |
| 872 | |
Mel Gorman | d100313 | 2007-10-16 01:26:00 -0700 | [diff] [blame] | 873 | return pages_moved; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 874 | } |
| 875 | |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 876 | static int move_freepages_block(struct zone *zone, struct page *page, |
| 877 | int migratetype) |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 878 | { |
| 879 | unsigned long start_pfn, end_pfn; |
| 880 | struct page *start_page, *end_page; |
| 881 | |
| 882 | start_pfn = page_to_pfn(page); |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 883 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 884 | start_page = pfn_to_page(start_pfn); |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 885 | end_page = start_page + pageblock_nr_pages - 1; |
| 886 | end_pfn = start_pfn + pageblock_nr_pages - 1; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 887 | |
| 888 | /* Do not cross zone boundaries */ |
| 889 | if (start_pfn < zone->zone_start_pfn) |
| 890 | start_page = page; |
| 891 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) |
| 892 | return 0; |
| 893 | |
| 894 | return move_freepages(zone, start_page, end_page, migratetype); |
| 895 | } |
| 896 | |
Mel Gorman | 2f66a68 | 2009-09-21 17:02:31 -0700 | [diff] [blame] | 897 | static void change_pageblock_range(struct page *pageblock_page, |
| 898 | int start_order, int migratetype) |
| 899 | { |
| 900 | int nr_pageblocks = 1 << (start_order - pageblock_order); |
| 901 | |
| 902 | while (nr_pageblocks--) { |
| 903 | set_pageblock_migratetype(pageblock_page, migratetype); |
| 904 | pageblock_page += pageblock_nr_pages; |
| 905 | } |
| 906 | } |
| 907 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 908 | /* Remove an element from the buddy allocator from the fallback list */ |
Mel Gorman | 0ac3a40 | 2009-06-16 15:32:06 -0700 | [diff] [blame] | 909 | static inline struct page * |
| 910 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 911 | { |
| 912 | struct free_area * area; |
| 913 | int current_order; |
| 914 | struct page *page; |
| 915 | int migratetype, i; |
| 916 | |
| 917 | /* Find the largest possible block of pages in the other list */ |
| 918 | for (current_order = MAX_ORDER-1; current_order >= order; |
| 919 | --current_order) { |
| 920 | for (i = 0; i < MIGRATE_TYPES - 1; i++) { |
| 921 | migratetype = fallbacks[start_migratetype][i]; |
| 922 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 923 | /* MIGRATE_RESERVE handled later if necessary */ |
| 924 | if (migratetype == MIGRATE_RESERVE) |
| 925 | continue; |
Mel Gorman | e010487 | 2007-10-16 01:25:53 -0700 | [diff] [blame] | 926 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 927 | area = &(zone->free_area[current_order]); |
| 928 | if (list_empty(&area->free_list[migratetype])) |
| 929 | continue; |
| 930 | |
| 931 | page = list_entry(area->free_list[migratetype].next, |
| 932 | struct page, lru); |
| 933 | area->nr_free--; |
| 934 | |
| 935 | /* |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 936 | * If breaking a large block of pages, move all free |
Mel Gorman | 46dafbc | 2007-10-16 01:25:55 -0700 | [diff] [blame] | 937 | * pages to the preferred allocation list. If falling |
| 938 | * back for a reclaimable kernel allocation, be more |
| 939 | * agressive about taking ownership of free pages |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 940 | */ |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 941 | if (unlikely(current_order >= (pageblock_order >> 1)) || |
Mel Gorman | dd5d241 | 2009-09-05 11:17:11 -0700 | [diff] [blame] | 942 | start_migratetype == MIGRATE_RECLAIMABLE || |
| 943 | page_group_by_mobility_disabled) { |
Mel Gorman | 46dafbc | 2007-10-16 01:25:55 -0700 | [diff] [blame] | 944 | unsigned long pages; |
| 945 | pages = move_freepages_block(zone, page, |
| 946 | start_migratetype); |
| 947 | |
| 948 | /* Claim the whole block if over half of it is free */ |
Mel Gorman | dd5d241 | 2009-09-05 11:17:11 -0700 | [diff] [blame] | 949 | if (pages >= (1 << (pageblock_order-1)) || |
| 950 | page_group_by_mobility_disabled) |
Mel Gorman | 46dafbc | 2007-10-16 01:25:55 -0700 | [diff] [blame] | 951 | set_pageblock_migratetype(page, |
| 952 | start_migratetype); |
| 953 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 954 | migratetype = start_migratetype; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 955 | } |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 956 | |
| 957 | /* Remove the page from the freelists */ |
| 958 | list_del(&page->lru); |
| 959 | rmv_page_order(page); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 960 | |
Mel Gorman | 2f66a68 | 2009-09-21 17:02:31 -0700 | [diff] [blame] | 961 | /* Take ownership for orders >= pageblock_order */ |
| 962 | if (current_order >= pageblock_order) |
| 963 | change_pageblock_range(page, current_order, |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 964 | start_migratetype); |
| 965 | |
| 966 | expand(zone, page, order, current_order, area, migratetype); |
Mel Gorman | e0fff1b | 2009-09-21 17:02:42 -0700 | [diff] [blame] | 967 | |
| 968 | trace_mm_page_alloc_extfrag(page, order, current_order, |
| 969 | start_migratetype, migratetype); |
| 970 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 971 | return page; |
| 972 | } |
| 973 | } |
| 974 | |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 975 | return NULL; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 976 | } |
| 977 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 978 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 979 | * Do the hard work of removing an element from the buddy allocator. |
| 980 | * Call me with the zone->lock already held. |
| 981 | */ |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 982 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
| 983 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 984 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 985 | struct page *page; |
| 986 | |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 987 | retry_reserve: |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 988 | page = __rmqueue_smallest(zone, order, migratetype); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 989 | |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 990 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 991 | page = __rmqueue_fallback(zone, order, migratetype); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 992 | |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 993 | /* |
| 994 | * Use MIGRATE_RESERVE rather than fail an allocation. goto |
| 995 | * is used because __rmqueue_smallest is an inline function |
| 996 | * and we want just one call site |
| 997 | */ |
| 998 | if (!page) { |
| 999 | migratetype = MIGRATE_RESERVE; |
| 1000 | goto retry_reserve; |
| 1001 | } |
| 1002 | } |
| 1003 | |
Mel Gorman | 0d3d062 | 2009-09-21 17:02:44 -0700 | [diff] [blame] | 1004 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1005 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1006 | } |
| 1007 | |
| 1008 | /* |
| 1009 | * Obtain a specified number of elements from the buddy allocator, all under |
| 1010 | * a single hold of the lock, for efficiency. Add them to the supplied list. |
| 1011 | * Returns the number of new pages which were placed at *list. |
| 1012 | */ |
| 1013 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1014 | unsigned long count, struct list_head *list, |
Mel Gorman | e084b2d | 2009-07-29 15:02:04 -0700 | [diff] [blame] | 1015 | int migratetype, int cold) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1016 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1017 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1018 | |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 1019 | spin_lock(&zone->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1020 | for (i = 0; i < count; ++i) { |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1021 | struct page *page = __rmqueue(zone, order, migratetype); |
Nick Piggin | 085cc7d5 | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 1022 | if (unlikely(page == NULL)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1023 | break; |
Mel Gorman | 81eabcb | 2007-12-17 16:20:05 -0800 | [diff] [blame] | 1024 | |
| 1025 | /* |
| 1026 | * Split buddy pages returned by expand() are received here |
| 1027 | * in physical page order. The page is added to the callers and |
| 1028 | * list and the list head then moves forward. From the callers |
| 1029 | * perspective, the linked list is ordered by page number in |
| 1030 | * some conditions. This is useful for IO devices that can |
| 1031 | * merge IO requests if the physical pages are ordered |
| 1032 | * properly. |
| 1033 | */ |
Mel Gorman | e084b2d | 2009-07-29 15:02:04 -0700 | [diff] [blame] | 1034 | if (likely(cold == 0)) |
| 1035 | list_add(&page->lru, list); |
| 1036 | else |
| 1037 | list_add_tail(&page->lru, list); |
Mel Gorman | 535131e6 | 2007-10-16 01:25:49 -0700 | [diff] [blame] | 1038 | set_page_private(page, migratetype); |
Mel Gorman | 81eabcb | 2007-12-17 16:20:05 -0800 | [diff] [blame] | 1039 | list = &page->lru; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1040 | } |
Mel Gorman | f2260e6 | 2009-06-16 15:32:13 -0700 | [diff] [blame] | 1041 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 1042 | spin_unlock(&zone->lock); |
Nick Piggin | 085cc7d5 | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 1043 | return i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1044 | } |
| 1045 | |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 1046 | #ifdef CONFIG_NUMA |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 1047 | /* |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 1048 | * Called from the vmstat counter updater to drain pagesets of this |
| 1049 | * currently executing processor on remote nodes after they have |
| 1050 | * expired. |
| 1051 | * |
Christoph Lameter | 879336c | 2006-03-22 00:09:08 -0800 | [diff] [blame] | 1052 | * Note that this function must be called with the thread pinned to |
| 1053 | * a single processor. |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 1054 | */ |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 1055 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 1056 | { |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 1057 | unsigned long flags; |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 1058 | int to_drain; |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 1059 | |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 1060 | local_irq_save(flags); |
| 1061 | if (pcp->count >= pcp->batch) |
| 1062 | to_drain = pcp->batch; |
| 1063 | else |
| 1064 | to_drain = pcp->count; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1065 | free_pcppages_bulk(zone, to_drain, pcp); |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 1066 | pcp->count -= to_drain; |
| 1067 | local_irq_restore(flags); |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 1068 | } |
| 1069 | #endif |
| 1070 | |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 1071 | /* |
| 1072 | * Drain pages of the indicated processor. |
| 1073 | * |
| 1074 | * The processor must either be the current processor and the |
| 1075 | * thread pinned to the current processor or a processor that |
| 1076 | * is not online. |
| 1077 | */ |
| 1078 | static void drain_pages(unsigned int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1079 | { |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 1080 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1081 | struct zone *zone; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1082 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 1083 | for_each_populated_zone(zone) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1084 | struct per_cpu_pageset *pset; |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 1085 | struct per_cpu_pages *pcp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1086 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 1087 | local_irq_save(flags); |
| 1088 | pset = per_cpu_ptr(zone->pageset, cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1089 | |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 1090 | pcp = &pset->pcp; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1091 | free_pcppages_bulk(zone, pcp->count, pcp); |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 1092 | pcp->count = 0; |
| 1093 | local_irq_restore(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1094 | } |
| 1095 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1096 | |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 1097 | /* |
| 1098 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. |
| 1099 | */ |
| 1100 | void drain_local_pages(void *arg) |
| 1101 | { |
| 1102 | drain_pages(smp_processor_id()); |
| 1103 | } |
| 1104 | |
| 1105 | /* |
| 1106 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator |
| 1107 | */ |
| 1108 | void drain_all_pages(void) |
| 1109 | { |
Jens Axboe | 15c8b6c | 2008-05-09 09:39:44 +0200 | [diff] [blame] | 1110 | on_each_cpu(drain_local_pages, NULL, 1); |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 1111 | } |
| 1112 | |
Rafael J. Wysocki | 296699d | 2007-07-29 23:27:18 +0200 | [diff] [blame] | 1113 | #ifdef CONFIG_HIBERNATION |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1114 | |
| 1115 | void mark_free_pages(struct zone *zone) |
| 1116 | { |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 1117 | unsigned long pfn, max_zone_pfn; |
| 1118 | unsigned long flags; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1119 | int order, t; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1120 | struct list_head *curr; |
| 1121 | |
| 1122 | if (!zone->spanned_pages) |
| 1123 | return; |
| 1124 | |
| 1125 | spin_lock_irqsave(&zone->lock, flags); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 1126 | |
| 1127 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
| 1128 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
| 1129 | if (pfn_valid(pfn)) { |
| 1130 | struct page *page = pfn_to_page(pfn); |
| 1131 | |
Rafael J. Wysocki | 7be9823 | 2007-05-06 14:50:42 -0700 | [diff] [blame] | 1132 | if (!swsusp_page_is_forbidden(page)) |
| 1133 | swsusp_unset_page_free(page); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 1134 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1135 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1136 | for_each_migratetype_order(order, t) { |
| 1137 | list_for_each(curr, &zone->free_area[order].free_list[t]) { |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 1138 | unsigned long i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1139 | |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 1140 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
| 1141 | for (i = 0; i < (1UL << order); i++) |
Rafael J. Wysocki | 7be9823 | 2007-05-06 14:50:42 -0700 | [diff] [blame] | 1142 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 1143 | } |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1144 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1145 | spin_unlock_irqrestore(&zone->lock, flags); |
| 1146 | } |
Mel Gorman | e2c55dc | 2007-10-16 01:25:50 -0700 | [diff] [blame] | 1147 | #endif /* CONFIG_PM */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1148 | |
| 1149 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1150 | * Free a 0-order page |
Li Hong | fc91668 | 2010-03-05 13:41:54 -0800 | [diff] [blame] | 1151 | * cold == 1 ? free a cold page : free a hot page |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1152 | */ |
Li Hong | fc91668 | 2010-03-05 13:41:54 -0800 | [diff] [blame] | 1153 | void free_hot_cold_page(struct page *page, int cold) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1154 | { |
| 1155 | struct zone *zone = page_zone(page); |
| 1156 | struct per_cpu_pages *pcp; |
| 1157 | unsigned long flags; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1158 | int migratetype; |
Johannes Weiner | 451ea25 | 2009-09-21 17:01:48 -0700 | [diff] [blame] | 1159 | int wasMlocked = __TestClearPageMlocked(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1160 | |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 1161 | if (!free_pages_prepare(page, 0)) |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 1162 | return; |
| 1163 | |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1164 | migratetype = get_pageblock_migratetype(page); |
| 1165 | set_page_private(page, migratetype); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1166 | local_irq_save(flags); |
Johannes Weiner | c277331 | 2009-06-19 19:30:56 +0200 | [diff] [blame] | 1167 | if (unlikely(wasMlocked)) |
Mel Gorman | da456f1 | 2009-06-16 15:32:08 -0700 | [diff] [blame] | 1168 | free_page_mlock(page); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1169 | __count_vm_event(PGFREE); |
Mel Gorman | da456f1 | 2009-06-16 15:32:08 -0700 | [diff] [blame] | 1170 | |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1171 | /* |
| 1172 | * We only track unmovable, reclaimable and movable on pcp lists. |
| 1173 | * Free ISOLATE pages back to the allocator because they are being |
| 1174 | * offlined but treat RESERVE as movable pages so we can get those |
| 1175 | * areas back if necessary. Otherwise, we may have to free |
| 1176 | * excessively into the page allocator |
| 1177 | */ |
| 1178 | if (migratetype >= MIGRATE_PCPTYPES) { |
| 1179 | if (unlikely(migratetype == MIGRATE_ISOLATE)) { |
| 1180 | free_one_page(zone, page, 0, migratetype); |
| 1181 | goto out; |
| 1182 | } |
| 1183 | migratetype = MIGRATE_MOVABLE; |
| 1184 | } |
| 1185 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 1186 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 1187 | if (cold) |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1188 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 1189 | else |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1190 | list_add(&page->lru, &pcp->lists[migratetype]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1191 | pcp->count++; |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 1192 | if (pcp->count >= pcp->high) { |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1193 | free_pcppages_bulk(zone, pcp->batch, pcp); |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 1194 | pcp->count -= pcp->batch; |
| 1195 | } |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1196 | |
| 1197 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1198 | local_irq_restore(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1199 | } |
| 1200 | |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 1201 | /* |
| 1202 | * split_page takes a non-compound higher-order page, and splits it into |
| 1203 | * n (1<<order) sub-pages: page[0..n] |
| 1204 | * Each sub-page must be freed individually. |
| 1205 | * |
| 1206 | * Note: this is probably too low level an operation for use in drivers. |
| 1207 | * Please consult with lkml before using this in your driver. |
| 1208 | */ |
| 1209 | void split_page(struct page *page, unsigned int order) |
| 1210 | { |
| 1211 | int i; |
| 1212 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 1213 | VM_BUG_ON(PageCompound(page)); |
| 1214 | VM_BUG_ON(!page_count(page)); |
Vegard Nossum | b1eeab6 | 2008-11-25 16:55:53 +0100 | [diff] [blame] | 1215 | |
| 1216 | #ifdef CONFIG_KMEMCHECK |
| 1217 | /* |
| 1218 | * Split shadow pages too, because free(page[0]) would |
| 1219 | * otherwise free the whole shadow. |
| 1220 | */ |
| 1221 | if (kmemcheck_page_is_tracked(page)) |
| 1222 | split_page(virt_to_page(page[0].shadow), order); |
| 1223 | #endif |
| 1224 | |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 1225 | for (i = 1; i < (1 << order); i++) |
| 1226 | set_page_refcounted(page + i); |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 1227 | } |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 1228 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1229 | /* |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1230 | * Similar to split_page except the page is already free. As this is only |
| 1231 | * being used for migration, the migratetype of the block also changes. |
| 1232 | * As this is called with interrupts disabled, the caller is responsible |
| 1233 | * for calling arch_alloc_page() and kernel_map_page() after interrupts |
| 1234 | * are enabled. |
| 1235 | * |
| 1236 | * Note: this is probably too low level an operation for use in drivers. |
| 1237 | * Please consult with lkml before using this in your driver. |
| 1238 | */ |
| 1239 | int split_free_page(struct page *page) |
| 1240 | { |
| 1241 | unsigned int order; |
| 1242 | unsigned long watermark; |
| 1243 | struct zone *zone; |
| 1244 | |
| 1245 | BUG_ON(!PageBuddy(page)); |
| 1246 | |
| 1247 | zone = page_zone(page); |
| 1248 | order = page_order(page); |
| 1249 | |
| 1250 | /* Obey watermarks as if the page was being allocated */ |
| 1251 | watermark = low_wmark_pages(zone) + (1 << order); |
| 1252 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) |
| 1253 | return 0; |
| 1254 | |
| 1255 | /* Remove page from free list */ |
| 1256 | list_del(&page->lru); |
| 1257 | zone->free_area[order].nr_free--; |
| 1258 | rmv_page_order(page); |
| 1259 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order)); |
| 1260 | |
| 1261 | /* Split into individual pages */ |
| 1262 | set_page_refcounted(page); |
| 1263 | split_page(page, order); |
| 1264 | |
| 1265 | if (order >= pageblock_order - 1) { |
| 1266 | struct page *endpage = page + (1 << order) - 1; |
| 1267 | for (; page < endpage; page += pageblock_nr_pages) |
| 1268 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
| 1269 | } |
| 1270 | |
| 1271 | return 1 << order; |
| 1272 | } |
| 1273 | |
| 1274 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1275 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But |
| 1276 | * we cheat by calling it from here, in the order > 0 path. Saves a branch |
| 1277 | * or two. |
| 1278 | */ |
Mel Gorman | 0a15c3e | 2009-06-16 15:32:05 -0700 | [diff] [blame] | 1279 | static inline |
| 1280 | struct page *buffered_rmqueue(struct zone *preferred_zone, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1281 | struct zone *zone, int order, gfp_t gfp_flags, |
| 1282 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1283 | { |
| 1284 | unsigned long flags; |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 1285 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1286 | int cold = !!(gfp_flags & __GFP_COLD); |
| 1287 | |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 1288 | again: |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 1289 | if (likely(order == 0)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1290 | struct per_cpu_pages *pcp; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1291 | struct list_head *list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1292 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1293 | local_irq_save(flags); |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 1294 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
| 1295 | list = &pcp->lists[migratetype]; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1296 | if (list_empty(list)) { |
Mel Gorman | 535131e6 | 2007-10-16 01:25:49 -0700 | [diff] [blame] | 1297 | pcp->count += rmqueue_bulk(zone, 0, |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1298 | pcp->batch, list, |
Mel Gorman | e084b2d | 2009-07-29 15:02:04 -0700 | [diff] [blame] | 1299 | migratetype, cold); |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1300 | if (unlikely(list_empty(list))) |
Shaohua Li | 6fb332f | 2009-09-21 17:01:17 -0700 | [diff] [blame] | 1301 | goto failed; |
Mel Gorman | 535131e6 | 2007-10-16 01:25:49 -0700 | [diff] [blame] | 1302 | } |
Mel Gorman | b92a6ed | 2007-10-16 01:25:50 -0700 | [diff] [blame] | 1303 | |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1304 | if (cold) |
| 1305 | page = list_entry(list->prev, struct page, lru); |
| 1306 | else |
| 1307 | page = list_entry(list->next, struct page, lru); |
| 1308 | |
Mel Gorman | b92a6ed | 2007-10-16 01:25:50 -0700 | [diff] [blame] | 1309 | list_del(&page->lru); |
| 1310 | pcp->count--; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1311 | } else { |
Andrew Morton | dab48da | 2009-06-16 15:32:37 -0700 | [diff] [blame] | 1312 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
| 1313 | /* |
| 1314 | * __GFP_NOFAIL is not to be used in new code. |
| 1315 | * |
| 1316 | * All __GFP_NOFAIL callers should be fixed so that they |
| 1317 | * properly detect and handle allocation failures. |
| 1318 | * |
| 1319 | * We most definitely don't want callers attempting to |
Linus Torvalds | 4923abf | 2009-06-24 12:16:49 -0700 | [diff] [blame] | 1320 | * allocate greater than order-1 page units with |
Andrew Morton | dab48da | 2009-06-16 15:32:37 -0700 | [diff] [blame] | 1321 | * __GFP_NOFAIL. |
| 1322 | */ |
Linus Torvalds | 4923abf | 2009-06-24 12:16:49 -0700 | [diff] [blame] | 1323 | WARN_ON_ONCE(order > 1); |
Andrew Morton | dab48da | 2009-06-16 15:32:37 -0700 | [diff] [blame] | 1324 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1325 | spin_lock_irqsave(&zone->lock, flags); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1326 | page = __rmqueue(zone, order, migratetype); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1327 | spin_unlock(&zone->lock); |
| 1328 | if (!page) |
| 1329 | goto failed; |
KOSAKI Motohiro | 6ccf80e | 2010-01-15 17:01:18 -0800 | [diff] [blame] | 1330 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1331 | } |
| 1332 | |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1333 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
Mel Gorman | 18ea7e7 | 2008-04-28 02:12:14 -0700 | [diff] [blame] | 1334 | zone_statistics(preferred_zone, zone); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1335 | local_irq_restore(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1336 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 1337 | VM_BUG_ON(bad_range(zone, page)); |
Nick Piggin | 17cf440 | 2006-03-22 00:08:41 -0800 | [diff] [blame] | 1338 | if (prep_new_page(page, order, gfp_flags)) |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1339 | goto again; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1340 | return page; |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1341 | |
| 1342 | failed: |
| 1343 | local_irq_restore(flags); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1344 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1345 | } |
| 1346 | |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 1347 | /* The ALLOC_WMARK bits are used as an index to zone->watermark */ |
| 1348 | #define ALLOC_WMARK_MIN WMARK_MIN |
| 1349 | #define ALLOC_WMARK_LOW WMARK_LOW |
| 1350 | #define ALLOC_WMARK_HIGH WMARK_HIGH |
| 1351 | #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ |
| 1352 | |
| 1353 | /* Mask to get the watermark bits */ |
| 1354 | #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) |
| 1355 | |
Nick Piggin | 3148890 | 2005-11-28 13:44:03 -0800 | [diff] [blame] | 1356 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ |
| 1357 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ |
| 1358 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1359 | |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1360 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
| 1361 | |
| 1362 | static struct fail_page_alloc_attr { |
| 1363 | struct fault_attr attr; |
| 1364 | |
| 1365 | u32 ignore_gfp_highmem; |
| 1366 | u32 ignore_gfp_wait; |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 1367 | u32 min_order; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1368 | |
| 1369 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS |
| 1370 | |
| 1371 | struct dentry *ignore_gfp_highmem_file; |
| 1372 | struct dentry *ignore_gfp_wait_file; |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 1373 | struct dentry *min_order_file; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1374 | |
| 1375 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ |
| 1376 | |
| 1377 | } fail_page_alloc = { |
| 1378 | .attr = FAULT_ATTR_INITIALIZER, |
Don Mullis | 6b1b60f | 2006-12-08 02:39:53 -0800 | [diff] [blame] | 1379 | .ignore_gfp_wait = 1, |
| 1380 | .ignore_gfp_highmem = 1, |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 1381 | .min_order = 1, |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1382 | }; |
| 1383 | |
| 1384 | static int __init setup_fail_page_alloc(char *str) |
| 1385 | { |
| 1386 | return setup_fault_attr(&fail_page_alloc.attr, str); |
| 1387 | } |
| 1388 | __setup("fail_page_alloc=", setup_fail_page_alloc); |
| 1389 | |
| 1390 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
| 1391 | { |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 1392 | if (order < fail_page_alloc.min_order) |
| 1393 | return 0; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1394 | if (gfp_mask & __GFP_NOFAIL) |
| 1395 | return 0; |
| 1396 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
| 1397 | return 0; |
| 1398 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) |
| 1399 | return 0; |
| 1400 | |
| 1401 | return should_fail(&fail_page_alloc.attr, 1 << order); |
| 1402 | } |
| 1403 | |
| 1404 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS |
| 1405 | |
| 1406 | static int __init fail_page_alloc_debugfs(void) |
| 1407 | { |
