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> |
Yinghai Lu | edbe7d2 | 2010-08-25 13:39:16 -0700 | [diff] [blame] | 24 | #include <linux/memblock.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include <linux/compiler.h> |
Randy Dunlap | 9f15833 | 2005-09-13 01:25:16 -0700 | [diff] [blame] | 26 | #include <linux/kernel.h> |
Vegard Nossum | b1eeab6 | 2008-11-25 16:55:53 +0100 | [diff] [blame] | 27 | #include <linux/kmemcheck.h> |
Andrey Ryabinin | b8c73fc | 2015-02-13 14:39:28 -0800 | [diff] [blame] | 28 | #include <linux/kasan.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | #include <linux/module.h> |
| 30 | #include <linux/suspend.h> |
| 31 | #include <linux/pagevec.h> |
| 32 | #include <linux/blkdev.h> |
| 33 | #include <linux/slab.h> |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 34 | #include <linux/ratelimit.h> |
David Rientjes | 5a3135c2 | 2007-10-16 23:25:53 -0700 | [diff] [blame] | 35 | #include <linux/oom.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | #include <linux/notifier.h> |
| 37 | #include <linux/topology.h> |
| 38 | #include <linux/sysctl.h> |
| 39 | #include <linux/cpu.h> |
| 40 | #include <linux/cpuset.h> |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 41 | #include <linux/memory_hotplug.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 42 | #include <linux/nodemask.h> |
| 43 | #include <linux/vmalloc.h> |
KOSAKI Motohiro | a6cccdc | 2011-05-24 17:11:33 -0700 | [diff] [blame] | 44 | #include <linux/vmstat.h> |
Christoph Lameter | 4be38e3 | 2006-01-06 00:11:17 -0800 | [diff] [blame] | 45 | #include <linux/mempolicy.h> |
Dan Williams | 4b94ffd | 2016-01-15 16:56:22 -0800 | [diff] [blame] | 46 | #include <linux/memremap.h> |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 47 | #include <linux/stop_machine.h> |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 48 | #include <linux/sort.h> |
| 49 | #include <linux/pfn.h> |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 50 | #include <linux/backing-dev.h> |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 51 | #include <linux/fault-inject.h> |
KAMEZAWA Hiroyuki | a5d76b5 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 52 | #include <linux/page-isolation.h> |
Joonsoo Kim | eefa864 | 2014-12-12 16:55:46 -0800 | [diff] [blame] | 53 | #include <linux/page_ext.h> |
Thomas Gleixner | 3ac7fe5 | 2008-04-30 00:55:01 -0700 | [diff] [blame] | 54 | #include <linux/debugobjects.h> |
Catalin Marinas | dbb1f81 | 2009-06-11 13:23:19 +0100 | [diff] [blame] | 55 | #include <linux/kmemleak.h> |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 56 | #include <linux/compaction.h> |
Mel Gorman | 0d3d062 | 2009-09-21 17:02:44 -0700 | [diff] [blame] | 57 | #include <trace/events/kmem.h> |
Linus Torvalds | 268bb0c | 2011-05-20 12:50:29 -0700 | [diff] [blame] | 58 | #include <linux/prefetch.h> |
Lisa Du | 6e543d5 | 2013-09-11 14:22:36 -0700 | [diff] [blame] | 59 | #include <linux/mm_inline.h> |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 60 | #include <linux/migrate.h> |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 61 | #include <linux/page_ext.h> |
David Rientjes | 949f7ec | 2013-04-29 15:07:48 -0700 | [diff] [blame] | 62 | #include <linux/hugetlb.h> |
Clark Williams | 8bd75c7 | 2013-02-07 09:47:07 -0600 | [diff] [blame] | 63 | #include <linux/sched/rt.h> |
Joonsoo Kim | 48c96a3 | 2014-12-12 16:56:01 -0800 | [diff] [blame] | 64 | #include <linux/page_owner.h> |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 65 | #include <linux/kthread.h> |
Vladimir Davydov | 4949148 | 2016-07-26 15:24:24 -0700 | [diff] [blame] | 66 | #include <linux/memcontrol.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 67 | |
Jiang Liu | 7ee3d4e | 2013-07-03 15:03:41 -0700 | [diff] [blame] | 68 | #include <asm/sections.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 69 | #include <asm/tlbflush.h> |
Andrew Morton | ac924c6 | 2006-05-15 09:43:59 -0700 | [diff] [blame] | 70 | #include <asm/div64.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 71 | #include "internal.h" |
| 72 | |
Cody P Schafer | c8e251f | 2013-07-03 15:01:29 -0700 | [diff] [blame] | 73 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
| 74 | static DEFINE_MUTEX(pcp_batch_high_lock); |
David Rientjes | 7cd2b0a | 2014-06-23 13:22:04 -0700 | [diff] [blame] | 75 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
Cody P Schafer | c8e251f | 2013-07-03 15:01:29 -0700 | [diff] [blame] | 76 | |
Lee Schermerhorn | 7281201 | 2010-05-26 14:44:56 -0700 | [diff] [blame] | 77 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
| 78 | DEFINE_PER_CPU(int, numa_node); |
| 79 | EXPORT_PER_CPU_SYMBOL(numa_node); |
| 80 | #endif |
| 81 | |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 82 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
| 83 | /* |
| 84 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. |
| 85 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. |
| 86 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() |
| 87 | * defined in <linux/topology.h>. |
| 88 | */ |
| 89 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ |
| 90 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); |
Joonsoo Kim | ad2c814 | 2014-10-09 15:26:13 -0700 | [diff] [blame] | 91 | int _node_numa_mem_[MAX_NUMNODES]; |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 92 | #endif |
| 93 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 94 | /* |
Christoph Lameter | 1380891 | 2007-10-16 01:25:27 -0700 | [diff] [blame] | 95 | * Array of node states. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | */ |
Christoph Lameter | 1380891 | 2007-10-16 01:25:27 -0700 | [diff] [blame] | 97 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
| 98 | [N_POSSIBLE] = NODE_MASK_ALL, |
| 99 | [N_ONLINE] = { { [0] = 1UL } }, |
| 100 | #ifndef CONFIG_NUMA |
| 101 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, |
| 102 | #ifdef CONFIG_HIGHMEM |
| 103 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, |
| 104 | #endif |
Lai Jiangshan | 20b2f52 | 2012-12-12 13:52:00 -0800 | [diff] [blame] | 105 | #ifdef CONFIG_MOVABLE_NODE |
| 106 | [N_MEMORY] = { { [0] = 1UL } }, |
| 107 | #endif |
Christoph Lameter | 1380891 | 2007-10-16 01:25:27 -0700 | [diff] [blame] | 108 | [N_CPU] = { { [0] = 1UL } }, |
| 109 | #endif /* NUMA */ |
| 110 | }; |
| 111 | EXPORT_SYMBOL(node_states); |
| 112 | |
Jiang Liu | c3d5f5f | 2013-07-03 15:03:14 -0700 | [diff] [blame] | 113 | /* Protect totalram_pages and zone->managed_pages */ |
| 114 | static DEFINE_SPINLOCK(managed_page_count_lock); |
| 115 | |
Ravikiran G Thirumalai | 6c231b7 | 2005-09-06 15:17:45 -0700 | [diff] [blame] | 116 | unsigned long totalram_pages __read_mostly; |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 117 | unsigned long totalreserve_pages __read_mostly; |
Pintu Kumar | e48322a | 2014-12-18 16:17:15 -0800 | [diff] [blame] | 118 | unsigned long totalcma_pages __read_mostly; |
Johannes Weiner | ab8fabd | 2012-01-10 15:07:42 -0800 | [diff] [blame] | 119 | |
Hugh Dickins | 1b76b02 | 2012-05-11 01:00:07 -0700 | [diff] [blame] | 120 | int percpu_pagelist_fraction; |
Benjamin Herrenschmidt | dcce284 | 2009-06-18 13:24:12 +1000 | [diff] [blame] | 121 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 122 | |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 123 | /* |
| 124 | * A cached value of the page's pageblock's migratetype, used when the page is |
| 125 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when |
| 126 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. |
| 127 | * Also the migratetype set in the page does not necessarily match the pcplist |
| 128 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any |
| 129 | * other index - this ensures that it will be put on the correct CMA freelist. |
| 130 | */ |
| 131 | static inline int get_pcppage_migratetype(struct page *page) |
| 132 | { |
| 133 | return page->index; |
| 134 | } |
| 135 | |
| 136 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) |
| 137 | { |
| 138 | page->index = migratetype; |
| 139 | } |
| 140 | |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 141 | #ifdef CONFIG_PM_SLEEP |
| 142 | /* |
| 143 | * The following functions are used by the suspend/hibernate code to temporarily |
| 144 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations |
| 145 | * while devices are suspended. To avoid races with the suspend/hibernate code, |
| 146 | * they should always be called with pm_mutex held (gfp_allowed_mask also should |
| 147 | * only be modified with pm_mutex held, unless the suspend/hibernate code is |
| 148 | * guaranteed not to run in parallel with that modification). |
| 149 | */ |
Rafael J. Wysocki | c9e664f | 2010-12-03 22:57:45 +0100 | [diff] [blame] | 150 | |
| 151 | static gfp_t saved_gfp_mask; |
| 152 | |
| 153 | void pm_restore_gfp_mask(void) |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 154 | { |
| 155 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
Rafael J. Wysocki | c9e664f | 2010-12-03 22:57:45 +0100 | [diff] [blame] | 156 | if (saved_gfp_mask) { |
| 157 | gfp_allowed_mask = saved_gfp_mask; |
| 158 | saved_gfp_mask = 0; |
| 159 | } |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 160 | } |
| 161 | |
Rafael J. Wysocki | c9e664f | 2010-12-03 22:57:45 +0100 | [diff] [blame] | 162 | void pm_restrict_gfp_mask(void) |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 163 | { |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 164 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
Rafael J. Wysocki | c9e664f | 2010-12-03 22:57:45 +0100 | [diff] [blame] | 165 | WARN_ON(saved_gfp_mask); |
| 166 | saved_gfp_mask = gfp_allowed_mask; |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 167 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 168 | } |
Mel Gorman | f90ac39 | 2012-01-10 15:07:15 -0800 | [diff] [blame] | 169 | |
| 170 | bool pm_suspended_storage(void) |
| 171 | { |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 172 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
Mel Gorman | f90ac39 | 2012-01-10 15:07:15 -0800 | [diff] [blame] | 173 | return false; |
| 174 | return true; |
| 175 | } |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 176 | #endif /* CONFIG_PM_SLEEP */ |
| 177 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 178 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 179 | unsigned int pageblock_order __read_mostly; |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 180 | #endif |
| 181 | |
Hugh Dickins | d98c7a0 | 2006-02-14 13:52:59 -0800 | [diff] [blame] | 182 | static void __free_pages_ok(struct page *page, unsigned int order); |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 183 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | /* |
| 185 | * results with 256, 32 in the lowmem_reserve sysctl: |
| 186 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) |
| 187 | * 1G machine -> (16M dma, 784M normal, 224M high) |
| 188 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA |
| 189 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL |
Yaowei Bai | 84109e1 | 2015-02-12 15:00:22 -0800 | [diff] [blame] | 190 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
Andi Kleen | a2f1b42 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 191 | * |
| 192 | * TBD: should special case ZONE_DMA32 machines here - in those we normally |
| 193 | * don't need any ZONE_NORMAL reservation |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | */ |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 195 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 196 | #ifdef CONFIG_ZONE_DMA |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 197 | 256, |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 198 | #endif |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 199 | #ifdef CONFIG_ZONE_DMA32 |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 200 | 256, |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 201 | #endif |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 202 | #ifdef CONFIG_HIGHMEM |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 203 | 32, |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 204 | #endif |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 205 | 32, |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 206 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 207 | |
| 208 | EXPORT_SYMBOL(totalram_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 209 | |
Helge Deller | 15ad7cd | 2006-12-06 20:40:36 -0800 | [diff] [blame] | 210 | static char * const zone_names[MAX_NR_ZONES] = { |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 211 | #ifdef CONFIG_ZONE_DMA |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 212 | "DMA", |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 213 | #endif |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 214 | #ifdef CONFIG_ZONE_DMA32 |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 215 | "DMA32", |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 216 | #endif |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 217 | "Normal", |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 218 | #ifdef CONFIG_HIGHMEM |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 219 | "HighMem", |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 220 | #endif |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 221 | "Movable", |
Dan Williams | 033fbae | 2015-08-09 15:29:06 -0400 | [diff] [blame] | 222 | #ifdef CONFIG_ZONE_DEVICE |
| 223 | "Device", |
| 224 | #endif |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 225 | }; |
| 226 | |
Vlastimil Babka | 60f3035 | 2016-03-15 14:56:08 -0700 | [diff] [blame] | 227 | char * const migratetype_names[MIGRATE_TYPES] = { |
| 228 | "Unmovable", |
| 229 | "Movable", |
| 230 | "Reclaimable", |
| 231 | "HighAtomic", |
| 232 | #ifdef CONFIG_CMA |
| 233 | "CMA", |
| 234 | #endif |
| 235 | #ifdef CONFIG_MEMORY_ISOLATION |
| 236 | "Isolate", |
| 237 | #endif |
| 238 | }; |
| 239 | |
Kirill A. Shutemov | f1e6155 | 2015-11-06 16:29:50 -0800 | [diff] [blame] | 240 | compound_page_dtor * const compound_page_dtors[] = { |
| 241 | NULL, |
| 242 | free_compound_page, |
| 243 | #ifdef CONFIG_HUGETLB_PAGE |
| 244 | free_huge_page, |
| 245 | #endif |
Kirill A. Shutemov | 9a98225 | 2016-01-15 16:54:17 -0800 | [diff] [blame] | 246 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 247 | free_transhuge_page, |
| 248 | #endif |
Kirill A. Shutemov | f1e6155 | 2015-11-06 16:29:50 -0800 | [diff] [blame] | 249 | }; |
| 250 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 251 | int min_free_kbytes = 1024; |
Han Pingtian | 42aa83c | 2014-01-23 15:53:28 -0800 | [diff] [blame] | 252 | int user_min_free_kbytes = -1; |
Johannes Weiner | 795ae7a | 2016-03-17 14:19:14 -0700 | [diff] [blame] | 253 | int watermark_scale_factor = 10; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 254 | |
Jan Beulich | 2c85f51 | 2009-09-21 17:03:07 -0700 | [diff] [blame] | 255 | static unsigned long __meminitdata nr_kernel_pages; |
| 256 | static unsigned long __meminitdata nr_all_pages; |
Yasunori Goto | a3142c8 | 2007-05-08 00:23:07 -0700 | [diff] [blame] | 257 | static unsigned long __meminitdata dma_reserve; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 258 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 259 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
| 260 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; |
| 261 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; |
| 262 | static unsigned long __initdata required_kernelcore; |
| 263 | static unsigned long __initdata required_movablecore; |
| 264 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 265 | static bool mirrored_kernelcore; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 266 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 267 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ |
| 268 | int movable_zone; |
| 269 | EXPORT_SYMBOL(movable_zone); |
| 270 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 271 | |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 272 | #if MAX_NUMNODES > 1 |
| 273 | int nr_node_ids __read_mostly = MAX_NUMNODES; |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 274 | int nr_online_nodes __read_mostly = 1; |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 275 | EXPORT_SYMBOL(nr_node_ids); |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 276 | EXPORT_SYMBOL(nr_online_nodes); |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 277 | #endif |
| 278 | |
Mel Gorman | 9ef9acb | 2007-10-16 01:25:54 -0700 | [diff] [blame] | 279 | int page_group_by_mobility_disabled __read_mostly; |
| 280 | |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 281 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
| 282 | static inline void reset_deferred_meminit(pg_data_t *pgdat) |
| 283 | { |
| 284 | pgdat->first_deferred_pfn = ULONG_MAX; |
| 285 | } |
| 286 | |
| 287 | /* Returns true if the struct page for the pfn is uninitialised */ |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 288 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 289 | { |
Mel Gorman | ef70b6f | 2016-07-14 12:07:23 -0700 | [diff] [blame] | 290 | int nid = early_pfn_to_nid(pfn); |
| 291 | |
| 292 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 293 | return true; |
| 294 | |
| 295 | return false; |
| 296 | } |
| 297 | |
| 298 | /* |
| 299 | * Returns false when the remaining initialisation should be deferred until |
| 300 | * later in the boot cycle when it can be parallelised. |
| 301 | */ |
| 302 | static inline bool update_defer_init(pg_data_t *pgdat, |
| 303 | unsigned long pfn, unsigned long zone_end, |
| 304 | unsigned long *nr_initialised) |
| 305 | { |
Li Zhang | 987b309 | 2016-03-17 14:20:16 -0700 | [diff] [blame] | 306 | unsigned long max_initialise; |
| 307 | |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 308 | /* Always populate low zones for address-contrained allocations */ |
| 309 | if (zone_end < pgdat_end_pfn(pgdat)) |
| 310 | return true; |
Li Zhang | 987b309 | 2016-03-17 14:20:16 -0700 | [diff] [blame] | 311 | /* |
| 312 | * Initialise at least 2G of a node but also take into account that |
| 313 | * two large system hashes that can take up 1GB for 0.25TB/node. |
| 314 | */ |
| 315 | max_initialise = max(2UL << (30 - PAGE_SHIFT), |
| 316 | (pgdat->node_spanned_pages >> 8)); |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 317 | |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 318 | (*nr_initialised)++; |
Li Zhang | 987b309 | 2016-03-17 14:20:16 -0700 | [diff] [blame] | 319 | if ((*nr_initialised > max_initialise) && |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 320 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
| 321 | pgdat->first_deferred_pfn = pfn; |
| 322 | return false; |
| 323 | } |
| 324 | |
| 325 | return true; |
| 326 | } |
| 327 | #else |
| 328 | static inline void reset_deferred_meminit(pg_data_t *pgdat) |
| 329 | { |
| 330 | } |
| 331 | |
| 332 | static inline bool early_page_uninitialised(unsigned long pfn) |
| 333 | { |
| 334 | return false; |
| 335 | } |
| 336 | |