| 1408 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
| 1409 | struct dentry *dir; |
| 1410 | int err; |
| 1411 | |
| 1412 | err = init_fault_attr_dentries(&fail_page_alloc.attr, |
| 1413 | "fail_page_alloc"); |
| 1414 | if (err) |
| 1415 | return err; |
| 1416 | dir = fail_page_alloc.attr.dentries.dir; |
| 1417 | |
| 1418 | fail_page_alloc.ignore_gfp_wait_file = |
| 1419 | debugfs_create_bool("ignore-gfp-wait", mode, dir, |
| 1420 | &fail_page_alloc.ignore_gfp_wait); |
| 1421 | |
| 1422 | fail_page_alloc.ignore_gfp_highmem_file = |
| 1423 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, |
| 1424 | &fail_page_alloc.ignore_gfp_highmem); |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 1425 | fail_page_alloc.min_order_file = |
| 1426 | debugfs_create_u32("min-order", mode, dir, |
| 1427 | &fail_page_alloc.min_order); |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1428 | |
| 1429 | if (!fail_page_alloc.ignore_gfp_wait_file || |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 1430 | !fail_page_alloc.ignore_gfp_highmem_file || |
| 1431 | !fail_page_alloc.min_order_file) { |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1432 | err = -ENOMEM; |
| 1433 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); |
| 1434 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 1435 | debugfs_remove(fail_page_alloc.min_order_file); |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1436 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); |
| 1437 | } |
| 1438 | |
| 1439 | return err; |
| 1440 | } |
| 1441 | |
| 1442 | late_initcall(fail_page_alloc_debugfs); |
| 1443 | |
| 1444 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ |
| 1445 | |
| 1446 | #else /* CONFIG_FAIL_PAGE_ALLOC */ |
| 1447 | |
| 1448 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
| 1449 | { |
| 1450 | return 0; |
| 1451 | } |
| 1452 | |
| 1453 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ |
| 1454 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1455 | /* |
| 1456 | * Return 1 if free pages are above 'mark'. This takes into account the order |
| 1457 | * of the allocation. |
| 1458 | */ |
| 1459 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1460 | int classzone_idx, int alloc_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1461 | { |
| 1462 | /* free_pages my go negative - that's OK */ |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 1463 | long min = mark; |
| 1464 | long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1465 | int o; |
| 1466 | |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1467 | if (alloc_flags & ALLOC_HIGH) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1468 | min -= min / 2; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1469 | if (alloc_flags & ALLOC_HARDER) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1470 | min -= min / 4; |
| 1471 | |
| 1472 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) |
| 1473 | return 0; |
| 1474 | for (o = 0; o < order; o++) { |
| 1475 | /* At the next order, this order's pages become unavailable */ |
| 1476 | free_pages -= z->free_area[o].nr_free << o; |
| 1477 | |
| 1478 | /* Require fewer higher order pages to be free */ |
| 1479 | min >>= 1; |
| 1480 | |
| 1481 | if (free_pages <= min) |
| 1482 | return 0; |
| 1483 | } |
| 1484 | return 1; |
| 1485 | } |
| 1486 | |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1487 | #ifdef CONFIG_NUMA |
| 1488 | /* |
| 1489 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to |
| 1490 | * skip over zones that are not allowed by the cpuset, or that have |
| 1491 | * been recently (in last second) found to be nearly full. See further |
| 1492 | * comments in mmzone.h. Reduces cache footprint of zonelist scans |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 1493 | * that have to skip over a lot of full or unallowed zones. |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1494 | * |
| 1495 | * If the zonelist cache is present in the passed in zonelist, then |
| 1496 | * returns a pointer to the allowed node mask (either the current |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 1497 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1498 | * |
| 1499 | * If the zonelist cache is not available for this zonelist, does |
| 1500 | * nothing and returns NULL. |
| 1501 | * |
| 1502 | * If the fullzones BITMAP in the zonelist cache is stale (more than |
| 1503 | * a second since last zap'd) then we zap it out (clear its bits.) |
| 1504 | * |
| 1505 | * We hold off even calling zlc_setup, until after we've checked the |
| 1506 | * first zone in the zonelist, on the theory that most allocations will |
| 1507 | * be satisfied from that first zone, so best to examine that zone as |
| 1508 | * quickly as we can. |
| 1509 | */ |
| 1510 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) |
| 1511 | { |
| 1512 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ |
| 1513 | nodemask_t *allowednodes; /* zonelist_cache approximation */ |
| 1514 | |
| 1515 | zlc = zonelist->zlcache_ptr; |
| 1516 | if (!zlc) |
| 1517 | return NULL; |
| 1518 | |
S.Caglar Onur | f05111f | 2008-04-28 02:12:38 -0700 | [diff] [blame] | 1519 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1520 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
| 1521 | zlc->last_full_zap = jiffies; |
| 1522 | } |
| 1523 | |
| 1524 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? |
| 1525 | &cpuset_current_mems_allowed : |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 1526 | &node_states[N_HIGH_MEMORY]; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1527 | return allowednodes; |
| 1528 | } |
| 1529 | |
| 1530 | /* |
| 1531 | * Given 'z' scanning a zonelist, run a couple of quick checks to see |
| 1532 | * if it is worth looking at further for free memory: |
| 1533 | * 1) Check that the zone isn't thought to be full (doesn't have its |
| 1534 | * bit set in the zonelist_cache fullzones BITMAP). |
| 1535 | * 2) Check that the zones node (obtained from the zonelist_cache |
| 1536 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. |
| 1537 | * Return true (non-zero) if zone is worth looking at further, or |
| 1538 | * else return false (zero) if it is not. |
| 1539 | * |
| 1540 | * This check -ignores- the distinction between various watermarks, |
| 1541 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is |
| 1542 | * found to be full for any variation of these watermarks, it will |
| 1543 | * be considered full for up to one second by all requests, unless |
| 1544 | * we are so low on memory on all allowed nodes that we are forced |
| 1545 | * into the second scan of the zonelist. |
| 1546 | * |
| 1547 | * In the second scan we ignore this zonelist cache and exactly |
| 1548 | * apply the watermarks to all zones, even it is slower to do so. |
| 1549 | * We are low on memory in the second scan, and should leave no stone |
| 1550 | * unturned looking for a free page. |
| 1551 | */ |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1552 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1553 | nodemask_t *allowednodes) |
| 1554 | { |
| 1555 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ |
| 1556 | int i; /* index of *z in zonelist zones */ |
| 1557 | int n; /* node that zone *z is on */ |
| 1558 | |
| 1559 | zlc = zonelist->zlcache_ptr; |
| 1560 | if (!zlc) |
| 1561 | return 1; |
| 1562 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1563 | i = z - zonelist->_zonerefs; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1564 | n = zlc->z_to_n[i]; |
| 1565 | |
| 1566 | /* This zone is worth trying if it is allowed but not full */ |
| 1567 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); |
| 1568 | } |
| 1569 | |
| 1570 | /* |
| 1571 | * Given 'z' scanning a zonelist, set the corresponding bit in |
| 1572 | * zlc->fullzones, so that subsequent attempts to allocate a page |
| 1573 | * from that zone don't waste time re-examining it. |
| 1574 | */ |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1575 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1576 | { |
| 1577 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ |
| 1578 | int i; /* index of *z in zonelist zones */ |
| 1579 | |
| 1580 | zlc = zonelist->zlcache_ptr; |
| 1581 | if (!zlc) |
| 1582 | return; |
| 1583 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1584 | i = z - zonelist->_zonerefs; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1585 | |
| 1586 | set_bit(i, zlc->fullzones); |
| 1587 | } |
| 1588 | |
| 1589 | #else /* CONFIG_NUMA */ |
| 1590 | |
| 1591 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) |
| 1592 | { |
| 1593 | return NULL; |
| 1594 | } |
| 1595 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1596 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1597 | nodemask_t *allowednodes) |
| 1598 | { |
| 1599 | return 1; |
| 1600 | } |
| 1601 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1602 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1603 | { |
| 1604 | } |
| 1605 | #endif /* CONFIG_NUMA */ |
| 1606 | |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1607 | /* |
Paul Jackson | 0798e51 | 2006-12-06 20:31:38 -0800 | [diff] [blame] | 1608 | * get_page_from_freelist goes through the zonelist trying to allocate |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1609 | * a page. |
| 1610 | */ |
| 1611 | static struct page * |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1612 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 1613 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1614 | struct zone *preferred_zone, int migratetype) |
Martin Hicks | 753ee72 | 2005-06-21 17:14:41 -0700 | [diff] [blame] | 1615 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1616 | struct zoneref *z; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1617 | struct page *page = NULL; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1618 | int classzone_idx; |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 1619 | struct zone *zone; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1620 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
| 1621 | int zlc_active = 0; /* set if using zonelist_cache */ |
| 1622 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1623 | |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1624 | classzone_idx = zone_idx(preferred_zone); |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1625 | zonelist_scan: |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1626 | /* |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1627 | * Scan zonelist, looking for a zone with enough free. |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1628 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
| 1629 | */ |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1630 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
| 1631 | high_zoneidx, nodemask) { |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1632 | if (NUMA_BUILD && zlc_active && |
| 1633 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
| 1634 | continue; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1635 | if ((alloc_flags & ALLOC_CPUSET) && |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 1636 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1637 | goto try_next_zone; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1638 | |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 1639 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1640 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { |
Nick Piggin | 3148890 | 2005-11-28 13:44:03 -0800 | [diff] [blame] | 1641 | unsigned long mark; |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 1642 | int ret; |
| 1643 | |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 1644 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 1645 | if (zone_watermark_ok(zone, order, mark, |
| 1646 | classzone_idx, alloc_flags)) |
| 1647 | goto try_this_zone; |
| 1648 | |
| 1649 | if (zone_reclaim_mode == 0) |
| 1650 | goto this_zone_full; |
| 1651 | |
| 1652 | ret = zone_reclaim(zone, gfp_mask, order); |
| 1653 | switch (ret) { |
| 1654 | case ZONE_RECLAIM_NOSCAN: |
| 1655 | /* did not scan */ |
| 1656 | goto try_next_zone; |
| 1657 | case ZONE_RECLAIM_FULL: |
| 1658 | /* scanned but unreclaimable */ |
| 1659 | goto this_zone_full; |
| 1660 | default: |
| 1661 | /* did we reclaim enough */ |
| 1662 | if (!zone_watermark_ok(zone, order, mark, |
| 1663 | classzone_idx, alloc_flags)) |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1664 | goto this_zone_full; |
Paul Jackson | 0798e51 | 2006-12-06 20:31:38 -0800 | [diff] [blame] | 1665 | } |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1666 | } |
| 1667 | |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 1668 | try_this_zone: |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1669 | page = buffered_rmqueue(preferred_zone, zone, order, |
| 1670 | gfp_mask, migratetype); |
Paul Jackson | 0798e51 | 2006-12-06 20:31:38 -0800 | [diff] [blame] | 1671 | if (page) |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1672 | break; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1673 | this_zone_full: |
| 1674 | if (NUMA_BUILD) |
| 1675 | zlc_mark_zone_full(zonelist, z); |
| 1676 | try_next_zone: |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 1677 | if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) { |
Mel Gorman | d395b73 | 2009-06-16 15:32:09 -0700 | [diff] [blame] | 1678 | /* |
| 1679 | * we do zlc_setup after the first zone is tried but only |
| 1680 | * if there are multiple nodes make it worthwhile |
| 1681 | */ |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1682 | allowednodes = zlc_setup(zonelist, alloc_flags); |
| 1683 | zlc_active = 1; |
| 1684 | did_zlc_setup = 1; |
| 1685 | } |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1686 | } |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1687 | |
| 1688 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { |
| 1689 | /* Disable zlc cache for second zonelist scan */ |
| 1690 | zlc_active = 0; |
| 1691 | goto zonelist_scan; |
| 1692 | } |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1693 | return page; |
Martin Hicks | 753ee72 | 2005-06-21 17:14:41 -0700 | [diff] [blame] | 1694 | } |
| 1695 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1696 | static inline int |
| 1697 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, |
| 1698 | unsigned long pages_reclaimed) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1699 | { |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1700 | /* Do not loop if specifically requested */ |
| 1701 | if (gfp_mask & __GFP_NORETRY) |
| 1702 | return 0; |
| 1703 | |
| 1704 | /* |
| 1705 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER |
| 1706 | * means __GFP_NOFAIL, but that may not be true in other |
| 1707 | * implementations. |
| 1708 | */ |
| 1709 | if (order <= PAGE_ALLOC_COSTLY_ORDER) |
| 1710 | return 1; |
| 1711 | |
| 1712 | /* |
| 1713 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is |
| 1714 | * specified, then we retry until we no longer reclaim any pages |
| 1715 | * (above), or we've reclaimed an order of pages at least as |
| 1716 | * large as the allocation's order. In both cases, if the |
| 1717 | * allocation still fails, we stop retrying. |
| 1718 | */ |
| 1719 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) |
| 1720 | return 1; |
| 1721 | |
| 1722 | /* |
| 1723 | * Don't let big-order allocations loop unless the caller |
| 1724 | * explicitly requests that. |
| 1725 | */ |
| 1726 | if (gfp_mask & __GFP_NOFAIL) |
| 1727 | return 1; |
| 1728 | |
| 1729 | return 0; |
| 1730 | } |
| 1731 | |
| 1732 | static inline struct page * |
| 1733 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, |
| 1734 | struct zonelist *zonelist, enum zone_type high_zoneidx, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1735 | nodemask_t *nodemask, struct zone *preferred_zone, |
| 1736 | int migratetype) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1737 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1738 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1739 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1740 | /* Acquire the OOM killer lock for the zones in zonelist */ |
| 1741 | if (!try_set_zone_oom(zonelist, gfp_mask)) { |
| 1742 | schedule_timeout_uninterruptible(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1743 | return NULL; |
| 1744 | } |
Jens Axboe | 6b1de91 | 2005-11-17 21:35:02 +0100 | [diff] [blame] | 1745 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1746 | /* |
| 1747 | * Go through the zonelist yet one more time, keep very high watermark |
| 1748 | * here, this is only to catch a parallel oom killing, we must fail if |
| 1749 | * we're still under heavy pressure. |
| 1750 | */ |
| 1751 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, |
| 1752 | order, zonelist, high_zoneidx, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 1753 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1754 | preferred_zone, migratetype); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1755 | if (page) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1756 | goto out; |
| 1757 | |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 1758 | if (!(gfp_mask & __GFP_NOFAIL)) { |
| 1759 | /* The OOM killer will not help higher order allocs */ |
| 1760 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
| 1761 | goto out; |
| 1762 | /* |
| 1763 | * GFP_THISNODE contains __GFP_NORETRY and we never hit this. |
| 1764 | * Sanity check for bare calls of __GFP_THISNODE, not real OOM. |
| 1765 | * The caller should handle page allocation failure by itself if |
| 1766 | * it specifies __GFP_THISNODE. |
| 1767 | * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER. |
| 1768 | */ |
| 1769 | if (gfp_mask & __GFP_THISNODE) |
| 1770 | goto out; |
| 1771 | } |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1772 | /* Exhausted what can be done so it's blamo time */ |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 1773 | out_of_memory(zonelist, gfp_mask, order, nodemask); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1774 | |
| 1775 | out: |
| 1776 | clear_zonelist_oom(zonelist, gfp_mask); |
| 1777 | return page; |
| 1778 | } |
| 1779 | |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1780 | #ifdef CONFIG_COMPACTION |
| 1781 | /* Try memory compaction for high-order allocations before reclaim */ |
| 1782 | static struct page * |
| 1783 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, |
| 1784 | struct zonelist *zonelist, enum zone_type high_zoneidx, |
| 1785 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, |
| 1786 | int migratetype, unsigned long *did_some_progress) |
| 1787 | { |
| 1788 | struct page *page; |
| 1789 | |
Mel Gorman | 4f92e25 | 2010-05-24 14:32:32 -0700 | [diff] [blame] | 1790 | if (!order || compaction_deferred(preferred_zone)) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1791 | return NULL; |
| 1792 | |
| 1793 | *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask, |
| 1794 | nodemask); |
| 1795 | if (*did_some_progress != COMPACT_SKIPPED) { |
| 1796 | |
| 1797 | /* Page migration frees to the PCP lists but we want merging */ |
| 1798 | drain_pages(get_cpu()); |
| 1799 | put_cpu(); |
| 1800 | |
| 1801 | page = get_page_from_freelist(gfp_mask, nodemask, |
| 1802 | order, zonelist, high_zoneidx, |
| 1803 | alloc_flags, preferred_zone, |
| 1804 | migratetype); |
| 1805 | if (page) { |
Mel Gorman | 4f92e25 | 2010-05-24 14:32:32 -0700 | [diff] [blame] | 1806 | preferred_zone->compact_considered = 0; |
| 1807 | preferred_zone->compact_defer_shift = 0; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1808 | count_vm_event(COMPACTSUCCESS); |
| 1809 | return page; |
| 1810 | } |
| 1811 | |
| 1812 | /* |
| 1813 | * It's bad if compaction run occurs and fails. |
| 1814 | * The most likely reason is that pages exist, |
| 1815 | * but not enough to satisfy watermarks. |
| 1816 | */ |
| 1817 | count_vm_event(COMPACTFAIL); |
Mel Gorman | 4f92e25 | 2010-05-24 14:32:32 -0700 | [diff] [blame] | 1818 | defer_compaction(preferred_zone); |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1819 | |
| 1820 | cond_resched(); |
| 1821 | } |
| 1822 | |
| 1823 | return NULL; |
| 1824 | } |
| 1825 | #else |
| 1826 | static inline struct page * |
| 1827 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, |
| 1828 | struct zonelist *zonelist, enum zone_type high_zoneidx, |
| 1829 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, |
| 1830 | int migratetype, unsigned long *did_some_progress) |
| 1831 | { |
| 1832 | return NULL; |
| 1833 | } |
| 1834 | #endif /* CONFIG_COMPACTION */ |
| 1835 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1836 | /* The really slow allocator path where we enter direct reclaim */ |
| 1837 | static inline struct page * |
| 1838 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, |
| 1839 | struct zonelist *zonelist, enum zone_type high_zoneidx, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 1840 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1841 | int migratetype, unsigned long *did_some_progress) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1842 | { |
| 1843 | struct page *page = NULL; |
| 1844 | struct reclaim_state reclaim_state; |
| 1845 | struct task_struct *p = current; |
| 1846 | |
| 1847 | cond_resched(); |
| 1848 | |
| 1849 | /* We now go into synchronous reclaim */ |
| 1850 | cpuset_memory_pressure_bump(); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1851 | p->flags |= PF_MEMALLOC; |
| 1852 | lockdep_set_current_reclaim_state(gfp_mask); |
| 1853 | reclaim_state.reclaimed_slab = 0; |
| 1854 | p->reclaim_state = &reclaim_state; |
| 1855 | |
| 1856 | *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); |
| 1857 | |
| 1858 | p->reclaim_state = NULL; |
| 1859 | lockdep_clear_current_reclaim_state(); |
| 1860 | p->flags &= ~PF_MEMALLOC; |
| 1861 | |
| 1862 | cond_resched(); |
| 1863 | |
| 1864 | if (order != 0) |
| 1865 | drain_all_pages(); |
| 1866 | |
| 1867 | if (likely(*did_some_progress)) |
| 1868 | page = get_page_from_freelist(gfp_mask, nodemask, order, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 1869 | zonelist, high_zoneidx, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1870 | alloc_flags, preferred_zone, |
| 1871 | migratetype); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1872 | return page; |
| 1873 | } |
| 1874 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1875 | /* |
| 1876 | * This is called in the allocator slow-path if the allocation request is of |
| 1877 | * sufficient urgency to ignore watermarks and take other desperate measures |
| 1878 | */ |
| 1879 | static inline struct page * |
| 1880 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, |
| 1881 | struct zonelist *zonelist, enum zone_type high_zoneidx, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1882 | nodemask_t *nodemask, struct zone *preferred_zone, |
| 1883 | int migratetype) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1884 | { |
| 1885 | struct page *page; |
| 1886 | |
| 1887 | do { |
| 1888 | page = get_page_from_freelist(gfp_mask, nodemask, order, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 1889 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1890 | preferred_zone, migratetype); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1891 | |
| 1892 | if (!page && gfp_mask & __GFP_NOFAIL) |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 1893 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1894 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
| 1895 | |
| 1896 | return page; |
| 1897 | } |
| 1898 | |
| 1899 | static inline |
| 1900 | void wake_all_kswapd(unsigned int order, struct zonelist *zonelist, |
| 1901 | enum zone_type high_zoneidx) |
| 1902 | { |
| 1903 | struct zoneref *z; |
| 1904 | struct zone *zone; |
| 1905 | |
| 1906 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) |
| 1907 | wakeup_kswapd(zone, order); |
| 1908 | } |
| 1909 | |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 1910 | static inline int |
| 1911 | gfp_to_alloc_flags(gfp_t gfp_mask) |
| 1912 | { |
| 1913 | struct task_struct *p = current; |
| 1914 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
| 1915 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
| 1916 | |
Mel Gorman | a56f57f | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 1917 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
| 1918 | BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH); |
| 1919 | |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 1920 | /* |
| 1921 | * The caller may dip into page reserves a bit more if the caller |
| 1922 | * cannot run direct reclaim, or if the caller has realtime scheduling |
| 1923 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
| 1924 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). |
| 1925 | */ |
Mel Gorman | a56f57f | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 1926 | alloc_flags |= (gfp_mask & __GFP_HIGH); |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 1927 | |
| 1928 | if (!wait) { |
| 1929 | alloc_flags |= ALLOC_HARDER; |
| 1930 | /* |
| 1931 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. |
| 1932 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
| 1933 | */ |
| 1934 | alloc_flags &= ~ALLOC_CPUSET; |
Mel Gorman | 9d0ed60 | 2009-11-11 14:26:17 -0800 | [diff] [blame] | 1935 | } else if (unlikely(rt_task(p)) && !in_interrupt()) |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 1936 | alloc_flags |= ALLOC_HARDER; |
| 1937 | |
| 1938 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
| 1939 | if (!in_interrupt() && |
| 1940 | ((p->flags & PF_MEMALLOC) || |
| 1941 | unlikely(test_thread_flag(TIF_MEMDIE)))) |
| 1942 | alloc_flags |= ALLOC_NO_WATERMARKS; |
| 1943 | } |
| 1944 | |
| 1945 | return alloc_flags; |
| 1946 | } |
| 1947 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1948 | static inline struct page * |
| 1949 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, |
| 1950 | struct zonelist *zonelist, enum zone_type high_zoneidx, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1951 | nodemask_t *nodemask, struct zone *preferred_zone, |
| 1952 | int migratetype) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1953 | { |
| 1954 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
| 1955 | struct page *page = NULL; |
| 1956 | int alloc_flags; |
| 1957 | unsigned long pages_reclaimed = 0; |
| 1958 | unsigned long did_some_progress; |
| 1959 | struct task_struct *p = current; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1960 | |
Christoph Lameter | 952f3b5 | 2006-12-06 20:33:26 -0800 | [diff] [blame] | 1961 | /* |
Mel Gorman | 72807a7 | 2009-06-16 15:32:18 -0700 | [diff] [blame] | 1962 | * In the slowpath, we sanity check order to avoid ever trying to |
| 1963 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may |
| 1964 | * be using allocators in order of preference for an area that is |
| 1965 | * too large. |
| 1966 | */ |
Mel Gorman | 1fc28b7 | 2009-07-29 15:04:08 -0700 | [diff] [blame] | 1967 | if (order >= MAX_ORDER) { |
| 1968 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); |
Mel Gorman | 72807a7 | 2009-06-16 15:32:18 -0700 | [diff] [blame] | 1969 | return NULL; |
Mel Gorman | 1fc28b7 | 2009-07-29 15:04:08 -0700 | [diff] [blame] | 1970 | } |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1971 | |
Christoph Lameter | 952f3b5 | 2006-12-06 20:33:26 -0800 | [diff] [blame] | 1972 | /* |
| 1973 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and |
| 1974 | * __GFP_NOWARN set) should not cause reclaim since the subsystem |
| 1975 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim |
| 1976 | * using a larger set of nodes after it has established that the |
| 1977 | * allowed per node queues are empty and that nodes are |
| 1978 | * over allocated. |
| 1979 | */ |
| 1980 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) |
| 1981 | goto nopage; |
| 1982 | |
Mel Gorman | cc4a685 | 2009-11-11 14:26:14 -0800 | [diff] [blame] | 1983 | restart: |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1984 | wake_all_kswapd(order, zonelist, high_zoneidx); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1985 | |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 1986 | /* |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1987 | * OK, we're below the kswapd watermark and have kicked background |
| 1988 | * reclaim. Now things get more complex, so set up alloc_flags according |
| 1989 | * to how we want to proceed. |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 1990 | */ |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 1991 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1992 | |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 1993 | /* This is the last chance, in general, before the goto nopage. */ |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1994 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 1995 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, |
| 1996 | preferred_zone, migratetype); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1997 | if (page) |
| 1998 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1999 | |
Kirill Korotaev | b43a57b | 2006-12-06 20:32:27 -0800 | [diff] [blame] | 2000 | rebalance: |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2001 | /* Allocate without watermarks if the context allows */ |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2002 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
| 2003 | page = __alloc_pages_high_priority(gfp_mask, order, |
| 2004 | zonelist, high_zoneidx, nodemask, |
| 2005 | preferred_zone, migratetype); |
| 2006 | if (page) |
| 2007 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2008 | } |
| 2009 | |
| 2010 | /* Atomic allocations - we can't balance anything */ |
| 2011 | if (!wait) |
| 2012 | goto nopage; |
| 2013 | |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2014 | /* Avoid recursion of direct reclaim */ |
| 2015 | if (p->flags & PF_MEMALLOC) |
| 2016 | goto nopage; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2017 | |
David Rientjes | 6583bb6 | 2009-07-29 15:02:06 -0700 | [diff] [blame] | 2018 | /* Avoid allocations with no watermarks from looping endlessly */ |
| 2019 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) |
| 2020 | goto nopage; |
| 2021 | |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 2022 | /* Try direct compaction */ |
| 2023 | page = __alloc_pages_direct_compact(gfp_mask, order, |
| 2024 | zonelist, high_zoneidx, |
| 2025 | nodemask, |
| 2026 | alloc_flags, preferred_zone, |
| 2027 | migratetype, &did_some_progress); |
| 2028 | if (page) |
| 2029 | goto got_pg; |
| 2030 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2031 | /* Try direct reclaim and then allocating */ |
| 2032 | page = __alloc_pages_direct_reclaim(gfp_mask, order, |
| 2033 | zonelist, high_zoneidx, |
| 2034 | nodemask, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2035 | alloc_flags, preferred_zone, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2036 | migratetype, &did_some_progress); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2037 | if (page) |
| 2038 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2039 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2040 | /* |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2041 | * If we failed to make any progress reclaiming, then we are |
| 2042 | * running out of options and have to consider going OOM |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2043 | */ |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2044 | if (!did_some_progress) { |
| 2045 | if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
Rafael J. Wysocki | 7f33d49 | 2009-06-16 15:32:41 -0700 | [diff] [blame] | 2046 | if (oom_killer_disabled) |
| 2047 | goto nopage; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2048 | page = __alloc_pages_may_oom(gfp_mask, order, |
| 2049 | zonelist, high_zoneidx, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2050 | nodemask, preferred_zone, |
| 2051 | migratetype); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2052 | if (page) |
| 2053 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2054 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2055 | /* |
David Rientjes | 82553a9 | 2009-06-16 15:32:58 -0700 | [diff] [blame] | 2056 | * The OOM killer does not trigger for high-order |
| 2057 | * ~__GFP_NOFAIL allocations so if no progress is being |
| 2058 | * made, there are no other options and retrying is |
| 2059 | * unlikely to help. |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2060 | */ |
David Rientjes | 82553a9 | 2009-06-16 15:32:58 -0700 | [diff] [blame] | 2061 | if (order > PAGE_ALLOC_COSTLY_ORDER && |
| 2062 | !(gfp_mask & __GFP_NOFAIL)) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2063 | goto nopage; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2064 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2065 | goto restart; |
| 2066 | } |
| 2067 | } |
| 2068 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2069 | /* Check if we should retry the allocation */ |
Nishanth Aravamudan | a41f24e | 2008-04-29 00:58:25 -0700 | [diff] [blame] | 2070 | pages_reclaimed += did_some_progress; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2071 | if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) { |
| 2072 | /* Wait for some write requests to complete then retry */ |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 2073 | congestion_wait(BLK_RW_ASYNC, HZ/50); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2074 | goto rebalance; |
| 2075 | } |
| 2076 | |
| 2077 | nopage: |
| 2078 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { |
| 2079 | printk(KERN_WARNING "%s: page allocation failure." |
| 2080 | " order:%d, mode:0x%x\n", |
| 2081 | p->comm, order, gfp_mask); |
| 2082 | dump_stack(); |
Janet Morgan | 578c2fd | 2005-06-21 17:14:56 -0700 | [diff] [blame] | 2083 | show_mem(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2084 | } |
Vegard Nossum | b1eeab6 | 2008-11-25 16:55:53 +0100 | [diff] [blame] | 2085 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2086 | got_pg: |
Vegard Nossum | b1eeab6 | 2008-11-25 16:55:53 +0100 | [diff] [blame] | 2087 | if (kmemcheck_enabled) |
| 2088 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2089 | return page; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2090 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2091 | } |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2092 | |
| 2093 | /* |
| 2094 | * This is the 'heart' of the zoned buddy allocator. |
| 2095 | */ |
| 2096 | struct page * |
| 2097 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, |
| 2098 | struct zonelist *zonelist, nodemask_t *nodemask) |
| 2099 | { |
| 2100 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2101 | struct zone *preferred_zone; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2102 | struct page *page; |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2103 | int migratetype = allocflags_to_migratetype(gfp_mask); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2104 | |
Benjamin Herrenschmidt | dcce284 | 2009-06-18 13:24:12 +1000 | [diff] [blame] | 2105 | gfp_mask &= gfp_allowed_mask; |
| 2106 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2107 | lockdep_trace_alloc(gfp_mask); |
| 2108 | |
| 2109 | might_sleep_if(gfp_mask & __GFP_WAIT); |
| 2110 | |
| 2111 | if (should_fail_alloc_page(gfp_mask, order)) |
| 2112 | return NULL; |
| 2113 | |
| 2114 | /* |
| 2115 | * Check the zones suitable for the gfp_mask contain at least one |
| 2116 | * valid zone. It's possible to have an empty zonelist as a result |
| 2117 | * of GFP_THISNODE and a memoryless node |
| 2118 | */ |
| 2119 | if (unlikely(!zonelist->_zonerefs->zone)) |
| 2120 | return NULL; |
| 2121 | |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2122 | get_mems_allowed(); |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2123 | /* The preferred zone is used for statistics later */ |
| 2124 | first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone); |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2125 | if (!preferred_zone) { |
| 2126 | put_mems_allowed(); |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2127 | return NULL; |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2128 | } |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2129 | |
| 2130 | /* First allocation attempt */ |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2131 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2132 | zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2133 | preferred_zone, migratetype); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2134 | if (unlikely(!page)) |
| 2135 | page = __alloc_pages_slowpath(gfp_mask, order, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2136 | zonelist, high_zoneidx, nodemask, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2137 | preferred_zone, migratetype); |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2138 | put_mems_allowed(); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2139 | |
Mel Gorman | 4b4f278 | 2009-09-21 17:02:41 -0700 | [diff] [blame] | 2140 | trace_mm_page_alloc(page, order, gfp_mask, migratetype); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2141 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2142 | } |
Mel Gorman | d239171 | 2009-06-16 15:31:52 -0700 | [diff] [blame] | 2143 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2144 | |
| 2145 | /* |
| 2146 | * Common helper functions. |
| 2147 | */ |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 2148 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2149 | { |
Akinobu Mita | 945a111 | 2009-09-21 17:01:47 -0700 | [diff] [blame] | 2150 | struct page *page; |
| 2151 | |
| 2152 | /* |
| 2153 | * __get_free_pages() returns a 32-bit address, which cannot represent |
| 2154 | * a highmem page |
| 2155 | */ |
| 2156 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
| 2157 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2158 | page = alloc_pages(gfp_mask, order); |
| 2159 | if (!page) |
| 2160 | return 0; |
| 2161 | return (unsigned long) page_address(page); |
| 2162 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2163 | EXPORT_SYMBOL(__get_free_pages); |
| 2164 | |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 2165 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2166 | { |
Akinobu Mita | 945a111 | 2009-09-21 17:01:47 -0700 | [diff] [blame] | 2167 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2168 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2169 | EXPORT_SYMBOL(get_zeroed_page); |
| 2170 | |
| 2171 | void __pagevec_free(struct pagevec *pvec) |
| 2172 | { |
| 2173 | int i = pagevec_count(pvec); |
| 2174 | |
Mel Gorman | 4b4f278 | 2009-09-21 17:02:41 -0700 | [diff] [blame] | 2175 | while (--i >= 0) { |
| 2176 | trace_mm_pagevec_free(pvec->pages[i], pvec->cold); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2177 | free_hot_cold_page(pvec->pages[i], pvec->cold); |
Mel Gorman | 4b4f278 | 2009-09-21 17:02:41 -0700 | [diff] [blame] | 2178 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2179 | } |
| 2180 | |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 2181 | void __free_pages(struct page *page, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2182 | { |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 2183 | if (put_page_testzero(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2184 | if (order == 0) |
Li Hong | fc91668 | 2010-03-05 13:41:54 -0800 | [diff] [blame] | 2185 | free_hot_cold_page(page, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2186 | else |
| 2187 | __free_pages_ok(page, order); |
| 2188 | } |
| 2189 | } |
| 2190 | |
| 2191 | EXPORT_SYMBOL(__free_pages); |
| 2192 | |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 2193 | void free_pages(unsigned long addr, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2194 | { |
| 2195 | if (addr != 0) { |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 2196 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2197 | __free_pages(virt_to_page((void *)addr), order); |
| 2198 | } |
| 2199 | } |
| 2200 | |
| 2201 | EXPORT_SYMBOL(free_pages); |
| 2202 | |
Timur Tabi | 2be0ffe | 2008-07-23 21:28:11 -0700 | [diff] [blame] | 2203 | /** |
| 2204 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. |
| 2205 | * @size: the number of bytes to allocate |
| 2206 | * @gfp_mask: GFP flags for the allocation |
| 2207 | * |
| 2208 | * This function is similar to alloc_pages(), except that it allocates the |
| 2209 | * minimum number of pages to satisfy the request. alloc_pages() can only |
| 2210 | * allocate memory in power-of-two pages. |
| 2211 | * |
| 2212 | * This function is also limited by MAX_ORDER. |
| 2213 | * |
| 2214 | * Memory allocated by this function must be released by free_pages_exact(). |
| 2215 | */ |
| 2216 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) |
| 2217 | { |
| 2218 | unsigned int order = get_order(size); |
| 2219 | unsigned long addr; |
| 2220 | |
| 2221 | addr = __get_free_pages(gfp_mask, order); |
| 2222 | if (addr) { |
| 2223 | unsigned long alloc_end = addr + (PAGE_SIZE << order); |
| 2224 | unsigned long used = addr + PAGE_ALIGN(size); |
| 2225 | |
Kevin Cernekee | 5bfd756 | 2009-07-05 12:08:19 -0700 | [diff] [blame] | 2226 | split_page(virt_to_page((void *)addr), order); |
Timur Tabi | 2be0ffe | 2008-07-23 21:28:11 -0700 | [diff] [blame] | 2227 | while (used < alloc_end) { |
| 2228 | free_page(used); |
| 2229 | used += PAGE_SIZE; |
| 2230 | } |
| 2231 | } |
| 2232 | |
| 2233 | return (void *)addr; |
| 2234 | } |
| 2235 | EXPORT_SYMBOL(alloc_pages_exact); |
| 2236 | |
| 2237 | /** |
| 2238 | * free_pages_exact - release memory allocated via alloc_pages_exact() |
| 2239 | * @virt: the value returned by alloc_pages_exact. |
| 2240 | * @size: size of allocation, same value as passed to alloc_pages_exact(). |
| 2241 | * |
| 2242 | * Release the memory allocated by a previous call to alloc_pages_exact. |
| 2243 | */ |
| 2244 | void free_pages_exact(void *virt, size_t size) |
| 2245 | { |
| 2246 | unsigned long addr = (unsigned long)virt; |
| 2247 | unsigned long end = addr + PAGE_ALIGN(size); |
| 2248 | |
| 2249 | while (addr < end) { |
| 2250 | free_page(addr); |
| 2251 | addr += PAGE_SIZE; |
| 2252 | } |
| 2253 | } |
| 2254 | EXPORT_SYMBOL(free_pages_exact); |
| 2255 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2256 | static unsigned int nr_free_zone_pages(int offset) |
| 2257 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2258 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2259 | struct zone *zone; |
| 2260 | |
Martin J. Bligh | e310fd4 | 2005-07-29 22:59:18 -0700 | [diff] [blame] | 2261 | /* Just pick one node, since fallback list is circular */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2262 | unsigned int sum = 0; |
| 2263 | |
Mel Gorman | 0e88460 | 2008-04-28 02:12:14 -0700 | [diff] [blame] | 2264 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2265 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2266 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
Martin J. Bligh | e310fd4 | 2005-07-29 22:59:18 -0700 | [diff] [blame] | 2267 | unsigned long size = zone->present_pages; |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2268 | unsigned long high = high_wmark_pages(zone); |
Martin J. Bligh | e310fd4 | 2005-07-29 22:59:18 -0700 | [diff] [blame] | 2269 | if (size > high) |
| 2270 | sum += size - high; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2271 | } |
| 2272 | |
| 2273 | return sum; |
| 2274 | } |
| 2275 | |
| 2276 | /* |
| 2277 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL |
| 2278 | */ |
| 2279 | unsigned int nr_free_buffer_pages(void) |
| 2280 | { |
Al Viro | af4ca45 | 2005-10-21 02:55:38 -0400 | [diff] [blame] | 2281 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2282 | } |
Meelap Shah | c2f1a55 | 2007-07-17 04:04:39 -0700 | [diff] [blame] | 2283 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2284 | |
| 2285 | /* |
| 2286 | * Amount of free RAM allocatable within all zones |
| 2287 | */ |
| 2288 | unsigned int nr_free_pagecache_pages(void) |
| 2289 | { |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 2290 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2291 | } |
Christoph Lameter | 08e0f6a | 2006-09-27 01:50:06 -0700 | [diff] [blame] | 2292 | |
| 2293 | static inline void show_node(struct zone *zone) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2294 | { |
Christoph Lameter | 08e0f6a | 2006-09-27 01:50:06 -0700 | [diff] [blame] | 2295 | if (NUMA_BUILD) |
Andy Whitcroft | 25ba77c | 2006-12-06 20:33:03 -0800 | [diff] [blame] | 2296 | printk("Node %d ", zone_to_nid(zone)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2297 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2298 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2299 | void si_meminfo(struct sysinfo *val) |
| 2300 | { |
| 2301 | val->totalram = totalram_pages; |
| 2302 | val->sharedram = 0; |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 2303 | val->freeram = global_page_state(NR_FREE_PAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2304 | val->bufferram = nr_blockdev_pages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2305 | val->totalhigh = totalhigh_pages; |
| 2306 | val->freehigh = nr_free_highpages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2307 | val->mem_unit = PAGE_SIZE; |
| 2308 | } |
| 2309 | |
| 2310 | EXPORT_SYMBOL(si_meminfo); |
| 2311 | |
| 2312 | #ifdef CONFIG_NUMA |
| 2313 | void si_meminfo_node(struct sysinfo *val, int nid) |
| 2314 | { |
| 2315 | pg_data_t *pgdat = NODE_DATA(nid); |
| 2316 | |
| 2317 | val->totalram = pgdat->node_present_pages; |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 2318 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 2319 | #ifdef CONFIG_HIGHMEM |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2320 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 2321 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
| 2322 | NR_FREE_PAGES); |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 2323 | #else |
| 2324 | val->totalhigh = 0; |
| 2325 | val->freehigh = 0; |
| 2326 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2327 | val->mem_unit = PAGE_SIZE; |
| 2328 | } |
| 2329 | #endif |
| 2330 | |
| 2331 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
| 2332 | |
| 2333 | /* |
| 2334 | * Show free area list (used inside shift_scroll-lock stuff) |
| 2335 | * We also calculate the percentage fragmentation. We do this by counting the |
| 2336 | * memory on each free list with the exception of the first item on the list. |
| 2337 | */ |
| 2338 | void show_free_areas(void) |
| 2339 | { |
Jes Sorensen | c724191 | 2006-09-27 01:50:05 -0700 | [diff] [blame] | 2340 | int cpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2341 | struct zone *zone; |
| 2342 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 2343 | for_each_populated_zone(zone) { |
Jes Sorensen | c724191 | 2006-09-27 01:50:05 -0700 | [diff] [blame] | 2344 | show_node(zone); |
| 2345 | printk("%s per-cpu:\n", zone->name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2346 | |
Dave Jones | 6b482c6 | 2005-11-10 15:45:56 -0500 | [diff] [blame] | 2347 | for_each_online_cpu(cpu) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2348 | struct per_cpu_pageset *pageset; |
| 2349 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 2350 | pageset = per_cpu_ptr(zone->pageset, cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2351 | |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 2352 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
| 2353 | cpu, pageset->pcp.high, |
| 2354 | pageset->pcp.batch, pageset->pcp.count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2355 | } |
| 2356 | } |
| 2357 | |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 2358 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
| 2359 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" |
Lee Schermerhorn | 7b85412 | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 2360 | " unevictable:%lu" |
Andrew Morton | b76146e | 2009-10-26 16:49:52 -0700 | [diff] [blame] | 2361 | " dirty:%lu writeback:%lu unstable:%lu\n" |
KOSAKI Motohiro | 3701b03 | 2009-09-21 17:01:29 -0700 | [diff] [blame] | 2362 | " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
KOSAKI Motohiro | 4b02108 | 2009-09-21 17:01:33 -0700 | [diff] [blame] | 2363 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n", |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2364 | global_page_state(NR_ACTIVE_ANON), |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2365 | global_page_state(NR_INACTIVE_ANON), |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 2366 | global_page_state(NR_ISOLATED_ANON), |
| 2367 | global_page_state(NR_ACTIVE_FILE), |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2368 | global_page_state(NR_INACTIVE_FILE), |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 2369 | global_page_state(NR_ISOLATED_FILE), |
Lee Schermerhorn | 7b85412 | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 2370 | global_page_state(NR_UNEVICTABLE), |
Christoph Lameter | b1e7a8f | 2006-06-30 01:55:39 -0700 | [diff] [blame] | 2371 | global_page_state(NR_FILE_DIRTY), |
Christoph Lameter | ce866b3 | 2006-06-30 01:55:40 -0700 | [diff] [blame] | 2372 | global_page_state(NR_WRITEBACK), |
Christoph Lameter | fd39fc85 | 2006-06-30 01:55:40 -0700 | [diff] [blame] | 2373 | global_page_state(NR_UNSTABLE_NFS), |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 2374 | global_page_state(NR_FREE_PAGES), |
KOSAKI Motohiro | 3701b03 | 2009-09-21 17:01:29 -0700 | [diff] [blame] | 2375 | global_page_state(NR_SLAB_RECLAIMABLE), |
| 2376 | global_page_state(NR_SLAB_UNRECLAIMABLE), |
Christoph Lameter | 65ba55f | 2006-06-30 01:55:34 -0700 | [diff] [blame] | 2377 | global_page_state(NR_FILE_MAPPED), |
KOSAKI Motohiro | 4b02108 | 2009-09-21 17:01:33 -0700 | [diff] [blame] | 2378 | global_page_state(NR_SHMEM), |
Andrew Morton | a25700a | 2007-02-08 14:20:40 -0800 | [diff] [blame] | 2379 | global_page_state(NR_PAGETABLE), |
| 2380 | global_page_state(NR_BOUNCE)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2381 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 2382 | for_each_populated_zone(zone) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2383 | int i; |
| 2384 | |
| 2385 | show_node(zone); |
| 2386 | printk("%s" |
| 2387 | " free:%lukB" |
| 2388 | " min:%lukB" |
| 2389 | " low:%lukB" |
| 2390 | " high:%lukB" |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2391 | " active_anon:%lukB" |
| 2392 | " inactive_anon:%lukB" |
| 2393 | " active_file:%lukB" |
| 2394 | " inactive_file:%lukB" |
Lee Schermerhorn | 7b85412 | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 2395 | " unevictable:%lukB" |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 2396 | " isolated(anon):%lukB" |
| 2397 | " isolated(file):%lukB" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2398 | " present:%lukB" |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 2399 | " mlocked:%lukB" |
| 2400 | " dirty:%lukB" |
| 2401 | " writeback:%lukB" |
| 2402 | " mapped:%lukB" |
KOSAKI Motohiro | 4b02108 | 2009-09-21 17:01:33 -0700 | [diff] [blame] | 2403 | " shmem:%lukB" |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 2404 | " slab_reclaimable:%lukB" |
| 2405 | " slab_unreclaimable:%lukB" |
KOSAKI Motohiro | c6a7f57 | 2009-09-21 17:01:32 -0700 | [diff] [blame] | 2406 | " kernel_stack:%lukB" |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 2407 | " pagetables:%lukB" |
| 2408 | " unstable:%lukB" |
| 2409 | " bounce:%lukB" |
| 2410 | " writeback_tmp:%lukB" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2411 | " pages_scanned:%lu" |
| 2412 | " all_unreclaimable? %s" |
| 2413 | "\n", |
| 2414 | zone->name, |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 2415 | K(zone_page_state(zone, NR_FREE_PAGES)), |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2416 | K(min_wmark_pages(zone)), |
| 2417 | K(low_wmark_pages(zone)), |
| 2418 | K(high_wmark_pages(zone)), |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2419 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
| 2420 | K(zone_page_state(zone, NR_INACTIVE_ANON)), |
| 2421 | K(zone_page_state(zone, NR_ACTIVE_FILE)), |
| 2422 | K(zone_page_state(zone, NR_INACTIVE_FILE)), |
Lee Schermerhorn | 7b85412 | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 2423 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 2424 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
| 2425 | K(zone_page_state(zone, NR_ISOLATED_FILE)), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2426 | K(zone->present_pages), |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 2427 | K(zone_page_state(zone, NR_MLOCK)), |
| 2428 | K(zone_page_state(zone, NR_FILE_DIRTY)), |
| 2429 | K(zone_page_state(zone, NR_WRITEBACK)), |
| 2430 | K(zone_page_state(zone, NR_FILE_MAPPED)), |
KOSAKI Motohiro | 4b02108 | 2009-09-21 17:01:33 -0700 | [diff] [blame] | 2431 | K(zone_page_state(zone, NR_SHMEM)), |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 2432 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
| 2433 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), |
KOSAKI Motohiro | c6a7f57 | 2009-09-21 17:01:32 -0700 | [diff] [blame] | 2434 | zone_page_state(zone, NR_KERNEL_STACK) * |
| 2435 | THREAD_SIZE / 1024, |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 2436 | K(zone_page_state(zone, NR_PAGETABLE)), |
| 2437 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), |
| 2438 | K(zone_page_state(zone, NR_BOUNCE)), |
| 2439 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2440 | zone->pages_scanned, |
KOSAKI Motohiro | 93e4a89a | 2010-03-05 13:41:55 -0800 | [diff] [blame] | 2441 | (zone->all_unreclaimable ? "yes" : "no") |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2442 | ); |
| 2443 | printk("lowmem_reserve[]:"); |
| 2444 | for (i = 0; i < MAX_NR_ZONES; i++) |
| 2445 | printk(" %lu", zone->lowmem_reserve[i]); |
| 2446 | printk("\n"); |
| 2447 | } |
| 2448 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 2449 | for_each_populated_zone(zone) { |
Kirill Korotaev | 8f9de51 | 2006-06-23 02:03:50 -0700 | [diff] [blame] | 2450 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2451 | |
| 2452 | show_node(zone); |
| 2453 | printk("%s: ", zone->name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2454 | |
| 2455 | spin_lock_irqsave(&zone->lock, flags); |
| 2456 | for (order = 0; order < MAX_ORDER; order++) { |
Kirill Korotaev | 8f9de51 | 2006-06-23 02:03:50 -0700 | [diff] [blame] | 2457 | nr[order] = zone->free_area[order].nr_free; |
| 2458 | total += nr[order] << order; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2459 | } |
| 2460 | spin_unlock_irqrestore(&zone->lock, flags); |
Kirill Korotaev | 8f9de51 | 2006-06-23 02:03:50 -0700 | [diff] [blame] | 2461 | for (order = 0; order < MAX_ORDER; order++) |
| 2462 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2463 | printk("= %lukB\n", K(total)); |
| 2464 | } |
| 2465 | |
Larry Woodman | e6f3602 | 2008-02-04 22:29:30 -0800 | [diff] [blame] | 2466 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
| 2467 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2468 | show_swap_cache_info(); |
| 2469 | } |
| 2470 | |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 2471 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
| 2472 | { |
| 2473 | zoneref->zone = zone; |
| 2474 | zoneref->zone_idx = zone_idx(zone); |
| 2475 | } |
| 2476 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2477 | /* |
| 2478 | * Builds allocation fallback zone lists. |
Christoph Lameter | 1a93205 | 2006-01-06 00:11:16 -0800 | [diff] [blame] | 2479 | * |
| 2480 | * Add all populated zones of a node to the zonelist. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2481 | */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2482 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
| 2483 | int nr_zones, enum zone_type zone_type) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2484 | { |
Christoph Lameter | 1a93205 | 2006-01-06 00:11:16 -0800 | [diff] [blame] | 2485 | struct zone *zone; |
| 2486 | |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 2487 | BUG_ON(zone_type >= MAX_NR_ZONES); |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 2488 | zone_type++; |
Christoph Lameter | 02a68a5 | 2006-01-06 00:11:18 -0800 | [diff] [blame] | 2489 | |
| 2490 | do { |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 2491 | zone_type--; |
Christoph Lameter | 070f803 | 2006-01-06 00:11:19 -0800 | [diff] [blame] | 2492 | zone = pgdat->node_zones + zone_type; |
Christoph Lameter | 1a93205 | 2006-01-06 00:11:16 -0800 | [diff] [blame] | 2493 | if (populated_zone(zone)) { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2494 | zoneref_set_zone(zone, |
| 2495 | &zonelist->_zonerefs[nr_zones++]); |
Christoph Lameter | 070f803 | 2006-01-06 00:11:19 -0800 | [diff] [blame] | 2496 | check_highest_zone(zone_type); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2497 | } |
Christoph Lameter | 02a68a5 | 2006-01-06 00:11:18 -0800 | [diff] [blame] | 2498 | |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 2499 | } while (zone_type); |
Christoph Lameter | 070f803 | 2006-01-06 00:11:19 -0800 | [diff] [blame] | 2500 | return nr_zones; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2501 | } |
| 2502 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2503 | |
| 2504 | /* |
| 2505 | * zonelist_order: |
| 2506 | * 0 = automatic detection of better ordering. |
| 2507 | * 1 = order by ([node] distance, -zonetype) |
| 2508 | * 2 = order by (-zonetype, [node] distance) |
| 2509 | * |
| 2510 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create |
| 2511 | * the same zonelist. So only NUMA can configure this param. |
| 2512 | */ |
| 2513 | #define ZONELIST_ORDER_DEFAULT 0 |
| 2514 | #define ZONELIST_ORDER_NODE 1 |
| 2515 | #define ZONELIST_ORDER_ZONE 2 |
| 2516 | |
| 2517 | /* zonelist order in the kernel. |
| 2518 | * set_zonelist_order() will set this to NODE or ZONE. |
| 2519 | */ |
| 2520 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; |
| 2521 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; |
| 2522 | |
| 2523 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2524 | #ifdef CONFIG_NUMA |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2525 | /* The value user specified ....changed by config */ |
| 2526 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; |
| 2527 | /* string for sysctl */ |
| 2528 | #define NUMA_ZONELIST_ORDER_LEN 16 |
| 2529 | char numa_zonelist_order[16] = "default"; |
| 2530 | |
| 2531 | /* |
| 2532 | * interface for configure zonelist ordering. |