| 337 | static inline bool update_defer_init(pg_data_t *pgdat, |
| 338 | unsigned long pfn, unsigned long zone_end, |
| 339 | unsigned long *nr_initialised) |
| 340 | { |
| 341 | return true; |
| 342 | } |
| 343 | #endif |
| 344 | |
Mel Gorman | 0b423ca | 2016-05-19 17:14:27 -0700 | [diff] [blame] | 345 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
| 346 | static inline unsigned long *get_pageblock_bitmap(struct page *page, |
| 347 | unsigned long pfn) |
| 348 | { |
| 349 | #ifdef CONFIG_SPARSEMEM |
| 350 | return __pfn_to_section(pfn)->pageblock_flags; |
| 351 | #else |
| 352 | return page_zone(page)->pageblock_flags; |
| 353 | #endif /* CONFIG_SPARSEMEM */ |
| 354 | } |
| 355 | |
| 356 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) |
| 357 | { |
| 358 | #ifdef CONFIG_SPARSEMEM |
| 359 | pfn &= (PAGES_PER_SECTION-1); |
| 360 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
| 361 | #else |
| 362 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); |
| 363 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
| 364 | #endif /* CONFIG_SPARSEMEM */ |
| 365 | } |
| 366 | |
| 367 | /** |
| 368 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages |
| 369 | * @page: The page within the block of interest |
| 370 | * @pfn: The target page frame number |
| 371 | * @end_bitidx: The last bit of interest to retrieve |
| 372 | * @mask: mask of bits that the caller is interested in |
| 373 | * |
| 374 | * Return: pageblock_bits flags |
| 375 | */ |
| 376 | static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page, |
| 377 | unsigned long pfn, |
| 378 | unsigned long end_bitidx, |
| 379 | unsigned long mask) |
| 380 | { |
| 381 | unsigned long *bitmap; |
| 382 | unsigned long bitidx, word_bitidx; |
| 383 | unsigned long word; |
| 384 | |
| 385 | bitmap = get_pageblock_bitmap(page, pfn); |
| 386 | bitidx = pfn_to_bitidx(page, pfn); |
| 387 | word_bitidx = bitidx / BITS_PER_LONG; |
| 388 | bitidx &= (BITS_PER_LONG-1); |
| 389 | |
| 390 | word = bitmap[word_bitidx]; |
| 391 | bitidx += end_bitidx; |
| 392 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; |
| 393 | } |
| 394 | |
| 395 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, |
| 396 | unsigned long end_bitidx, |
| 397 | unsigned long mask) |
| 398 | { |
| 399 | return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask); |
| 400 | } |
| 401 | |
| 402 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) |
| 403 | { |
| 404 | return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK); |
| 405 | } |
| 406 | |
| 407 | /** |
| 408 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages |
| 409 | * @page: The page within the block of interest |
| 410 | * @flags: The flags to set |
| 411 | * @pfn: The target page frame number |
| 412 | * @end_bitidx: The last bit of interest |
| 413 | * @mask: mask of bits that the caller is interested in |
| 414 | */ |
| 415 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, |
| 416 | unsigned long pfn, |
| 417 | unsigned long end_bitidx, |
| 418 | unsigned long mask) |
| 419 | { |
| 420 | unsigned long *bitmap; |
| 421 | unsigned long bitidx, word_bitidx; |
| 422 | unsigned long old_word, word; |
| 423 | |
| 424 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); |
| 425 | |
| 426 | bitmap = get_pageblock_bitmap(page, pfn); |
| 427 | bitidx = pfn_to_bitidx(page, pfn); |
| 428 | word_bitidx = bitidx / BITS_PER_LONG; |
| 429 | bitidx &= (BITS_PER_LONG-1); |
| 430 | |
| 431 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); |
| 432 | |
| 433 | bitidx += end_bitidx; |
| 434 | mask <<= (BITS_PER_LONG - bitidx - 1); |
| 435 | flags <<= (BITS_PER_LONG - bitidx - 1); |
| 436 | |
| 437 | word = READ_ONCE(bitmap[word_bitidx]); |
| 438 | for (;;) { |
| 439 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); |
| 440 | if (word == old_word) |
| 441 | break; |
| 442 | word = old_word; |
| 443 | } |
| 444 | } |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 445 | |
Minchan Kim | ee6f509 | 2012-07-31 16:43:50 -0700 | [diff] [blame] | 446 | void set_pageblock_migratetype(struct page *page, int migratetype) |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 447 | { |
KOSAKI Motohiro | 5d0f3f7 | 2013-11-12 15:08:18 -0800 | [diff] [blame] | 448 | if (unlikely(page_group_by_mobility_disabled && |
| 449 | migratetype < MIGRATE_PCPTYPES)) |
Mel Gorman | 49255c6 | 2009-06-16 15:31:58 -0700 | [diff] [blame] | 450 | migratetype = MIGRATE_UNMOVABLE; |
| 451 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 452 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
| 453 | PB_migrate, PB_migrate_end); |
| 454 | } |
| 455 | |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 456 | #ifdef CONFIG_DEBUG_VM |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 457 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 458 | { |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 459 | int ret = 0; |
| 460 | unsigned seq; |
| 461 | unsigned long pfn = page_to_pfn(page); |
Cody P Schafer | b5e6a5a | 2013-02-22 16:35:28 -0800 | [diff] [blame] | 462 | unsigned long sp, start_pfn; |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 463 | |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 464 | do { |
| 465 | seq = zone_span_seqbegin(zone); |
Cody P Schafer | b5e6a5a | 2013-02-22 16:35:28 -0800 | [diff] [blame] | 466 | start_pfn = zone->zone_start_pfn; |
| 467 | sp = zone->spanned_pages; |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 468 | if (!zone_spans_pfn(zone, pfn)) |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 469 | ret = 1; |
| 470 | } while (zone_span_seqretry(zone, seq)); |
| 471 | |
Cody P Schafer | b5e6a5a | 2013-02-22 16:35:28 -0800 | [diff] [blame] | 472 | if (ret) |
Dave Hansen | 613813e | 2014-06-04 16:07:27 -0700 | [diff] [blame] | 473 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
| 474 | pfn, zone_to_nid(zone), zone->name, |
| 475 | start_pfn, start_pfn + sp); |
Cody P Schafer | b5e6a5a | 2013-02-22 16:35:28 -0800 | [diff] [blame] | 476 | |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 477 | return ret; |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 478 | } |
| 479 | |
| 480 | static int page_is_consistent(struct zone *zone, struct page *page) |
| 481 | { |
Andy Whitcroft | 14e0729 | 2007-05-06 14:49:14 -0700 | [diff] [blame] | 482 | if (!pfn_valid_within(page_to_pfn(page))) |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 483 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 484 | if (zone != page_zone(page)) |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 485 | return 0; |
| 486 | |
| 487 | return 1; |
| 488 | } |
| 489 | /* |
| 490 | * Temporary debugging check for pages not lying within a given zone. |
| 491 | */ |
| 492 | static int bad_range(struct zone *zone, struct page *page) |
| 493 | { |
| 494 | if (page_outside_zone_boundaries(zone, page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 495 | return 1; |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 496 | if (!page_is_consistent(zone, page)) |
| 497 | return 1; |
| 498 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 499 | return 0; |
| 500 | } |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 501 | #else |
| 502 | static inline int bad_range(struct zone *zone, struct page *page) |
| 503 | { |
| 504 | return 0; |
| 505 | } |
| 506 | #endif |
| 507 | |
Kirill A. Shutemov | d230dec | 2014-04-07 15:37:38 -0700 | [diff] [blame] | 508 | static void bad_page(struct page *page, const char *reason, |
| 509 | unsigned long bad_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 510 | { |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 511 | static unsigned long resume; |
| 512 | static unsigned long nr_shown; |
| 513 | static unsigned long nr_unshown; |
| 514 | |
| 515 | /* |
| 516 | * Allow a burst of 60 reports, then keep quiet for that minute; |
| 517 | * or allow a steady drip of one report per second. |
| 518 | */ |
| 519 | if (nr_shown == 60) { |
| 520 | if (time_before(jiffies, resume)) { |
| 521 | nr_unshown++; |
| 522 | goto out; |
| 523 | } |
| 524 | if (nr_unshown) { |
Vlastimil Babka | ff8e811 | 2016-03-15 14:56:24 -0700 | [diff] [blame] | 525 | pr_alert( |
Hugh Dickins | 1e9e636 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 526 | "BUG: Bad page state: %lu messages suppressed\n", |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 527 | nr_unshown); |
| 528 | nr_unshown = 0; |
| 529 | } |
| 530 | nr_shown = 0; |
| 531 | } |
| 532 | if (nr_shown++ == 0) |
| 533 | resume = jiffies + 60 * HZ; |
| 534 | |
Vlastimil Babka | ff8e811 | 2016-03-15 14:56:24 -0700 | [diff] [blame] | 535 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 536 | current->comm, page_to_pfn(page)); |
Vlastimil Babka | ff8e811 | 2016-03-15 14:56:24 -0700 | [diff] [blame] | 537 | __dump_page(page, reason); |
| 538 | bad_flags &= page->flags; |
| 539 | if (bad_flags) |
| 540 | pr_alert("bad because of flags: %#lx(%pGp)\n", |
| 541 | bad_flags, &bad_flags); |
Vlastimil Babka | 4e46211 | 2016-03-15 14:56:21 -0700 | [diff] [blame] | 542 | dump_page_owner(page); |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 543 | |
Dave Jones | 4f31888 | 2011-10-31 17:07:24 -0700 | [diff] [blame] | 544 | print_modules(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 545 | dump_stack(); |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 546 | out: |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 547 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
Mel Gorman | 22b751c | 2013-02-22 16:34:59 -0800 | [diff] [blame] | 548 | page_mapcount_reset(page); /* remove PageBuddy */ |
Rusty Russell | 373d4d0 | 2013-01-21 17:17:39 +1030 | [diff] [blame] | 549 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 550 | } |
| 551 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 552 | /* |
| 553 | * Higher-order pages are called "compound pages". They are structured thusly: |
| 554 | * |
Kirill A. Shutemov | 1d798ca | 2015-11-06 16:29:54 -0800 | [diff] [blame] | 555 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 556 | * |
Kirill A. Shutemov | 1d798ca | 2015-11-06 16:29:54 -0800 | [diff] [blame] | 557 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
| 558 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 559 | * |
Kirill A. Shutemov | 1d798ca | 2015-11-06 16:29:54 -0800 | [diff] [blame] | 560 | * The first tail page's ->compound_dtor holds the offset in array of compound |
| 561 | * page destructors. See compound_page_dtors. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 562 | * |
Kirill A. Shutemov | 1d798ca | 2015-11-06 16:29:54 -0800 | [diff] [blame] | 563 | * The first tail page's ->compound_order holds the order of allocation. |
Hugh Dickins | 41d78ba | 2006-02-14 13:52:58 -0800 | [diff] [blame] | 564 | * This usage means that zero-order pages may not be compound. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 565 | */ |
Hugh Dickins | d98c7a0 | 2006-02-14 13:52:59 -0800 | [diff] [blame] | 566 | |
Kirill A. Shutemov | 9a98225 | 2016-01-15 16:54:17 -0800 | [diff] [blame] | 567 | void free_compound_page(struct page *page) |
Hugh Dickins | d98c7a0 | 2006-02-14 13:52:59 -0800 | [diff] [blame] | 568 | { |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 569 | __free_pages_ok(page, compound_order(page)); |
Hugh Dickins | d98c7a0 | 2006-02-14 13:52:59 -0800 | [diff] [blame] | 570 | } |
| 571 | |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 572 | void prep_compound_page(struct page *page, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 573 | { |
| 574 | int i; |
| 575 | int nr_pages = 1 << order; |
| 576 | |
Kirill A. Shutemov | f1e6155 | 2015-11-06 16:29:50 -0800 | [diff] [blame] | 577 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 578 | set_compound_order(page, order); |
Christoph Lameter | 6d77795 | 2007-05-06 14:49:40 -0700 | [diff] [blame] | 579 | __SetPageHead(page); |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 580 | for (i = 1; i < nr_pages; i++) { |
| 581 | struct page *p = page + i; |
Youquan Song | 58a84aa | 2011-12-08 14:34:18 -0800 | [diff] [blame] | 582 | set_page_count(p, 0); |
Kirill A. Shutemov | 1c290f6 | 2016-01-15 16:52:07 -0800 | [diff] [blame] | 583 | p->mapping = TAIL_MAPPING; |
Kirill A. Shutemov | 1d798ca | 2015-11-06 16:29:54 -0800 | [diff] [blame] | 584 | set_compound_head(p, page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 585 | } |
Kirill A. Shutemov | 53f9263 | 2016-01-15 16:53:42 -0800 | [diff] [blame] | 586 | atomic_set(compound_mapcount_ptr(page), -1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 587 | } |
| 588 | |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 589 | #ifdef CONFIG_DEBUG_PAGEALLOC |
| 590 | unsigned int _debug_guardpage_minorder; |
Christian Borntraeger | ea6eabb | 2016-03-15 14:55:30 -0700 | [diff] [blame] | 591 | bool _debug_pagealloc_enabled __read_mostly |
| 592 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); |
Joonsoo Kim | 505f6d2 | 2016-03-17 14:17:56 -0700 | [diff] [blame] | 593 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 594 | bool _debug_guardpage_enabled __read_mostly; |
| 595 | |
Joonsoo Kim | 031bc57 | 2014-12-12 16:55:52 -0800 | [diff] [blame] | 596 | static int __init early_debug_pagealloc(char *buf) |
| 597 | { |
| 598 | if (!buf) |
| 599 | return -EINVAL; |
Minfei Huang | 2a138dc | 2016-05-20 16:58:13 -0700 | [diff] [blame] | 600 | return kstrtobool(buf, &_debug_pagealloc_enabled); |
Joonsoo Kim | 031bc57 | 2014-12-12 16:55:52 -0800 | [diff] [blame] | 601 | } |
| 602 | early_param("debug_pagealloc", early_debug_pagealloc); |
| 603 | |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 604 | static bool need_debug_guardpage(void) |
| 605 | { |
Joonsoo Kim | 031bc57 | 2014-12-12 16:55:52 -0800 | [diff] [blame] | 606 | /* If we don't use debug_pagealloc, we don't need guard page */ |
| 607 | if (!debug_pagealloc_enabled()) |
| 608 | return false; |
| 609 | |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 610 | return true; |
| 611 | } |
| 612 | |
| 613 | static void init_debug_guardpage(void) |
| 614 | { |
Joonsoo Kim | 031bc57 | 2014-12-12 16:55:52 -0800 | [diff] [blame] | 615 | if (!debug_pagealloc_enabled()) |
| 616 | return; |
| 617 | |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 618 | _debug_guardpage_enabled = true; |
| 619 | } |
| 620 | |
| 621 | struct page_ext_operations debug_guardpage_ops = { |
| 622 | .need = need_debug_guardpage, |
| 623 | .init = init_debug_guardpage, |
| 624 | }; |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 625 | |
| 626 | static int __init debug_guardpage_minorder_setup(char *buf) |
| 627 | { |
| 628 | unsigned long res; |
| 629 | |
| 630 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 631 | pr_err("Bad debug_guardpage_minorder value\n"); |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 632 | return 0; |
| 633 | } |
| 634 | _debug_guardpage_minorder = res; |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 635 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 636 | return 0; |
| 637 | } |
| 638 | __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup); |
| 639 | |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 640 | static inline void set_page_guard(struct zone *zone, struct page *page, |
| 641 | unsigned int order, int migratetype) |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 642 | { |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 643 | struct page_ext *page_ext; |
| 644 | |
| 645 | if (!debug_guardpage_enabled()) |
| 646 | return; |
| 647 | |
| 648 | page_ext = lookup_page_ext(page); |
Yang Shi | f86e427 | 2016-06-03 14:55:38 -0700 | [diff] [blame] | 649 | if (unlikely(!page_ext)) |
| 650 | return; |
| 651 | |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 652 | __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
| 653 | |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 654 | INIT_LIST_HEAD(&page->lru); |
| 655 | set_page_private(page, order); |
| 656 | /* Guard pages are not available for any usage */ |
| 657 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 658 | } |
| 659 | |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 660 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
| 661 | unsigned int order, int migratetype) |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 662 | { |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 663 | struct page_ext *page_ext; |
| 664 | |
| 665 | if (!debug_guardpage_enabled()) |
| 666 | return; |
| 667 | |
| 668 | page_ext = lookup_page_ext(page); |
Yang Shi | f86e427 | 2016-06-03 14:55:38 -0700 | [diff] [blame] | 669 | if (unlikely(!page_ext)) |
| 670 | return; |
| 671 | |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 672 | __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
| 673 | |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 674 | set_page_private(page, 0); |
| 675 | if (!is_migrate_isolate(migratetype)) |
| 676 | __mod_zone_freepage_state(zone, (1 << order), migratetype); |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 677 | } |
| 678 | #else |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 679 | struct page_ext_operations debug_guardpage_ops = { NULL, }; |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 680 | static inline void set_page_guard(struct zone *zone, struct page *page, |
| 681 | unsigned int order, int migratetype) {} |
| 682 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
| 683 | unsigned int order, int migratetype) {} |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 684 | #endif |
| 685 | |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 686 | static inline void set_page_order(struct page *page, unsigned int order) |
Andrew Morton | 6aa3001b2 | 2006-04-18 22:20:52 -0700 | [diff] [blame] | 687 | { |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 688 | set_page_private(page, order); |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 689 | __SetPageBuddy(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 690 | } |
| 691 | |
| 692 | static inline void rmv_page_order(struct page *page) |
| 693 | { |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 694 | __ClearPageBuddy(page); |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 695 | set_page_private(page, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 696 | } |
| 697 | |
| 698 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 699 | * This function checks whether a page is free && is the buddy |
| 700 | * we can do coalesce a page and its buddy if |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 701 | * (a) the buddy is not in a hole && |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 702 | * (b) the buddy is in the buddy system && |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 703 | * (c) a page and its buddy have the same order && |
| 704 | * (d) a page and its buddy are in the same zone. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 705 | * |
Wang Sheng-Hui | cf6fe94 | 2013-09-11 14:22:48 -0700 | [diff] [blame] | 706 | * For recording whether a page is in the buddy system, we set ->_mapcount |
| 707 | * PAGE_BUDDY_MAPCOUNT_VALUE. |
| 708 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is |
| 709 | * serialized by zone->lock. |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 710 | * |
| 711 | * For recording page's order, we use page_private(page). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 712 | */ |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 713 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 714 | unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 715 | { |
Andy Whitcroft | 14e0729 | 2007-05-06 14:49:14 -0700 | [diff] [blame] | 716 | if (!pfn_valid_within(page_to_pfn(buddy))) |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 717 | return 0; |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 718 | |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 719 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
Mel Gorman | d34c5fa | 2014-06-04 16:10:10 -0700 | [diff] [blame] | 720 | if (page_zone_id(page) != page_zone_id(buddy)) |
| 721 | return 0; |
| 722 | |
Weijie Yang | 4c5018c | 2015-02-10 14:11:39 -0800 | [diff] [blame] | 723 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
| 724 | |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 725 | return 1; |
| 726 | } |
| 727 | |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 728 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
Mel Gorman | d34c5fa | 2014-06-04 16:10:10 -0700 | [diff] [blame] | 729 | /* |
| 730 | * zone check is done late to avoid uselessly |
| 731 | * calculating zone/node ids for pages that could |
| 732 | * never merge. |
| 733 | */ |
| 734 | if (page_zone_id(page) != page_zone_id(buddy)) |
| 735 | return 0; |
| 736 | |
Weijie Yang | 4c5018c | 2015-02-10 14:11:39 -0800 | [diff] [blame] | 737 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
| 738 | |
Andrew Morton | 6aa3001b2 | 2006-04-18 22:20:52 -0700 | [diff] [blame] | 739 | return 1; |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 740 | } |
Andrew Morton | 6aa3001b2 | 2006-04-18 22:20:52 -0700 | [diff] [blame] | 741 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 742 | } |
| 743 | |
| 744 | /* |
| 745 | * Freeing function for a buddy system allocator. |
| 746 | * |
| 747 | * The concept of a buddy system is to maintain direct-mapped table |
| 748 | * (containing bit values) for memory blocks of various "orders". |
| 749 | * The bottom level table contains the map for the smallest allocatable |
| 750 | * units of memory (here, pages), and each level above it describes |
| 751 | * pairs of units from the levels below, hence, "buddies". |
| 752 | * At a high level, all that happens here is marking the table entry |
| 753 | * at the bottom level available, and propagating the changes upward |
| 754 | * as necessary, plus some accounting needed to play nicely with other |
| 755 | * parts of the VM system. |
| 756 | * At each level, we keep a list of pages, which are heads of continuous |
Wang Sheng-Hui | cf6fe94 | 2013-09-11 14:22:48 -0700 | [diff] [blame] | 757 | * free pages of length of (1 << order) and marked with _mapcount |
| 758 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) |
| 759 | * field. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 760 | * So when we are allocating or freeing one, we can derive the state of the |
Michal Nazarewicz | 5f63b72 | 2012-01-11 15:16:11 +0100 | [diff] [blame] | 761 | * other. That is, if we allocate a small block, and both were |
| 762 | * free, the remainder of the region must be split into blocks. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 763 | * If a block is freed, and its buddy is also free, then this |
Michal Nazarewicz | 5f63b72 | 2012-01-11 15:16:11 +0100 | [diff] [blame] | 764 | * triggers coalescing into a block of larger size. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 765 | * |
Nadia Yvette Chambers | 6d49e35 | 2012-12-06 10:39:54 +0100 | [diff] [blame] | 766 | * -- nyc |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 767 | */ |
| 768 | |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 769 | static inline void __free_one_page(struct page *page, |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 770 | unsigned long pfn, |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 771 | struct zone *zone, unsigned int order, |
| 772 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 773 | { |
| 774 | unsigned long page_idx; |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 775 | unsigned long combined_idx; |
KyongHo Cho | 43506fa | 2011-01-13 15:47:24 -0800 | [diff] [blame] | 776 | unsigned long uninitialized_var(buddy_idx); |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 777 | struct page *buddy; |
Vlastimil Babka | d9dddbf | 2016-03-25 14:21:50 -0700 | [diff] [blame] | 778 | unsigned int max_order; |
| 779 | |
| 780 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 781 | |
Cody P Schafer | d29bb97 | 2013-02-22 16:35:25 -0800 | [diff] [blame] | 782 | VM_BUG_ON(!zone_is_initialized(zone)); |
Kirill A. Shutemov | 6e9f0d5 | 2015-02-11 15:25:50 -0800 | [diff] [blame] | 783 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 784 | |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 785 | VM_BUG_ON(migratetype == -1); |
Vlastimil Babka | d9dddbf | 2016-03-25 14:21:50 -0700 | [diff] [blame] | 786 | if (likely(!is_migrate_isolate(migratetype))) |
Joonsoo Kim | 8f82b55 | 2014-11-13 15:19:18 -0800 | [diff] [blame] | 787 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 788 | |
Vlastimil Babka | d9dddbf | 2016-03-25 14:21:50 -0700 | [diff] [blame] | 789 | page_idx = pfn & ((1 << MAX_ORDER) - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 790 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 791 | VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page); |
| 792 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 793 | |
Vlastimil Babka | d9dddbf | 2016-03-25 14:21:50 -0700 | [diff] [blame] | 794 | continue_merging: |
Joonsoo Kim | 3c60509 | 2014-11-13 15:19:21 -0800 | [diff] [blame] | 795 | while (order < max_order - 1) { |
KyongHo Cho | 43506fa | 2011-01-13 15:47:24 -0800 | [diff] [blame] | 796 | buddy_idx = __find_buddy_index(page_idx, order); |
| 797 | buddy = page + (buddy_idx - page_idx); |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 798 | if (!page_is_buddy(page, buddy, order)) |
Vlastimil Babka | d9dddbf | 2016-03-25 14:21:50 -0700 | [diff] [blame] | 799 | goto done_merging; |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 800 | /* |
| 801 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, |
| 802 | * merge with it and move up one order. |
| 803 | */ |
| 804 | if (page_is_guard(buddy)) { |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 805 | clear_page_guard(zone, buddy, order, migratetype); |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 806 | } else { |
| 807 | list_del(&buddy->lru); |
| 808 | zone->free_area[order].nr_free--; |
| 809 | rmv_page_order(buddy); |
| 810 | } |
KyongHo Cho | 43506fa | 2011-01-13 15:47:24 -0800 | [diff] [blame] | 811 | combined_idx = buddy_idx & page_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 812 | page = page + (combined_idx - page_idx); |
| 813 | page_idx = combined_idx; |
| 814 | order++; |
| 815 | } |
Vlastimil Babka | d9dddbf | 2016-03-25 14:21:50 -0700 | [diff] [blame] | 816 | if (max_order < MAX_ORDER) { |
| 817 | /* If we are here, it means order is >= pageblock_order. |
| 818 | * We want to prevent merge between freepages on isolate |
| 819 | * pageblock and normal pageblock. Without this, pageblock |
| 820 | * isolation could cause incorrect freepage or CMA accounting. |
| 821 | * |
| 822 | * We don't want to hit this code for the more frequent |
| 823 | * low-order merging. |
| 824 | */ |
| 825 | if (unlikely(has_isolate_pageblock(zone))) { |
| 826 | int buddy_mt; |
| 827 | |
| 828 | buddy_idx = __find_buddy_index(page_idx, order); |
| 829 | buddy = page + (buddy_idx - page_idx); |
| 830 | buddy_mt = get_pageblock_migratetype(buddy); |
| 831 | |
| 832 | if (migratetype != buddy_mt |
| 833 | && (is_migrate_isolate(migratetype) || |
| 834 | is_migrate_isolate(buddy_mt))) |
| 835 | goto done_merging; |
| 836 | } |
| 837 | max_order++; |
| 838 | goto continue_merging; |
| 839 | } |
| 840 | |
| 841 | done_merging: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 842 | set_page_order(page, order); |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 843 | |
| 844 | /* |
| 845 | * If this is not the largest possible page, check if the buddy |
| 846 | * of the next-highest order is free. If it is, it's possible |
| 847 | * that pages are being freed that will coalesce soon. In case, |
| 848 | * that is happening, add the free page to the tail of the list |
| 849 | * so it's less likely to be used soon and more likely to be merged |
| 850 | * as a higher order page |
| 851 | */ |
Mel Gorman | b7f50cf | 2010-10-26 14:21:11 -0700 | [diff] [blame] | 852 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 853 | struct page *higher_page, *higher_buddy; |
KyongHo Cho | 43506fa | 2011-01-13 15:47:24 -0800 | [diff] [blame] | 854 | combined_idx = buddy_idx & page_idx; |
| 855 | higher_page = page + (combined_idx - page_idx); |
| 856 | buddy_idx = __find_buddy_index(combined_idx, order + 1); |
Li Haifeng | 0ba8f2d | 2012-09-17 14:09:21 -0700 | [diff] [blame] | 857 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 858 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
| 859 | list_add_tail(&page->lru, |
| 860 | &zone->free_area[order].free_list[migratetype]); |
| 861 | goto out; |
| 862 | } |
| 863 | } |
| 864 | |
| 865 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); |
| 866 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 867 | zone->free_area[order].nr_free++; |
| 868 | } |
| 869 | |
Mel Gorman | 7bfec6f | 2016-05-19 17:14:15 -0700 | [diff] [blame] | 870 | /* |
| 871 | * A bad page could be due to a number of fields. Instead of multiple branches, |
| 872 | * try and check multiple fields with one check. The caller must do a detailed |
| 873 | * check if necessary. |
| 874 | */ |
| 875 | static inline bool page_expected_state(struct page *page, |
| 876 | unsigned long check_flags) |
| 877 | { |
| 878 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
| 879 | return false; |
| 880 | |
| 881 | if (unlikely((unsigned long)page->mapping | |
| 882 | page_ref_count(page) | |
| 883 | #ifdef CONFIG_MEMCG |
| 884 | (unsigned long)page->mem_cgroup | |
| 885 | #endif |
| 886 | (page->flags & check_flags))) |
| 887 | return false; |
| 888 | |
| 889 | return true; |
| 890 | } |
| 891 | |
Mel Gorman | bb552ac | 2016-05-19 17:14:18 -0700 | [diff] [blame] | 892 | static void free_pages_check_bad(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 893 | { |
Mel Gorman | 7bfec6f | 2016-05-19 17:14:15 -0700 | [diff] [blame] | 894 | const char *bad_reason; |
| 895 | unsigned long bad_flags; |
| 896 | |
Mel Gorman | 7bfec6f | 2016-05-19 17:14:15 -0700 | [diff] [blame] | 897 | bad_reason = NULL; |
| 898 | bad_flags = 0; |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 899 | |
Kirill A. Shutemov | 53f9263 | 2016-01-15 16:53:42 -0800 | [diff] [blame] | 900 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 901 | bad_reason = "nonzero mapcount"; |
| 902 | if (unlikely(page->mapping != NULL)) |
| 903 | bad_reason = "non-NULL mapping"; |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 904 | if (unlikely(page_ref_count(page) != 0)) |
Joonsoo Kim | 0139aa7 | 2016-05-19 17:10:49 -0700 | [diff] [blame] | 905 | bad_reason = "nonzero _refcount"; |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 906 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { |
| 907 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; |
| 908 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; |
| 909 | } |
Johannes Weiner | 9edad6e | 2014-12-10 15:44:58 -0800 | [diff] [blame] | 910 | #ifdef CONFIG_MEMCG |
| 911 | if (unlikely(page->mem_cgroup)) |
| 912 | bad_reason = "page still charged to cgroup"; |
| 913 | #endif |
Mel Gorman | 7bfec6f | 2016-05-19 17:14:15 -0700 | [diff] [blame] | 914 | bad_page(page, bad_reason, bad_flags); |
Mel Gorman | bb552ac | 2016-05-19 17:14:18 -0700 | [diff] [blame] | 915 | } |
| 916 | |
| 917 | static inline int free_pages_check(struct page *page) |
| 918 | { |
Mel Gorman | da838d4 | 2016-05-19 17:14:21 -0700 | [diff] [blame] | 919 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
Mel Gorman | bb552ac | 2016-05-19 17:14:18 -0700 | [diff] [blame] | 920 | return 0; |
Mel Gorman | bb552ac | 2016-05-19 17:14:18 -0700 | [diff] [blame] | 921 | |
| 922 | /* Something has gone sideways, find it */ |
| 923 | free_pages_check_bad(page); |
Mel Gorman | 7bfec6f | 2016-05-19 17:14:15 -0700 | [diff] [blame] | 924 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 925 | } |
| 926 | |
Mel Gorman | 4db7548 | 2016-05-19 17:14:32 -0700 | [diff] [blame] | 927 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
| 928 | { |
| 929 | int ret = 1; |
| 930 | |
| 931 | /* |
| 932 | * We rely page->lru.next never has bit 0 set, unless the page |
| 933 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. |
| 934 | */ |
| 935 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); |
| 936 | |
| 937 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { |
| 938 | ret = 0; |
| 939 | goto out; |
| 940 | } |
| 941 | switch (page - head_page) { |
| 942 | case 1: |
| 943 | /* the first tail page: ->mapping is compound_mapcount() */ |
| 944 | if (unlikely(compound_mapcount(page))) { |
| 945 | bad_page(page, "nonzero compound_mapcount", 0); |
| 946 | goto out; |
| 947 | } |
| 948 | break; |
| 949 | case 2: |
| 950 | /* |
| 951 | * the second tail page: ->mapping is |
| 952 | * page_deferred_list().next -- ignore value. |
| 953 | */ |
| 954 | break; |
| 955 | default: |
| 956 | if (page->mapping != TAIL_MAPPING) { |
| 957 | bad_page(page, "corrupted mapping in tail page", 0); |
| 958 | goto out; |
| 959 | } |
| 960 | break; |
| 961 | } |
| 962 | if (unlikely(!PageTail(page))) { |
| 963 | bad_page(page, "PageTail not set", 0); |
| 964 | goto out; |
| 965 | } |
| 966 | if (unlikely(compound_head(page) != head_page)) { |
| 967 | bad_page(page, "compound_head not consistent", 0); |
| 968 | goto out; |
| 969 | } |
| 970 | ret = 0; |
| 971 | out: |
| 972 | page->mapping = NULL; |
| 973 | clear_compound_head(page); |
| 974 | return ret; |
| 975 | } |
| 976 | |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 977 | static __always_inline bool free_pages_prepare(struct page *page, |
| 978 | unsigned int order, bool check_free) |
| 979 | { |
| 980 | int bad = 0; |
| 981 | |
| 982 | VM_BUG_ON_PAGE(PageTail(page), page); |
| 983 | |
| 984 | trace_mm_page_free(page, order); |
| 985 | kmemcheck_free_shadow(page, order); |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 986 | |
| 987 | /* |
| 988 | * Check tail pages before head page information is cleared to |
| 989 | * avoid checking PageCompound for order-0 pages. |
| 990 | */ |
| 991 | if (unlikely(order)) { |
| 992 | bool compound = PageCompound(page); |
| 993 | int i; |
| 994 | |
| 995 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); |
| 996 | |
Kirill A. Shutemov | 9a73f61 | 2016-07-26 15:25:53 -0700 | [diff] [blame] | 997 | if (compound) |
| 998 | ClearPageDoubleMap(page); |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 999 | for (i = 1; i < (1 << order); i++) { |
| 1000 | if (compound) |
| 1001 | bad += free_tail_pages_check(page, page + i); |
| 1002 | if (unlikely(free_pages_check(page + i))) { |
| 1003 | bad++; |
| 1004 | continue; |
| 1005 | } |
| 1006 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; |
| 1007 | } |
| 1008 | } |
Minchan Kim | bda807d | 2016-07-26 15:23:05 -0700 | [diff] [blame] | 1009 | if (PageMappingFlags(page)) |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 1010 | page->mapping = NULL; |
Vladimir Davydov | 4949148 | 2016-07-26 15:24:24 -0700 | [diff] [blame] | 1011 | if (memcg_kmem_enabled() && PageKmemcg(page)) { |
| 1012 | memcg_kmem_uncharge(page, order); |
| 1013 | __ClearPageKmemcg(page); |
| 1014 | } |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 1015 | if (check_free) |
| 1016 | bad += free_pages_check(page); |
| 1017 | if (bad) |
| 1018 | return false; |
| 1019 | |
| 1020 | page_cpupid_reset_last(page); |
| 1021 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; |
| 1022 | reset_page_owner(page, order); |
| 1023 | |
| 1024 | if (!PageHighMem(page)) { |
| 1025 | debug_check_no_locks_freed(page_address(page), |
| 1026 | PAGE_SIZE << order); |
| 1027 | debug_check_no_obj_freed(page_address(page), |
| 1028 | PAGE_SIZE << order); |
| 1029 | } |
| 1030 | arch_free_page(page, order); |
| 1031 | kernel_poison_pages(page, 1 << order, 0); |
| 1032 | kernel_map_pages(page, 1 << order, 0); |
seokhoon.yoon | 29b52de | 2016-05-20 16:58:47 -0700 | [diff] [blame] | 1033 | kasan_free_pages(page, order); |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 1034 | |
| 1035 | return true; |
| 1036 | } |
Mel Gorman | 4db7548 | 2016-05-19 17:14:32 -0700 | [diff] [blame] | 1037 | |
| 1038 | #ifdef CONFIG_DEBUG_VM |
| 1039 | static inline bool free_pcp_prepare(struct page *page) |
| 1040 | { |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 1041 | return free_pages_prepare(page, 0, true); |
Mel Gorman | 4db7548 | 2016-05-19 17:14:32 -0700 | [diff] [blame] | 1042 | } |
| 1043 | |
| 1044 | static inline bool bulkfree_pcp_prepare(struct page *page) |
| 1045 | { |
| 1046 | return false; |
| 1047 | } |
| 1048 | #else |
| 1049 | static bool free_pcp_prepare(struct page *page) |
| 1050 | { |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 1051 | return free_pages_prepare(page, 0, false); |
Mel Gorman | 4db7548 | 2016-05-19 17:14:32 -0700 | [diff] [blame] | 1052 | } |
| 1053 | |
| 1054 | static bool bulkfree_pcp_prepare(struct page *page) |
| 1055 | { |
| 1056 | return free_pages_check(page); |
| 1057 | } |
| 1058 | #endif /* CONFIG_DEBUG_VM */ |
| 1059 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1060 | /* |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1061 | * Frees a number of pages from the PCP lists |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1062 | * 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] | 1063 | * count is the number of pages to free. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1064 | * |
| 1065 | * If the zone was previously in an "all pages pinned" state then look to |
| 1066 | * see if this freeing clears that state. |
| 1067 | * |
| 1068 | * And clear the zone's pages_scanned counter, to hold off the "all pages are |
| 1069 | * pinned" detection logic. |
| 1070 | */ |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1071 | static void free_pcppages_bulk(struct zone *zone, int count, |
| 1072 | struct per_cpu_pages *pcp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1073 | { |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1074 | int migratetype = 0; |
Mel Gorman | a6f9edd | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 1075 | int batch_free = 0; |
Mel Gorman | 0d5d823 | 2014-08-06 16:07:16 -0700 | [diff] [blame] | 1076 | unsigned long nr_scanned; |