| 2533 | * command line option "numa_zonelist_order" |
| 2534 | * = "[dD]efault - default, automatic configuration. |
| 2535 | * = "[nN]ode - order by node locality, then by zone within node |
| 2536 | * = "[zZ]one - order by zone, then by locality within zone |
| 2537 | */ |
| 2538 | |
| 2539 | static int __parse_numa_zonelist_order(char *s) |
| 2540 | { |
| 2541 | if (*s == 'd' || *s == 'D') { |
| 2542 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; |
| 2543 | } else if (*s == 'n' || *s == 'N') { |
| 2544 | user_zonelist_order = ZONELIST_ORDER_NODE; |
| 2545 | } else if (*s == 'z' || *s == 'Z') { |
| 2546 | user_zonelist_order = ZONELIST_ORDER_ZONE; |
| 2547 | } else { |
| 2548 | printk(KERN_WARNING |
| 2549 | "Ignoring invalid numa_zonelist_order value: " |
| 2550 | "%s\n", s); |
| 2551 | return -EINVAL; |
| 2552 | } |
| 2553 | return 0; |
| 2554 | } |
| 2555 | |
| 2556 | static __init int setup_numa_zonelist_order(char *s) |
| 2557 | { |
| 2558 | if (s) |
| 2559 | return __parse_numa_zonelist_order(s); |
| 2560 | return 0; |
| 2561 | } |
| 2562 | early_param("numa_zonelist_order", setup_numa_zonelist_order); |
| 2563 | |
| 2564 | /* |
| 2565 | * sysctl handler for numa_zonelist_order |
| 2566 | */ |
| 2567 | int numa_zonelist_order_handler(ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 2568 | void __user *buffer, size_t *length, |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2569 | loff_t *ppos) |
| 2570 | { |
| 2571 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; |
| 2572 | int ret; |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 2573 | static DEFINE_MUTEX(zl_order_mutex); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2574 | |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 2575 | mutex_lock(&zl_order_mutex); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2576 | if (write) |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 2577 | strcpy(saved_string, (char*)table->data); |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 2578 | ret = proc_dostring(table, write, buffer, length, ppos); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2579 | if (ret) |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 2580 | goto out; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2581 | if (write) { |
| 2582 | int oldval = user_zonelist_order; |
| 2583 | if (__parse_numa_zonelist_order((char*)table->data)) { |
| 2584 | /* |
| 2585 | * bogus value. restore saved string |
| 2586 | */ |
| 2587 | strncpy((char*)table->data, saved_string, |
| 2588 | NUMA_ZONELIST_ORDER_LEN); |
| 2589 | user_zonelist_order = oldval; |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 2590 | } else if (oldval != user_zonelist_order) { |
| 2591 | mutex_lock(&zonelists_mutex); |
Haicheng Li | 1f52250 | 2010-05-24 14:32:51 -0700 | [diff] [blame] | 2592 | build_all_zonelists(NULL); |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 2593 | mutex_unlock(&zonelists_mutex); |
| 2594 | } |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2595 | } |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 2596 | out: |
| 2597 | mutex_unlock(&zl_order_mutex); |
| 2598 | return ret; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2599 | } |
| 2600 | |
| 2601 | |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 2602 | #define MAX_NODE_LOAD (nr_online_nodes) |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2603 | static int node_load[MAX_NUMNODES]; |
| 2604 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2605 | /** |
Pavel Pisa | 4dc3b16 | 2005-05-01 08:59:25 -0700 | [diff] [blame] | 2606 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2607 | * @node: node whose fallback list we're appending |
| 2608 | * @used_node_mask: nodemask_t of already used nodes |
| 2609 | * |
| 2610 | * We use a number of factors to determine which is the next node that should |
| 2611 | * appear on a given node's fallback list. The node should not have appeared |
| 2612 | * already in @node's fallback list, and it should be the next closest node |
| 2613 | * according to the distance array (which contains arbitrary distance values |
| 2614 | * from each node to each node in the system), and should also prefer nodes |
| 2615 | * with no CPUs, since presumably they'll have very little allocation pressure |
| 2616 | * on them otherwise. |
| 2617 | * It returns -1 if no node is found. |
| 2618 | */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2619 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2620 | { |
Linus Torvalds | 4cf808eb | 2006-02-17 20:38:21 +0100 | [diff] [blame] | 2621 | int n, val; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2622 | int min_val = INT_MAX; |
| 2623 | int best_node = -1; |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 2624 | const struct cpumask *tmp = cpumask_of_node(0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2625 | |
Linus Torvalds | 4cf808eb | 2006-02-17 20:38:21 +0100 | [diff] [blame] | 2626 | /* Use the local node if we haven't already */ |
| 2627 | if (!node_isset(node, *used_node_mask)) { |
| 2628 | node_set(node, *used_node_mask); |
| 2629 | return node; |
| 2630 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2631 | |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 2632 | for_each_node_state(n, N_HIGH_MEMORY) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2633 | |
| 2634 | /* Don't want a node to appear more than once */ |
| 2635 | if (node_isset(n, *used_node_mask)) |
| 2636 | continue; |
| 2637 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2638 | /* Use the distance array to find the distance */ |
| 2639 | val = node_distance(node, n); |
| 2640 | |
Linus Torvalds | 4cf808eb | 2006-02-17 20:38:21 +0100 | [diff] [blame] | 2641 | /* Penalize nodes under us ("prefer the next node") */ |
| 2642 | val += (n < node); |
| 2643 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2644 | /* Give preference to headless and unused nodes */ |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 2645 | tmp = cpumask_of_node(n); |
| 2646 | if (!cpumask_empty(tmp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2647 | val += PENALTY_FOR_NODE_WITH_CPUS; |
| 2648 | |
| 2649 | /* Slight preference for less loaded node */ |
| 2650 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); |
| 2651 | val += node_load[n]; |
| 2652 | |
| 2653 | if (val < min_val) { |
| 2654 | min_val = val; |
| 2655 | best_node = n; |
| 2656 | } |
| 2657 | } |
| 2658 | |
| 2659 | if (best_node >= 0) |
| 2660 | node_set(best_node, *used_node_mask); |
| 2661 | |
| 2662 | return best_node; |
| 2663 | } |
| 2664 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2665 | |
| 2666 | /* |
| 2667 | * Build zonelists ordered by node and zones within node. |
| 2668 | * This results in maximum locality--normal zone overflows into local |
| 2669 | * DMA zone, if any--but risks exhausting DMA zone. |
| 2670 | */ |
| 2671 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2672 | { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2673 | int j; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2674 | struct zonelist *zonelist; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2675 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2676 | zonelist = &pgdat->node_zonelists[0]; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2677 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2678 | ; |
| 2679 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, |
| 2680 | MAX_NR_ZONES - 1); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2681 | zonelist->_zonerefs[j].zone = NULL; |
| 2682 | zonelist->_zonerefs[j].zone_idx = 0; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2683 | } |
| 2684 | |
| 2685 | /* |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 2686 | * Build gfp_thisnode zonelists |
| 2687 | */ |
| 2688 | static void build_thisnode_zonelists(pg_data_t *pgdat) |
| 2689 | { |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 2690 | int j; |
| 2691 | struct zonelist *zonelist; |
| 2692 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2693 | zonelist = &pgdat->node_zonelists[1]; |
| 2694 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2695 | zonelist->_zonerefs[j].zone = NULL; |
| 2696 | zonelist->_zonerefs[j].zone_idx = 0; |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 2697 | } |
| 2698 | |
| 2699 | /* |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2700 | * Build zonelists ordered by zone and nodes within zones. |
| 2701 | * This results in conserving DMA zone[s] until all Normal memory is |
| 2702 | * exhausted, but results in overflowing to remote node while memory |
| 2703 | * may still exist in local DMA zone. |
| 2704 | */ |
| 2705 | static int node_order[MAX_NUMNODES]; |
| 2706 | |
| 2707 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) |
| 2708 | { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2709 | int pos, j, node; |
| 2710 | int zone_type; /* needs to be signed */ |
| 2711 | struct zone *z; |
| 2712 | struct zonelist *zonelist; |
| 2713 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2714 | zonelist = &pgdat->node_zonelists[0]; |
| 2715 | pos = 0; |
| 2716 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { |
| 2717 | for (j = 0; j < nr_nodes; j++) { |
| 2718 | node = node_order[j]; |
| 2719 | z = &NODE_DATA(node)->node_zones[zone_type]; |
| 2720 | if (populated_zone(z)) { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2721 | zoneref_set_zone(z, |
| 2722 | &zonelist->_zonerefs[pos++]); |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2723 | check_highest_zone(zone_type); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2724 | } |
| 2725 | } |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2726 | } |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2727 | zonelist->_zonerefs[pos].zone = NULL; |
| 2728 | zonelist->_zonerefs[pos].zone_idx = 0; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2729 | } |
| 2730 | |
| 2731 | static int default_zonelist_order(void) |
| 2732 | { |
| 2733 | int nid, zone_type; |
| 2734 | unsigned long low_kmem_size,total_size; |
| 2735 | struct zone *z; |
| 2736 | int average_size; |
| 2737 | /* |
Thomas Weber | 8839316 | 2010-03-16 11:47:56 +0100 | [diff] [blame] | 2738 | * ZONE_DMA and ZONE_DMA32 can be very small area in the system. |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2739 | * If they are really small and used heavily, the system can fall |
| 2740 | * into OOM very easily. |
David Rientjes | e325c90 | 2010-05-24 14:32:13 -0700 | [diff] [blame] | 2741 | * This function detect ZONE_DMA/DMA32 size and configures zone order. |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2742 | */ |
| 2743 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ |
| 2744 | low_kmem_size = 0; |
| 2745 | total_size = 0; |
| 2746 | for_each_online_node(nid) { |
| 2747 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
| 2748 | z = &NODE_DATA(nid)->node_zones[zone_type]; |
| 2749 | if (populated_zone(z)) { |
| 2750 | if (zone_type < ZONE_NORMAL) |
| 2751 | low_kmem_size += z->present_pages; |
| 2752 | total_size += z->present_pages; |
David Rientjes | e325c90 | 2010-05-24 14:32:13 -0700 | [diff] [blame] | 2753 | } else if (zone_type == ZONE_NORMAL) { |
| 2754 | /* |
| 2755 | * If any node has only lowmem, then node order |
| 2756 | * is preferred to allow kernel allocations |
| 2757 | * locally; otherwise, they can easily infringe |
| 2758 | * on other nodes when there is an abundance of |
| 2759 | * lowmem available to allocate from. |
| 2760 | */ |
| 2761 | return ZONELIST_ORDER_NODE; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2762 | } |
| 2763 | } |
| 2764 | } |
| 2765 | if (!low_kmem_size || /* there are no DMA area. */ |
| 2766 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ |
| 2767 | return ZONELIST_ORDER_NODE; |
| 2768 | /* |
| 2769 | * look into each node's config. |
| 2770 | * If there is a node whose DMA/DMA32 memory is very big area on |
| 2771 | * local memory, NODE_ORDER may be suitable. |
| 2772 | */ |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 2773 | average_size = total_size / |
| 2774 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2775 | for_each_online_node(nid) { |
| 2776 | low_kmem_size = 0; |
| 2777 | total_size = 0; |
| 2778 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
| 2779 | z = &NODE_DATA(nid)->node_zones[zone_type]; |
| 2780 | if (populated_zone(z)) { |
| 2781 | if (zone_type < ZONE_NORMAL) |
| 2782 | low_kmem_size += z->present_pages; |
| 2783 | total_size += z->present_pages; |
| 2784 | } |
| 2785 | } |
| 2786 | if (low_kmem_size && |
| 2787 | total_size > average_size && /* ignore small node */ |
| 2788 | low_kmem_size > total_size * 70/100) |
| 2789 | return ZONELIST_ORDER_NODE; |
| 2790 | } |
| 2791 | return ZONELIST_ORDER_ZONE; |
| 2792 | } |
| 2793 | |
| 2794 | static void set_zonelist_order(void) |
| 2795 | { |
| 2796 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) |
| 2797 | current_zonelist_order = default_zonelist_order(); |
| 2798 | else |
| 2799 | current_zonelist_order = user_zonelist_order; |
| 2800 | } |
| 2801 | |
| 2802 | static void build_zonelists(pg_data_t *pgdat) |
| 2803 | { |
| 2804 | int j, node, load; |
| 2805 | enum zone_type i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2806 | nodemask_t used_mask; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2807 | int local_node, prev_node; |
| 2808 | struct zonelist *zonelist; |
| 2809 | int order = current_zonelist_order; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2810 | |
| 2811 | /* initialize zonelists */ |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 2812 | for (i = 0; i < MAX_ZONELISTS; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2813 | zonelist = pgdat->node_zonelists + i; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2814 | zonelist->_zonerefs[0].zone = NULL; |
| 2815 | zonelist->_zonerefs[0].zone_idx = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2816 | } |
| 2817 | |
| 2818 | /* NUMA-aware ordering of nodes */ |
| 2819 | local_node = pgdat->node_id; |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 2820 | load = nr_online_nodes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2821 | prev_node = local_node; |
| 2822 | nodes_clear(used_mask); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2823 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2824 | memset(node_order, 0, sizeof(node_order)); |
| 2825 | j = 0; |
| 2826 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2827 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2828 | int distance = node_distance(local_node, node); |
| 2829 | |
| 2830 | /* |
| 2831 | * If another node is sufficiently far away then it is better |
| 2832 | * to reclaim pages in a zone before going off node. |
| 2833 | */ |
| 2834 | if (distance > RECLAIM_DISTANCE) |
| 2835 | zone_reclaim_mode = 1; |
| 2836 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2837 | /* |
| 2838 | * We don't want to pressure a particular node. |
| 2839 | * So adding penalty to the first node in same |
| 2840 | * distance group to make it round-robin. |
| 2841 | */ |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 2842 | if (distance != node_distance(local_node, prev_node)) |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2843 | node_load[node] = load; |
| 2844 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2845 | prev_node = node; |
| 2846 | load--; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2847 | if (order == ZONELIST_ORDER_NODE) |
| 2848 | build_zonelists_in_node_order(pgdat, node); |
| 2849 | else |
| 2850 | node_order[j++] = node; /* remember order */ |
| 2851 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2852 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2853 | if (order == ZONELIST_ORDER_ZONE) { |
| 2854 | /* calculate node order -- i.e., DMA last! */ |
| 2855 | build_zonelists_in_zone_order(pgdat, j); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2856 | } |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 2857 | |
| 2858 | build_thisnode_zonelists(pgdat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2859 | } |
| 2860 | |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2861 | /* Construct the zonelist performance cache - see further mmzone.h */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2862 | static void build_zonelist_cache(pg_data_t *pgdat) |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2863 | { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2864 | struct zonelist *zonelist; |
| 2865 | struct zonelist_cache *zlc; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2866 | struct zoneref *z; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2867 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2868 | zonelist = &pgdat->node_zonelists[0]; |
| 2869 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; |
| 2870 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2871 | for (z = zonelist->_zonerefs; z->zone; z++) |
| 2872 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2873 | } |
| 2874 | |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 2875 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
| 2876 | /* |
| 2877 | * Return node id of node used for "local" allocations. |
| 2878 | * I.e., first node id of first zone in arg node's generic zonelist. |
| 2879 | * Used for initializing percpu 'numa_mem', which is used primarily |
| 2880 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. |
| 2881 | */ |
| 2882 | int local_memory_node(int node) |
| 2883 | { |
| 2884 | struct zone *zone; |
| 2885 | |
| 2886 | (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
| 2887 | gfp_zone(GFP_KERNEL), |
| 2888 | NULL, |
| 2889 | &zone); |
| 2890 | return zone->node; |
| 2891 | } |
| 2892 | #endif |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2893 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2894 | #else /* CONFIG_NUMA */ |
| 2895 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2896 | static void set_zonelist_order(void) |
| 2897 | { |
| 2898 | current_zonelist_order = ZONELIST_ORDER_ZONE; |
| 2899 | } |
| 2900 | |
| 2901 | static void build_zonelists(pg_data_t *pgdat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2902 | { |
Christoph Lameter | 19655d3 | 2006-09-25 23:31:19 -0700 | [diff] [blame] | 2903 | int node, local_node; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2904 | enum zone_type j; |
| 2905 | struct zonelist *zonelist; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2906 | |
| 2907 | local_node = pgdat->node_id; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2908 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2909 | zonelist = &pgdat->node_zonelists[0]; |
| 2910 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2911 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2912 | /* |
| 2913 | * Now we build the zonelist so that it contains the zones |
| 2914 | * of all the other nodes. |
| 2915 | * We don't want to pressure a particular node, so when |
| 2916 | * building the zones for node N, we make sure that the |
| 2917 | * zones coming right after the local ones are those from |
| 2918 | * node N+1 (modulo N) |
| 2919 | */ |
| 2920 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { |
| 2921 | if (!node_online(node)) |
| 2922 | continue; |
| 2923 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, |
| 2924 | MAX_NR_ZONES - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2925 | } |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2926 | for (node = 0; node < local_node; node++) { |
| 2927 | if (!node_online(node)) |
| 2928 | continue; |
| 2929 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, |
| 2930 | MAX_NR_ZONES - 1); |
| 2931 | } |
| 2932 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2933 | zonelist->_zonerefs[j].zone = NULL; |
| 2934 | zonelist->_zonerefs[j].zone_idx = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2935 | } |
| 2936 | |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2937 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2938 | static void build_zonelist_cache(pg_data_t *pgdat) |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2939 | { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2940 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2941 | } |
| 2942 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2943 | #endif /* CONFIG_NUMA */ |
| 2944 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 2945 | /* |
| 2946 | * Boot pageset table. One per cpu which is going to be used for all |
| 2947 | * zones and all nodes. The parameters will be set in such a way |
| 2948 | * that an item put on a list will immediately be handed over to |
| 2949 | * the buddy list. This is safe since pageset manipulation is done |
| 2950 | * with interrupts disabled. |
| 2951 | * |
| 2952 | * The boot_pagesets must be kept even after bootup is complete for |
| 2953 | * unused processors and/or zones. They do play a role for bootstrapping |
| 2954 | * hotplugged processors. |
| 2955 | * |
| 2956 | * zoneinfo_show() and maybe other functions do |
| 2957 | * not check if the processor is online before following the pageset pointer. |
| 2958 | * Other parts of the kernel may not check if the zone is available. |
| 2959 | */ |
| 2960 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); |
| 2961 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); |
Haicheng Li | 1f52250 | 2010-05-24 14:32:51 -0700 | [diff] [blame] | 2962 | static void setup_zone_pageset(struct zone *zone); |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 2963 | |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 2964 | /* |
| 2965 | * Global mutex to protect against size modification of zonelists |
| 2966 | * as well as to serialize pageset setup for the new populated zone. |
| 2967 | */ |
| 2968 | DEFINE_MUTEX(zonelists_mutex); |
| 2969 | |
Rusty Russell | 9b1a4d3 | 2008-07-28 12:16:30 -0500 | [diff] [blame] | 2970 | /* return values int ....just for stop_machine() */ |
Haicheng Li | 1f52250 | 2010-05-24 14:32:51 -0700 | [diff] [blame] | 2971 | static __init_refok int __build_all_zonelists(void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2972 | { |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 2973 | int nid; |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 2974 | int cpu; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2975 | |
Bo Liu | 7f9cfb3 | 2009-08-18 14:11:19 -0700 | [diff] [blame] | 2976 | #ifdef CONFIG_NUMA |
| 2977 | memset(node_load, 0, sizeof(node_load)); |
| 2978 | #endif |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2979 | for_each_online_node(nid) { |
Christoph Lameter | 7ea1530 | 2007-10-16 01:25:29 -0700 | [diff] [blame] | 2980 | pg_data_t *pgdat = NODE_DATA(nid); |
| 2981 | |
| 2982 | build_zonelists(pgdat); |
| 2983 | build_zonelist_cache(pgdat); |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2984 | } |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 2985 | |
Haicheng Li | 1f52250 | 2010-05-24 14:32:51 -0700 | [diff] [blame] | 2986 | #ifdef CONFIG_MEMORY_HOTPLUG |
| 2987 | /* Setup real pagesets for the new zone */ |
| 2988 | if (data) { |
| 2989 | struct zone *zone = data; |
| 2990 | setup_zone_pageset(zone); |
| 2991 | } |
| 2992 | #endif |
| 2993 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 2994 | /* |
| 2995 | * Initialize the boot_pagesets that are going to be used |
| 2996 | * for bootstrapping processors. The real pagesets for |
| 2997 | * each zone will be allocated later when the per cpu |
| 2998 | * allocator is available. |
| 2999 | * |
| 3000 | * boot_pagesets are used also for bootstrapping offline |
| 3001 | * cpus if the system is already booted because the pagesets |
| 3002 | * are needed to initialize allocators on a specific cpu too. |
| 3003 | * F.e. the percpu allocator needs the page allocator which |
| 3004 | * needs the percpu allocator in order to allocate its pagesets |
| 3005 | * (a chicken-egg dilemma). |
| 3006 | */ |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 3007 | for_each_possible_cpu(cpu) { |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3008 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
| 3009 | |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 3010 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
| 3011 | /* |
| 3012 | * We now know the "local memory node" for each node-- |
| 3013 | * i.e., the node of the first zone in the generic zonelist. |
| 3014 | * Set up numa_mem percpu variable for on-line cpus. During |
| 3015 | * boot, only the boot cpu should be on-line; we'll init the |
| 3016 | * secondary cpus' numa_mem as they come on-line. During |
| 3017 | * node/memory hotplug, we'll fixup all on-line cpus. |
| 3018 | */ |
| 3019 | if (cpu_online(cpu)) |
| 3020 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
| 3021 | #endif |
| 3022 | } |
| 3023 | |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 3024 | return 0; |
| 3025 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3026 | |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 3027 | /* |
| 3028 | * Called with zonelists_mutex held always |
| 3029 | * unless system_state == SYSTEM_BOOTING. |
| 3030 | */ |
Haicheng Li | 1f52250 | 2010-05-24 14:32:51 -0700 | [diff] [blame] | 3031 | void build_all_zonelists(void *data) |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 3032 | { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3033 | set_zonelist_order(); |
| 3034 | |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 3035 | if (system_state == SYSTEM_BOOTING) { |
Randy Dunlap | 423b41d | 2006-09-27 01:50:12 -0700 | [diff] [blame] | 3036 | __build_all_zonelists(NULL); |
Mel Gorman | 68ad8df | 2008-07-23 21:26:52 -0700 | [diff] [blame] | 3037 | mminit_verify_zonelist(); |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 3038 | cpuset_init_current_mems_allowed(); |
| 3039 | } else { |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 3040 | /* we have to stop all cpus to guarantee there is no user |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 3041 | of zonelist */ |
Haicheng Li | 1f52250 | 2010-05-24 14:32:51 -0700 | [diff] [blame] | 3042 | stop_machine(__build_all_zonelists, data, NULL); |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 3043 | /* cpuset refresh routine should be here */ |
| 3044 | } |
Andrew Morton | bd1e22b | 2006-06-23 02:03:47 -0700 | [diff] [blame] | 3045 | vm_total_pages = nr_free_pagecache_pages(); |
Mel Gorman | 9ef9acb | 2007-10-16 01:25:54 -0700 | [diff] [blame] | 3046 | /* |
| 3047 | * Disable grouping by mobility if the number of pages in the |
| 3048 | * system is too low to allow the mechanism to work. It would be |
| 3049 | * more accurate, but expensive to check per-zone. This check is |
| 3050 | * made on memory-hotadd so a system can start with mobility |
| 3051 | * disabled and enable it later |
| 3052 | */ |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 3053 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
Mel Gorman | 9ef9acb | 2007-10-16 01:25:54 -0700 | [diff] [blame] | 3054 | page_group_by_mobility_disabled = 1; |
| 3055 | else |
| 3056 | page_group_by_mobility_disabled = 0; |
| 3057 | |
| 3058 | printk("Built %i zonelists in %s order, mobility grouping %s. " |
| 3059 | "Total pages: %ld\n", |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 3060 | nr_online_nodes, |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3061 | zonelist_order_name[current_zonelist_order], |
Mel Gorman | 9ef9acb | 2007-10-16 01:25:54 -0700 | [diff] [blame] | 3062 | page_group_by_mobility_disabled ? "off" : "on", |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3063 | vm_total_pages); |
| 3064 | #ifdef CONFIG_NUMA |
| 3065 | printk("Policy zone: %s\n", zone_names[policy_zone]); |
| 3066 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3067 | } |
| 3068 | |
| 3069 | /* |
| 3070 | * Helper functions to size the waitqueue hash table. |
| 3071 | * Essentially these want to choose hash table sizes sufficiently |
| 3072 | * large so that collisions trying to wait on pages are rare. |
| 3073 | * But in fact, the number of active page waitqueues on typical |
| 3074 | * systems is ridiculously low, less than 200. So this is even |
| 3075 | * conservative, even though it seems large. |
| 3076 | * |
| 3077 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to |
| 3078 | * waitqueues, i.e. the size of the waitq table given the number of pages. |
| 3079 | */ |
| 3080 | #define PAGES_PER_WAITQUEUE 256 |
| 3081 | |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3082 | #ifndef CONFIG_MEMORY_HOTPLUG |
Yasunori Goto | 02b694d | 2006-06-23 02:03:08 -0700 | [diff] [blame] | 3083 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3084 | { |
| 3085 | unsigned long size = 1; |
| 3086 | |
| 3087 | pages /= PAGES_PER_WAITQUEUE; |
| 3088 | |
| 3089 | while (size < pages) |
| 3090 | size <<= 1; |
| 3091 | |
| 3092 | /* |
| 3093 | * Once we have dozens or even hundreds of threads sleeping |
| 3094 | * on IO we've got bigger problems than wait queue collision. |
| 3095 | * Limit the size of the wait table to a reasonable size. |
| 3096 | */ |
| 3097 | size = min(size, 4096UL); |
| 3098 | |
| 3099 | return max(size, 4UL); |
| 3100 | } |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3101 | #else |
| 3102 | /* |
| 3103 | * A zone's size might be changed by hot-add, so it is not possible to determine |