Mel Gorman | 3777999 | 2016-05-19 17:13:58 -0700 | [diff] [blame] | 1077 | bool isolated_pageblocks; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1078 | |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 1079 | spin_lock(&zone->lock); |
Mel Gorman | 3777999 | 2016-05-19 17:13:58 -0700 | [diff] [blame] | 1080 | isolated_pageblocks = has_isolate_pageblock(zone); |
Mel Gorman | 0d5d823 | 2014-08-06 16:07:16 -0700 | [diff] [blame] | 1081 | nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED); |
| 1082 | if (nr_scanned) |
| 1083 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned); |
Mel Gorman | f2260e6 | 2009-06-16 15:32:13 -0700 | [diff] [blame] | 1084 | |
Mel Gorman | e5b31ac | 2016-05-19 17:14:24 -0700 | [diff] [blame] | 1085 | while (count) { |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 1086 | struct page *page; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1087 | struct list_head *list; |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 1088 | |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1089 | /* |
Mel Gorman | a6f9edd | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 1090 | * Remove pages from lists in a round-robin fashion. A |
| 1091 | * batch_free count is maintained that is incremented when an |
| 1092 | * empty list is encountered. This is so more pages are freed |
| 1093 | * off fuller lists instead of spinning excessively around empty |
| 1094 | * lists |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1095 | */ |
| 1096 | do { |
Mel Gorman | a6f9edd | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 1097 | batch_free++; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1098 | if (++migratetype == MIGRATE_PCPTYPES) |
| 1099 | migratetype = 0; |
| 1100 | list = &pcp->lists[migratetype]; |
| 1101 | } while (list_empty(list)); |
| 1102 | |
Namhyung Kim | 1d16871 | 2011-03-22 16:32:45 -0700 | [diff] [blame] | 1103 | /* This is the only non-empty list. Free them all. */ |
| 1104 | if (batch_free == MIGRATE_PCPTYPES) |
Mel Gorman | e5b31ac | 2016-05-19 17:14:24 -0700 | [diff] [blame] | 1105 | batch_free = count; |
Namhyung Kim | 1d16871 | 2011-03-22 16:32:45 -0700 | [diff] [blame] | 1106 | |
Mel Gorman | a6f9edd | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 1107 | do { |
Bartlomiej Zolnierkiewicz | 770c8aa | 2012-10-08 16:31:57 -0700 | [diff] [blame] | 1108 | int mt; /* migratetype of the to-be-freed page */ |
| 1109 | |
Geliang Tang | a16601c | 2016-01-14 15:20:30 -0800 | [diff] [blame] | 1110 | page = list_last_entry(list, struct page, lru); |
Mel Gorman | a6f9edd | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 1111 | /* must delete as __free_one_page list manipulates */ |
| 1112 | list_del(&page->lru); |
Vlastimil Babka | aa016d1 | 2015-09-08 15:01:22 -0700 | [diff] [blame] | 1113 | |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 1114 | mt = get_pcppage_migratetype(page); |
Vlastimil Babka | aa016d1 | 2015-09-08 15:01:22 -0700 | [diff] [blame] | 1115 | /* MIGRATE_ISOLATE page should not go to pcplists */ |
| 1116 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); |
| 1117 | /* Pageblock could have been isolated meanwhile */ |
Mel Gorman | 3777999 | 2016-05-19 17:13:58 -0700 | [diff] [blame] | 1118 | if (unlikely(isolated_pageblocks)) |
Joonsoo Kim | 51bb1a4 | 2014-11-13 15:19:14 -0800 | [diff] [blame] | 1119 | mt = get_pageblock_migratetype(page); |
Joonsoo Kim | 51bb1a4 | 2014-11-13 15:19:14 -0800 | [diff] [blame] | 1120 | |
Mel Gorman | 4db7548 | 2016-05-19 17:14:32 -0700 | [diff] [blame] | 1121 | if (bulkfree_pcp_prepare(page)) |
| 1122 | continue; |
| 1123 | |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 1124 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); |
Bartlomiej Zolnierkiewicz | 770c8aa | 2012-10-08 16:31:57 -0700 | [diff] [blame] | 1125 | trace_mm_page_pcpu_drain(page, 0, mt); |
Mel Gorman | e5b31ac | 2016-05-19 17:14:24 -0700 | [diff] [blame] | 1126 | } while (--count && --batch_free && !list_empty(list)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1127 | } |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 1128 | spin_unlock(&zone->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1129 | } |
| 1130 | |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 1131 | static void free_one_page(struct zone *zone, |
| 1132 | struct page *page, unsigned long pfn, |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 1133 | unsigned int order, |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 1134 | int migratetype) |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 1135 | { |
Mel Gorman | 0d5d823 | 2014-08-06 16:07:16 -0700 | [diff] [blame] | 1136 | unsigned long nr_scanned; |
Christoph Lameter | 006d22d | 2006-09-25 23:31:48 -0700 | [diff] [blame] | 1137 | spin_lock(&zone->lock); |
Mel Gorman | 0d5d823 | 2014-08-06 16:07:16 -0700 | [diff] [blame] | 1138 | nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED); |
| 1139 | if (nr_scanned) |
| 1140 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned); |
Mel Gorman | f2260e6 | 2009-06-16 15:32:13 -0700 | [diff] [blame] | 1141 | |
Joonsoo Kim | ad53f92 | 2014-11-13 15:19:11 -0800 | [diff] [blame] | 1142 | if (unlikely(has_isolate_pageblock(zone) || |
| 1143 | is_migrate_isolate(migratetype))) { |
| 1144 | migratetype = get_pfnblock_migratetype(page, pfn); |
Joonsoo Kim | ad53f92 | 2014-11-13 15:19:11 -0800 | [diff] [blame] | 1145 | } |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 1146 | __free_one_page(page, pfn, zone, order, migratetype); |
Christoph Lameter | 006d22d | 2006-09-25 23:31:48 -0700 | [diff] [blame] | 1147 | spin_unlock(&zone->lock); |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 1148 | } |
| 1149 | |
Robin Holt | 1e8ce83 | 2015-06-30 14:56:45 -0700 | [diff] [blame] | 1150 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
| 1151 | unsigned long zone, int nid) |
| 1152 | { |
Robin Holt | 1e8ce83 | 2015-06-30 14:56:45 -0700 | [diff] [blame] | 1153 | set_page_links(page, zone, nid, pfn); |
Robin Holt | 1e8ce83 | 2015-06-30 14:56:45 -0700 | [diff] [blame] | 1154 | init_page_count(page); |
| 1155 | page_mapcount_reset(page); |
| 1156 | page_cpupid_reset_last(page); |
Robin Holt | 1e8ce83 | 2015-06-30 14:56:45 -0700 | [diff] [blame] | 1157 | |
Robin Holt | 1e8ce83 | 2015-06-30 14:56:45 -0700 | [diff] [blame] | 1158 | INIT_LIST_HEAD(&page->lru); |
| 1159 | #ifdef WANT_PAGE_VIRTUAL |
| 1160 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ |
| 1161 | if (!is_highmem_idx(zone)) |
| 1162 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
| 1163 | #endif |
| 1164 | } |
| 1165 | |
| 1166 | static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone, |
| 1167 | int nid) |
| 1168 | { |
| 1169 | return __init_single_page(pfn_to_page(pfn), pfn, zone, nid); |
| 1170 | } |
| 1171 | |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1172 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
| 1173 | static void init_reserved_page(unsigned long pfn) |
| 1174 | { |
| 1175 | pg_data_t *pgdat; |
| 1176 | int nid, zid; |
| 1177 | |
| 1178 | if (!early_page_uninitialised(pfn)) |
| 1179 | return; |
| 1180 | |
| 1181 | nid = early_pfn_to_nid(pfn); |
| 1182 | pgdat = NODE_DATA(nid); |
| 1183 | |
| 1184 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 1185 | struct zone *zone = &pgdat->node_zones[zid]; |
| 1186 | |
| 1187 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) |
| 1188 | break; |
| 1189 | } |
| 1190 | __init_single_pfn(pfn, zid, nid); |
| 1191 | } |
| 1192 | #else |
| 1193 | static inline void init_reserved_page(unsigned long pfn) |
| 1194 | { |
| 1195 | } |
| 1196 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
| 1197 | |
Nathan Zimmer | 92923ca | 2015-06-30 14:56:48 -0700 | [diff] [blame] | 1198 | /* |
| 1199 | * Initialised pages do not have PageReserved set. This function is |
| 1200 | * called for each range allocated by the bootmem allocator and |
| 1201 | * marks the pages PageReserved. The remaining valid pages are later |
| 1202 | * sent to the buddy page allocator. |
| 1203 | */ |
Stefan Bader | 4b50bcc | 2016-05-20 16:58:38 -0700 | [diff] [blame] | 1204 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
Nathan Zimmer | 92923ca | 2015-06-30 14:56:48 -0700 | [diff] [blame] | 1205 | { |
| 1206 | unsigned long start_pfn = PFN_DOWN(start); |
| 1207 | unsigned long end_pfn = PFN_UP(end); |
| 1208 | |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1209 | for (; start_pfn < end_pfn; start_pfn++) { |
| 1210 | if (pfn_valid(start_pfn)) { |
| 1211 | struct page *page = pfn_to_page(start_pfn); |
| 1212 | |
| 1213 | init_reserved_page(start_pfn); |
Kirill A. Shutemov | 1d798ca | 2015-11-06 16:29:54 -0800 | [diff] [blame] | 1214 | |
| 1215 | /* Avoid false-positive PageTail() */ |
| 1216 | INIT_LIST_HEAD(&page->lru); |
| 1217 | |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1218 | SetPageReserved(page); |
| 1219 | } |
| 1220 | } |
Nathan Zimmer | 92923ca | 2015-06-30 14:56:48 -0700 | [diff] [blame] | 1221 | } |
| 1222 | |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 1223 | static void __free_pages_ok(struct page *page, unsigned int order) |
| 1224 | { |
| 1225 | unsigned long flags; |
Minchan Kim | 95e3441 | 2012-10-08 16:32:11 -0700 | [diff] [blame] | 1226 | int migratetype; |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 1227 | unsigned long pfn = page_to_pfn(page); |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 1228 | |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 1229 | if (!free_pages_prepare(page, order, true)) |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 1230 | return; |
| 1231 | |
Mel Gorman | cfc47a2 | 2014-06-04 16:10:19 -0700 | [diff] [blame] | 1232 | migratetype = get_pfnblock_migratetype(page, pfn); |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 1233 | local_irq_save(flags); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1234 | __count_vm_events(PGFREE, 1 << order); |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 1235 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 1236 | local_irq_restore(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1237 | } |
| 1238 | |
Li Zhang | 949698a | 2016-05-19 17:11:37 -0700 | [diff] [blame] | 1239 | static void __init __free_pages_boot_core(struct page *page, unsigned int order) |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 1240 | { |
Johannes Weiner | c399307 | 2012-01-10 15:08:10 -0800 | [diff] [blame] | 1241 | unsigned int nr_pages = 1 << order; |
Yinghai Lu | e2d0bd2 | 2013-09-11 14:20:37 -0700 | [diff] [blame] | 1242 | struct page *p = page; |
Johannes Weiner | c399307 | 2012-01-10 15:08:10 -0800 | [diff] [blame] | 1243 | unsigned int loop; |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 1244 | |
Yinghai Lu | e2d0bd2 | 2013-09-11 14:20:37 -0700 | [diff] [blame] | 1245 | prefetchw(p); |
| 1246 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { |
| 1247 | prefetchw(p + 1); |
Johannes Weiner | c399307 | 2012-01-10 15:08:10 -0800 | [diff] [blame] | 1248 | __ClearPageReserved(p); |
| 1249 | set_page_count(p, 0); |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 1250 | } |
Yinghai Lu | e2d0bd2 | 2013-09-11 14:20:37 -0700 | [diff] [blame] | 1251 | __ClearPageReserved(p); |
| 1252 | set_page_count(p, 0); |
Johannes Weiner | c399307 | 2012-01-10 15:08:10 -0800 | [diff] [blame] | 1253 | |
Yinghai Lu | e2d0bd2 | 2013-09-11 14:20:37 -0700 | [diff] [blame] | 1254 | page_zone(page)->managed_pages += nr_pages; |
Johannes Weiner | c399307 | 2012-01-10 15:08:10 -0800 | [diff] [blame] | 1255 | set_page_refcounted(page); |
| 1256 | __free_pages(page, order); |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 1257 | } |
| 1258 | |
Mel Gorman | 75a592a | 2015-06-30 14:56:59 -0700 | [diff] [blame] | 1259 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
| 1260 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) |
Mel Gorman | 7ace991 | 2015-08-06 15:46:13 -0700 | [diff] [blame] | 1261 | |
Mel Gorman | 75a592a | 2015-06-30 14:56:59 -0700 | [diff] [blame] | 1262 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
| 1263 | |
| 1264 | int __meminit early_pfn_to_nid(unsigned long pfn) |
| 1265 | { |
Mel Gorman | 7ace991 | 2015-08-06 15:46:13 -0700 | [diff] [blame] | 1266 | static DEFINE_SPINLOCK(early_pfn_lock); |
Mel Gorman | 75a592a | 2015-06-30 14:56:59 -0700 | [diff] [blame] | 1267 | int nid; |
| 1268 | |
Mel Gorman | 7ace991 | 2015-08-06 15:46:13 -0700 | [diff] [blame] | 1269 | spin_lock(&early_pfn_lock); |
Mel Gorman | 75a592a | 2015-06-30 14:56:59 -0700 | [diff] [blame] | 1270 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
Mel Gorman | 7ace991 | 2015-08-06 15:46:13 -0700 | [diff] [blame] | 1271 | if (nid < 0) |
Mel Gorman | e4568d3 | 2016-07-14 12:07:20 -0700 | [diff] [blame] | 1272 | nid = first_online_node; |
Mel Gorman | 7ace991 | 2015-08-06 15:46:13 -0700 | [diff] [blame] | 1273 | spin_unlock(&early_pfn_lock); |
| 1274 | |
| 1275 | return nid; |
Mel Gorman | 75a592a | 2015-06-30 14:56:59 -0700 | [diff] [blame] | 1276 | } |
| 1277 | #endif |
| 1278 | |
| 1279 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
| 1280 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, |
| 1281 | struct mminit_pfnnid_cache *state) |
| 1282 | { |
| 1283 | int nid; |
| 1284 | |
| 1285 | nid = __early_pfn_to_nid(pfn, state); |
| 1286 | if (nid >= 0 && nid != node) |
| 1287 | return false; |
| 1288 | return true; |
| 1289 | } |
| 1290 | |
| 1291 | /* Only safe to use early in boot when initialisation is single-threaded */ |
| 1292 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) |
| 1293 | { |
| 1294 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); |
| 1295 | } |
| 1296 | |
| 1297 | #else |
| 1298 | |
| 1299 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) |
| 1300 | { |
| 1301 | return true; |
| 1302 | } |
| 1303 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, |
| 1304 | struct mminit_pfnnid_cache *state) |
| 1305 | { |
| 1306 | return true; |
| 1307 | } |
| 1308 | #endif |
| 1309 | |
| 1310 | |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1311 | void __init __free_pages_bootmem(struct page *page, unsigned long pfn, |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 1312 | unsigned int order) |
| 1313 | { |
| 1314 | if (early_page_uninitialised(pfn)) |
| 1315 | return; |
Li Zhang | 949698a | 2016-05-19 17:11:37 -0700 | [diff] [blame] | 1316 | return __free_pages_boot_core(page, order); |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 1317 | } |
| 1318 | |
Joonsoo Kim | 7cf91a9 | 2016-03-15 14:57:51 -0700 | [diff] [blame] | 1319 | /* |
| 1320 | * Check that the whole (or subset of) a pageblock given by the interval of |
| 1321 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it |
| 1322 | * with the migration of free compaction scanner. The scanners then need to |
| 1323 | * use only pfn_valid_within() check for arches that allow holes within |
| 1324 | * pageblocks. |
| 1325 | * |
| 1326 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. |
| 1327 | * |
| 1328 | * It's possible on some configurations to have a setup like node0 node1 node0 |
| 1329 | * i.e. it's possible that all pages within a zones range of pages do not |
| 1330 | * belong to a single zone. We assume that a border between node0 and node1 |
| 1331 | * can occur within a single pageblock, but not a node0 node1 node0 |
| 1332 | * interleaving within a single pageblock. It is therefore sufficient to check |
| 1333 | * the first and last page of a pageblock and avoid checking each individual |
| 1334 | * page in a pageblock. |
| 1335 | */ |
| 1336 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, |
| 1337 | unsigned long end_pfn, struct zone *zone) |
| 1338 | { |
| 1339 | struct page *start_page; |
| 1340 | struct page *end_page; |
| 1341 | |
| 1342 | /* end_pfn is one past the range we are checking */ |
| 1343 | end_pfn--; |
| 1344 | |
| 1345 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) |
| 1346 | return NULL; |
| 1347 | |
| 1348 | start_page = pfn_to_page(start_pfn); |
| 1349 | |
| 1350 | if (page_zone(start_page) != zone) |
| 1351 | return NULL; |
| 1352 | |
| 1353 | end_page = pfn_to_page(end_pfn); |
| 1354 | |
| 1355 | /* This gives a shorter code than deriving page_zone(end_page) */ |
| 1356 | if (page_zone_id(start_page) != page_zone_id(end_page)) |
| 1357 | return NULL; |
| 1358 | |
| 1359 | return start_page; |
| 1360 | } |
| 1361 | |
| 1362 | void set_zone_contiguous(struct zone *zone) |
| 1363 | { |
| 1364 | unsigned long block_start_pfn = zone->zone_start_pfn; |
| 1365 | unsigned long block_end_pfn; |
| 1366 | |
| 1367 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); |
| 1368 | for (; block_start_pfn < zone_end_pfn(zone); |
| 1369 | block_start_pfn = block_end_pfn, |
| 1370 | block_end_pfn += pageblock_nr_pages) { |
| 1371 | |
| 1372 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); |
| 1373 | |
| 1374 | if (!__pageblock_pfn_to_page(block_start_pfn, |
| 1375 | block_end_pfn, zone)) |
| 1376 | return; |
| 1377 | } |
| 1378 | |
| 1379 | /* We confirm that there is no hole */ |
| 1380 | zone->contiguous = true; |
| 1381 | } |
| 1382 | |
| 1383 | void clear_zone_contiguous(struct zone *zone) |
| 1384 | { |
| 1385 | zone->contiguous = false; |
| 1386 | } |
| 1387 | |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1388 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1389 | static void __init deferred_free_range(struct page *page, |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1390 | unsigned long pfn, int nr_pages) |
| 1391 | { |
| 1392 | int i; |
| 1393 | |
| 1394 | if (!page) |
| 1395 | return; |
| 1396 | |
| 1397 | /* Free a large naturally-aligned chunk if possible */ |
| 1398 | if (nr_pages == MAX_ORDER_NR_PAGES && |
| 1399 | (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) { |
Mel Gorman | ac5d253 | 2015-06-30 14:57:20 -0700 | [diff] [blame] | 1400 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
Li Zhang | 949698a | 2016-05-19 17:11:37 -0700 | [diff] [blame] | 1401 | __free_pages_boot_core(page, MAX_ORDER-1); |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1402 | return; |
| 1403 | } |
| 1404 | |
Li Zhang | 949698a | 2016-05-19 17:11:37 -0700 | [diff] [blame] | 1405 | for (i = 0; i < nr_pages; i++, page++) |
| 1406 | __free_pages_boot_core(page, 0); |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1407 | } |
| 1408 | |
Nicolai Stange | d3cd131 | 2015-08-06 15:46:16 -0700 | [diff] [blame] | 1409 | /* Completion tracking for deferred_init_memmap() threads */ |
| 1410 | static atomic_t pgdat_init_n_undone __initdata; |
| 1411 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); |
| 1412 | |
| 1413 | static inline void __init pgdat_init_report_one_done(void) |
| 1414 | { |
| 1415 | if (atomic_dec_and_test(&pgdat_init_n_undone)) |
| 1416 | complete(&pgdat_init_all_done_comp); |
| 1417 | } |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1418 | |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1419 | /* Initialise remaining memory on a node */ |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1420 | static int __init deferred_init_memmap(void *data) |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1421 | { |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1422 | pg_data_t *pgdat = data; |
| 1423 | int nid = pgdat->node_id; |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1424 | struct mminit_pfnnid_cache nid_init_state = { }; |
| 1425 | unsigned long start = jiffies; |
| 1426 | unsigned long nr_pages = 0; |
| 1427 | unsigned long walk_start, walk_end; |
| 1428 | int i, zid; |
| 1429 | struct zone *zone; |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1430 | unsigned long first_init_pfn = pgdat->first_deferred_pfn; |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1431 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1432 | |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1433 | if (first_init_pfn == ULONG_MAX) { |
Nicolai Stange | d3cd131 | 2015-08-06 15:46:16 -0700 | [diff] [blame] | 1434 | pgdat_init_report_one_done(); |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1435 | return 0; |
| 1436 | } |
| 1437 | |
| 1438 | /* Bind memory initialisation thread to a local node if possible */ |
| 1439 | if (!cpumask_empty(cpumask)) |
| 1440 | set_cpus_allowed_ptr(current, cpumask); |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1441 | |
| 1442 | /* Sanity check boundaries */ |
| 1443 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); |
| 1444 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); |
| 1445 | pgdat->first_deferred_pfn = ULONG_MAX; |
| 1446 | |
| 1447 | /* Only the highest zone is deferred so find it */ |
| 1448 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 1449 | zone = pgdat->node_zones + zid; |
| 1450 | if (first_init_pfn < zone_end_pfn(zone)) |
| 1451 | break; |
| 1452 | } |
| 1453 | |
| 1454 | for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) { |
| 1455 | unsigned long pfn, end_pfn; |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1456 | struct page *page = NULL; |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1457 | struct page *free_base_page = NULL; |
| 1458 | unsigned long free_base_pfn = 0; |
| 1459 | int nr_to_free = 0; |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1460 | |
| 1461 | end_pfn = min(walk_end, zone_end_pfn(zone)); |
| 1462 | pfn = first_init_pfn; |
| 1463 | if (pfn < walk_start) |
| 1464 | pfn = walk_start; |
| 1465 | if (pfn < zone->zone_start_pfn) |
| 1466 | pfn = zone->zone_start_pfn; |
| 1467 | |
| 1468 | for (; pfn < end_pfn; pfn++) { |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1469 | if (!pfn_valid_within(pfn)) |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1470 | goto free_range; |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1471 | |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1472 | /* |
| 1473 | * Ensure pfn_valid is checked every |
| 1474 | * MAX_ORDER_NR_PAGES for memory holes |
| 1475 | */ |
| 1476 | if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) { |
| 1477 | if (!pfn_valid(pfn)) { |
| 1478 | page = NULL; |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1479 | goto free_range; |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1480 | } |
| 1481 | } |
| 1482 | |
| 1483 | if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) { |
| 1484 | page = NULL; |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1485 | goto free_range; |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1486 | } |
| 1487 | |
| 1488 | /* Minimise pfn page lookups and scheduler checks */ |
| 1489 | if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) { |
| 1490 | page++; |
| 1491 | } else { |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1492 | nr_pages += nr_to_free; |
| 1493 | deferred_free_range(free_base_page, |
| 1494 | free_base_pfn, nr_to_free); |
| 1495 | free_base_page = NULL; |
| 1496 | free_base_pfn = nr_to_free = 0; |
| 1497 | |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1498 | page = pfn_to_page(pfn); |
| 1499 | cond_resched(); |
| 1500 | } |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1501 | |
| 1502 | if (page->flags) { |
| 1503 | VM_BUG_ON(page_zone(page) != zone); |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1504 | goto free_range; |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1505 | } |
| 1506 | |
| 1507 | __init_single_page(page, pfn, zid, nid); |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1508 | if (!free_base_page) { |
| 1509 | free_base_page = page; |
| 1510 | free_base_pfn = pfn; |
| 1511 | nr_to_free = 0; |
| 1512 | } |
| 1513 | nr_to_free++; |
| 1514 | |
| 1515 | /* Where possible, batch up pages for a single free */ |
| 1516 | continue; |
| 1517 | free_range: |
| 1518 | /* Free the current block of pages to allocator */ |
| 1519 | nr_pages += nr_to_free; |
| 1520 | deferred_free_range(free_base_page, free_base_pfn, |
| 1521 | nr_to_free); |
| 1522 | free_base_page = NULL; |
| 1523 | free_base_pfn = nr_to_free = 0; |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1524 | } |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1525 | |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1526 | first_init_pfn = max(end_pfn, first_init_pfn); |
| 1527 | } |
| 1528 | |
| 1529 | /* Sanity check that the next zone really is unpopulated */ |
| 1530 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); |
| 1531 | |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1532 | pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages, |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1533 | jiffies_to_msecs(jiffies - start)); |
Nicolai Stange | d3cd131 | 2015-08-06 15:46:16 -0700 | [diff] [blame] | 1534 | |
| 1535 | pgdat_init_report_one_done(); |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1536 | return 0; |
| 1537 | } |
Joonsoo Kim | 7cf91a9 | 2016-03-15 14:57:51 -0700 | [diff] [blame] | 1538 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1539 | |
| 1540 | void __init page_alloc_init_late(void) |
| 1541 | { |
Joonsoo Kim | 7cf91a9 | 2016-03-15 14:57:51 -0700 | [diff] [blame] | 1542 | struct zone *zone; |
| 1543 | |
| 1544 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1545 | int nid; |
| 1546 | |
Nicolai Stange | d3cd131 | 2015-08-06 15:46:16 -0700 | [diff] [blame] | 1547 | /* There will be num_node_state(N_MEMORY) threads */ |
| 1548 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1549 | for_each_node_state(nid, N_MEMORY) { |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1550 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
| 1551 | } |
| 1552 | |
| 1553 | /* Block until all are initialised */ |
Nicolai Stange | d3cd131 | 2015-08-06 15:46:16 -0700 | [diff] [blame] | 1554 | wait_for_completion(&pgdat_init_all_done_comp); |
Mel Gorman | 4248b0d | 2015-08-06 15:46:20 -0700 | [diff] [blame] | 1555 | |
| 1556 | /* Reinit limits that are based on free pages after the kernel is up */ |
| 1557 | files_maxfiles_init(); |
Joonsoo Kim | 7cf91a9 | 2016-03-15 14:57:51 -0700 | [diff] [blame] | 1558 | #endif |
| 1559 | |
| 1560 | for_each_populated_zone(zone) |
| 1561 | set_zone_contiguous(zone); |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1562 | } |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1563 | |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1564 | #ifdef CONFIG_CMA |
Li Zhong | 9cf510a | 2013-08-23 13:52:52 +0800 | [diff] [blame] | 1565 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1566 | void __init init_cma_reserved_pageblock(struct page *page) |
| 1567 | { |
| 1568 | unsigned i = pageblock_nr_pages; |
| 1569 | struct page *p = page; |
| 1570 | |
| 1571 | do { |
| 1572 | __ClearPageReserved(p); |
| 1573 | set_page_count(p, 0); |
| 1574 | } while (++p, --i); |
| 1575 | |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1576 | set_pageblock_migratetype(page, MIGRATE_CMA); |
Michal Nazarewicz | dc78327 | 2014-07-02 15:22:35 -0700 | [diff] [blame] | 1577 | |
| 1578 | if (pageblock_order >= MAX_ORDER) { |
| 1579 | i = pageblock_nr_pages; |
| 1580 | p = page; |
| 1581 | do { |
| 1582 | set_page_refcounted(p); |
| 1583 | __free_pages(p, MAX_ORDER - 1); |
| 1584 | p += MAX_ORDER_NR_PAGES; |
| 1585 | } while (i -= MAX_ORDER_NR_PAGES); |
| 1586 | } else { |
| 1587 | set_page_refcounted(page); |
| 1588 | __free_pages(page, pageblock_order); |
| 1589 | } |
| 1590 | |
Jiang Liu | 3dcc057 | 2013-07-03 15:03:21 -0700 | [diff] [blame] | 1591 | adjust_managed_page_count(page, pageblock_nr_pages); |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1592 | } |
| 1593 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1594 | |
| 1595 | /* |
| 1596 | * The order of subdivision here is critical for the IO subsystem. |
| 1597 | * Please do not alter this order without good reasons and regression |
| 1598 | * testing. Specifically, as large blocks of memory are subdivided, |
| 1599 | * the order in which smaller blocks are delivered depends on the order |
| 1600 | * they're subdivided in this function. This is the primary factor |
| 1601 | * influencing the order in which pages are delivered to the IO |
| 1602 | * subsystem according to empirical testing, and this is also justified |
| 1603 | * by considering the behavior of a buddy system containing a single |
| 1604 | * large block of memory acted on by a series of small allocations. |
| 1605 | * This behavior is a critical factor in sglist merging's success. |
| 1606 | * |
Nadia Yvette Chambers | 6d49e35 | 2012-12-06 10:39:54 +0100 | [diff] [blame] | 1607 | * -- nyc |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1608 | */ |
Nick Piggin | 085cc7d | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 1609 | static inline void expand(struct zone *zone, struct page *page, |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1610 | int low, int high, struct free_area *area, |
| 1611 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1612 | { |
| 1613 | unsigned long size = 1 << high; |
| 1614 | |
| 1615 | while (high > low) { |
| 1616 | area--; |
| 1617 | high--; |
| 1618 | size >>= 1; |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 1619 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 1620 | |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 1621 | if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) && |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 1622 | debug_guardpage_enabled() && |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 1623 | high < debug_guardpage_minorder()) { |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 1624 | /* |
| 1625 | * Mark as guard pages (or page), that will allow to |
| 1626 | * merge back to allocator when buddy will be freed. |
| 1627 | * Corresponding page table entries will not be touched, |
| 1628 | * pages will stay not present in virtual address space |
| 1629 | */ |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 1630 | set_page_guard(zone, &page[size], high, migratetype); |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 1631 | continue; |
| 1632 | } |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1633 | list_add(&page[size].lru, &area->free_list[migratetype]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1634 | area->nr_free++; |
| 1635 | set_page_order(&page[size], high); |
| 1636 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1637 | } |
| 1638 | |
Vlastimil Babka | 4e61180 | 2016-05-19 17:14:41 -0700 | [diff] [blame] | 1639 | static void check_new_page_bad(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1640 | { |
Vlastimil Babka | 4e61180 | 2016-05-19 17:14:41 -0700 | [diff] [blame] | 1641 | const char *bad_reason = NULL; |
| 1642 | unsigned long bad_flags = 0; |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 1643 | |
Kirill A. Shutemov | 53f9263 | 2016-01-15 16:53:42 -0800 | [diff] [blame] | 1644 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 1645 | bad_reason = "nonzero mapcount"; |
| 1646 | if (unlikely(page->mapping != NULL)) |
| 1647 | bad_reason = "non-NULL mapping"; |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 1648 | if (unlikely(page_ref_count(page) != 0)) |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 1649 | bad_reason = "nonzero _count"; |
Naoya Horiguchi | f4c18e6 | 2015-08-06 15:47:08 -0700 | [diff] [blame] | 1650 | if (unlikely(page->flags & __PG_HWPOISON)) { |
| 1651 | bad_reason = "HWPoisoned (hardware-corrupted)"; |
| 1652 | bad_flags = __PG_HWPOISON; |
Naoya Horiguchi | e570f56 | 2016-05-20 16:58:50 -0700 | [diff] [blame] | 1653 | /* Don't complain about hwpoisoned pages */ |
| 1654 | page_mapcount_reset(page); /* remove PageBuddy */ |
| 1655 | return; |
Naoya Horiguchi | f4c18e6 | 2015-08-06 15:47:08 -0700 | [diff] [blame] | 1656 | } |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 1657 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
| 1658 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; |
| 1659 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; |
| 1660 | } |
Johannes Weiner | 9edad6e | 2014-12-10 15:44:58 -0800 | [diff] [blame] | 1661 | #ifdef CONFIG_MEMCG |
| 1662 | if (unlikely(page->mem_cgroup)) |
| 1663 | bad_reason = "page still charged to cgroup"; |
| 1664 | #endif |
Vlastimil Babka | 4e61180 | 2016-05-19 17:14:41 -0700 | [diff] [blame] | 1665 | bad_page(page, bad_reason, bad_flags); |
| 1666 | } |
| 1667 | |
| 1668 | /* |
| 1669 | * This page is about to be returned from the page allocator |
| 1670 | */ |
| 1671 | static inline int check_new_page(struct page *page) |
| 1672 | { |
| 1673 | if (likely(page_expected_state(page, |
| 1674 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) |
| 1675 | return 0; |
| 1676 | |
| 1677 | check_new_page_bad(page); |
| 1678 | return 1; |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 1679 | } |
| 1680 | |
Laura Abbott | 1414c7f | 2016-03-15 14:56:30 -0700 | [diff] [blame] | 1681 | static inline bool free_pages_prezeroed(bool poisoned) |
| 1682 | { |
| 1683 | return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && |
| 1684 | page_poisoning_enabled() && poisoned; |
| 1685 | } |
| 1686 | |
Mel Gorman | 479f854 | 2016-05-19 17:14:35 -0700 | [diff] [blame] | 1687 | #ifdef CONFIG_DEBUG_VM |
| 1688 | static bool check_pcp_refill(struct page *page) |
| 1689 | { |
| 1690 | return false; |
| 1691 | } |
| 1692 | |
| 1693 | static bool check_new_pcp(struct page *page) |
| 1694 | { |
| 1695 | return check_new_page(page); |
| 1696 | } |
| 1697 | #else |
| 1698 | static bool check_pcp_refill(struct page *page) |
| 1699 | { |
| 1700 | return check_new_page(page); |
| 1701 | } |
| 1702 | static bool check_new_pcp(struct page *page) |
| 1703 | { |
| 1704 | return false; |
| 1705 | } |
| 1706 | #endif /* CONFIG_DEBUG_VM */ |
| 1707 | |
| 1708 | static bool check_new_pages(struct page *page, unsigned int order) |
| 1709 | { |
| 1710 | int i; |
| 1711 | for (i = 0; i < (1 << order); i++) { |
| 1712 | struct page *p = page + i; |
| 1713 | |
| 1714 | if (unlikely(check_new_page(p))) |
| 1715 | return true; |
| 1716 | } |
| 1717 | |
| 1718 | return false; |
| 1719 | } |
| 1720 | |
Joonsoo Kim | 46f24fd | 2016-07-26 15:23:58 -0700 | [diff] [blame] | 1721 | inline void post_alloc_hook(struct page *page, unsigned int order, |
| 1722 | gfp_t gfp_flags) |
| 1723 | { |
| 1724 | set_page_private(page, 0); |
| 1725 | set_page_refcounted(page); |
| 1726 | |
| 1727 | arch_alloc_page(page, order); |
| 1728 | kernel_map_pages(page, 1 << order, 1); |
| 1729 | kernel_poison_pages(page, 1 << order, 1); |
| 1730 | kasan_alloc_pages(page, order); |
| 1731 | set_page_owner(page, order, gfp_flags); |
| 1732 | } |
| 1733 | |
Mel Gorman | 479f854 | 2016-05-19 17:14:35 -0700 | [diff] [blame] | 1734 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 1735 | unsigned int alloc_flags) |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 1736 | { |
| 1737 | int i; |
Laura Abbott | 1414c7f | 2016-03-15 14:56:30 -0700 | [diff] [blame] | 1738 | bool poisoned = true; |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 1739 | |
| 1740 | for (i = 0; i < (1 << order); i++) { |
| 1741 | struct page *p = page + i; |
Laura Abbott | 1414c7f | 2016-03-15 14:56:30 -0700 | [diff] [blame] | 1742 | if (poisoned) |
| 1743 | poisoned &= page_is_poisoned(p); |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 1744 | } |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 1745 | |
Joonsoo Kim | 46f24fd | 2016-07-26 15:23:58 -0700 | [diff] [blame] | 1746 | post_alloc_hook(page, order, gfp_flags); |
Nick Piggin | 17cf440 | 2006-03-22 00:08:41 -0800 | [diff] [blame] | 1747 | |
Laura Abbott | 1414c7f | 2016-03-15 14:56:30 -0700 | [diff] [blame] | 1748 | if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO)) |
Anisse Astier | f4d2897 | 2015-06-24 16:56:36 -0700 | [diff] [blame] | 1749 | for (i = 0; i < (1 << order); i++) |
| 1750 | clear_highpage(page + i); |
Nick Piggin | 17cf440 | 2006-03-22 00:08:41 -0800 | [diff] [blame] | 1751 | |
| 1752 | if (order && (gfp_flags & __GFP_COMP)) |
| 1753 | prep_compound_page(page, order); |
| 1754 | |
Vlastimil Babka | 7537919 | 2015-02-11 15:25:38 -0800 | [diff] [blame] | 1755 | /* |
Michal Hocko | 2f064f3 | 2015-08-21 14:11:51 -0700 | [diff] [blame] | 1756 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
Vlastimil Babka | 7537919 | 2015-02-11 15:25:38 -0800 | [diff] [blame] | 1757 | * allocate the page. The expectation is that the caller is taking |
| 1758 | * steps that will free more memory. The caller should avoid the page |
| 1759 | * being used for !PFMEMALLOC purposes. |
| 1760 | */ |
Michal Hocko | 2f064f3 | 2015-08-21 14:11:51 -0700 | [diff] [blame] | 1761 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
| 1762 | set_page_pfmemalloc(page); |
| 1763 | else |
| 1764 | clear_page_pfmemalloc(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1765 | } |
| 1766 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1767 | /* |
| 1768 | * Go through the free lists for the given migratetype and remove |
| 1769 | * the smallest available page from the freelists |
| 1770 | */ |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 1771 | static inline |
| 1772 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1773 | int migratetype) |
| 1774 | { |
| 1775 | unsigned int current_order; |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 1776 | struct free_area *area; |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1777 | struct page *page; |
| 1778 | |
| 1779 | /* Find a page of the appropriate size in the preferred list */ |
| 1780 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { |
| 1781 | area = &(zone->free_area[current_order]); |
Geliang Tang | a16601c | 2016-01-14 15:20:30 -0800 | [diff] [blame] | 1782 | page = list_first_entry_or_null(&area->free_list[migratetype], |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1783 | struct page, lru); |
Geliang Tang | a16601c | 2016-01-14 15:20:30 -0800 | [diff] [blame] | 1784 | if (!page) |
| 1785 | continue; |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1786 | list_del(&page->lru); |
| 1787 | rmv_page_order(page); |
| 1788 | area->nr_free--; |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1789 | expand(zone, page, order, current_order, area, migratetype); |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 1790 | set_pcppage_migratetype(page, migratetype); |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1791 | return page; |