| 3104 | * a suitable size for its wait_table. So we use the maximum size now. |
| 3105 | * |
| 3106 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: |
| 3107 | * |
| 3108 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. |
| 3109 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. |
| 3110 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. |
| 3111 | * |
| 3112 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages |
| 3113 | * or more by the traditional way. (See above). It equals: |
| 3114 | * |
| 3115 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. |
| 3116 | * ia64(16K page size) : = ( 8G + 4M)byte. |
| 3117 | * powerpc (64K page size) : = (32G +16M)byte. |
| 3118 | */ |
| 3119 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
| 3120 | { |
| 3121 | return 4096UL; |
| 3122 | } |
| 3123 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3124 | |
| 3125 | /* |
| 3126 | * This is an integer logarithm so that shifts can be used later |
| 3127 | * to extract the more random high bits from the multiplicative |
| 3128 | * hash function before the remainder is taken. |
| 3129 | */ |
| 3130 | static inline unsigned long wait_table_bits(unsigned long size) |
| 3131 | { |
| 3132 | return ffz(~size); |
| 3133 | } |
| 3134 | |
| 3135 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) |
| 3136 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3137 | /* |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 3138 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 3139 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
| 3140 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3141 | * higher will lead to a bigger reserve which will get freed as contiguous |
| 3142 | * blocks as reclaim kicks in |
| 3143 | */ |
| 3144 | static void setup_zone_migrate_reserve(struct zone *zone) |
| 3145 | { |
| 3146 | unsigned long start_pfn, pfn, end_pfn; |
| 3147 | struct page *page; |
Mel Gorman | 78986a6 | 2009-09-21 17:03:02 -0700 | [diff] [blame] | 3148 | unsigned long block_migratetype; |
| 3149 | int reserve; |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3150 | |
| 3151 | /* Get the start pfn, end pfn and the number of blocks to reserve */ |
| 3152 | start_pfn = zone->zone_start_pfn; |
| 3153 | end_pfn = start_pfn + zone->spanned_pages; |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 3154 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 3155 | pageblock_order; |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3156 | |
Mel Gorman | 78986a6 | 2009-09-21 17:03:02 -0700 | [diff] [blame] | 3157 | /* |
| 3158 | * Reserve blocks are generally in place to help high-order atomic |
| 3159 | * allocations that are short-lived. A min_free_kbytes value that |
| 3160 | * would result in more than 2 reserve blocks for atomic allocations |
| 3161 | * is assumed to be in place to help anti-fragmentation for the |
| 3162 | * future allocation of hugepages at runtime. |
| 3163 | */ |
| 3164 | reserve = min(2, reserve); |
| 3165 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 3166 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3167 | if (!pfn_valid(pfn)) |
| 3168 | continue; |
| 3169 | page = pfn_to_page(pfn); |
| 3170 | |
Adam Litke | 344c790 | 2008-09-02 14:35:38 -0700 | [diff] [blame] | 3171 | /* Watch out for overlapping nodes */ |
| 3172 | if (page_to_nid(page) != zone_to_nid(zone)) |
| 3173 | continue; |
| 3174 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3175 | /* Blocks with reserved pages will never free, skip them. */ |
| 3176 | if (PageReserved(page)) |
| 3177 | continue; |
| 3178 | |
| 3179 | block_migratetype = get_pageblock_migratetype(page); |
| 3180 | |
| 3181 | /* If this block is reserved, account for it */ |
| 3182 | if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) { |
| 3183 | reserve--; |
| 3184 | continue; |
| 3185 | } |
| 3186 | |
| 3187 | /* Suitable for reserving if this block is movable */ |
| 3188 | if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) { |
| 3189 | set_pageblock_migratetype(page, MIGRATE_RESERVE); |
| 3190 | move_freepages_block(zone, page, MIGRATE_RESERVE); |
| 3191 | reserve--; |
| 3192 | continue; |
| 3193 | } |
| 3194 | |
| 3195 | /* |
| 3196 | * If the reserve is met and this is a previous reserved block, |
| 3197 | * take it back |
| 3198 | */ |
| 3199 | if (block_migratetype == MIGRATE_RESERVE) { |
| 3200 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
| 3201 | move_freepages_block(zone, page, MIGRATE_MOVABLE); |
| 3202 | } |
| 3203 | } |
| 3204 | } |
Mel Gorman | ac0e5b7 | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3205 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3206 | /* |
| 3207 | * Initially all pages are reserved - free ones are freed |
| 3208 | * up by free_all_bootmem() once the early boot process is |
| 3209 | * done. Non-atomic initialization, single-pass. |
| 3210 | */ |
Matt Tolentino | c09b424 | 2006-01-17 07:03:44 +0100 | [diff] [blame] | 3211 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 3212 | unsigned long start_pfn, enum memmap_context context) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3213 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3214 | struct page *page; |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 3215 | unsigned long end_pfn = start_pfn + size; |
| 3216 | unsigned long pfn; |
KAMEZAWA Hiroyuki | 86051ca | 2008-04-29 00:58:21 -0700 | [diff] [blame] | 3217 | struct zone *z; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3218 | |
Hugh Dickins | 22b31ee | 2009-01-06 14:40:09 -0800 | [diff] [blame] | 3219 | if (highest_memmap_pfn < end_pfn - 1) |
| 3220 | highest_memmap_pfn = end_pfn - 1; |
| 3221 | |
KAMEZAWA Hiroyuki | 86051ca | 2008-04-29 00:58:21 -0700 | [diff] [blame] | 3222 | z = &NODE_DATA(nid)->node_zones[zone]; |
Greg Ungerer | cbe8dd4 | 2006-01-12 01:05:24 -0800 | [diff] [blame] | 3223 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 3224 | /* |
| 3225 | * There can be holes in boot-time mem_map[]s |
| 3226 | * handed to this function. They do not |
| 3227 | * exist on hotplugged memory. |
| 3228 | */ |
| 3229 | if (context == MEMMAP_EARLY) { |
| 3230 | if (!early_pfn_valid(pfn)) |
| 3231 | continue; |
| 3232 | if (!early_pfn_in_nid(pfn, nid)) |
| 3233 | continue; |
| 3234 | } |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 3235 | page = pfn_to_page(pfn); |
| 3236 | set_page_links(page, zone, nid, pfn); |
Mel Gorman | 708614e | 2008-07-23 21:26:51 -0700 | [diff] [blame] | 3237 | mminit_verify_page_links(page, zone, nid, pfn); |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 3238 | init_page_count(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3239 | reset_page_mapcount(page); |
| 3240 | SetPageReserved(page); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 3241 | /* |
| 3242 | * Mark the block movable so that blocks are reserved for |
| 3243 | * movable at startup. This will force kernel allocations |
| 3244 | * to reserve their blocks rather than leaking throughout |
| 3245 | * the address space during boot when many long-lived |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3246 | * kernel allocations are made. Later some blocks near |
| 3247 | * the start are marked MIGRATE_RESERVE by |
| 3248 | * setup_zone_migrate_reserve() |
KAMEZAWA Hiroyuki | 86051ca | 2008-04-29 00:58:21 -0700 | [diff] [blame] | 3249 | * |
| 3250 | * bitmap is created for zone's valid pfn range. but memmap |
| 3251 | * can be created for invalid pages (for alignment) |
| 3252 | * check here not to call set_pageblock_migratetype() against |
| 3253 | * pfn out of zone. |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 3254 | */ |
KAMEZAWA Hiroyuki | 86051ca | 2008-04-29 00:58:21 -0700 | [diff] [blame] | 3255 | if ((z->zone_start_pfn <= pfn) |
| 3256 | && (pfn < z->zone_start_pfn + z->spanned_pages) |
| 3257 | && !(pfn & (pageblock_nr_pages - 1))) |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3258 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 3259 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3260 | INIT_LIST_HEAD(&page->lru); |
| 3261 | #ifdef WANT_PAGE_VIRTUAL |
| 3262 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ |
| 3263 | if (!is_highmem_idx(zone)) |
Bob Picco | 3212c6b | 2005-06-27 14:36:28 -0700 | [diff] [blame] | 3264 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3265 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3266 | } |
| 3267 | } |
| 3268 | |
Andi Kleen | 1e548de | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 3269 | static void __meminit zone_init_free_lists(struct zone *zone) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3270 | { |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 3271 | int order, t; |
| 3272 | for_each_migratetype_order(order, t) { |
| 3273 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3274 | zone->free_area[order].nr_free = 0; |
| 3275 | } |
| 3276 | } |
| 3277 | |
| 3278 | #ifndef __HAVE_ARCH_MEMMAP_INIT |
| 3279 | #define memmap_init(size, nid, zone, start_pfn) \ |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 3280 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3281 | #endif |
| 3282 | |
Sam Ravnborg | 1d6f4e6 | 2008-01-17 15:21:12 -0800 | [diff] [blame] | 3283 | static int zone_batchsize(struct zone *zone) |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3284 | { |
David Howells | 3a6be87 | 2009-05-06 16:03:03 -0700 | [diff] [blame] | 3285 | #ifdef CONFIG_MMU |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3286 | int batch; |
| 3287 | |
| 3288 | /* |
| 3289 | * The per-cpu-pages pools are set to around 1000th of the |
Seth, Rohit | ba56e91 | 2005-10-29 18:15:47 -0700 | [diff] [blame] | 3290 | * size of the zone. But no more than 1/2 of a meg. |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3291 | * |
| 3292 | * OK, so we don't know how big the cache is. So guess. |
| 3293 | */ |
| 3294 | batch = zone->present_pages / 1024; |
Seth, Rohit | ba56e91 | 2005-10-29 18:15:47 -0700 | [diff] [blame] | 3295 | if (batch * PAGE_SIZE > 512 * 1024) |
| 3296 | batch = (512 * 1024) / PAGE_SIZE; |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3297 | batch /= 4; /* We effectively *= 4 below */ |
| 3298 | if (batch < 1) |
| 3299 | batch = 1; |
| 3300 | |
| 3301 | /* |
Nick Piggin | 0ceaacc | 2005-12-04 13:55:25 +1100 | [diff] [blame] | 3302 | * Clamp the batch to a 2^n - 1 value. Having a power |
| 3303 | * of 2 value was found to be more likely to have |
| 3304 | * suboptimal cache aliasing properties in some cases. |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3305 | * |
Nick Piggin | 0ceaacc | 2005-12-04 13:55:25 +1100 | [diff] [blame] | 3306 | * For example if 2 tasks are alternately allocating |
| 3307 | * batches of pages, one task can end up with a lot |
| 3308 | * of pages of one half of the possible page colors |
| 3309 | * and the other with pages of the other colors. |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3310 | */ |
David Howells | 9155203 | 2009-05-06 16:03:02 -0700 | [diff] [blame] | 3311 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
Seth, Rohit | ba56e91 | 2005-10-29 18:15:47 -0700 | [diff] [blame] | 3312 | |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3313 | return batch; |
David Howells | 3a6be87 | 2009-05-06 16:03:03 -0700 | [diff] [blame] | 3314 | |
| 3315 | #else |
| 3316 | /* The deferral and batching of frees should be suppressed under NOMMU |
| 3317 | * conditions. |
| 3318 | * |
| 3319 | * The problem is that NOMMU needs to be able to allocate large chunks |
| 3320 | * of contiguous memory as there's no hardware page translation to |
| 3321 | * assemble apparent contiguous memory from discontiguous pages. |
| 3322 | * |
| 3323 | * Queueing large contiguous runs of pages for batching, however, |
| 3324 | * causes the pages to actually be freed in smaller chunks. As there |
| 3325 | * can be a significant delay between the individual batches being |
| 3326 | * recycled, this leads to the once large chunks of space being |
| 3327 | * fragmented and becoming unavailable for high-order allocations. |
| 3328 | */ |
| 3329 | return 0; |
| 3330 | #endif |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3331 | } |
| 3332 | |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 3333 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 3334 | { |
| 3335 | struct per_cpu_pages *pcp; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 3336 | int migratetype; |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 3337 | |
Magnus Damm | 1c6fe94 | 2005-10-26 01:58:59 -0700 | [diff] [blame] | 3338 | memset(p, 0, sizeof(*p)); |
| 3339 | |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 3340 | pcp = &p->pcp; |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 3341 | pcp->count = 0; |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 3342 | pcp->high = 6 * batch; |
| 3343 | pcp->batch = max(1UL, 1 * batch); |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 3344 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
| 3345 | INIT_LIST_HEAD(&pcp->lists[migratetype]); |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 3346 | } |
| 3347 | |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 3348 | /* |
| 3349 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist |
| 3350 | * to the value high for the pageset p. |
| 3351 | */ |
| 3352 | |
| 3353 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, |
| 3354 | unsigned long high) |
| 3355 | { |
| 3356 | struct per_cpu_pages *pcp; |
| 3357 | |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 3358 | pcp = &p->pcp; |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 3359 | pcp->high = high; |
| 3360 | pcp->batch = max(1UL, high/4); |
| 3361 | if ((high/4) > (PAGE_SHIFT * 8)) |
| 3362 | pcp->batch = PAGE_SHIFT * 8; |
| 3363 | } |
| 3364 | |
Wu Fengguang | 319774e | 2010-05-24 14:32:49 -0700 | [diff] [blame] | 3365 | static __meminit void setup_zone_pageset(struct zone *zone) |
| 3366 | { |
| 3367 | int cpu; |
| 3368 | |
| 3369 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
| 3370 | |
| 3371 | for_each_possible_cpu(cpu) { |
| 3372 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); |
| 3373 | |
| 3374 | setup_pageset(pcp, zone_batchsize(zone)); |
| 3375 | |
| 3376 | if (percpu_pagelist_fraction) |
| 3377 | setup_pagelist_highmark(pcp, |
| 3378 | (zone->present_pages / |
| 3379 | percpu_pagelist_fraction)); |
| 3380 | } |
| 3381 | } |
| 3382 | |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3383 | /* |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3384 | * Allocate per cpu pagesets and initialize them. |
| 3385 | * Before this call only boot pagesets were available. |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 3386 | */ |
Al Viro | 78d9955 | 2005-12-15 09:18:25 +0000 | [diff] [blame] | 3387 | void __init setup_per_cpu_pageset(void) |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3388 | { |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3389 | struct zone *zone; |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3390 | |
Wu Fengguang | 319774e | 2010-05-24 14:32:49 -0700 | [diff] [blame] | 3391 | for_each_populated_zone(zone) |
| 3392 | setup_zone_pageset(zone); |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3393 | } |
| 3394 | |
Sam Ravnborg | 577a32f | 2007-05-17 23:29:25 +0200 | [diff] [blame] | 3395 | static noinline __init_refok |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3396 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3397 | { |
| 3398 | int i; |
| 3399 | struct pglist_data *pgdat = zone->zone_pgdat; |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3400 | size_t alloc_size; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3401 | |
| 3402 | /* |
| 3403 | * The per-page waitqueue mechanism uses hashed waitqueues |
| 3404 | * per zone. |
| 3405 | */ |
Yasunori Goto | 02b694d | 2006-06-23 02:03:08 -0700 | [diff] [blame] | 3406 | zone->wait_table_hash_nr_entries = |
| 3407 | wait_table_hash_nr_entries(zone_size_pages); |
| 3408 | zone->wait_table_bits = |
| 3409 | wait_table_bits(zone->wait_table_hash_nr_entries); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3410 | alloc_size = zone->wait_table_hash_nr_entries |
| 3411 | * sizeof(wait_queue_head_t); |
| 3412 | |
Heiko Carstens | cd94b9d | 2008-05-23 13:04:52 -0700 | [diff] [blame] | 3413 | if (!slab_is_available()) { |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3414 | zone->wait_table = (wait_queue_head_t *) |
| 3415 | alloc_bootmem_node(pgdat, alloc_size); |
| 3416 | } else { |
| 3417 | /* |
| 3418 | * This case means that a zone whose size was 0 gets new memory |
| 3419 | * via memory hot-add. |
| 3420 | * But it may be the case that a new node was hot-added. In |
| 3421 | * this case vmalloc() will not be able to use this new node's |
| 3422 | * memory - this wait_table must be initialized to use this new |
| 3423 | * node itself as well. |
| 3424 | * To use this new node's memory, further consideration will be |
| 3425 | * necessary. |
| 3426 | */ |
Jesper Juhl | 8691f3a | 2007-10-16 01:24:49 -0700 | [diff] [blame] | 3427 | zone->wait_table = vmalloc(alloc_size); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3428 | } |
| 3429 | if (!zone->wait_table) |
| 3430 | return -ENOMEM; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3431 | |
Yasunori Goto | 02b694d | 2006-06-23 02:03:08 -0700 | [diff] [blame] | 3432 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3433 | init_waitqueue_head(zone->wait_table + i); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3434 | |
| 3435 | return 0; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3436 | } |
| 3437 | |
Shaohua Li | 112067f | 2009-09-21 17:01:16 -0700 | [diff] [blame] | 3438 | static int __zone_pcp_update(void *data) |
| 3439 | { |
| 3440 | struct zone *zone = data; |
| 3441 | int cpu; |
| 3442 | unsigned long batch = zone_batchsize(zone), flags; |
| 3443 | |
Thomas Gleixner | 2d30a1f | 2010-03-10 15:20:40 -0800 | [diff] [blame] | 3444 | for_each_possible_cpu(cpu) { |
Shaohua Li | 112067f | 2009-09-21 17:01:16 -0700 | [diff] [blame] | 3445 | struct per_cpu_pageset *pset; |
| 3446 | struct per_cpu_pages *pcp; |
| 3447 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3448 | pset = per_cpu_ptr(zone->pageset, cpu); |
Shaohua Li | 112067f | 2009-09-21 17:01:16 -0700 | [diff] [blame] | 3449 | pcp = &pset->pcp; |
| 3450 | |
| 3451 | local_irq_save(flags); |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 3452 | free_pcppages_bulk(zone, pcp->count, pcp); |
Shaohua Li | 112067f | 2009-09-21 17:01:16 -0700 | [diff] [blame] | 3453 | setup_pageset(pset, batch); |
| 3454 | local_irq_restore(flags); |
| 3455 | } |
| 3456 | return 0; |
| 3457 | } |
| 3458 | |
| 3459 | void zone_pcp_update(struct zone *zone) |
| 3460 | { |
| 3461 | stop_machine(__zone_pcp_update, zone, NULL); |
| 3462 | } |
| 3463 | |
Matt Tolentino | c09b424 | 2006-01-17 07:03:44 +0100 | [diff] [blame] | 3464 | static __meminit void zone_pcp_init(struct zone *zone) |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3465 | { |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3466 | /* |
| 3467 | * per cpu subsystem is not up at this point. The following code |
| 3468 | * relies on the ability of the linker to provide the |
| 3469 | * offset of a (static) per cpu variable into the per cpu area. |
| 3470 | */ |
| 3471 | zone->pageset = &boot_pageset; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3472 | |
Anton Blanchard | f5335c0 | 2006-03-25 03:06:49 -0800 | [diff] [blame] | 3473 | if (zone->present_pages) |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3474 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
| 3475 | zone->name, zone->present_pages, |
| 3476 | zone_batchsize(zone)); |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3477 | } |
| 3478 | |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3479 | __meminit int init_currently_empty_zone(struct zone *zone, |
| 3480 | unsigned long zone_start_pfn, |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 3481 | unsigned long size, |
| 3482 | enum memmap_context context) |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3483 | { |
| 3484 | struct pglist_data *pgdat = zone->zone_pgdat; |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3485 | int ret; |
| 3486 | ret = zone_wait_table_init(zone, size); |
| 3487 | if (ret) |
| 3488 | return ret; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3489 | pgdat->nr_zones = zone_idx(zone) + 1; |
| 3490 | |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3491 | zone->zone_start_pfn = zone_start_pfn; |
| 3492 | |
Mel Gorman | 708614e | 2008-07-23 21:26:51 -0700 | [diff] [blame] | 3493 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
| 3494 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", |
| 3495 | pgdat->node_id, |
| 3496 | (unsigned long)zone_idx(zone), |
| 3497 | zone_start_pfn, (zone_start_pfn + size)); |
| 3498 | |
Andi Kleen | 1e548de | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 3499 | zone_init_free_lists(zone); |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3500 | |
| 3501 | return 0; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3502 | } |
| 3503 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3504 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
| 3505 | /* |
| 3506 | * Basic iterator support. Return the first range of PFNs for a node |
| 3507 | * Note: nid == MAX_NUMNODES returns first region regardless of node |
| 3508 | */ |
Yasunori Goto | a3142c8 | 2007-05-08 00:23:07 -0700 | [diff] [blame] | 3509 | static int __meminit first_active_region_index_in_nid(int nid) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3510 | { |
| 3511 | int i; |
| 3512 | |
| 3513 | for (i = 0; i < nr_nodemap_entries; i++) |
| 3514 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) |
| 3515 | return i; |
| 3516 | |
| 3517 | return -1; |
| 3518 | } |
| 3519 | |
| 3520 | /* |
| 3521 | * Basic iterator support. Return the next active range of PFNs for a node |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 3522 | * Note: nid == MAX_NUMNODES returns next region regardless of node |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3523 | */ |
Yasunori Goto | a3142c8 | 2007-05-08 00:23:07 -0700 | [diff] [blame] | 3524 | static int __meminit next_active_region_index_in_nid(int index, int nid) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3525 | { |
| 3526 | for (index = index + 1; index < nr_nodemap_entries; index++) |
| 3527 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) |
| 3528 | return index; |
| 3529 | |
| 3530 | return -1; |
| 3531 | } |
| 3532 | |
| 3533 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
| 3534 | /* |
| 3535 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. |
| 3536 | * Architectures may implement their own version but if add_active_range() |
| 3537 | * was used and there are no special requirements, this is a convenient |
| 3538 | * alternative |
| 3539 | */ |
KAMEZAWA Hiroyuki | f2dbcfa | 2009-02-18 14:48:32 -0800 | [diff] [blame] | 3540 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3541 | { |
| 3542 | int i; |
| 3543 | |
| 3544 | for (i = 0; i < nr_nodemap_entries; i++) { |
| 3545 | unsigned long start_pfn = early_node_map[i].start_pfn; |
| 3546 | unsigned long end_pfn = early_node_map[i].end_pfn; |
| 3547 | |
| 3548 | if (start_pfn <= pfn && pfn < end_pfn) |
| 3549 | return early_node_map[i].nid; |
| 3550 | } |
KAMEZAWA Hiroyuki | cc2559b | 2009-02-18 14:48:33 -0800 | [diff] [blame] | 3551 | /* This is a memory hole */ |
| 3552 | return -1; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3553 | } |
| 3554 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ |
| 3555 | |
KAMEZAWA Hiroyuki | f2dbcfa | 2009-02-18 14:48:32 -0800 | [diff] [blame] | 3556 | int __meminit early_pfn_to_nid(unsigned long pfn) |
| 3557 | { |
KAMEZAWA Hiroyuki | cc2559b | 2009-02-18 14:48:33 -0800 | [diff] [blame] | 3558 | int nid; |
| 3559 | |
| 3560 | nid = __early_pfn_to_nid(pfn); |
| 3561 | if (nid >= 0) |
| 3562 | return nid; |
| 3563 | /* just returns 0 */ |
| 3564 | return 0; |
KAMEZAWA Hiroyuki | f2dbcfa | 2009-02-18 14:48:32 -0800 | [diff] [blame] | 3565 | } |
| 3566 | |
KAMEZAWA Hiroyuki | cc2559b | 2009-02-18 14:48:33 -0800 | [diff] [blame] | 3567 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
| 3568 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) |
| 3569 | { |
| 3570 | int nid; |
| 3571 | |
| 3572 | nid = __early_pfn_to_nid(pfn); |
| 3573 | if (nid >= 0 && nid != node) |
| 3574 | return false; |
| 3575 | return true; |
| 3576 | } |
| 3577 | #endif |
KAMEZAWA Hiroyuki | f2dbcfa | 2009-02-18 14:48:32 -0800 | [diff] [blame] | 3578 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3579 | /* Basic iterator support to walk early_node_map[] */ |
| 3580 | #define for_each_active_range_index_in_nid(i, nid) \ |
| 3581 | for (i = first_active_region_index_in_nid(nid); i != -1; \ |
| 3582 | i = next_active_region_index_in_nid(i, nid)) |
| 3583 | |
| 3584 | /** |
| 3585 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 3586 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
| 3587 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3588 | * |
| 3589 | * If an architecture guarantees that all ranges registered with |
| 3590 | * add_active_ranges() contain no holes and may be freed, this |
| 3591 | * this function may be used instead of calling free_bootmem() manually. |
| 3592 | */ |
| 3593 | void __init free_bootmem_with_active_regions(int nid, |
| 3594 | unsigned long max_low_pfn) |
| 3595 | { |
| 3596 | int i; |
| 3597 | |
| 3598 | for_each_active_range_index_in_nid(i, nid) { |
| 3599 | unsigned long size_pages = 0; |
| 3600 | unsigned long end_pfn = early_node_map[i].end_pfn; |
| 3601 | |
| 3602 | if (early_node_map[i].start_pfn >= max_low_pfn) |
| 3603 | continue; |
| 3604 | |
| 3605 | if (end_pfn > max_low_pfn) |
| 3606 | end_pfn = max_low_pfn; |
| 3607 | |
| 3608 | size_pages = end_pfn - early_node_map[i].start_pfn; |
| 3609 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), |
| 3610 | PFN_PHYS(early_node_map[i].start_pfn), |
| 3611 | size_pages << PAGE_SHIFT); |
| 3612 | } |
| 3613 | } |
| 3614 | |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 3615 | int __init add_from_early_node_map(struct range *range, int az, |
| 3616 | int nr_range, int nid) |
| 3617 | { |
| 3618 | int i; |
| 3619 | u64 start, end; |
| 3620 | |
| 3621 | /* need to go over early_node_map to find out good range for node */ |
| 3622 | for_each_active_range_index_in_nid(i, nid) { |
| 3623 | start = early_node_map[i].start_pfn; |
| 3624 | end = early_node_map[i].end_pfn; |
| 3625 | nr_range = add_range(range, az, nr_range, start, end); |
| 3626 | } |
| 3627 | return nr_range; |
| 3628 | } |
| 3629 | |
Yinghai Lu | 2ee78f7 | 2010-02-17 11:29:49 -0800 | [diff] [blame] | 3630 | #ifdef CONFIG_NO_BOOTMEM |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 3631 | void * __init __alloc_memory_core_early(int nid, u64 size, u64 align, |
| 3632 | u64 goal, u64 limit) |
| 3633 | { |
| 3634 | int i; |
| 3635 | void *ptr; |
| 3636 | |
| 3637 | /* need to go over early_node_map to find out good range for node */ |
| 3638 | for_each_active_range_index_in_nid(i, nid) { |
| 3639 | u64 addr; |
| 3640 | u64 ei_start, ei_last; |
| 3641 | |
| 3642 | ei_last = early_node_map[i].end_pfn; |
| 3643 | ei_last <<= PAGE_SHIFT; |
| 3644 | ei_start = early_node_map[i].start_pfn; |
| 3645 | ei_start <<= PAGE_SHIFT; |
| 3646 | addr = find_early_area(ei_start, ei_last, |
| 3647 | goal, limit, size, align); |
| 3648 | |
| 3649 | if (addr == -1ULL) |
| 3650 | continue; |
| 3651 | |
| 3652 | #if 0 |
| 3653 | printk(KERN_DEBUG "alloc (nid=%d %llx - %llx) (%llx - %llx) %llx %llx => %llx\n", |
| 3654 | nid, |
| 3655 | ei_start, ei_last, goal, limit, size, |
| 3656 | align, addr); |
| 3657 | #endif |
| 3658 | |
| 3659 | ptr = phys_to_virt(addr); |
| 3660 | memset(ptr, 0, size); |
| 3661 | reserve_early_without_check(addr, addr + size, "BOOTMEM"); |
Catalin Marinas | 9078370 | 2010-07-19 11:54:15 +0100 | [diff] [blame] | 3662 | /* |
| 3663 | * The min_count is set to 0 so that bootmem allocated blocks |
| 3664 | * are never reported as leaks. |
| 3665 | */ |
| 3666 | kmemleak_alloc(ptr, size, 0, 0); |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 3667 | return ptr; |
| 3668 | } |
| 3669 | |
| 3670 | return NULL; |
| 3671 | } |
Yinghai Lu | 2ee78f7 | 2010-02-17 11:29:49 -0800 | [diff] [blame] | 3672 | #endif |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 3673 | |
| 3674 | |
Yinghai Lu | b5bc6c0 | 2008-06-14 18:32:52 -0700 | [diff] [blame] | 3675 | void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data) |
| 3676 | { |
| 3677 | int i; |
Yinghai Lu | d52d53b | 2008-06-16 20:10:55 -0700 | [diff] [blame] | 3678 | int ret; |
Yinghai Lu | b5bc6c0 | 2008-06-14 18:32:52 -0700 | [diff] [blame] | 3679 | |
Yinghai Lu | d52d53b | 2008-06-16 20:10:55 -0700 | [diff] [blame] | 3680 | for_each_active_range_index_in_nid(i, nid) { |
| 3681 | ret = work_fn(early_node_map[i].start_pfn, |
| 3682 | early_node_map[i].end_pfn, data); |
| 3683 | if (ret) |
| 3684 | break; |
| 3685 | } |
Yinghai Lu | b5bc6c0 | 2008-06-14 18:32:52 -0700 | [diff] [blame] | 3686 | } |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3687 | /** |