| 1792 | } |
| 1793 | |
| 1794 | return NULL; |
| 1795 | } |
| 1796 | |
| 1797 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1798 | /* |
| 1799 | * This array describes the order lists are fallen back to when |
| 1800 | * the free lists for the desirable migrate type are depleted |
| 1801 | */ |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1802 | static int fallbacks[MIGRATE_TYPES][4] = { |
Mel Gorman | 974a786 | 2015-11-06 16:28:34 -0800 | [diff] [blame] | 1803 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
| 1804 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
| 1805 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, |
Joonsoo Kim | dc67647 | 2015-04-14 15:45:15 -0700 | [diff] [blame] | 1806 | #ifdef CONFIG_CMA |
Mel Gorman | 974a786 | 2015-11-06 16:28:34 -0800 | [diff] [blame] | 1807 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1808 | #endif |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 1809 | #ifdef CONFIG_MEMORY_ISOLATION |
Mel Gorman | 974a786 | 2015-11-06 16:28:34 -0800 | [diff] [blame] | 1810 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 1811 | #endif |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1812 | }; |
| 1813 | |
Joonsoo Kim | dc67647 | 2015-04-14 15:45:15 -0700 | [diff] [blame] | 1814 | #ifdef CONFIG_CMA |
| 1815 | static struct page *__rmqueue_cma_fallback(struct zone *zone, |
| 1816 | unsigned int order) |
| 1817 | { |
| 1818 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); |
| 1819 | } |
| 1820 | #else |
| 1821 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
| 1822 | unsigned int order) { return NULL; } |
| 1823 | #endif |
| 1824 | |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1825 | /* |
| 1826 | * 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] | 1827 | * 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] | 1828 | * boundary. If alignment is required, use move_freepages_block() |
| 1829 | */ |
Minchan Kim | 435b405 | 2012-10-08 16:32:16 -0700 | [diff] [blame] | 1830 | int move_freepages(struct zone *zone, |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 1831 | struct page *start_page, struct page *end_page, |
| 1832 | int migratetype) |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1833 | { |
| 1834 | struct page *page; |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 1835 | unsigned int order; |
Mel Gorman | d100313 | 2007-10-16 01:26:00 -0700 | [diff] [blame] | 1836 | int pages_moved = 0; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1837 | |
| 1838 | #ifndef CONFIG_HOLES_IN_ZONE |
| 1839 | /* |
| 1840 | * page_zone is not safe to call in this context when |
| 1841 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant |
| 1842 | * anyway as we check zone boundaries in move_freepages_block(). |
| 1843 | * Remove at a later date when no bug reports exist related to |
Mel Gorman | ac0e5b7 | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1844 | * grouping pages by mobility |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1845 | */ |
Mel Gorman | 97ee4ba | 2014-10-09 15:28:28 -0700 | [diff] [blame] | 1846 | VM_BUG_ON(page_zone(start_page) != page_zone(end_page)); |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1847 | #endif |
| 1848 | |
| 1849 | for (page = start_page; page <= end_page;) { |
Adam Litke | 344c790 | 2008-09-02 14:35:38 -0700 | [diff] [blame] | 1850 | /* Make sure we are not inadvertently changing nodes */ |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 1851 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); |
Adam Litke | 344c790 | 2008-09-02 14:35:38 -0700 | [diff] [blame] | 1852 | |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1853 | if (!pfn_valid_within(page_to_pfn(page))) { |
| 1854 | page++; |
| 1855 | continue; |
| 1856 | } |
| 1857 | |
| 1858 | if (!PageBuddy(page)) { |
| 1859 | page++; |
| 1860 | continue; |
| 1861 | } |
| 1862 | |
| 1863 | order = page_order(page); |
Kirill A. Shutemov | 84be48d | 2011-03-22 16:33:41 -0700 | [diff] [blame] | 1864 | list_move(&page->lru, |
| 1865 | &zone->free_area[order].free_list[migratetype]); |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1866 | page += 1 << order; |
Mel Gorman | d100313 | 2007-10-16 01:26:00 -0700 | [diff] [blame] | 1867 | pages_moved += 1 << order; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1868 | } |
| 1869 | |
Mel Gorman | d100313 | 2007-10-16 01:26:00 -0700 | [diff] [blame] | 1870 | return pages_moved; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1871 | } |
| 1872 | |
Minchan Kim | ee6f509 | 2012-07-31 16:43:50 -0700 | [diff] [blame] | 1873 | int move_freepages_block(struct zone *zone, struct page *page, |
Linus Torvalds | 68e3e92 | 2012-06-03 20:05:57 -0700 | [diff] [blame] | 1874 | int migratetype) |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1875 | { |
| 1876 | unsigned long start_pfn, end_pfn; |
| 1877 | struct page *start_page, *end_page; |
| 1878 | |
| 1879 | start_pfn = page_to_pfn(page); |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 1880 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1881 | start_page = pfn_to_page(start_pfn); |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 1882 | end_page = start_page + pageblock_nr_pages - 1; |
| 1883 | end_pfn = start_pfn + pageblock_nr_pages - 1; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1884 | |
| 1885 | /* Do not cross zone boundaries */ |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 1886 | if (!zone_spans_pfn(zone, start_pfn)) |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1887 | start_page = page; |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 1888 | if (!zone_spans_pfn(zone, end_pfn)) |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1889 | return 0; |
| 1890 | |
| 1891 | return move_freepages(zone, start_page, end_page, migratetype); |
| 1892 | } |
| 1893 | |
Mel Gorman | 2f66a68 | 2009-09-21 17:02:31 -0700 | [diff] [blame] | 1894 | static void change_pageblock_range(struct page *pageblock_page, |
| 1895 | int start_order, int migratetype) |
| 1896 | { |
| 1897 | int nr_pageblocks = 1 << (start_order - pageblock_order); |
| 1898 | |
| 1899 | while (nr_pageblocks--) { |
| 1900 | set_pageblock_migratetype(pageblock_page, migratetype); |
| 1901 | pageblock_page += pageblock_nr_pages; |
| 1902 | } |
| 1903 | } |
| 1904 | |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1905 | /* |
Vlastimil Babka | 9c0415e | 2015-02-11 15:28:21 -0800 | [diff] [blame] | 1906 | * When we are falling back to another migratetype during allocation, try to |
| 1907 | * steal extra free pages from the same pageblocks to satisfy further |
| 1908 | * allocations, instead of polluting multiple pageblocks. |
| 1909 | * |
| 1910 | * If we are stealing a relatively large buddy page, it is likely there will |
| 1911 | * be more free pages in the pageblock, so try to steal them all. For |
| 1912 | * reclaimable and unmovable allocations, we steal regardless of page size, |
| 1913 | * as fragmentation caused by those allocations polluting movable pageblocks |
| 1914 | * is worse than movable allocations stealing from unmovable and reclaimable |
| 1915 | * pageblocks. |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1916 | */ |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 1917 | static bool can_steal_fallback(unsigned int order, int start_mt) |
| 1918 | { |
| 1919 | /* |
| 1920 | * Leaving this order check is intended, although there is |
| 1921 | * relaxed order check in next check. The reason is that |
| 1922 | * we can actually steal whole pageblock if this condition met, |
| 1923 | * but, below check doesn't guarantee it and that is just heuristic |
| 1924 | * so could be changed anytime. |
| 1925 | */ |
| 1926 | if (order >= pageblock_order) |
| 1927 | return true; |
| 1928 | |
| 1929 | if (order >= pageblock_order / 2 || |
| 1930 | start_mt == MIGRATE_RECLAIMABLE || |
| 1931 | start_mt == MIGRATE_UNMOVABLE || |
| 1932 | page_group_by_mobility_disabled) |
| 1933 | return true; |
| 1934 | |
| 1935 | return false; |
| 1936 | } |
| 1937 | |
| 1938 | /* |
| 1939 | * This function implements actual steal behaviour. If order is large enough, |
| 1940 | * we can steal whole pageblock. If not, we first move freepages in this |
| 1941 | * pageblock and check whether half of pages are moved or not. If half of |
| 1942 | * pages are moved, we can change migratetype of pageblock and permanently |
| 1943 | * use it's pages as requested migratetype in the future. |
| 1944 | */ |
| 1945 | static void steal_suitable_fallback(struct zone *zone, struct page *page, |
| 1946 | int start_type) |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1947 | { |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 1948 | unsigned int current_order = page_order(page); |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 1949 | int pages; |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1950 | |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1951 | /* Take ownership for orders >= pageblock_order */ |
| 1952 | if (current_order >= pageblock_order) { |
| 1953 | change_pageblock_range(page, current_order, start_type); |
Vlastimil Babka | 3a1086f | 2015-02-11 15:28:18 -0800 | [diff] [blame] | 1954 | return; |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1955 | } |
| 1956 | |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 1957 | pages = move_freepages_block(zone, page, start_type); |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1958 | |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 1959 | /* Claim the whole block if over half of it is free */ |
| 1960 | if (pages >= (1 << (pageblock_order-1)) || |
| 1961 | page_group_by_mobility_disabled) |
| 1962 | set_pageblock_migratetype(page, start_type); |
| 1963 | } |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1964 | |
Joonsoo Kim | 2149cda | 2015-04-14 15:45:21 -0700 | [diff] [blame] | 1965 | /* |
| 1966 | * Check whether there is a suitable fallback freepage with requested order. |
| 1967 | * If only_stealable is true, this function returns fallback_mt only if |
| 1968 | * we can steal other freepages all together. This would help to reduce |
| 1969 | * fragmentation due to mixed migratetype pages in one pageblock. |
| 1970 | */ |
| 1971 | int find_suitable_fallback(struct free_area *area, unsigned int order, |
| 1972 | int migratetype, bool only_stealable, bool *can_steal) |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 1973 | { |
| 1974 | int i; |
| 1975 | int fallback_mt; |
| 1976 | |
| 1977 | if (area->nr_free == 0) |
| 1978 | return -1; |
| 1979 | |
| 1980 | *can_steal = false; |
| 1981 | for (i = 0;; i++) { |
| 1982 | fallback_mt = fallbacks[migratetype][i]; |
Mel Gorman | 974a786 | 2015-11-06 16:28:34 -0800 | [diff] [blame] | 1983 | if (fallback_mt == MIGRATE_TYPES) |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 1984 | break; |
| 1985 | |
| 1986 | if (list_empty(&area->free_list[fallback_mt])) |
| 1987 | continue; |
| 1988 | |
| 1989 | if (can_steal_fallback(order, migratetype)) |
| 1990 | *can_steal = true; |
| 1991 | |
Joonsoo Kim | 2149cda | 2015-04-14 15:45:21 -0700 | [diff] [blame] | 1992 | if (!only_stealable) |
| 1993 | return fallback_mt; |
| 1994 | |
| 1995 | if (*can_steal) |
| 1996 | return fallback_mt; |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1997 | } |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 1998 | |
| 1999 | return -1; |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 2000 | } |
| 2001 | |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2002 | /* |
| 2003 | * Reserve a pageblock for exclusive use of high-order atomic allocations if |
| 2004 | * there are no empty page blocks that contain a page with a suitable order |
| 2005 | */ |
| 2006 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, |
| 2007 | unsigned int alloc_order) |
| 2008 | { |
| 2009 | int mt; |
| 2010 | unsigned long max_managed, flags; |
| 2011 | |
| 2012 | /* |
| 2013 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. |
| 2014 | * Check is race-prone but harmless. |
| 2015 | */ |
| 2016 | max_managed = (zone->managed_pages / 100) + pageblock_nr_pages; |
| 2017 | if (zone->nr_reserved_highatomic >= max_managed) |
| 2018 | return; |
| 2019 | |
| 2020 | spin_lock_irqsave(&zone->lock, flags); |
| 2021 | |
| 2022 | /* Recheck the nr_reserved_highatomic limit under the lock */ |
| 2023 | if (zone->nr_reserved_highatomic >= max_managed) |
| 2024 | goto out_unlock; |
| 2025 | |
| 2026 | /* Yoink! */ |
| 2027 | mt = get_pageblock_migratetype(page); |
| 2028 | if (mt != MIGRATE_HIGHATOMIC && |
| 2029 | !is_migrate_isolate(mt) && !is_migrate_cma(mt)) { |
| 2030 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
| 2031 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); |
| 2032 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC); |
| 2033 | } |
| 2034 | |
| 2035 | out_unlock: |
| 2036 | spin_unlock_irqrestore(&zone->lock, flags); |
| 2037 | } |
| 2038 | |
| 2039 | /* |
| 2040 | * Used when an allocation is about to fail under memory pressure. This |
| 2041 | * potentially hurts the reliability of high-order allocations when under |
| 2042 | * intense memory pressure but failed atomic allocations should be easier |
| 2043 | * to recover from than an OOM. |
| 2044 | */ |
| 2045 | static void unreserve_highatomic_pageblock(const struct alloc_context *ac) |
| 2046 | { |
| 2047 | struct zonelist *zonelist = ac->zonelist; |
| 2048 | unsigned long flags; |
| 2049 | struct zoneref *z; |
| 2050 | struct zone *zone; |
| 2051 | struct page *page; |
| 2052 | int order; |
| 2053 | |
| 2054 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, |
| 2055 | ac->nodemask) { |
| 2056 | /* Preserve at least one pageblock */ |
| 2057 | if (zone->nr_reserved_highatomic <= pageblock_nr_pages) |
| 2058 | continue; |
| 2059 | |
| 2060 | spin_lock_irqsave(&zone->lock, flags); |
| 2061 | for (order = 0; order < MAX_ORDER; order++) { |
| 2062 | struct free_area *area = &(zone->free_area[order]); |
| 2063 | |
Geliang Tang | a16601c | 2016-01-14 15:20:30 -0800 | [diff] [blame] | 2064 | page = list_first_entry_or_null( |
| 2065 | &area->free_list[MIGRATE_HIGHATOMIC], |
| 2066 | struct page, lru); |
| 2067 | if (!page) |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2068 | continue; |
| 2069 | |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2070 | /* |
| 2071 | * It should never happen but changes to locking could |
| 2072 | * inadvertently allow a per-cpu drain to add pages |
| 2073 | * to MIGRATE_HIGHATOMIC while unreserving so be safe |
| 2074 | * and watch for underflows. |
| 2075 | */ |
| 2076 | zone->nr_reserved_highatomic -= min(pageblock_nr_pages, |
| 2077 | zone->nr_reserved_highatomic); |
| 2078 | |
| 2079 | /* |
| 2080 | * Convert to ac->migratetype and avoid the normal |
| 2081 | * pageblock stealing heuristics. Minimally, the caller |
| 2082 | * is doing the work and needs the pages. More |
| 2083 | * importantly, if the block was always converted to |
| 2084 | * MIGRATE_UNMOVABLE or another type then the number |
| 2085 | * of pageblocks that cannot be completely freed |
| 2086 | * may increase. |
| 2087 | */ |
| 2088 | set_pageblock_migratetype(page, ac->migratetype); |
| 2089 | move_freepages_block(zone, page, ac->migratetype); |
| 2090 | spin_unlock_irqrestore(&zone->lock, flags); |
| 2091 | return; |
| 2092 | } |
| 2093 | spin_unlock_irqrestore(&zone->lock, flags); |
| 2094 | } |
| 2095 | } |
| 2096 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2097 | /* Remove an element from the buddy allocator from the fallback list */ |
Mel Gorman | 0ac3a40 | 2009-06-16 15:32:06 -0700 | [diff] [blame] | 2098 | static inline struct page * |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 2099 | __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2100 | { |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 2101 | struct free_area *area; |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 2102 | unsigned int current_order; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2103 | struct page *page; |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2104 | int fallback_mt; |
| 2105 | bool can_steal; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2106 | |
| 2107 | /* Find the largest possible block of pages in the other list */ |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 2108 | for (current_order = MAX_ORDER-1; |
| 2109 | current_order >= order && current_order <= MAX_ORDER-1; |
| 2110 | --current_order) { |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2111 | area = &(zone->free_area[current_order]); |
| 2112 | fallback_mt = find_suitable_fallback(area, current_order, |
Joonsoo Kim | 2149cda | 2015-04-14 15:45:21 -0700 | [diff] [blame] | 2113 | start_migratetype, false, &can_steal); |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2114 | if (fallback_mt == -1) |
| 2115 | continue; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2116 | |
Geliang Tang | a16601c | 2016-01-14 15:20:30 -0800 | [diff] [blame] | 2117 | page = list_first_entry(&area->free_list[fallback_mt], |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2118 | struct page, lru); |
| 2119 | if (can_steal) |
| 2120 | steal_suitable_fallback(zone, page, start_migratetype); |
Mel Gorman | e010487 | 2007-10-16 01:25:53 -0700 | [diff] [blame] | 2121 | |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2122 | /* Remove the page from the freelists */ |
| 2123 | area->nr_free--; |
| 2124 | list_del(&page->lru); |
| 2125 | rmv_page_order(page); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2126 | |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2127 | expand(zone, page, order, current_order, area, |
| 2128 | start_migratetype); |
| 2129 | /* |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 2130 | * The pcppage_migratetype may differ from pageblock's |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2131 | * migratetype depending on the decisions in |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 2132 | * find_suitable_fallback(). This is OK as long as it does not |
| 2133 | * differ for MIGRATE_CMA pageblocks. Those can be used as |
| 2134 | * fallback only via special __rmqueue_cma_fallback() function |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2135 | */ |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 2136 | set_pcppage_migratetype(page, start_migratetype); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2137 | |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2138 | trace_mm_page_alloc_extfrag(page, order, current_order, |
| 2139 | start_migratetype, fallback_mt); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2140 | |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2141 | return page; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2142 | } |
| 2143 | |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 2144 | return NULL; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2145 | } |
| 2146 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 2147 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2148 | * Do the hard work of removing an element from the buddy allocator. |
| 2149 | * Call me with the zone->lock already held. |
| 2150 | */ |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2151 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
Mel Gorman | 6ac0206 | 2016-01-14 15:20:28 -0800 | [diff] [blame] | 2152 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2153 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2154 | struct page *page; |
| 2155 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 2156 | page = __rmqueue_smallest(zone, order, migratetype); |
Mel Gorman | 974a786 | 2015-11-06 16:28:34 -0800 | [diff] [blame] | 2157 | if (unlikely(!page)) { |
Joonsoo Kim | dc67647 | 2015-04-14 15:45:15 -0700 | [diff] [blame] | 2158 | if (migratetype == MIGRATE_MOVABLE) |
| 2159 | page = __rmqueue_cma_fallback(zone, order); |
| 2160 | |
| 2161 | if (!page) |
| 2162 | page = __rmqueue_fallback(zone, order, migratetype); |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 2163 | } |
| 2164 | |
Mel Gorman | 0d3d062 | 2009-09-21 17:02:44 -0700 | [diff] [blame] | 2165 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2166 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2167 | } |
| 2168 | |
Michal Nazarewicz | 5f63b72 | 2012-01-11 15:16:11 +0100 | [diff] [blame] | 2169 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2170 | * Obtain a specified number of elements from the buddy allocator, all under |
| 2171 | * a single hold of the lock, for efficiency. Add them to the supplied list. |
| 2172 | * Returns the number of new pages which were placed at *list. |
| 2173 | */ |
Michal Nazarewicz | 5f63b72 | 2012-01-11 15:16:11 +0100 | [diff] [blame] | 2174 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2175 | unsigned long count, struct list_head *list, |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2176 | int migratetype, bool cold) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2177 | { |
Vlastimil Babka | 5bcc9f8 | 2014-06-04 16:07:22 -0700 | [diff] [blame] | 2178 | int i; |
Michal Nazarewicz | 5f63b72 | 2012-01-11 15:16:11 +0100 | [diff] [blame] | 2179 | |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 2180 | spin_lock(&zone->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2181 | for (i = 0; i < count; ++i) { |
Mel Gorman | 6ac0206 | 2016-01-14 15:20:28 -0800 | [diff] [blame] | 2182 | struct page *page = __rmqueue(zone, order, migratetype); |
Nick Piggin | 085cc7d | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 2183 | if (unlikely(page == NULL)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2184 | break; |
Mel Gorman | 81eabcb | 2007-12-17 16:20:05 -0800 | [diff] [blame] | 2185 | |
Mel Gorman | 479f854 | 2016-05-19 17:14:35 -0700 | [diff] [blame] | 2186 | if (unlikely(check_pcp_refill(page))) |
| 2187 | continue; |
| 2188 | |
Mel Gorman | 81eabcb | 2007-12-17 16:20:05 -0800 | [diff] [blame] | 2189 | /* |
| 2190 | * Split buddy pages returned by expand() are received here |
| 2191 | * in physical page order. The page is added to the callers and |
| 2192 | * list and the list head then moves forward. From the callers |
| 2193 | * perspective, the linked list is ordered by page number in |
| 2194 | * some conditions. This is useful for IO devices that can |
| 2195 | * merge IO requests if the physical pages are ordered |
| 2196 | * properly. |
| 2197 | */ |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2198 | if (likely(!cold)) |
Mel Gorman | e084b2d | 2009-07-29 15:02:04 -0700 | [diff] [blame] | 2199 | list_add(&page->lru, list); |
| 2200 | else |
| 2201 | list_add_tail(&page->lru, list); |
Mel Gorman | 81eabcb | 2007-12-17 16:20:05 -0800 | [diff] [blame] | 2202 | list = &page->lru; |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 2203 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
Bartlomiej Zolnierkiewicz | d1ce749 | 2012-10-08 16:32:02 -0700 | [diff] [blame] | 2204 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
| 2205 | -(1 << order)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2206 | } |
Mel Gorman | f2260e6 | 2009-06-16 15:32:13 -0700 | [diff] [blame] | 2207 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 2208 | spin_unlock(&zone->lock); |
Nick Piggin | 085cc7d | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 2209 | return i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2210 | } |
| 2211 | |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 2212 | #ifdef CONFIG_NUMA |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 2213 | /* |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 2214 | * Called from the vmstat counter updater to drain pagesets of this |
| 2215 | * currently executing processor on remote nodes after they have |
| 2216 | * expired. |
| 2217 | * |
Christoph Lameter | 879336c | 2006-03-22 00:09:08 -0800 | [diff] [blame] | 2218 | * Note that this function must be called with the thread pinned to |
| 2219 | * a single processor. |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 2220 | */ |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 2221 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 2222 | { |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 2223 | unsigned long flags; |
Michal Nazarewicz | 7be12fc | 2014-08-06 16:05:15 -0700 | [diff] [blame] | 2224 | int to_drain, batch; |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 2225 | |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 2226 | local_irq_save(flags); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 2227 | batch = READ_ONCE(pcp->batch); |
Michal Nazarewicz | 7be12fc | 2014-08-06 16:05:15 -0700 | [diff] [blame] | 2228 | to_drain = min(pcp->count, batch); |
KOSAKI Motohiro | 2a13515 | 2012-07-31 16:42:53 -0700 | [diff] [blame] | 2229 | if (to_drain > 0) { |
| 2230 | free_pcppages_bulk(zone, to_drain, pcp); |
| 2231 | pcp->count -= to_drain; |
| 2232 | } |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 2233 | local_irq_restore(flags); |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 2234 | } |
| 2235 | #endif |
| 2236 | |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2237 | /* |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2238 | * Drain pcplists of the indicated processor and zone. |
| 2239 | * |
| 2240 | * The processor must either be the current processor and the |
| 2241 | * thread pinned to the current processor or a processor that |
| 2242 | * is not online. |
| 2243 | */ |
| 2244 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
| 2245 | { |
| 2246 | unsigned long flags; |
| 2247 | struct per_cpu_pageset *pset; |
| 2248 | struct per_cpu_pages *pcp; |
| 2249 | |
| 2250 | local_irq_save(flags); |
| 2251 | pset = per_cpu_ptr(zone->pageset, cpu); |
| 2252 | |
| 2253 | pcp = &pset->pcp; |
| 2254 | if (pcp->count) { |
| 2255 | free_pcppages_bulk(zone, pcp->count, pcp); |
| 2256 | pcp->count = 0; |
| 2257 | } |
| 2258 | local_irq_restore(flags); |
| 2259 | } |
| 2260 | |
| 2261 | /* |
| 2262 | * Drain pcplists of all zones on the indicated processor. |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2263 | * |
| 2264 | * The processor must either be the current processor and the |
| 2265 | * thread pinned to the current processor or a processor that |
| 2266 | * is not online. |
| 2267 | */ |
| 2268 | static void drain_pages(unsigned int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2269 | { |
| 2270 | struct zone *zone; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2271 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 2272 | for_each_populated_zone(zone) { |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2273 | drain_pages_zone(cpu, zone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2274 | } |
| 2275 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2276 | |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2277 | /* |
| 2278 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2279 | * |
| 2280 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just |
| 2281 | * the single zone's pages. |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2282 | */ |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2283 | void drain_local_pages(struct zone *zone) |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2284 | { |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2285 | int cpu = smp_processor_id(); |
| 2286 | |
| 2287 | if (zone) |
| 2288 | drain_pages_zone(cpu, zone); |
| 2289 | else |
| 2290 | drain_pages(cpu); |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2291 | } |
| 2292 | |
| 2293 | /* |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2294 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
| 2295 | * |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2296 | * When zone parameter is non-NULL, spill just the single zone's pages. |
| 2297 | * |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2298 | * Note that this code is protected against sending an IPI to an offline |
| 2299 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: |
| 2300 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but |
| 2301 | * nothing keeps CPUs from showing up after we populated the cpumask and |
| 2302 | * before the call to on_each_cpu_mask(). |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2303 | */ |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2304 | void drain_all_pages(struct zone *zone) |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2305 | { |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2306 | int cpu; |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2307 | |
| 2308 | /* |
| 2309 | * Allocate in the BSS so we wont require allocation in |
| 2310 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y |
| 2311 | */ |
| 2312 | static cpumask_t cpus_with_pcps; |
| 2313 | |
| 2314 | /* |
| 2315 | * We don't care about racing with CPU hotplug event |
| 2316 | * as offline notification will cause the notified |
| 2317 | * cpu to drain that CPU pcps and on_each_cpu_mask |
| 2318 | * disables preemption as part of its processing |
| 2319 | */ |
| 2320 | for_each_online_cpu(cpu) { |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2321 | struct per_cpu_pageset *pcp; |
| 2322 | struct zone *z; |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2323 | bool has_pcps = false; |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2324 | |
| 2325 | if (zone) { |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2326 | pcp = per_cpu_ptr(zone->pageset, cpu); |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2327 | if (pcp->pcp.count) |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2328 | has_pcps = true; |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2329 | } else { |
| 2330 | for_each_populated_zone(z) { |
| 2331 | pcp = per_cpu_ptr(z->pageset, cpu); |
| 2332 | if (pcp->pcp.count) { |
| 2333 | has_pcps = true; |
| 2334 | break; |
| 2335 | } |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2336 | } |
| 2337 | } |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2338 | |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2339 | if (has_pcps) |
| 2340 | cpumask_set_cpu(cpu, &cpus_with_pcps); |
| 2341 | else |
| 2342 | cpumask_clear_cpu(cpu, &cpus_with_pcps); |
| 2343 | } |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2344 | on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages, |
| 2345 | zone, 1); |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2346 | } |
| 2347 | |
Rafael J. Wysocki | 296699d | 2007-07-29 23:27:18 +0200 | [diff] [blame] | 2348 | #ifdef CONFIG_HIBERNATION |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2349 | |
| 2350 | void mark_free_pages(struct zone *zone) |
| 2351 | { |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 2352 | unsigned long pfn, max_zone_pfn; |
| 2353 | unsigned long flags; |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 2354 | unsigned int order, t; |
Geliang Tang | 86760a2 | 2016-01-14 15:20:33 -0800 | [diff] [blame] | 2355 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2356 | |
Xishi Qiu | 8080fc0 | 2013-09-11 14:21:45 -0700 | [diff] [blame] | 2357 | if (zone_is_empty(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2358 | return; |
| 2359 | |
| 2360 | spin_lock_irqsave(&zone->lock, flags); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 2361 | |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 2362 | max_zone_pfn = zone_end_pfn(zone); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 2363 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
| 2364 | if (pfn_valid(pfn)) { |
Geliang Tang | 86760a2 | 2016-01-14 15:20:33 -0800 | [diff] [blame] | 2365 | page = pfn_to_page(pfn); |
Joonsoo Kim | ba6b097 | 2016-05-19 17:12:16 -0700 | [diff] [blame] | 2366 | |
| 2367 | if (page_zone(page) != zone) |
| 2368 | continue; |
| 2369 | |
Rafael J. Wysocki | 7be9823 | 2007-05-06 14:50:42 -0700 | [diff] [blame] | 2370 | if (!swsusp_page_is_forbidden(page)) |
| 2371 | swsusp_unset_page_free(page); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 2372 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2373 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2374 | for_each_migratetype_order(order, t) { |
Geliang Tang | 86760a2 | 2016-01-14 15:20:33 -0800 | [diff] [blame] | 2375 | list_for_each_entry(page, |
| 2376 | &zone->free_area[order].free_list[t], lru) { |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 2377 | unsigned long i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2378 | |
Geliang Tang | 86760a2 | 2016-01-14 15:20:33 -0800 | [diff] [blame] | 2379 | pfn = page_to_pfn(page); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 2380 | for (i = 0; i < (1UL << order); i++) |
Rafael J. Wysocki | 7be9823 | 2007-05-06 14:50:42 -0700 | [diff] [blame] | 2381 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 2382 | } |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2383 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2384 | spin_unlock_irqrestore(&zone->lock, flags); |
| 2385 | } |
Mel Gorman | e2c55dc | 2007-10-16 01:25:50 -0700 | [diff] [blame] | 2386 | #endif /* CONFIG_PM */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2387 | |
| 2388 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2389 | * Free a 0-order page |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2390 | * cold == true ? free a cold page : free a hot page |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2391 | */ |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2392 | void free_hot_cold_page(struct page *page, bool cold) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2393 | { |
| 2394 | struct zone *zone = page_zone(page); |
| 2395 | struct per_cpu_pages *pcp; |
| 2396 | unsigned long flags; |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 2397 | unsigned long pfn = page_to_pfn(page); |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 2398 | int migratetype; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2399 | |
Mel Gorman | 4db7548 | 2016-05-19 17:14:32 -0700 | [diff] [blame] | 2400 | if (!free_pcp_prepare(page)) |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 2401 | return; |
| 2402 | |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 2403 | migratetype = get_pfnblock_migratetype(page, pfn); |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 2404 | set_pcppage_migratetype(page, migratetype); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2405 | local_irq_save(flags); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2406 | __count_vm_event(PGFREE); |
Mel Gorman | da456f1 | 2009-06-16 15:32:08 -0700 | [diff] [blame] | 2407 | |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 2408 | /* |
| 2409 | * We only track unmovable, reclaimable and movable on pcp lists. |
| 2410 | * Free ISOLATE pages back to the allocator because they are being |
| 2411 | * offlined but treat RESERVE as movable pages so we can get those |
| 2412 | * areas back if necessary. Otherwise, we may have to free |
| 2413 | * excessively into the page allocator |
| 2414 | */ |
| 2415 | if (migratetype >= MIGRATE_PCPTYPES) { |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 2416 | if (unlikely(is_migrate_isolate(migratetype))) { |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 2417 | free_one_page(zone, page, pfn, 0, migratetype); |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 2418 | goto out; |
| 2419 | } |
| 2420 | migratetype = MIGRATE_MOVABLE; |
| 2421 | } |
| 2422 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 2423 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2424 | if (!cold) |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 2425 | list_add(&page->lru, &pcp->lists[migratetype]); |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2426 | else |
| 2427 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2428 | pcp->count++; |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 2429 | if (pcp->count >= pcp->high) { |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 2430 | unsigned long batch = READ_ONCE(pcp->batch); |
Cody P Schafer | 998d39c | 2013-07-03 15:01:32 -0700 | [diff] [blame] | 2431 | free_pcppages_bulk(zone, batch, pcp); |
| 2432 | pcp->count -= batch; |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 2433 | } |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 2434 | |
| 2435 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2436 | local_irq_restore(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2437 | } |
| 2438 | |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 2439 | /* |
Konstantin Khlebnikov | cc59850 | 2012-01-10 15:07:04 -0800 | [diff] [blame] | 2440 | * Free a list of 0-order pages |
| 2441 | */ |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2442 | void free_hot_cold_page_list(struct list_head *list, bool cold) |
Konstantin Khlebnikov | cc59850 | 2012-01-10 15:07:04 -0800 | [diff] [blame] | 2443 | { |
| 2444 | struct page *page, *next; |
| 2445 | |
| 2446 | list_for_each_entry_safe(page, next, list, lru) { |
Konstantin Khlebnikov | b413d48 | 2012-01-10 15:07:09 -0800 | [diff] [blame] | 2447 | trace_mm_page_free_batched(page, cold); |
Konstantin Khlebnikov | cc59850 | 2012-01-10 15:07:04 -0800 | [diff] [blame] | 2448 | free_hot_cold_page(page, cold); |
| 2449 | } |
| 2450 | } |
| 2451 | |
| 2452 | /* |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 2453 | * split_page takes a non-compound higher-order page, and splits it into |
| 2454 | * n (1<<order) sub-pages: page[0..n] |
| 2455 | * Each sub-page must be freed individually. |
| 2456 | * |
| 2457 | * Note: this is probably too low level an operation for use in drivers. |
| 2458 | * Please consult with lkml before using this in your driver. |
| 2459 | */ |
| 2460 | void split_page(struct page *page, unsigned int order) |
| 2461 | { |
| 2462 | int i; |
| 2463 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 2464 | VM_BUG_ON_PAGE(PageCompound(page), page); |
| 2465 | VM_BUG_ON_PAGE(!page_count(page), page); |
Vegard Nossum | b1eeab6 | 2008-11-25 16:55:53 +0100 | [diff] [blame] | 2466 | |
| 2467 | #ifdef CONFIG_KMEMCHECK |