| 3688 | * sparse_memory_present_with_active_regions - Call memory_present for each active range |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 3689 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3690 | * |
| 3691 | * If an architecture guarantees that all ranges registered with |
| 3692 | * add_active_ranges() contain no holes and may be freed, this |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 3693 | * function may be used instead of calling memory_present() manually. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3694 | */ |
| 3695 | void __init sparse_memory_present_with_active_regions(int nid) |
| 3696 | { |
| 3697 | int i; |
| 3698 | |
| 3699 | for_each_active_range_index_in_nid(i, nid) |
| 3700 | memory_present(early_node_map[i].nid, |
| 3701 | early_node_map[i].start_pfn, |
| 3702 | early_node_map[i].end_pfn); |
| 3703 | } |
| 3704 | |
| 3705 | /** |
| 3706 | * get_pfn_range_for_nid - Return the start and end page frames for a node |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 3707 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
| 3708 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. |
| 3709 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3710 | * |
| 3711 | * It returns the start and end page frame of a node based on information |
| 3712 | * provided by an arch calling add_active_range(). If called for a node |
| 3713 | * with no available memory, a warning is printed and the start and end |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 3714 | * PFNs will be 0. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3715 | */ |
Yasunori Goto | a3142c8 | 2007-05-08 00:23:07 -0700 | [diff] [blame] | 3716 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3717 | unsigned long *start_pfn, unsigned long *end_pfn) |
| 3718 | { |
| 3719 | int i; |
| 3720 | *start_pfn = -1UL; |
| 3721 | *end_pfn = 0; |
| 3722 | |
| 3723 | for_each_active_range_index_in_nid(i, nid) { |
| 3724 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); |
| 3725 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); |
| 3726 | } |
| 3727 | |
Christoph Lameter | 633c066 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 3728 | if (*start_pfn == -1UL) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3729 | *start_pfn = 0; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3730 | } |
| 3731 | |
| 3732 | /* |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 3733 | * This finds a zone that can be used for ZONE_MOVABLE pages. The |
| 3734 | * assumption is made that zones within a node are ordered in monotonic |
| 3735 | * increasing memory addresses so that the "highest" populated zone is used |
| 3736 | */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 3737 | static void __init find_usable_zone_for_movable(void) |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 3738 | { |
| 3739 | int zone_index; |
| 3740 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { |
| 3741 | if (zone_index == ZONE_MOVABLE) |
| 3742 | continue; |
| 3743 | |
| 3744 | if (arch_zone_highest_possible_pfn[zone_index] > |
| 3745 | arch_zone_lowest_possible_pfn[zone_index]) |
| 3746 | break; |
| 3747 | } |
| 3748 | |
| 3749 | VM_BUG_ON(zone_index == -1); |
| 3750 | movable_zone = zone_index; |
| 3751 | } |
| 3752 | |
| 3753 | /* |
| 3754 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE |
| 3755 | * because it is sized independant of architecture. Unlike the other zones, |
| 3756 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
| 3757 | * in each node depending on the size of each node and how evenly kernelcore |
| 3758 | * is distributed. This helper function adjusts the zone ranges |
| 3759 | * provided by the architecture for a given node by using the end of the |
| 3760 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that |
| 3761 | * zones within a node are in order of monotonic increases memory addresses |
| 3762 | */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 3763 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 3764 | unsigned long zone_type, |
| 3765 | unsigned long node_start_pfn, |
| 3766 | unsigned long node_end_pfn, |
| 3767 | unsigned long *zone_start_pfn, |
| 3768 | unsigned long *zone_end_pfn) |
| 3769 | { |
| 3770 | /* Only adjust if ZONE_MOVABLE is on this node */ |
| 3771 | if (zone_movable_pfn[nid]) { |
| 3772 | /* Size ZONE_MOVABLE */ |
| 3773 | if (zone_type == ZONE_MOVABLE) { |
| 3774 | *zone_start_pfn = zone_movable_pfn[nid]; |
| 3775 | *zone_end_pfn = min(node_end_pfn, |
| 3776 | arch_zone_highest_possible_pfn[movable_zone]); |
| 3777 | |
| 3778 | /* Adjust for ZONE_MOVABLE starting within this range */ |
| 3779 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && |
| 3780 | *zone_end_pfn > zone_movable_pfn[nid]) { |
| 3781 | *zone_end_pfn = zone_movable_pfn[nid]; |
| 3782 | |
| 3783 | /* Check if this whole range is within ZONE_MOVABLE */ |
| 3784 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) |
| 3785 | *zone_start_pfn = *zone_end_pfn; |
| 3786 | } |
| 3787 | } |
| 3788 | |
| 3789 | /* |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3790 | * Return the number of pages a zone spans in a node, including holes |
| 3791 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() |
| 3792 | */ |
Paul Mundt | 6ea6e68 | 2007-07-15 23:38:20 -0700 | [diff] [blame] | 3793 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3794 | unsigned long zone_type, |
| 3795 | unsigned long *ignored) |
| 3796 | { |
| 3797 | unsigned long node_start_pfn, node_end_pfn; |
| 3798 | unsigned long zone_start_pfn, zone_end_pfn; |
| 3799 | |
| 3800 | /* Get the start and end of the node and zone */ |
| 3801 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); |
| 3802 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
| 3803 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 3804 | adjust_zone_range_for_zone_movable(nid, zone_type, |
| 3805 | node_start_pfn, node_end_pfn, |
| 3806 | &zone_start_pfn, &zone_end_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3807 | |
| 3808 | /* Check that this node has pages within the zone's required range */ |
| 3809 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) |
| 3810 | return 0; |
| 3811 | |
| 3812 | /* Move the zone boundaries inside the node if necessary */ |
| 3813 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); |
| 3814 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); |
| 3815 | |
| 3816 | /* Return the spanned pages */ |
| 3817 | return zone_end_pfn - zone_start_pfn; |
| 3818 | } |
| 3819 | |
| 3820 | /* |
| 3821 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 3822 | * then all holes in the requested range will be accounted for. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3823 | */ |
Yinghai Lu | 3299625 | 2009-12-15 17:59:02 -0800 | [diff] [blame] | 3824 | unsigned long __meminit __absent_pages_in_range(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3825 | unsigned long range_start_pfn, |
| 3826 | unsigned long range_end_pfn) |
| 3827 | { |
| 3828 | int i = 0; |
| 3829 | unsigned long prev_end_pfn = 0, hole_pages = 0; |
| 3830 | unsigned long start_pfn; |
| 3831 | |
| 3832 | /* Find the end_pfn of the first active range of pfns in the node */ |
| 3833 | i = first_active_region_index_in_nid(nid); |
| 3834 | if (i == -1) |
| 3835 | return 0; |
| 3836 | |
Mel Gorman | b5445f9 | 2007-07-26 10:41:18 -0700 | [diff] [blame] | 3837 | prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn); |
| 3838 | |
Mel Gorman | 9c7cd68 | 2006-09-27 01:49:58 -0700 | [diff] [blame] | 3839 | /* Account for ranges before physical memory on this node */ |
| 3840 | if (early_node_map[i].start_pfn > range_start_pfn) |
Mel Gorman | b5445f9 | 2007-07-26 10:41:18 -0700 | [diff] [blame] | 3841 | hole_pages = prev_end_pfn - range_start_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3842 | |
| 3843 | /* Find all holes for the zone within the node */ |
| 3844 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { |
| 3845 | |
| 3846 | /* No need to continue if prev_end_pfn is outside the zone */ |
| 3847 | if (prev_end_pfn >= range_end_pfn) |
| 3848 | break; |
| 3849 | |
| 3850 | /* Make sure the end of the zone is not within the hole */ |
| 3851 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); |
| 3852 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); |
| 3853 | |
| 3854 | /* Update the hole size cound and move on */ |
| 3855 | if (start_pfn > range_start_pfn) { |
| 3856 | BUG_ON(prev_end_pfn > start_pfn); |
| 3857 | hole_pages += start_pfn - prev_end_pfn; |
| 3858 | } |
| 3859 | prev_end_pfn = early_node_map[i].end_pfn; |
| 3860 | } |
| 3861 | |
Mel Gorman | 9c7cd68 | 2006-09-27 01:49:58 -0700 | [diff] [blame] | 3862 | /* Account for ranges past physical memory on this node */ |
| 3863 | if (range_end_pfn > prev_end_pfn) |
Mel Gorman | 0c6cb97 | 2006-10-28 10:38:59 -0700 | [diff] [blame] | 3864 | hole_pages += range_end_pfn - |
Mel Gorman | 9c7cd68 | 2006-09-27 01:49:58 -0700 | [diff] [blame] | 3865 | max(range_start_pfn, prev_end_pfn); |
| 3866 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3867 | return hole_pages; |
| 3868 | } |
| 3869 | |
| 3870 | /** |
| 3871 | * absent_pages_in_range - Return number of page frames in holes within a range |
| 3872 | * @start_pfn: The start PFN to start searching for holes |
| 3873 | * @end_pfn: The end PFN to stop searching for holes |
| 3874 | * |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 3875 | * It returns the number of pages frames in memory holes within a range. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3876 | */ |
| 3877 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, |
| 3878 | unsigned long end_pfn) |
| 3879 | { |
| 3880 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); |
| 3881 | } |
| 3882 | |
| 3883 | /* Return the number of page frames in holes in a zone on a node */ |
Paul Mundt | 6ea6e68 | 2007-07-15 23:38:20 -0700 | [diff] [blame] | 3884 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3885 | unsigned long zone_type, |
| 3886 | unsigned long *ignored) |
| 3887 | { |
Mel Gorman | 9c7cd68 | 2006-09-27 01:49:58 -0700 | [diff] [blame] | 3888 | unsigned long node_start_pfn, node_end_pfn; |
| 3889 | unsigned long zone_start_pfn, zone_end_pfn; |
| 3890 | |
| 3891 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); |
| 3892 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], |
| 3893 | node_start_pfn); |
| 3894 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], |
| 3895 | node_end_pfn); |
| 3896 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 3897 | adjust_zone_range_for_zone_movable(nid, zone_type, |
| 3898 | node_start_pfn, node_end_pfn, |
| 3899 | &zone_start_pfn, &zone_end_pfn); |
Mel Gorman | 9c7cd68 | 2006-09-27 01:49:58 -0700 | [diff] [blame] | 3900 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3901 | } |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 3902 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3903 | #else |
Paul Mundt | 6ea6e68 | 2007-07-15 23:38:20 -0700 | [diff] [blame] | 3904 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3905 | unsigned long zone_type, |
| 3906 | unsigned long *zones_size) |
| 3907 | { |
| 3908 | return zones_size[zone_type]; |
| 3909 | } |
| 3910 | |
Paul Mundt | 6ea6e68 | 2007-07-15 23:38:20 -0700 | [diff] [blame] | 3911 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3912 | unsigned long zone_type, |
| 3913 | unsigned long *zholes_size) |
| 3914 | { |
| 3915 | if (!zholes_size) |
| 3916 | return 0; |
| 3917 | |
| 3918 | return zholes_size[zone_type]; |
| 3919 | } |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 3920 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3921 | #endif |
| 3922 | |
Yasunori Goto | a3142c8 | 2007-05-08 00:23:07 -0700 | [diff] [blame] | 3923 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3924 | unsigned long *zones_size, unsigned long *zholes_size) |
| 3925 | { |
| 3926 | unsigned long realtotalpages, totalpages = 0; |
| 3927 | enum zone_type i; |
| 3928 | |
| 3929 | for (i = 0; i < MAX_NR_ZONES; i++) |
| 3930 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, |
| 3931 | zones_size); |
| 3932 | pgdat->node_spanned_pages = totalpages; |
| 3933 | |
| 3934 | realtotalpages = totalpages; |
| 3935 | for (i = 0; i < MAX_NR_ZONES; i++) |
| 3936 | realtotalpages -= |
| 3937 | zone_absent_pages_in_node(pgdat->node_id, i, |
| 3938 | zholes_size); |
| 3939 | pgdat->node_present_pages = realtotalpages; |
| 3940 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, |
| 3941 | realtotalpages); |
| 3942 | } |
| 3943 | |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 3944 | #ifndef CONFIG_SPARSEMEM |
| 3945 | /* |
| 3946 | * Calculate the size of the zone->blockflags rounded to an unsigned long |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 3947 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
| 3948 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 3949 | * round what is now in bits to nearest long in bits, then return it in |
| 3950 | * bytes. |
| 3951 | */ |
| 3952 | static unsigned long __init usemap_size(unsigned long zonesize) |
| 3953 | { |
| 3954 | unsigned long usemapsize; |
| 3955 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 3956 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
| 3957 | usemapsize = usemapsize >> pageblock_order; |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 3958 | usemapsize *= NR_PAGEBLOCK_BITS; |
| 3959 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); |
| 3960 | |
| 3961 | return usemapsize / 8; |
| 3962 | } |
| 3963 | |
| 3964 | static void __init setup_usemap(struct pglist_data *pgdat, |
| 3965 | struct zone *zone, unsigned long zonesize) |
| 3966 | { |
| 3967 | unsigned long usemapsize = usemap_size(zonesize); |
| 3968 | zone->pageblock_flags = NULL; |
Julia Lawall | 58a01a4 | 2009-01-06 14:39:28 -0800 | [diff] [blame] | 3969 | if (usemapsize) |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 3970 | zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize); |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 3971 | } |
| 3972 | #else |
| 3973 | static void inline setup_usemap(struct pglist_data *pgdat, |
| 3974 | struct zone *zone, unsigned long zonesize) {} |
| 3975 | #endif /* CONFIG_SPARSEMEM */ |
| 3976 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 3977 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 3978 | |
| 3979 | /* Return a sensible default order for the pageblock size. */ |
| 3980 | static inline int pageblock_default_order(void) |
| 3981 | { |
| 3982 | if (HPAGE_SHIFT > PAGE_SHIFT) |
| 3983 | return HUGETLB_PAGE_ORDER; |
| 3984 | |
| 3985 | return MAX_ORDER-1; |
| 3986 | } |
| 3987 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 3988 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
| 3989 | static inline void __init set_pageblock_order(unsigned int order) |
| 3990 | { |
| 3991 | /* Check that pageblock_nr_pages has not already been setup */ |
| 3992 | if (pageblock_order) |
| 3993 | return; |
| 3994 | |
| 3995 | /* |
| 3996 | * Assume the largest contiguous order of interest is a huge page. |
| 3997 | * This value may be variable depending on boot parameters on IA64 |
| 3998 | */ |
| 3999 | pageblock_order = order; |
| 4000 | } |
| 4001 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ |
| 4002 | |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 4003 | /* |
| 4004 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() |
| 4005 | * and pageblock_default_order() are unused as pageblock_order is set |
| 4006 | * at compile-time. See include/linux/pageblock-flags.h for the values of |
| 4007 | * pageblock_order based on the kernel config |
| 4008 | */ |
| 4009 | static inline int pageblock_default_order(unsigned int order) |
| 4010 | { |
| 4011 | return MAX_ORDER-1; |
| 4012 | } |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 4013 | #define set_pageblock_order(x) do {} while (0) |
| 4014 | |
| 4015 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ |
| 4016 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4017 | /* |
| 4018 | * Set up the zone data structures: |
| 4019 | * - mark all pages reserved |
| 4020 | * - mark all memory queues empty |
| 4021 | * - clear the memory bitmaps |
| 4022 | */ |
Alexander van Heukelum | b5a0e01 | 2008-02-23 15:24:06 -0800 | [diff] [blame] | 4023 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4024 | unsigned long *zones_size, unsigned long *zholes_size) |
| 4025 | { |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 4026 | enum zone_type j; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 4027 | int nid = pgdat->node_id; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4028 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 4029 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4030 | |
Dave Hansen | 208d54e | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 4031 | pgdat_resize_init(pgdat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4032 | pgdat->nr_zones = 0; |
| 4033 | init_waitqueue_head(&pgdat->kswapd_wait); |
| 4034 | pgdat->kswapd_max_order = 0; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 4035 | pgdat_page_cgroup_init(pgdat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4036 | |
| 4037 | for (j = 0; j < MAX_NR_ZONES; j++) { |
| 4038 | struct zone *zone = pgdat->node_zones + j; |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4039 | unsigned long size, realsize, memmap_pages; |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 4040 | enum lru_list l; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4041 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4042 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
| 4043 | realsize = size - zone_absent_pages_in_node(nid, j, |
| 4044 | zholes_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4045 | |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4046 | /* |
| 4047 | * Adjust realsize so that it accounts for how much memory |
| 4048 | * is used by this zone for memmap. This affects the watermark |
| 4049 | * and per-cpu initialisations |
| 4050 | */ |
Johannes Weiner | f723215 | 2008-05-23 13:04:21 -0700 | [diff] [blame] | 4051 | memmap_pages = |
| 4052 | PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT; |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4053 | if (realsize >= memmap_pages) { |
| 4054 | realsize -= memmap_pages; |
Yinghai Lu | 5594c8c | 2009-01-06 14:39:14 -0800 | [diff] [blame] | 4055 | if (memmap_pages) |
| 4056 | printk(KERN_DEBUG |
| 4057 | " %s zone: %lu pages used for memmap\n", |
| 4058 | zone_names[j], memmap_pages); |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4059 | } else |
| 4060 | printk(KERN_WARNING |
| 4061 | " %s zone: %lu pages exceeds realsize %lu\n", |
| 4062 | zone_names[j], memmap_pages, realsize); |
| 4063 | |
Christoph Lameter | 6267276 | 2007-02-10 01:43:07 -0800 | [diff] [blame] | 4064 | /* Account for reserved pages */ |
| 4065 | if (j == 0 && realsize > dma_reserve) { |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4066 | realsize -= dma_reserve; |
Yinghai Lu | d903ef9 | 2008-10-18 20:27:06 -0700 | [diff] [blame] | 4067 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
Christoph Lameter | 6267276 | 2007-02-10 01:43:07 -0800 | [diff] [blame] | 4068 | zone_names[0], dma_reserve); |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4069 | } |
| 4070 | |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 4071 | if (!is_highmem_idx(j)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4072 | nr_kernel_pages += realsize; |
| 4073 | nr_all_pages += realsize; |
| 4074 | |
| 4075 | zone->spanned_pages = size; |
| 4076 | zone->present_pages = realsize; |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 4077 | #ifdef CONFIG_NUMA |
Christoph Lameter | d5f541e | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 4078 | zone->node = nid; |
Christoph Lameter | 8417bba | 2006-09-25 23:31:51 -0700 | [diff] [blame] | 4079 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 4080 | / 100; |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 4081 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 4082 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4083 | zone->name = zone_names[j]; |
| 4084 | spin_lock_init(&zone->lock); |
| 4085 | spin_lock_init(&zone->lru_lock); |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 4086 | zone_seqlock_init(zone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4087 | zone->zone_pgdat = pgdat; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4088 | |
Martin Bligh | 3bb1a852 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 4089 | zone->prev_priority = DEF_PRIORITY; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4090 | |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 4091 | zone_pcp_init(zone); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 4092 | for_each_lru(l) { |
| 4093 | INIT_LIST_HEAD(&zone->lru[l].list); |
Wu Fengguang | f862963 | 2009-09-21 17:03:11 -0700 | [diff] [blame] | 4094 | zone->reclaim_stat.nr_saved_scan[l] = 0; |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 4095 | } |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 4096 | zone->reclaim_stat.recent_rotated[0] = 0; |
| 4097 | zone->reclaim_stat.recent_rotated[1] = 0; |
| 4098 | zone->reclaim_stat.recent_scanned[0] = 0; |
| 4099 | zone->reclaim_stat.recent_scanned[1] = 0; |
Christoph Lameter | 2244b95 | 2006-06-30 01:55:33 -0700 | [diff] [blame] | 4100 | zap_zone_vm_stats(zone); |
David Rientjes | e815af9 | 2007-10-16 23:25:54 -0700 | [diff] [blame] | 4101 | zone->flags = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4102 | if (!size) |
| 4103 | continue; |
| 4104 | |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 4105 | set_pageblock_order(pageblock_default_order()); |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 4106 | setup_usemap(pgdat, zone, size); |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 4107 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
| 4108 | size, MEMMAP_EARLY); |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 4109 | BUG_ON(ret); |
Heiko Carstens | 76cdd58 | 2008-05-14 16:05:52 -0700 | [diff] [blame] | 4110 | memmap_init(size, nid, j, zone_start_pfn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4111 | zone_start_pfn += size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4112 | } |
| 4113 | } |
| 4114 | |
Sam Ravnborg | 577a32f | 2007-05-17 23:29:25 +0200 | [diff] [blame] | 4115 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4116 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4117 | /* Skip empty nodes */ |
| 4118 | if (!pgdat->node_spanned_pages) |
| 4119 | return; |
| 4120 | |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 4121 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4122 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
| 4123 | if (!pgdat->node_mem_map) { |
Bob Picco | e984bb4 | 2006-05-20 15:00:31 -0700 | [diff] [blame] | 4124 | unsigned long size, start, end; |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 4125 | struct page *map; |
| 4126 | |
Bob Picco | e984bb4 | 2006-05-20 15:00:31 -0700 | [diff] [blame] | 4127 | /* |
| 4128 | * The zone's endpoints aren't required to be MAX_ORDER |
| 4129 | * aligned but the node_mem_map endpoints must be in order |
| 4130 | * for the buddy allocator to function correctly. |
| 4131 | */ |
| 4132 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
| 4133 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; |
| 4134 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
| 4135 | size = (end - start) * sizeof(struct page); |
Dave Hansen | 6f167ec | 2005-06-23 00:07:39 -0700 | [diff] [blame] | 4136 | map = alloc_remap(pgdat->node_id, size); |
| 4137 | if (!map) |
| 4138 | map = alloc_bootmem_node(pgdat, size); |
Bob Picco | e984bb4 | 2006-05-20 15:00:31 -0700 | [diff] [blame] | 4139 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4140 | } |
Roman Zippel | 12d810c | 2007-05-31 00:40:54 -0700 | [diff] [blame] | 4141 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4142 | /* |
| 4143 | * With no DISCONTIG, the global mem_map is just set as node 0's |
| 4144 | */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4145 | if (pgdat == NODE_DATA(0)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4146 | mem_map = NODE_DATA(0)->node_mem_map; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4147 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
| 4148 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
Thomas Bogendoerfer | 467bc46 | 2008-01-08 15:33:11 -0800 | [diff] [blame] | 4149 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4150 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
| 4151 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4152 | #endif |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 4153 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4154 | } |
| 4155 | |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 4156 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
| 4157 | unsigned long node_start_pfn, unsigned long *zholes_size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4158 | { |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 4159 | pg_data_t *pgdat = NODE_DATA(nid); |
| 4160 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4161 | pgdat->node_id = nid; |
| 4162 | pgdat->node_start_pfn = node_start_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4163 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4164 | |
| 4165 | alloc_node_mem_map(pgdat); |
Yinghai Lu | e8c27ac | 2008-06-01 13:15:22 -0700 | [diff] [blame] | 4166 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
| 4167 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", |
| 4168 | nid, (unsigned long)pgdat, |
| 4169 | (unsigned long)pgdat->node_mem_map); |
| 4170 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4171 | |
| 4172 | free_area_init_core(pgdat, zones_size, zholes_size); |
| 4173 | } |
| 4174 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4175 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 4176 | |
| 4177 | #if MAX_NUMNODES > 1 |
| 4178 | /* |
| 4179 | * Figure out the number of possible node ids. |
| 4180 | */ |
| 4181 | static void __init setup_nr_node_ids(void) |
| 4182 | { |
| 4183 | unsigned int node; |
| 4184 | unsigned int highest = 0; |
| 4185 | |
| 4186 | for_each_node_mask(node, node_possible_map) |
| 4187 | highest = node; |
| 4188 | nr_node_ids = highest + 1; |
| 4189 | } |
| 4190 | #else |
| 4191 | static inline void setup_nr_node_ids(void) |
| 4192 | { |
| 4193 | } |
| 4194 | #endif |
| 4195 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4196 | /** |
| 4197 | * add_active_range - Register a range of PFNs backed by physical memory |
| 4198 | * @nid: The node ID the range resides on |
| 4199 | * @start_pfn: The start PFN of the available physical memory |
| 4200 | * @end_pfn: The end PFN of the available physical memory |
| 4201 | * |
| 4202 | * These ranges are stored in an early_node_map[] and later used by |
| 4203 | * free_area_init_nodes() to calculate zone sizes and holes. If the |
| 4204 | * range spans a memory hole, it is up to the architecture to ensure |
| 4205 | * the memory is not freed by the bootmem allocator. If possible |
| 4206 | * the range being registered will be merged with existing ranges. |
| 4207 | */ |
| 4208 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, |
| 4209 | unsigned long end_pfn) |
| 4210 | { |
| 4211 | int i; |
| 4212 | |
Mel Gorman | 6b74ab9 | 2008-07-23 21:26:49 -0700 | [diff] [blame] | 4213 | mminit_dprintk(MMINIT_TRACE, "memory_register", |
| 4214 | "Entering add_active_range(%d, %#lx, %#lx) " |
| 4215 | "%d entries of %d used\n", |
| 4216 | nid, start_pfn, end_pfn, |
| 4217 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4218 | |
Mel Gorman | 2dbb51c | 2008-07-23 21:26:52 -0700 | [diff] [blame] | 4219 | mminit_validate_memmodel_limits(&start_pfn, &end_pfn); |
| 4220 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4221 | /* Merge with existing active regions if possible */ |
| 4222 | for (i = 0; i < nr_nodemap_entries; i++) { |
| 4223 | if (early_node_map[i].nid != nid) |
| 4224 | continue; |
| 4225 | |
| 4226 | /* Skip if an existing region covers this new one */ |
| 4227 | if (start_pfn >= early_node_map[i].start_pfn && |
| 4228 | end_pfn <= early_node_map[i].end_pfn) |
| 4229 | return; |
| 4230 | |
| 4231 | /* Merge forward if suitable */ |
| 4232 | if (start_pfn <= early_node_map[i].end_pfn && |
| 4233 | end_pfn > early_node_map[i].end_pfn) { |
| 4234 | early_node_map[i].end_pfn = end_pfn; |
| 4235 | return; |
| 4236 | } |
| 4237 | |
| 4238 | /* Merge backward if suitable */ |
Kazuhisa Ichikawa | d2dbe08 | 2010-01-15 17:01:20 -0800 | [diff] [blame] | 4239 | if (start_pfn < early_node_map[i].start_pfn && |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4240 | end_pfn >= early_node_map[i].start_pfn) { |
| 4241 | early_node_map[i].start_pfn = start_pfn; |
| 4242 | return; |
| 4243 | } |
| 4244 | } |
| 4245 | |
| 4246 | /* Check that early_node_map is large enough */ |
| 4247 | if (i >= MAX_ACTIVE_REGIONS) { |
| 4248 | printk(KERN_CRIT "More than %d memory regions, truncating\n", |
| 4249 | MAX_ACTIVE_REGIONS); |
| 4250 | return; |
| 4251 | } |
| 4252 | |
| 4253 | early_node_map[i].nid = nid; |
| 4254 | early_node_map[i].start_pfn = start_pfn; |
| 4255 | early_node_map[i].end_pfn = end_pfn; |
| 4256 | nr_nodemap_entries = i + 1; |
| 4257 | } |
| 4258 | |
| 4259 | /** |
Yinghai Lu | cc1050b | 2008-06-13 19:08:52 -0700 | [diff] [blame] | 4260 | * remove_active_range - Shrink an existing registered range of PFNs |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4261 | * @nid: The node id the range is on that should be shrunk |
Yinghai Lu | cc1050b | 2008-06-13 19:08:52 -0700 | [diff] [blame] | 4262 | * @start_pfn: The new PFN of the range |
| 4263 | * @end_pfn: The new PFN of the range |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4264 | * |
| 4265 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. |
Yinghai Lu | cc1a9d8 | 2008-06-08 19:39:16 -0700 | [diff] [blame] | 4266 | * The map is kept near the end physical page range that has already been |
| 4267 | * registered. This function allows an arch to shrink an existing registered |
| 4268 | * range. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4269 | */ |
Yinghai Lu | cc1050b | 2008-06-13 19:08:52 -0700 | [diff] [blame] | 4270 | void __init remove_active_range(unsigned int nid, unsigned long start_pfn, |
| 4271 | unsigned long end_pfn) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4272 | { |
Yinghai Lu | cc1a9d8 | 2008-06-08 19:39:16 -0700 | [diff] [blame] | 4273 | int i, j; |
| 4274 | int removed = 0; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4275 | |
Yinghai Lu | cc1050b | 2008-06-13 19:08:52 -0700 | [diff] [blame] | 4276 | printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n", |
| 4277 | nid, start_pfn, end_pfn); |
| 4278 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4279 | /* Find the old active region end and shrink */ |
Yinghai Lu | cc1a9d8 | 2008-06-08 19:39:16 -0700 | [diff] [blame] | 4280 | for_each_active_range_index_in_nid(i, nid) { |
Yinghai Lu | cc1050b | 2008-06-13 19:08:52 -0700 | [diff] [blame] | 4281 | if (early_node_map[i].start_pfn >= start_pfn && |
| 4282 | early_node_map[i].end_pfn <= end_pfn) { |
Yinghai Lu | cc1a9d8 | 2008-06-08 19:39:16 -0700 | [diff] [blame] | 4283 | /* clear it */ |
Yinghai Lu | cc1050b | 2008-06-13 19:08:52 -0700 | [diff] [blame] | 4284 | early_node_map[i].start_pfn = 0; |
Yinghai Lu | cc1a9d8 | 2008-06-08 19:39:16 -0700 | [diff] [blame] | 4285 | early_node_map[i].end_pfn = 0; |
| 4286 | removed = 1; |
| 4287 | continue; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4288 | } |
Yinghai Lu | cc1050b | 2008-06-13 19:08:52 -0700 | [diff] [blame] | 4289 | if (early_node_map[i].start_pfn < start_pfn && |
| 4290 | early_node_map[i].end_pfn > start_pfn) { |
| 4291 | unsigned long temp_end_pfn = early_node_map[i].end_pfn; |
| 4292 | early_node_map[i].end_pfn = start_pfn; |
| 4293 | if (temp_end_pfn > end_pfn) |
| 4294 | add_active_range(nid, end_pfn, temp_end_pfn); |
| 4295 | continue; |
| 4296 | } |
| 4297 | if (early_node_map[i].start_pfn >= start_pfn && |
| 4298 | early_node_map[i].end_pfn > end_pfn && |
| 4299 | early_node_map[i].start_pfn < end_pfn) { |
| 4300 | early_node_map[i].start_pfn = end_pfn; |
Yinghai Lu | cc1a9d8 | 2008-06-08 19:39:16 -0700 | [diff] [blame] | 4301 | continue; |
| 4302 | } |
| 4303 | } |
| 4304 | |
| 4305 | if (!removed) |
| 4306 | return; |
| 4307 | |
| 4308 | /* remove the blank ones */ |
| 4309 | for (i = nr_nodemap_entries - 1; i > 0; i--) { |
| 4310 | if (early_node_map[i].nid != nid) |
| 4311 | continue; |
| 4312 | if (early_node_map[i].end_pfn) |
| 4313 | continue; |
| 4314 | /* we found it, get rid of it */ |
| 4315 | for (j = i; j < nr_nodemap_entries - 1; j++) |
| 4316 | memcpy(&early_node_map[j], &early_node_map[j+1], |
| 4317 | sizeof(early_node_map[j])); |
| 4318 | j = nr_nodemap_entries - 1; |
| 4319 | memset(&early_node_map[j], 0, sizeof(early_node_map[j])); |
| 4320 | nr_nodemap_entries--; |
| 4321 | } |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4322 | } |
| 4323 | |
| 4324 | /** |
| 4325 | * remove_all_active_ranges - Remove all currently registered regions |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4326 | * |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4327 | * During discovery, it may be found that a table like SRAT is invalid |
| 4328 | * and an alternative discovery method must be used. This function removes |
| 4329 | * all currently registered regions. |
| 4330 | */ |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4331 | void __init remove_all_active_ranges(void) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4332 | { |
| 4333 | memset(early_node_map, 0, sizeof(early_node_map)); |
| 4334 | nr_nodemap_entries = 0; |
| 4335 | } |
| 4336 | |
| 4337 | /* Compare two active node_active_regions */ |
| 4338 | static int __init cmp_node_active_region(const void *a, const void *b) |
| 4339 | { |
| 4340 | struct node_active_region *arange = (struct node_active_region *)a; |
| 4341 | struct node_active_region *brange = (struct node_active_region *)b; |
| 4342 | |
| 4343 | /* Done this way to avoid overflows */ |
| 4344 | if (arange->start_pfn > brange->start_pfn) |
| 4345 | return 1; |
| 4346 | if (arange->start_pfn < brange->start_pfn) |
| 4347 | return -1; |
| 4348 | |
| 4349 | return 0; |
| 4350 | } |
| 4351 | |
| 4352 | /* sort the node_map by start_pfn */ |
Yinghai Lu | 3299625 | 2009-12-15 17:59:02 -0800 | [diff] [blame] | 4353 | void __init sort_node_map(void) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4354 | { |
| 4355 | sort(early_node_map, (size_t)nr_nodemap_entries, |
| 4356 | sizeof(struct node_active_region), |
| 4357 | cmp_node_active_region, NULL); |
| 4358 | } |
| 4359 | |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 4360 | /* Find the lowest pfn for a node */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 4361 | static unsigned long __init find_min_pfn_for_node(int nid) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4362 | { |
| 4363 | int i; |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 4364 | unsigned long min_pfn = ULONG_MAX; |
Mel Gorman | 1abbfb4 | 2006-11-23 12:01:41 +0000 | [diff] [blame] | 4365 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4366 | /* Assuming a sorted map, the first range found has the starting pfn */ |
| 4367 | for_each_active_range_index_in_nid(i, nid) |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 4368 | min_pfn = min(min_pfn, early_node_map[i].start_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4369 | |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 4370 | if (min_pfn == ULONG_MAX) { |
| 4371 | printk(KERN_WARNING |
Paul Jackson | 2bc0d261 | 2008-06-22 07:22:17 -0700 | [diff] [blame] | 4372 | "Could not find start_pfn for node %d\n", nid); |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 4373 | return 0; |
| 4374 | } |
| 4375 | |
| 4376 | return min_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4377 | } |
| 4378 | |
| 4379 | /** |
| 4380 | * find_min_pfn_with_active_regions - Find the minimum PFN registered |
| 4381 | * |
| 4382 | * It returns the minimum PFN based on information provided via |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4383 | * add_active_range(). |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4384 | */ |
| 4385 | unsigned long __init find_min_pfn_with_active_regions(void) |
| 4386 | { |
| 4387 | return find_min_pfn_for_node(MAX_NUMNODES); |
| 4388 | } |
| 4389 | |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4390 | /* |
| 4391 | * early_calculate_totalpages() |
| 4392 | * Sum pages in active regions for movable zone. |
| 4393 | * Populate N_HIGH_MEMORY for calculating usable_nodes. |
| 4394 | */ |
Adrian Bunk | 484f51f | 2007-10-16 01:26:03 -0700 | [diff] [blame] | 4395 | static unsigned long __init early_calculate_totalpages(void) |
Mel Gorman | 7e63efef | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4396 | { |
| 4397 | int i; |
| 4398 | unsigned long totalpages = 0; |
| 4399 | |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4400 | for (i = 0; i < nr_nodemap_entries; i++) { |
| 4401 | unsigned long pages = early_node_map[i].end_pfn - |
Mel Gorman | 7e63efef | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4402 | early_node_map[i].start_pfn; |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4403 | totalpages += pages; |
| 4404 | if (pages) |
| 4405 | node_set_state(early_node_map[i].nid, N_HIGH_MEMORY); |
| 4406 | } |
| 4407 | return totalpages; |
Mel Gorman | 7e63efef | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4408 | } |
| 4409 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4410 | /* |
| 4411 | * Find the PFN the Movable zone begins in each node. Kernel memory |
| 4412 | * is spread evenly between nodes as long as the nodes have enough |
| 4413 | * memory. When they don't, some nodes will have more kernelcore than |
| 4414 | * others |
| 4415 | */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 4416 | static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4417 | { |
| 4418 | int i, nid; |
| 4419 | unsigned long usable_startpfn; |
| 4420 | unsigned long kernelcore_node, kernelcore_remaining; |
Yinghai Lu | 66918dc | 2009-06-30 11:41:37 -0700 | [diff] [blame] | 4421 | /* save the state before borrow the nodemask */ |
| 4422 | nodemask_t saved_node_state = node_states[N_HIGH_MEMORY]; |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4423 | unsigned long totalpages = early_calculate_totalpages(); |
| 4424 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4425 | |
Mel Gorman | 7e63efef | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4426 | /* |
| 4427 | * If movablecore was specified, calculate what size of |
| 4428 | * kernelcore that corresponds so that memory usable for |
| 4429 | * any allocation type is evenly spread. If both kernelcore |
| 4430 | * and movablecore are specified, then the value of kernelcore |
| 4431 | * will be used for required_kernelcore if it's greater than |
| 4432 | * what movablecore would have allowed. |
| 4433 | */ |
| 4434 | if (required_movablecore) { |
Mel Gorman | 7e63efef | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4435 | unsigned long corepages; |
| 4436 | |
| 4437 | /* |
| 4438 | * Round-up so that ZONE_MOVABLE is at least as large as what |
| 4439 | * was requested by the user |
| 4440 | */ |
| 4441 | required_movablecore = |
| 4442 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); |
| 4443 | corepages = totalpages - required_movablecore; |
| 4444 | |
| 4445 | required_kernelcore = max(required_kernelcore, corepages); |
| 4446 | } |
| 4447 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4448 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
| 4449 | if (!required_kernelcore) |
Yinghai Lu | 66918dc | 2009-06-30 11:41:37 -0700 | [diff] [blame] | 4450 | goto out; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4451 | |
| 4452 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ |
| 4453 | find_usable_zone_for_movable(); |
| 4454 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
| 4455 | |
| 4456 | restart: |
| 4457 | /* Spread kernelcore memory as evenly as possible throughout nodes */ |
| 4458 | kernelcore_node = required_kernelcore / usable_nodes; |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4459 | for_each_node_state(nid, N_HIGH_MEMORY) { |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4460 | /* |
| 4461 | * Recalculate kernelcore_node if the division per node |
| 4462 | * now exceeds what is necessary to satisfy the requested |
| 4463 | * amount of memory for the kernel |
| 4464 | */ |
| 4465 | if (required_kernelcore < kernelcore_node) |
| 4466 | kernelcore_node = required_kernelcore / usable_nodes; |
| 4467 | |
| 4468 | /* |
| 4469 | * As the map is walked, we track how much memory is usable |
| 4470 | * by the kernel using kernelcore_remaining. When it is |
| 4471 | * 0, the rest of the node is usable by ZONE_MOVABLE |
| 4472 | */ |
| 4473 | kernelcore_remaining = kernelcore_node; |
| 4474 | |
| 4475 | /* Go through each range of PFNs within this node */ |
| 4476 | for_each_active_range_index_in_nid(i, nid) { |
| 4477 | unsigned long start_pfn, end_pfn; |
| 4478 | unsigned long size_pages; |
| 4479 | |
| 4480 | start_pfn = max(early_node_map[i].start_pfn, |
| 4481 | zone_movable_pfn[nid]); |
| 4482 | end_pfn = early_node_map[i].end_pfn; |
| 4483 | if (start_pfn >= end_pfn) |
| 4484 | continue; |
| 4485 | |
| 4486 | /* Account for what is only usable for kernelcore */ |
| 4487 | if (start_pfn < usable_startpfn) { |
| 4488 | unsigned long kernel_pages; |
| 4489 | kernel_pages = min(end_pfn, usable_startpfn) |
| 4490 | - start_pfn; |
| 4491 | |
| 4492 | kernelcore_remaining -= min(kernel_pages, |
| 4493 | kernelcore_remaining); |
| 4494 | required_kernelcore -= min(kernel_pages, |
| 4495 | required_kernelcore); |
| 4496 | |
| 4497 | /* Continue if range is now fully accounted */ |
| 4498 | if (end_pfn <= usable_startpfn) { |
| 4499 | |
| 4500 | /* |
| 4501 | * Push zone_movable_pfn to the end so |
| 4502 | * that if we have to rebalance |
| 4503 | * kernelcore across nodes, we will |
| 4504 | * not double account here |
| 4505 | */ |
| 4506 | zone_movable_pfn[nid] = end_pfn; |
| 4507 | continue; |
| 4508 | } |
| 4509 | start_pfn = usable_startpfn; |
| 4510 | } |
| 4511 | |
| 4512 | /* |
| 4513 | * The usable PFN range for ZONE_MOVABLE is from |
| 4514 | * start_pfn->end_pfn. Calculate size_pages as the |
| 4515 | * number of pages used as kernelcore |
| 4516 | */ |
| 4517 | size_pages = end_pfn - start_pfn; |
| 4518 | if (size_pages > kernelcore_remaining) |
| 4519 | size_pages = kernelcore_remaining; |
| 4520 | zone_movable_pfn[nid] = start_pfn + size_pages; |
| 4521 | |
| 4522 | /* |
| 4523 | * Some kernelcore has been met, update counts and |
| 4524 | * break if the kernelcore for this node has been |
| 4525 | * satisified |
| 4526 | */ |
| 4527 | required_kernelcore -= min(required_kernelcore, |
| 4528 | size_pages); |
| 4529 | kernelcore_remaining -= size_pages; |
| 4530 | if (!kernelcore_remaining) |
| 4531 | break; |
| 4532 | } |
| 4533 | } |
| 4534 | |
| 4535 | /* |
| 4536 | * If there is still required_kernelcore, we do another pass with one |
| 4537 | * less node in the count. This will push zone_movable_pfn[nid] further |
| 4538 | * along on the nodes that still have memory until kernelcore is |
| 4539 | * satisified |
| 4540 | */ |
| 4541 | usable_nodes--; |
| 4542 | if (usable_nodes && required_kernelcore > usable_nodes) |
| 4543 | goto restart; |
| 4544 | |
| 4545 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
| 4546 | for (nid = 0; nid < MAX_NUMNODES; nid++) |
| 4547 | zone_movable_pfn[nid] = |
| 4548 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); |
Yinghai Lu | 66918dc | 2009-06-30 11:41:37 -0700 | [diff] [blame] | 4549 | |
| 4550 | out: |
| 4551 | /* restore the node_state */ |
| 4552 | node_states[N_HIGH_MEMORY] = saved_node_state; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4553 | } |
| 4554 | |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4555 | /* Any regular memory on that node ? */ |
| 4556 | static void check_for_regular_memory(pg_data_t *pgdat) |
| 4557 | { |
| 4558 | #ifdef CONFIG_HIGHMEM |
| 4559 | enum zone_type zone_type; |
| 4560 | |
| 4561 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { |
| 4562 | struct zone *zone = &pgdat->node_zones[zone_type]; |
| 4563 | if (zone->present_pages) |
| 4564 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); |
| 4565 | } |
| 4566 | #endif |
| 4567 | } |
| 4568 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4569 | /** |
| 4570 | * free_area_init_nodes - Initialise all pg_data_t and zone data |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4571 | * @max_zone_pfn: an array of max PFNs for each zone |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4572 | * |
| 4573 | * This will call free_area_init_node() for each active node in the system. |
| 4574 | * Using the page ranges provided by add_active_range(), the size of each |
| 4575 | * zone in each node and their holes is calculated. If the maximum PFN |
| 4576 | * between two adjacent zones match, it is assumed that the zone is empty. |
| 4577 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed |
| 4578 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone |
| 4579 | * starts where the previous one ended. For example, ZONE_DMA32 starts |
| 4580 | * at arch_max_dma_pfn. |
| 4581 | */ |
| 4582 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) |
| 4583 | { |
| 4584 | unsigned long nid; |
Andrew Morton | db99100 | 2008-10-15 22:01:07 -0700 | [diff] [blame] | 4585 | int i; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4586 | |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 4587 | /* Sort early_node_map as initialisation assumes it is sorted */ |
| 4588 | sort_node_map(); |
| 4589 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4590 | /* Record where the zone boundaries are */ |
| 4591 | memset(arch_zone_lowest_possible_pfn, 0, |
| 4592 | sizeof(arch_zone_lowest_possible_pfn)); |
| 4593 | memset(arch_zone_highest_possible_pfn, 0, |
| 4594 | sizeof(arch_zone_highest_possible_pfn)); |
| 4595 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); |
| 4596 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; |
| 4597 | for (i = 1; i < MAX_NR_ZONES; i++) { |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4598 | if (i == ZONE_MOVABLE) |
| 4599 | continue; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4600 | arch_zone_lowest_possible_pfn[i] = |
| 4601 | arch_zone_highest_possible_pfn[i-1]; |
| 4602 | arch_zone_highest_possible_pfn[i] = |
| 4603 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); |
| 4604 | } |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4605 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
| 4606 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; |
| 4607 | |
| 4608 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ |
| 4609 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); |
| 4610 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4611 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4612 | /* Print out the zone ranges */ |
| 4613 | printk("Zone PFN ranges:\n"); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4614 | for (i = 0; i < MAX_NR_ZONES; i++) { |
| 4615 | if (i == ZONE_MOVABLE) |
| 4616 | continue; |
David Rientjes | 72f0ba0 | 2010-03-05 13:42:14 -0800 | [diff] [blame] | 4617 | printk(" %-8s ", zone_names[i]); |
| 4618 | if (arch_zone_lowest_possible_pfn[i] == |
| 4619 | arch_zone_highest_possible_pfn[i]) |
| 4620 | printk("empty\n"); |
| 4621 | else |
| 4622 | printk("%0#10lx -> %0#10lx\n", |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4623 | arch_zone_lowest_possible_pfn[i], |
| 4624 | arch_zone_highest_possible_pfn[i]); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4625 | } |
| 4626 | |
| 4627 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ |
| 4628 | printk("Movable zone start PFN for each node\n"); |
| 4629 | for (i = 0; i < MAX_NUMNODES; i++) { |
| 4630 | if (zone_movable_pfn[i]) |
| 4631 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); |
| 4632 | } |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4633 | |
| 4634 | /* Print out the early_node_map[] */ |
| 4635 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); |
| 4636 | for (i = 0; i < nr_nodemap_entries; i++) |
Paul Jackson | 5dab8ec | 2008-06-25 05:44:40 -0700 | [diff] [blame] | 4637 | printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4638 | early_node_map[i].start_pfn, |
| 4639 | early_node_map[i].end_pfn); |
| 4640 | |
| 4641 | /* Initialise every node */ |
Mel Gorman | 708614e | 2008-07-23 21:26:51 -0700 | [diff] [blame] | 4642 | mminit_verify_pageflags_layout(); |
Christoph Lameter | 8ef8286 | 2007-02-20 13:57:52 -0800 | [diff] [blame] | 4643 | setup_nr_node_ids(); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4644 | for_each_online_node(nid) { |
| 4645 | pg_data_t *pgdat = NODE_DATA(nid); |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 4646 | free_area_init_node(nid, NULL, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4647 | find_min_pfn_for_node(nid), NULL); |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4648 | |
| 4649 | /* Any memory on that node */ |
| 4650 | if (pgdat->node_present_pages) |
| 4651 | node_set_state(nid, N_HIGH_MEMORY); |
| 4652 | check_for_regular_memory(pgdat); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4653 | } |
| 4654 | } |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4655 | |
Mel Gorman | 7e63efef | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4656 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4657 | { |
| 4658 | unsigned long long coremem; |
| 4659 | if (!p) |
| 4660 | return -EINVAL; |
| 4661 | |
| 4662 | coremem = memparse(p, &p); |
Mel Gorman | 7e63efef | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4663 | *core = coremem >> PAGE_SHIFT; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4664 | |
Mel Gorman | 7e63efef | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4665 | /* Paranoid check that UL is enough for the coremem value */ |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4666 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
| 4667 | |
| 4668 | return 0; |
| 4669 | } |
Mel Gorman | ed7ed36 | 2007-07-17 04:03:14 -0700 | [diff] [blame] | 4670 | |
Mel Gorman | 7e63efef | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4671 | /* |
| 4672 | * kernelcore=size sets the amount of memory for use for allocations that |
| 4673 | * cannot be reclaimed or migrated. |
| 4674 | */ |
| 4675 | static int __init cmdline_parse_kernelcore(char *p) |
| 4676 | { |
| 4677 | return cmdline_parse_core(p, &required_kernelcore); |
| 4678 | } |
| 4679 | |
| 4680 | /* |
| 4681 | * movablecore=size sets the amount of memory for use for allocations that |
| 4682 | * can be reclaimed or migrated. |
| 4683 | */ |
| 4684 | static int __init cmdline_parse_movablecore(char *p) |
| 4685 | { |
| 4686 | return cmdline_parse_core(p, &required_movablecore); |
| 4687 | } |
| 4688 | |
Mel Gorman | ed7ed36 | 2007-07-17 04:03:14 -0700 | [diff] [blame] | 4689 | early_param("kernelcore", cmdline_parse_kernelcore); |
Mel Gorman | 7e63efef | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4690 | early_param("movablecore", cmdline_parse_movablecore); |
Mel Gorman | ed7ed36 | 2007-07-17 04:03:14 -0700 | [diff] [blame] | 4691 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4692 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
| 4693 | |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4694 | /** |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4695 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
| 4696 | * @new_dma_reserve: The number of pages to mark reserved |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4697 | * |
| 4698 | * The per-cpu batchsize and zone watermarks are determined by present_pages. |
| 4699 | * In the DMA zone, a significant percentage may be consumed by kernel image |
| 4700 | * and other unfreeable allocations which can skew the watermarks badly. This |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4701 | * function may optionally be used to account for unfreeable pages in the |
| 4702 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and |
| 4703 | * smaller per-cpu batchsize. |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4704 | */ |
| 4705 | void __init set_dma_reserve(unsigned long new_dma_reserve) |
| 4706 | { |
| 4707 | dma_reserve = new_dma_reserve; |
| 4708 | } |
| 4709 | |
Dave Hansen | 93b7504 | 2005-06-23 00:07:47 -0700 | [diff] [blame] | 4710 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 4711 | struct pglist_data __refdata contig_page_data = { |
| 4712 | #ifndef CONFIG_NO_BOOTMEM |
| 4713 | .bdata = &bootmem_node_data[0] |
| 4714 | #endif |
| 4715 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4716 | EXPORT_SYMBOL(contig_page_data); |
Dave Hansen | 93b7504 | 2005-06-23 00:07:47 -0700 | [diff] [blame] | 4717 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4718 | |
| 4719 | void __init free_area_init(unsigned long *zones_size) |
| 4720 | { |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 4721 | free_area_init_node(0, zones_size, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4722 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
| 4723 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4724 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4725 | static int page_alloc_cpu_notify(struct notifier_block *self, |
| 4726 | unsigned long action, void *hcpu) |
| 4727 | { |
| 4728 | int cpu = (unsigned long)hcpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4729 | |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 4730 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 4731 | drain_pages(cpu); |
| 4732 | |
| 4733 | /* |
| 4734 | * Spill the event counters of the dead processor |
| 4735 | * into the current processors event counters. |
| 4736 | * This artificially elevates the count of the current |
| 4737 | * processor. |
| 4738 | */ |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 4739 | vm_events_fold_cpu(cpu); |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 4740 | |
| 4741 | /* |
| 4742 | * Zero the differential counters of the dead processor |
| 4743 | * so that the vm statistics are consistent. |
| 4744 | * |
| 4745 | * This is only okay since the processor is dead and cannot |
| 4746 | * race with what we are doing. |
| 4747 | */ |
Christoph Lameter | 2244b95 | 2006-06-30 01:55:33 -0700 | [diff] [blame] | 4748 | refresh_cpu_vm_stats(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4749 | } |
| 4750 | return NOTIFY_OK; |
| 4751 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4752 | |
| 4753 | void __init page_alloc_init(void) |
| 4754 | { |
| 4755 | hotcpu_notifier(page_alloc_cpu_notify, 0); |
| 4756 | } |
| 4757 | |
| 4758 | /* |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 4759 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio |
| 4760 | * or min_free_kbytes changes. |
| 4761 | */ |
| 4762 | static void calculate_totalreserve_pages(void) |
| 4763 | { |
| 4764 | struct pglist_data *pgdat; |
| 4765 | unsigned long reserve_pages = 0; |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 4766 | enum zone_type i, j; |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 4767 | |
| 4768 | for_each_online_pgdat(pgdat) { |
| 4769 | for (i = 0; i < MAX_NR_ZONES; i++) { |
| 4770 | struct zone *zone = pgdat->node_zones + i; |
| 4771 | unsigned long max = 0; |
| 4772 | |
| 4773 | /* Find valid and maximum lowmem_reserve in the zone */ |
| 4774 | for (j = i; j < MAX_NR_ZONES; j++) { |
| 4775 | if (zone->lowmem_reserve[j] > max) |
| 4776 | max = zone->lowmem_reserve[j]; |
| 4777 | } |
| 4778 | |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 4779 | /* we treat the high watermark as reserved pages. */ |
| 4780 | max += high_wmark_pages(zone); |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 4781 | |
| 4782 | if (max > zone->present_pages) |
| 4783 | max = zone->present_pages; |
| 4784 | reserve_pages += max; |
| 4785 | } |
| 4786 | } |
| 4787 | totalreserve_pages = reserve_pages; |
| 4788 | } |
| 4789 | |
| 4790 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4791 | * setup_per_zone_lowmem_reserve - called whenever |
| 4792 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone |
| 4793 | * has a correct pages reserved value, so an adequate number of |
| 4794 | * pages are left in the zone after a successful __alloc_pages(). |
| 4795 | */ |
| 4796 | static void setup_per_zone_lowmem_reserve(void) |
| 4797 | { |
| 4798 | struct pglist_data *pgdat; |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 4799 | enum zone_type j, idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4800 | |
KAMEZAWA Hiroyuki | ec936fc | 2006-03-27 01:15:59 -0800 | [diff] [blame] | 4801 | for_each_online_pgdat(pgdat) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4802 | for (j = 0; j < MAX_NR_ZONES; j++) { |
| 4803 | struct zone *zone = pgdat->node_zones + j; |
| 4804 | unsigned long present_pages = zone->present_pages; |
| 4805 | |
| 4806 | zone->lowmem_reserve[j] = 0; |
| 4807 | |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 4808 | idx = j; |
| 4809 | while (idx) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4810 | struct zone *lower_zone; |
| 4811 | |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 4812 | idx--; |
| 4813 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4814 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
| 4815 | sysctl_lowmem_reserve_ratio[idx] = 1; |
| 4816 | |
| 4817 | lower_zone = pgdat->node_zones + idx; |
| 4818 | lower_zone->lowmem_reserve[j] = present_pages / |
| 4819 | sysctl_lowmem_reserve_ratio[idx]; |
| 4820 | present_pages += lower_zone->present_pages; |
| 4821 | } |
| 4822 | } |
| 4823 | } |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 4824 | |
| 4825 | /* update totalreserve_pages */ |
| 4826 | calculate_totalreserve_pages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4827 | } |
| 4828 | |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4829 | /** |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 4830 | * setup_per_zone_wmarks - called when min_free_kbytes changes |
Minchan Kim | bce7394 | 2009-06-16 15:32:50 -0700 | [diff] [blame] | 4831 | * or when memory is hot-{added|removed} |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4832 | * |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 4833 | * Ensures that the watermark[min,low,high] values for each zone are set |
| 4834 | * correctly with respect to min_free_kbytes. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4835 | */ |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 4836 | void setup_per_zone_wmarks(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4837 | { |
| 4838 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); |
| 4839 | unsigned long lowmem_pages = 0; |
| 4840 | struct zone *zone; |
| 4841 | unsigned long flags; |
| 4842 | |
| 4843 | /* Calculate total number of !ZONE_HIGHMEM pages */ |
| 4844 | for_each_zone(zone) { |
| 4845 | if (!is_highmem(zone)) |
| 4846 | lowmem_pages += zone->present_pages; |
| 4847 | } |
| 4848 | |
| 4849 | for_each_zone(zone) { |
Andrew Morton | ac924c6 | 2006-05-15 09:43:59 -0700 | [diff] [blame] | 4850 | u64 tmp; |
| 4851 | |
Gerald Schaefer | 1125b4e | 2008-10-18 20:27:11 -0700 | [diff] [blame] | 4852 | spin_lock_irqsave(&zone->lock, flags); |
Andrew Morton | ac924c6 | 2006-05-15 09:43:59 -0700 | [diff] [blame] | 4853 | tmp = (u64)pages_min * zone->present_pages; |
| 4854 | do_div(tmp, lowmem_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4855 | if (is_highmem(zone)) { |
| 4856 | /* |
Nick Piggin | 669ed17 | 2005-11-13 16:06:45 -0800 | [diff] [blame] | 4857 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
| 4858 | * need highmem pages, so cap pages_min to a small |
| 4859 | * value here. |
| 4860 | * |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 4861 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
Nick Piggin | 669ed17 | 2005-11-13 16:06:45 -0800 | [diff] [blame] | 4862 | * deltas controls asynch page reclaim, and so should |
| 4863 | * not be capped for highmem. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4864 | */ |
| 4865 | int min_pages; |
| 4866 | |
| 4867 | min_pages = zone->present_pages / 1024; |
| 4868 | if (min_pages < SWAP_CLUSTER_MAX) |
| 4869 | min_pages = SWAP_CLUSTER_MAX; |
| 4870 | if (min_pages > 128) |
| 4871 | min_pages = 128; |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 4872 | zone->watermark[WMARK_MIN] = min_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4873 | } else { |
Nick Piggin | 669ed17 | 2005-11-13 16:06:45 -0800 | [diff] [blame] | 4874 | /* |
| 4875 | * If it's a lowmem zone, reserve a number of pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4876 | * proportionate to the zone's size. |
| 4877 | */ |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 4878 | zone->watermark[WMARK_MIN] = tmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4879 | } |
| 4880 | |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 4881 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
| 4882 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 4883 | setup_zone_migrate_reserve(zone); |
Gerald Schaefer | 1125b4e | 2008-10-18 20:27:11 -0700 | [diff] [blame] | 4884 | spin_unlock_irqrestore(&zone->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4885 | } |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 4886 | |
| 4887 | /* update totalreserve_pages */ |
| 4888 | calculate_totalreserve_pages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4889 | } |
| 4890 | |
Randy Dunlap | 55a4462a | 2009-09-21 17:01:20 -0700 | [diff] [blame] | 4891 | /* |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 4892 | * The inactive anon list should be small enough that the VM never has to |
| 4893 | * do too much work, but large enough that each inactive page has a chance |
| 4894 | * to be referenced again before it is swapped out. |
| 4895 | * |
| 4896 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to |
| 4897 | * INACTIVE_ANON pages on this zone's LRU, maintained by the |
| 4898 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of |
| 4899 | * the anonymous pages are kept on the inactive list. |
| 4900 | * |
| 4901 | * total target max |
| 4902 | * memory ratio inactive anon |
| 4903 | * ------------------------------------- |
| 4904 | * 10MB 1 5MB |
| 4905 | * 100MB 1 50MB |
| 4906 | * 1GB 3 250MB |
| 4907 | * 10GB 10 0.9GB |
| 4908 | * 100GB 31 3GB |
| 4909 | * 1TB 101 10GB |
| 4910 | * 10TB 320 32GB |
| 4911 | */ |
Minchan Kim | 96cb4df | 2009-06-16 15:32:49 -0700 | [diff] [blame] | 4912 | void calculate_zone_inactive_ratio(struct zone *zone) |
| 4913 | { |
| 4914 | unsigned int gb, ratio; |
| 4915 | |
| 4916 | /* Zone size in gigabytes */ |
| 4917 | gb = zone->present_pages >> (30 - PAGE_SHIFT); |
| 4918 | if (gb) |
| 4919 | ratio = int_sqrt(10 * gb); |
| 4920 | else |
| 4921 | ratio = 1; |
| 4922 | |
| 4923 | zone->inactive_ratio = ratio; |
| 4924 | } |
| 4925 | |
Cyrill Gorcunov | e9bb35d | 2009-06-16 15:32:32 -0700 | [diff] [blame] | 4926 | static void __init setup_per_zone_inactive_ratio(void) |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 4927 | { |
| 4928 | struct zone *zone; |
| 4929 | |
Minchan Kim | 96cb4df | 2009-06-16 15:32:49 -0700 | [diff] [blame] | 4930 | for_each_zone(zone) |
| 4931 | calculate_zone_inactive_ratio(zone); |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 4932 | } |
| 4933 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4934 | /* |
| 4935 | * Initialise min_free_kbytes. |
| 4936 | * |
| 4937 | * For small machines we want it small (128k min). For large machines |
| 4938 | * we want it large (64MB max). But it is not linear, because network |
| 4939 | * bandwidth does not increase linearly with machine size. We use |
| 4940 | * |
| 4941 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
| 4942 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
| 4943 | * |
| 4944 | * which yields |
| 4945 | * |
| 4946 | * 16MB: 512k |
| 4947 | * 32MB: 724k |
| 4948 | * 64MB: 1024k |
| 4949 | * 128MB: 1448k |
| 4950 | * 256MB: 2048k |
| 4951 | * 512MB: 2896k |
| 4952 | * 1024MB: 4096k |
| 4953 | * 2048MB: 5792k |
| 4954 | * 4096MB: 8192k |
| 4955 | * 8192MB: 11584k |
| 4956 | * 16384MB: 16384k |
| 4957 | */ |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 4958 | static int __init init_per_zone_wmark_min(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4959 | { |
| 4960 | unsigned long lowmem_kbytes; |
| 4961 | |
| 4962 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); |
| 4963 | |
| 4964 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
| 4965 | if (min_free_kbytes < 128) |
| 4966 | min_free_kbytes = 128; |
| 4967 | if (min_free_kbytes > 65536) |
| 4968 | min_free_kbytes = 65536; |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 4969 | setup_per_zone_wmarks(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4970 | setup_per_zone_lowmem_reserve(); |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 4971 | setup_per_zone_inactive_ratio(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4972 | return 0; |
| 4973 | } |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 4974 | module_init(init_per_zone_wmark_min) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4975 | |
| 4976 | /* |
| 4977 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
| 4978 | * that we can call two helper functions whenever min_free_kbytes |
| 4979 | * changes. |
| 4980 | */ |
| 4981 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 4982 | void __user *buffer, size_t *length, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4983 | { |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 4984 | proc_dointvec(table, write, buffer, length, ppos); |
Mel Gorman | 3b1d92c | 2007-05-06 14:49:30 -0700 | [diff] [blame] | 4985 | if (write) |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 4986 | setup_per_zone_wmarks(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4987 | return 0; |
| 4988 | } |
| 4989 | |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 4990 | #ifdef CONFIG_NUMA |
| 4991 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 4992 | void __user *buffer, size_t *length, loff_t *ppos) |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 4993 | { |
| 4994 | struct zone *zone; |
| 4995 | int rc; |
| 4996 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 4997 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 4998 | if (rc) |
| 4999 | return rc; |
| 5000 | |
| 5001 | for_each_zone(zone) |
Christoph Lameter | 8417bba | 2006-09-25 23:31:51 -0700 | [diff] [blame] | 5002 | zone->min_unmapped_pages = (zone->present_pages * |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 5003 | sysctl_min_unmapped_ratio) / 100; |
| 5004 | return 0; |
| 5005 | } |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 5006 | |
| 5007 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5008 | void __user *buffer, size_t *length, loff_t *ppos) |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 5009 | { |
| 5010 | struct zone *zone; |
| 5011 | int rc; |
| 5012 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5013 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 5014 | if (rc) |
| 5015 | return rc; |
| 5016 | |
| 5017 | for_each_zone(zone) |
| 5018 | zone->min_slab_pages = (zone->present_pages * |
| 5019 | sysctl_min_slab_ratio) / 100; |
| 5020 | return 0; |
| 5021 | } |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 5022 | #endif |
| 5023 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5024 | /* |
| 5025 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around |
| 5026 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() |
| 5027 | * whenever sysctl_lowmem_reserve_ratio changes. |
| 5028 | * |
| 5029 | * The reserve ratio obviously has absolutely no relation with the |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 5030 | * minimum watermarks. The lowmem reserve ratio can only make sense |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5031 | * if in function of the boot time zone sizes. |
| 5032 | */ |
| 5033 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5034 | void __user *buffer, size_t *length, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5035 | { |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5036 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5037 | setup_per_zone_lowmem_reserve(); |
| 5038 | return 0; |
| 5039 | } |
| 5040 | |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5041 | /* |
| 5042 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each |
| 5043 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist |
| 5044 | * can have before it gets flushed back to buddy allocator. |
| 5045 | */ |
| 5046 | |
| 5047 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5048 | void __user *buffer, size_t *length, loff_t *ppos) |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5049 | { |
| 5050 | struct zone *zone; |
| 5051 | unsigned int cpu; |
| 5052 | int ret; |
| 5053 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5054 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5055 | if (!write || (ret == -EINVAL)) |
| 5056 | return ret; |
Dimitri Sivanich | 364df0e | 2009-06-23 12:37:04 -0700 | [diff] [blame] | 5057 | for_each_populated_zone(zone) { |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 5058 | for_each_possible_cpu(cpu) { |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5059 | unsigned long high; |
| 5060 | high = zone->present_pages / percpu_pagelist_fraction; |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 5061 | setup_pagelist_highmark( |
| 5062 | per_cpu_ptr(zone->pageset, cpu), high); |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5063 | } |
| 5064 | } |
| 5065 | return 0; |
| 5066 | } |
| 5067 | |
David S. Miller | f034b5d | 2006-08-24 03:08:07 -0700 | [diff] [blame] | 5068 | int hashdist = HASHDIST_DEFAULT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5069 | |
| 5070 | #ifdef CONFIG_NUMA |
| 5071 | static int __init set_hashdist(char *str) |
| 5072 | { |
| 5073 | if (!str) |
| 5074 | return 0; |
| 5075 | hashdist = simple_strtoul(str, &str, 0); |
| 5076 | return 1; |
| 5077 | } |
| 5078 | __setup("hashdist=", set_hashdist); |
| 5079 | #endif |
| 5080 | |
| 5081 | /* |
| 5082 | * allocate a large system hash table from bootmem |
| 5083 | * - it is assumed that the hash table must contain an exact power-of-2 |
| 5084 | * quantity of entries |
| 5085 | * - limit is the number of hash buckets, not the total allocation size |
| 5086 | */ |
| 5087 | void *__init alloc_large_system_hash(const char *tablename, |
| 5088 | unsigned long bucketsize, |
| 5089 | unsigned long numentries, |
| 5090 | int scale, |
| 5091 | int flags, |
| 5092 | unsigned int *_hash_shift, |
| 5093 | unsigned int *_hash_mask, |
| 5094 | unsigned long limit) |
| 5095 | { |
| 5096 | unsigned long long max = limit; |
| 5097 | unsigned long log2qty, size; |
| 5098 | void *table = NULL; |
| 5099 | |
| 5100 | /* allow the kernel cmdline to have a say */ |
| 5101 | if (!numentries) { |
| 5102 | /* round applicable memory size up to nearest megabyte */ |
Andrew Morton | 0490366 | 2006-12-06 20:37:33 -0800 | [diff] [blame] | 5103 | numentries = nr_kernel_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5104 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
| 5105 | numentries >>= 20 - PAGE_SHIFT; |
| 5106 | numentries <<= 20 - PAGE_SHIFT; |
| 5107 | |
| 5108 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
| 5109 | if (scale > PAGE_SHIFT) |
| 5110 | numentries >>= (scale - PAGE_SHIFT); |
| 5111 | else |
| 5112 | numentries <<= (PAGE_SHIFT - scale); |
Paul Mundt | 9ab37b8 | 2007-01-05 16:36:30 -0800 | [diff] [blame] | 5113 | |
| 5114 | /* Make sure we've got at least a 0-order allocation.. */ |
Jan Beulich | 2c85f51 | 2009-09-21 17:03:07 -0700 | [diff] [blame] | 5115 | if (unlikely(flags & HASH_SMALL)) { |
| 5116 | /* Makes no sense without HASH_EARLY */ |
| 5117 | WARN_ON(!(flags & HASH_EARLY)); |
| 5118 | if (!(numentries >> *_hash_shift)) { |
| 5119 | numentries = 1UL << *_hash_shift; |
| 5120 | BUG_ON(!numentries); |
| 5121 | } |
| 5122 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) |
Paul Mundt | 9ab37b8 | 2007-01-05 16:36:30 -0800 | [diff] [blame] | 5123 | numentries = PAGE_SIZE / bucketsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5124 | } |
John Hawkes | 6e692ed | 2006-03-25 03:08:02 -0800 | [diff] [blame] | 5125 | numentries = roundup_pow_of_two(numentries); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5126 | |
| 5127 | /* limit allocation size to 1/16 total memory by default */ |
| 5128 | if (max == 0) { |
| 5129 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; |
| 5130 | do_div(max, bucketsize); |
| 5131 | } |
| 5132 | |
| 5133 | if (numentries > max) |
| 5134 | numentries = max; |
| 5135 | |
David Howells | f0d1b0b | 2006-12-08 02:37:49 -0800 | [diff] [blame] | 5136 | log2qty = ilog2(numentries); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5137 | |
| 5138 | do { |
| 5139 | size = bucketsize << log2qty; |
| 5140 | if (flags & HASH_EARLY) |
Jan Beulich | 74768ed | 2008-08-12 15:08:39 -0700 | [diff] [blame] | 5141 | table = alloc_bootmem_nopanic(size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5142 | else if (hashdist) |
| 5143 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); |
| 5144 | else { |
Eric Dumazet | 1037b83 | 2007-07-15 23:38:05 -0700 | [diff] [blame] | 5145 | /* |
| 5146 | * If bucketsize is not a power-of-two, we may free |
Mel Gorman | a1dd268 | 2009-06-16 15:32:19 -0700 | [diff] [blame] | 5147 | * some pages at the end of hash table which |
| 5148 | * alloc_pages_exact() automatically does |
Eric Dumazet | 1037b83 | 2007-07-15 23:38:05 -0700 | [diff] [blame] | 5149 | */ |
Catalin Marinas | 264ef8a | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 5150 | if (get_order(size) < MAX_ORDER) { |
Mel Gorman | a1dd268 | 2009-06-16 15:32:19 -0700 | [diff] [blame] | 5151 | table = alloc_pages_exact(size, GFP_ATOMIC); |
Catalin Marinas | 264ef8a | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 5152 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
| 5153 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5154 | } |
| 5155 | } while (!table && size > PAGE_SIZE && --log2qty); |
| 5156 | |
| 5157 | if (!table) |
| 5158 | panic("Failed to allocate %s hash table\n", tablename); |
| 5159 | |
Dan Aloni | b49ad48 | 2007-07-15 23:38:23 -0700 | [diff] [blame] | 5160 | printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5161 | tablename, |
| 5162 | (1U << log2qty), |
David Howells | f0d1b0b | 2006-12-08 02:37:49 -0800 | [diff] [blame] | 5163 | ilog2(size) - PAGE_SHIFT, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5164 | size); |
| 5165 | |
| 5166 | if (_hash_shift) |
| 5167 | *_hash_shift = log2qty; |
| 5168 | if (_hash_mask) |
| 5169 | *_hash_mask = (1 << log2qty) - 1; |
| 5170 | |
| 5171 | return table; |
| 5172 | } |
KAMEZAWA Hiroyuki | a117e66 | 2006-03-27 01:15:25 -0800 | [diff] [blame] | 5173 | |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5174 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
| 5175 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, |
| 5176 | unsigned long pfn) |
| 5177 | { |
| 5178 | #ifdef CONFIG_SPARSEMEM |
| 5179 | return __pfn_to_section(pfn)->pageblock_flags; |
| 5180 | #else |
| 5181 | return zone->pageblock_flags; |
| 5182 | #endif /* CONFIG_SPARSEMEM */ |
| 5183 | } |
Andrew Morton | 6220ec7 | 2006-10-19 23:29:05 -0700 | [diff] [blame] | 5184 | |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5185 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) |
| 5186 | { |
| 5187 | #ifdef CONFIG_SPARSEMEM |
| 5188 | pfn &= (PAGES_PER_SECTION-1); |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5189 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5190 | #else |
| 5191 | pfn = pfn - zone->zone_start_pfn; |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5192 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5193 | #endif /* CONFIG_SPARSEMEM */ |
| 5194 | } |
| 5195 | |
| 5196 | /** |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5197 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5198 | * @page: The page within the block of interest |
| 5199 | * @start_bitidx: The first bit of interest to retrieve |
| 5200 | * @end_bitidx: The last bit of interest |
| 5201 | * returns pageblock_bits flags |
| 5202 | */ |
| 5203 | unsigned long get_pageblock_flags_group(struct page *page, |
| 5204 | int start_bitidx, int end_bitidx) |
| 5205 | { |
| 5206 | struct zone *zone; |
| 5207 | unsigned long *bitmap; |
| 5208 | unsigned long pfn, bitidx; |
| 5209 | unsigned long flags = 0; |
| 5210 | unsigned long value = 1; |
| 5211 | |
| 5212 | zone = page_zone(page); |
| 5213 | pfn = page_to_pfn(page); |
| 5214 | bitmap = get_pageblock_bitmap(zone, pfn); |
| 5215 | bitidx = pfn_to_bitidx(zone, pfn); |
| 5216 | |
| 5217 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) |
| 5218 | if (test_bit(bitidx + start_bitidx, bitmap)) |
| 5219 | flags |= value; |
| 5220 | |
| 5221 | return flags; |
| 5222 | } |
| 5223 | |
| 5224 | /** |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5225 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5226 | * @page: The page within the block of interest |
| 5227 | * @start_bitidx: The first bit of interest |
| 5228 | * @end_bitidx: The last bit of interest |
| 5229 | * @flags: The flags to set |
| 5230 | */ |
| 5231 | void set_pageblock_flags_group(struct page *page, unsigned long flags, |
| 5232 | int start_bitidx, int end_bitidx) |
| 5233 | { |
| 5234 | struct zone *zone; |
| 5235 | unsigned long *bitmap; |
| 5236 | unsigned long pfn, bitidx; |
| 5237 | unsigned long value = 1; |
| 5238 | |
| 5239 | zone = page_zone(page); |
| 5240 | pfn = page_to_pfn(page); |
| 5241 | bitmap = get_pageblock_bitmap(zone, pfn); |
| 5242 | bitidx = pfn_to_bitidx(zone, pfn); |
KAMEZAWA Hiroyuki | 86051ca | 2008-04-29 00:58:21 -0700 | [diff] [blame] | 5243 | VM_BUG_ON(pfn < zone->zone_start_pfn); |
| 5244 | VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages); |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5245 | |
| 5246 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) |
| 5247 | if (flags & value) |
| 5248 | __set_bit(bitidx + start_bitidx, bitmap); |
| 5249 | else |
| 5250 | __clear_bit(bitidx + start_bitidx, bitmap); |
| 5251 | } |
KAMEZAWA Hiroyuki | a5d76b54 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5252 | |
| 5253 | /* |
| 5254 | * This is designed as sub function...plz see page_isolation.c also. |
| 5255 | * set/clear page block's type to be ISOLATE. |
| 5256 | * page allocater never alloc memory from ISOLATE block. |
| 5257 | */ |
| 5258 | |
| 5259 | int set_migratetype_isolate(struct page *page) |
| 5260 | { |
| 5261 | struct zone *zone; |
Robert Jennings | 925cc71 | 2009-12-17 14:44:38 +0000 | [diff] [blame] | 5262 | struct page *curr_page; |
| 5263 | unsigned long flags, pfn, iter; |
| 5264 | unsigned long immobile = 0; |
| 5265 | struct memory_isolate_notify arg; |
| 5266 | int notifier_ret; |
KAMEZAWA Hiroyuki | a5d76b54 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5267 | int ret = -EBUSY; |
Shaohua Li | 8e7e40d | 2009-09-21 17:01:18 -0700 | [diff] [blame] | 5268 | int zone_idx; |
KAMEZAWA Hiroyuki | a5d76b54 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5269 | |
| 5270 | zone = page_zone(page); |
Shaohua Li | 8e7e40d | 2009-09-21 17:01:18 -0700 | [diff] [blame] | 5271 | zone_idx = zone_idx(zone); |
Robert Jennings | 925cc71 | 2009-12-17 14:44:38 +0000 | [diff] [blame] | 5272 | |
KAMEZAWA Hiroyuki | a5d76b54 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5273 | spin_lock_irqsave(&zone->lock, flags); |
Robert Jennings | 925cc71 | 2009-12-17 14:44:38 +0000 | [diff] [blame] | 5274 | if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE || |
| 5275 | zone_idx == ZONE_MOVABLE) { |
| 5276 | ret = 0; |
KAMEZAWA Hiroyuki | a5d76b54 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5277 | goto out; |
Robert Jennings | 925cc71 | 2009-12-17 14:44:38 +0000 | [diff] [blame] | 5278 | } |
| 5279 | |
| 5280 | pfn = page_to_pfn(page); |
| 5281 | arg.start_pfn = pfn; |
| 5282 | arg.nr_pages = pageblock_nr_pages; |
| 5283 | arg.pages_found = 0; |
| 5284 | |
| 5285 | /* |
| 5286 | * It may be possible to isolate a pageblock even if the |
| 5287 | * migratetype is not MIGRATE_MOVABLE. The memory isolation |
| 5288 | * notifier chain is used by balloon drivers to return the |
| 5289 | * number of pages in a range that are held by the balloon |
| 5290 | * driver to shrink memory. If all the pages are accounted for |
| 5291 | * by balloons, are free, or on the LRU, isolation can continue. |
| 5292 | * Later, for example, when memory hotplug notifier runs, these |
| 5293 | * pages reported as "can be isolated" should be isolated(freed) |
| 5294 | * by the balloon driver through the memory notifier chain. |
| 5295 | */ |
| 5296 | notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg); |
| 5297 | notifier_ret = notifier_to_errno(notifier_ret); |
| 5298 | if (notifier_ret || !arg.pages_found) |
| 5299 | goto out; |
| 5300 | |
| 5301 | for (iter = pfn; iter < (pfn + pageblock_nr_pages); iter++) { |
| 5302 | if (!pfn_valid_within(pfn)) |
| 5303 | continue; |
| 5304 | |
| 5305 | curr_page = pfn_to_page(iter); |
| 5306 | if (!page_count(curr_page) || PageLRU(curr_page)) |
| 5307 | continue; |
| 5308 | |
| 5309 | immobile++; |
| 5310 | } |
| 5311 | |
| 5312 | if (arg.pages_found == immobile) |
| 5313 | ret = 0; |
| 5314 | |
KAMEZAWA Hiroyuki | a5d76b54 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5315 | out: |
Robert Jennings | 925cc71 | 2009-12-17 14:44:38 +0000 | [diff] [blame] | 5316 | if (!ret) { |
| 5317 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); |
| 5318 | move_freepages_block(zone, page, MIGRATE_ISOLATE); |
| 5319 | } |
| 5320 | |
KAMEZAWA Hiroyuki | a5d76b54 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5321 | spin_unlock_irqrestore(&zone->lock, flags); |
| 5322 | if (!ret) |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 5323 | drain_all_pages(); |
KAMEZAWA Hiroyuki | a5d76b54 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5324 | return ret; |
| 5325 | } |
| 5326 | |
| 5327 | void unset_migratetype_isolate(struct page *page) |
| 5328 | { |
| 5329 | struct zone *zone; |
| 5330 | unsigned long flags; |
| 5331 | zone = page_zone(page); |
| 5332 | spin_lock_irqsave(&zone->lock, flags); |
| 5333 | if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE) |
| 5334 | goto out; |
| 5335 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
| 5336 | move_freepages_block(zone, page, MIGRATE_MOVABLE); |
| 5337 | out: |
| 5338 | spin_unlock_irqrestore(&zone->lock, flags); |
| 5339 | } |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 5340 | |
| 5341 | #ifdef CONFIG_MEMORY_HOTREMOVE |
| 5342 | /* |
| 5343 | * All pages in the range must be isolated before calling this. |
| 5344 | */ |
| 5345 | void |
| 5346 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
| 5347 | { |
| 5348 | struct page *page; |
| 5349 | struct zone *zone; |
| 5350 | int order, i; |
| 5351 | unsigned long pfn; |
| 5352 | unsigned long flags; |
| 5353 | /* find the first valid pfn */ |
| 5354 | for (pfn = start_pfn; pfn < end_pfn; pfn++) |
| 5355 | if (pfn_valid(pfn)) |
| 5356 | break; |
| 5357 | if (pfn == end_pfn) |
| 5358 | return; |
| 5359 | zone = page_zone(pfn_to_page(pfn)); |
| 5360 | spin_lock_irqsave(&zone->lock, flags); |
| 5361 | pfn = start_pfn; |
| 5362 | while (pfn < end_pfn) { |
| 5363 | if (!pfn_valid(pfn)) { |
| 5364 | pfn++; |
| 5365 | continue; |
| 5366 | } |
| 5367 | page = pfn_to_page(pfn); |
| 5368 | BUG_ON(page_count(page)); |
| 5369 | BUG_ON(!PageBuddy(page)); |
| 5370 | order = page_order(page); |
| 5371 | #ifdef CONFIG_DEBUG_VM |
| 5372 | printk(KERN_INFO "remove from free list %lx %d %lx\n", |
| 5373 | pfn, 1 << order, end_pfn); |
| 5374 | #endif |
| 5375 | list_del(&page->lru); |
| 5376 | rmv_page_order(page); |
| 5377 | zone->free_area[order].nr_free--; |
| 5378 | __mod_zone_page_state(zone, NR_FREE_PAGES, |
| 5379 | - (1UL << order)); |
| 5380 | for (i = 0; i < (1 << order); i++) |
| 5381 | SetPageReserved((page+i)); |
| 5382 | pfn += (1 << order); |
| 5383 | } |
| 5384 | spin_unlock_irqrestore(&zone->lock, flags); |
| 5385 | } |
| 5386 | #endif |
Wu Fengguang | 8d22ba1 | 2009-12-16 12:19:58 +0100 | [diff] [blame] | 5387 | |
| 5388 | #ifdef CONFIG_MEMORY_FAILURE |
| 5389 | bool is_free_buddy_page(struct page *page) |
| 5390 | { |
| 5391 | struct zone *zone = page_zone(page); |
| 5392 | unsigned long pfn = page_to_pfn(page); |
| 5393 | unsigned long flags; |
| 5394 | int order; |
| 5395 | |
| 5396 | spin_lock_irqsave(&zone->lock, flags); |
| 5397 | for (order = 0; order < MAX_ORDER; order++) { |
| 5398 | struct page *page_head = page - (pfn & ((1 << order) - 1)); |
| 5399 | |
| 5400 | if (PageBuddy(page_head) && page_order(page_head) >= order) |
| 5401 | break; |
| 5402 | } |
| 5403 | spin_unlock_irqrestore(&zone->lock, flags); |
| 5404 | |
| 5405 | return order < MAX_ORDER; |
| 5406 | } |
| 5407 | #endif |
Wu Fengguang | 718a382 | 2010-03-10 15:20:43 -0800 | [diff] [blame] | 5408 | |
| 5409 | static struct trace_print_flags pageflag_names[] = { |
| 5410 | {1UL << PG_locked, "locked" }, |
| 5411 | {1UL << PG_error, "error" }, |
| 5412 | {1UL << PG_referenced, "referenced" }, |
| 5413 | {1UL << PG_uptodate, "uptodate" }, |
| 5414 | {1UL << PG_dirty, "dirty" }, |
| 5415 | {1UL << PG_lru, "lru" }, |
| 5416 | {1UL << PG_active, "active" }, |
| 5417 | {1UL << PG_slab, "slab" }, |
| 5418 | {1UL << PG_owner_priv_1, "owner_priv_1" }, |
| 5419 | {1UL << PG_arch_1, "arch_1" }, |
| 5420 | {1UL << PG_reserved, "reserved" }, |
| 5421 | {1UL << PG_private, "private" }, |
| 5422 | {1UL << PG_private_2, "private_2" }, |
| 5423 | {1UL << PG_writeback, "writeback" }, |
| 5424 | #ifdef CONFIG_PAGEFLAGS_EXTENDED |
| 5425 | {1UL << PG_head, "head" }, |
| 5426 | {1UL << PG_tail, "tail" }, |
| 5427 | #else |
| 5428 | {1UL << PG_compound, "compound" }, |
| 5429 | #endif |
| 5430 | {1UL << PG_swapcache, "swapcache" }, |
| 5431 | {1UL << PG_mappedtodisk, "mappedtodisk" }, |
| 5432 | {1UL << PG_reclaim, "reclaim" }, |
| 5433 | {1UL << PG_buddy, "buddy" }, |
| 5434 | {1UL << PG_swapbacked, "swapbacked" }, |
| 5435 | {1UL << PG_unevictable, "unevictable" }, |
| 5436 | #ifdef CONFIG_MMU |
| 5437 | {1UL << PG_mlocked, "mlocked" }, |
| 5438 | #endif |
| 5439 | #ifdef CONFIG_ARCH_USES_PG_UNCACHED |
| 5440 | {1UL << PG_uncached, "uncached" }, |
| 5441 | #endif |
| 5442 | #ifdef CONFIG_MEMORY_FAILURE |
| 5443 | {1UL << PG_hwpoison, "hwpoison" }, |
| 5444 | #endif |
| 5445 | {-1UL, NULL }, |
| 5446 | }; |
| 5447 | |
| 5448 | static void dump_page_flags(unsigned long flags) |
| 5449 | { |
| 5450 | const char *delim = ""; |
| 5451 | unsigned long mask; |
| 5452 | int i; |
| 5453 | |
| 5454 | printk(KERN_ALERT "page flags: %#lx(", flags); |
| 5455 | |
| 5456 | /* remove zone id */ |
| 5457 | flags &= (1UL << NR_PAGEFLAGS) - 1; |
| 5458 | |
| 5459 | for (i = 0; pageflag_names[i].name && flags; i++) { |
| 5460 | |
| 5461 | mask = pageflag_names[i].mask; |
| 5462 | if ((flags & mask) != mask) |
| 5463 | continue; |
| 5464 | |
| 5465 | flags &= ~mask; |
| 5466 | printk("%s%s", delim, pageflag_names[i].name); |
| 5467 | delim = "|"; |
| 5468 | } |
| 5469 | |
| 5470 | /* check for left over flags */ |
| 5471 | if (flags) |
| 5472 | printk("%s%#lx", delim, flags); |
| 5473 | |
| 5474 | printk(")\n"); |
| 5475 | } |
| 5476 | |
| 5477 | void dump_page(struct page *page) |
| 5478 | { |
| 5479 | printk(KERN_ALERT |
| 5480 | "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n", |
| 5481 | page, page_count(page), page_mapcount(page), |
| 5482 | page->mapping, page->index); |
| 5483 | dump_page_flags(page->flags); |
| 5484 | } |