| 2468 | /* |
| 2469 | * Split shadow pages too, because free(page[0]) would |
| 2470 | * otherwise free the whole shadow. |
| 2471 | */ |
| 2472 | if (kmemcheck_page_is_tracked(page)) |
| 2473 | split_page(virt_to_page(page[0].shadow), order); |
| 2474 | #endif |
| 2475 | |
Joonsoo Kim | a9627bc | 2016-07-26 15:23:49 -0700 | [diff] [blame] | 2476 | for (i = 1; i < (1 << order); i++) |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 2477 | set_page_refcounted(page + i); |
Joonsoo Kim | a9627bc | 2016-07-26 15:23:49 -0700 | [diff] [blame] | 2478 | split_page_owner(page, order); |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 2479 | } |
K. Y. Srinivasan | 5853ff2 | 2013-03-25 15:47:38 -0700 | [diff] [blame] | 2480 | EXPORT_SYMBOL_GPL(split_page); |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 2481 | |
Joonsoo Kim | 3c60509 | 2014-11-13 15:19:21 -0800 | [diff] [blame] | 2482 | int __isolate_free_page(struct page *page, unsigned int order) |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 2483 | { |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 2484 | unsigned long watermark; |
| 2485 | struct zone *zone; |
Bartlomiej Zolnierkiewicz | 2139cbe | 2012-10-08 16:32:00 -0700 | [diff] [blame] | 2486 | int mt; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 2487 | |
| 2488 | BUG_ON(!PageBuddy(page)); |
| 2489 | |
| 2490 | zone = page_zone(page); |
Marek Szyprowski | 2e30abd | 2012-12-11 16:02:57 -0800 | [diff] [blame] | 2491 | mt = get_pageblock_migratetype(page); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 2492 | |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 2493 | if (!is_migrate_isolate(mt)) { |
Marek Szyprowski | 2e30abd | 2012-12-11 16:02:57 -0800 | [diff] [blame] | 2494 | /* Obey watermarks as if the page was being allocated */ |
| 2495 | watermark = low_wmark_pages(zone) + (1 << order); |
| 2496 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) |
| 2497 | return 0; |
| 2498 | |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 2499 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
Marek Szyprowski | 2e30abd | 2012-12-11 16:02:57 -0800 | [diff] [blame] | 2500 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 2501 | |
| 2502 | /* Remove page from free list */ |
| 2503 | list_del(&page->lru); |
| 2504 | zone->free_area[order].nr_free--; |
| 2505 | rmv_page_order(page); |
Bartlomiej Zolnierkiewicz | 2139cbe | 2012-10-08 16:32:00 -0700 | [diff] [blame] | 2506 | |
zhong jiang | 400bc7f | 2016-07-28 15:45:07 -0700 | [diff] [blame] | 2507 | /* |
| 2508 | * Set the pageblock if the isolated page is at least half of a |
| 2509 | * pageblock |
| 2510 | */ |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 2511 | if (order >= pageblock_order - 1) { |
| 2512 | struct page *endpage = page + (1 << order) - 1; |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 2513 | for (; page < endpage; page += pageblock_nr_pages) { |
| 2514 | int mt = get_pageblock_migratetype(page); |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 2515 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)) |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 2516 | set_pageblock_migratetype(page, |
| 2517 | MIGRATE_MOVABLE); |
| 2518 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 2519 | } |
| 2520 | |
Joonsoo Kim | f3a14ce | 2015-07-17 16:24:15 -0700 | [diff] [blame] | 2521 | |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 2522 | return 1UL << order; |
Mel Gorman | 1fb3f8c | 2012-10-08 16:29:12 -0700 | [diff] [blame] | 2523 | } |
| 2524 | |
| 2525 | /* |
Mel Gorman | 060e741 | 2016-05-19 17:13:27 -0700 | [diff] [blame] | 2526 | * Update NUMA hit/miss statistics |
| 2527 | * |
| 2528 | * Must be called with interrupts disabled. |
| 2529 | * |
| 2530 | * When __GFP_OTHER_NODE is set assume the node of the preferred |
| 2531 | * zone is the local node. This is useful for daemons who allocate |
| 2532 | * memory on behalf of other processes. |
| 2533 | */ |
| 2534 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z, |
| 2535 | gfp_t flags) |
| 2536 | { |
| 2537 | #ifdef CONFIG_NUMA |
| 2538 | int local_nid = numa_node_id(); |
| 2539 | enum zone_stat_item local_stat = NUMA_LOCAL; |
| 2540 | |
| 2541 | if (unlikely(flags & __GFP_OTHER_NODE)) { |
| 2542 | local_stat = NUMA_OTHER; |
| 2543 | local_nid = preferred_zone->node; |
| 2544 | } |
| 2545 | |
| 2546 | if (z->node == local_nid) { |
| 2547 | __inc_zone_state(z, NUMA_HIT); |
| 2548 | __inc_zone_state(z, local_stat); |
| 2549 | } else { |
| 2550 | __inc_zone_state(z, NUMA_MISS); |
| 2551 | __inc_zone_state(preferred_zone, NUMA_FOREIGN); |
| 2552 | } |
| 2553 | #endif |
| 2554 | } |
| 2555 | |
| 2556 | /* |
Vlastimil Babka | 7537919 | 2015-02-11 15:25:38 -0800 | [diff] [blame] | 2557 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2558 | */ |
Mel Gorman | 0a15c3e | 2009-06-16 15:32:05 -0700 | [diff] [blame] | 2559 | static inline |
| 2560 | struct page *buffered_rmqueue(struct zone *preferred_zone, |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 2561 | struct zone *zone, unsigned int order, |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 2562 | gfp_t gfp_flags, unsigned int alloc_flags, |
| 2563 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2564 | { |
| 2565 | unsigned long flags; |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 2566 | struct page *page; |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2567 | bool cold = ((gfp_flags & __GFP_COLD) != 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2568 | |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 2569 | if (likely(order == 0)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2570 | struct per_cpu_pages *pcp; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 2571 | struct list_head *list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2572 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2573 | local_irq_save(flags); |
Mel Gorman | 479f854 | 2016-05-19 17:14:35 -0700 | [diff] [blame] | 2574 | do { |
| 2575 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
| 2576 | list = &pcp->lists[migratetype]; |
| 2577 | if (list_empty(list)) { |
| 2578 | pcp->count += rmqueue_bulk(zone, 0, |
| 2579 | pcp->batch, list, |
| 2580 | migratetype, cold); |
| 2581 | if (unlikely(list_empty(list))) |
| 2582 | goto failed; |
| 2583 | } |
Mel Gorman | b92a6ed | 2007-10-16 01:25:50 -0700 | [diff] [blame] | 2584 | |
Mel Gorman | 479f854 | 2016-05-19 17:14:35 -0700 | [diff] [blame] | 2585 | if (cold) |
| 2586 | page = list_last_entry(list, struct page, lru); |
| 2587 | else |
| 2588 | page = list_first_entry(list, struct page, lru); |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 2589 | |
Vlastimil Babka | 83b9355 | 2016-06-03 14:55:52 -0700 | [diff] [blame] | 2590 | __dec_zone_state(zone, NR_ALLOC_BATCH); |
| 2591 | list_del(&page->lru); |
| 2592 | pcp->count--; |
| 2593 | |
| 2594 | } while (check_new_pcp(page)); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2595 | } else { |
Michal Hocko | 0f352e5 | 2016-03-17 14:19:32 -0700 | [diff] [blame] | 2596 | /* |
| 2597 | * We most definitely don't want callers attempting to |
| 2598 | * allocate greater than order-1 page units with __GFP_NOFAIL. |
| 2599 | */ |
| 2600 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2601 | spin_lock_irqsave(&zone->lock, flags); |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2602 | |
Mel Gorman | 479f854 | 2016-05-19 17:14:35 -0700 | [diff] [blame] | 2603 | do { |
| 2604 | page = NULL; |
| 2605 | if (alloc_flags & ALLOC_HARDER) { |
| 2606 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); |
| 2607 | if (page) |
| 2608 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
| 2609 | } |
| 2610 | if (!page) |
| 2611 | page = __rmqueue(zone, order, migratetype); |
| 2612 | } while (page && check_new_pages(page, order)); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 2613 | spin_unlock(&zone->lock); |
| 2614 | if (!page) |
| 2615 | goto failed; |
Mel Gorman | 754078e | 2016-05-19 17:13:33 -0700 | [diff] [blame] | 2616 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order)); |
Bartlomiej Zolnierkiewicz | d1ce749 | 2012-10-08 16:32:02 -0700 | [diff] [blame] | 2617 | __mod_zone_freepage_state(zone, -(1 << order), |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 2618 | get_pcppage_migratetype(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2619 | } |
| 2620 | |
Johannes Weiner | abe5f97 | 2014-10-02 16:21:10 -0700 | [diff] [blame] | 2621 | if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 && |
Johannes Weiner | 5705465 | 2014-10-09 15:28:17 -0700 | [diff] [blame] | 2622 | !test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) |
| 2623 | set_bit(ZONE_FAIR_DEPLETED, &zone->flags); |
Johannes Weiner | 2732936 | 2014-03-03 15:38:41 -0800 | [diff] [blame] | 2624 | |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2625 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
Andi Kleen | 78afd56 | 2011-03-22 16:33:12 -0700 | [diff] [blame] | 2626 | zone_statistics(preferred_zone, zone, gfp_flags); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 2627 | local_irq_restore(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2628 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 2629 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2630 | return page; |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 2631 | |
| 2632 | failed: |
| 2633 | local_irq_restore(flags); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 2634 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2635 | } |
| 2636 | |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2637 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
| 2638 | |
Akinobu Mita | b2588c4 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 2639 | static struct { |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2640 | struct fault_attr attr; |
| 2641 | |
Viresh Kumar | 621a5f7 | 2015-09-26 15:04:07 -0700 | [diff] [blame] | 2642 | bool ignore_gfp_highmem; |
Mel Gorman | 71baba4 | 2015-11-06 16:28:28 -0800 | [diff] [blame] | 2643 | bool ignore_gfp_reclaim; |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 2644 | u32 min_order; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2645 | } fail_page_alloc = { |
| 2646 | .attr = FAULT_ATTR_INITIALIZER, |
Mel Gorman | 71baba4 | 2015-11-06 16:28:28 -0800 | [diff] [blame] | 2647 | .ignore_gfp_reclaim = true, |
Viresh Kumar | 621a5f7 | 2015-09-26 15:04:07 -0700 | [diff] [blame] | 2648 | .ignore_gfp_highmem = true, |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 2649 | .min_order = 1, |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2650 | }; |
| 2651 | |
| 2652 | static int __init setup_fail_page_alloc(char *str) |
| 2653 | { |
| 2654 | return setup_fault_attr(&fail_page_alloc.attr, str); |
| 2655 | } |
| 2656 | __setup("fail_page_alloc=", setup_fail_page_alloc); |
| 2657 | |
Gavin Shan | deaf386 | 2012-07-31 16:41:51 -0700 | [diff] [blame] | 2658 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2659 | { |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 2660 | if (order < fail_page_alloc.min_order) |
Gavin Shan | deaf386 | 2012-07-31 16:41:51 -0700 | [diff] [blame] | 2661 | return false; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2662 | if (gfp_mask & __GFP_NOFAIL) |
Gavin Shan | deaf386 | 2012-07-31 16:41:51 -0700 | [diff] [blame] | 2663 | return false; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2664 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
Gavin Shan | deaf386 | 2012-07-31 16:41:51 -0700 | [diff] [blame] | 2665 | return false; |
Mel Gorman | 71baba4 | 2015-11-06 16:28:28 -0800 | [diff] [blame] | 2666 | if (fail_page_alloc.ignore_gfp_reclaim && |
| 2667 | (gfp_mask & __GFP_DIRECT_RECLAIM)) |
Gavin Shan | deaf386 | 2012-07-31 16:41:51 -0700 | [diff] [blame] | 2668 | return false; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2669 | |
| 2670 | return should_fail(&fail_page_alloc.attr, 1 << order); |
| 2671 | } |
| 2672 | |
| 2673 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS |
| 2674 | |
| 2675 | static int __init fail_page_alloc_debugfs(void) |
| 2676 | { |
Al Viro | f4ae40a | 2011-07-24 04:33:43 -0400 | [diff] [blame] | 2677 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2678 | struct dentry *dir; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2679 | |
Akinobu Mita | dd48c08 | 2011-08-03 16:21:01 -0700 | [diff] [blame] | 2680 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
| 2681 | &fail_page_alloc.attr); |
| 2682 | if (IS_ERR(dir)) |
| 2683 | return PTR_ERR(dir); |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2684 | |
Akinobu Mita | b2588c4 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 2685 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
Mel Gorman | 71baba4 | 2015-11-06 16:28:28 -0800 | [diff] [blame] | 2686 | &fail_page_alloc.ignore_gfp_reclaim)) |
Akinobu Mita | b2588c4 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 2687 | goto fail; |
| 2688 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, |
| 2689 | &fail_page_alloc.ignore_gfp_highmem)) |
| 2690 | goto fail; |
| 2691 | if (!debugfs_create_u32("min-order", mode, dir, |
| 2692 | &fail_page_alloc.min_order)) |
| 2693 | goto fail; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2694 | |
Akinobu Mita | b2588c4 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 2695 | return 0; |
| 2696 | fail: |
Akinobu Mita | dd48c08 | 2011-08-03 16:21:01 -0700 | [diff] [blame] | 2697 | debugfs_remove_recursive(dir); |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2698 | |
Akinobu Mita | b2588c4 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 2699 | return -ENOMEM; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2700 | } |
| 2701 | |
| 2702 | late_initcall(fail_page_alloc_debugfs); |
| 2703 | |
| 2704 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ |
| 2705 | |
| 2706 | #else /* CONFIG_FAIL_PAGE_ALLOC */ |
| 2707 | |
Gavin Shan | deaf386 | 2012-07-31 16:41:51 -0700 | [diff] [blame] | 2708 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2709 | { |
Gavin Shan | deaf386 | 2012-07-31 16:41:51 -0700 | [diff] [blame] | 2710 | return false; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2711 | } |
| 2712 | |
| 2713 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ |
| 2714 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2715 | /* |
Mel Gorman | 97a16fc | 2015-11-06 16:28:40 -0800 | [diff] [blame] | 2716 | * Return true if free base pages are above 'mark'. For high-order checks it |
| 2717 | * will return true of the order-0 watermark is reached and there is at least |
| 2718 | * one free page of a suitable size. Checking now avoids taking the zone lock |
| 2719 | * to check in the allocation paths if no pages are free. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2720 | */ |
Michal Hocko | 86a294a | 2016-05-20 16:57:12 -0700 | [diff] [blame] | 2721 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
| 2722 | int classzone_idx, unsigned int alloc_flags, |
| 2723 | long free_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2724 | { |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 2725 | long min = mark; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2726 | int o; |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 2727 | const bool alloc_harder = (alloc_flags & ALLOC_HARDER); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2728 | |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2729 | /* free_pages may go negative - that's OK */ |
Michal Hocko | df0a6da | 2012-01-10 15:08:02 -0800 | [diff] [blame] | 2730 | free_pages -= (1 << order) - 1; |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2731 | |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2732 | if (alloc_flags & ALLOC_HIGH) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2733 | min -= min / 2; |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2734 | |
| 2735 | /* |
| 2736 | * If the caller does not have rights to ALLOC_HARDER then subtract |
| 2737 | * the high-atomic reserves. This will over-estimate the size of the |
| 2738 | * atomic reserve but it avoids a search. |
| 2739 | */ |
Mel Gorman | 97a16fc | 2015-11-06 16:28:40 -0800 | [diff] [blame] | 2740 | if (likely(!alloc_harder)) |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2741 | free_pages -= z->nr_reserved_highatomic; |
| 2742 | else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2743 | min -= min / 4; |
Mel Gorman | e2b1919 | 2015-11-06 16:28:09 -0800 | [diff] [blame] | 2744 | |
Bartlomiej Zolnierkiewicz | d95ea5d | 2012-10-08 16:32:05 -0700 | [diff] [blame] | 2745 | #ifdef CONFIG_CMA |
| 2746 | /* If allocation can't use CMA areas don't use free CMA pages */ |
| 2747 | if (!(alloc_flags & ALLOC_CMA)) |
Mel Gorman | 97a16fc | 2015-11-06 16:28:40 -0800 | [diff] [blame] | 2748 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); |
Bartlomiej Zolnierkiewicz | d95ea5d | 2012-10-08 16:32:05 -0700 | [diff] [blame] | 2749 | #endif |
Tomasz Stanislawski | 026b081 | 2013-06-12 14:05:02 -0700 | [diff] [blame] | 2750 | |
Mel Gorman | 97a16fc | 2015-11-06 16:28:40 -0800 | [diff] [blame] | 2751 | /* |
| 2752 | * Check watermarks for an order-0 allocation request. If these |
| 2753 | * are not met, then a high-order request also cannot go ahead |
| 2754 | * even if a suitable page happened to be free. |
| 2755 | */ |
| 2756 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2757 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2758 | |
Mel Gorman | 97a16fc | 2015-11-06 16:28:40 -0800 | [diff] [blame] | 2759 | /* If this is an order-0 request then the watermark is fine */ |
| 2760 | if (!order) |
| 2761 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2762 | |
Mel Gorman | 97a16fc | 2015-11-06 16:28:40 -0800 | [diff] [blame] | 2763 | /* For a high-order request, check at least one suitable page is free */ |
| 2764 | for (o = order; o < MAX_ORDER; o++) { |
| 2765 | struct free_area *area = &z->free_area[o]; |
| 2766 | int mt; |
| 2767 | |
| 2768 | if (!area->nr_free) |
| 2769 | continue; |
| 2770 | |
| 2771 | if (alloc_harder) |
| 2772 | return true; |
| 2773 | |
| 2774 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
| 2775 | if (!list_empty(&area->free_list[mt])) |
| 2776 | return true; |
| 2777 | } |
| 2778 | |
| 2779 | #ifdef CONFIG_CMA |
| 2780 | if ((alloc_flags & ALLOC_CMA) && |
| 2781 | !list_empty(&area->free_list[MIGRATE_CMA])) { |
| 2782 | return true; |
| 2783 | } |
| 2784 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2785 | } |
Mel Gorman | 97a16fc | 2015-11-06 16:28:40 -0800 | [diff] [blame] | 2786 | return false; |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2787 | } |
| 2788 | |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 2789 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 2790 | int classzone_idx, unsigned int alloc_flags) |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2791 | { |
| 2792 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, |
| 2793 | zone_page_state(z, NR_FREE_PAGES)); |
| 2794 | } |
| 2795 | |
Mel Gorman | 48ee5f3 | 2016-05-19 17:14:07 -0700 | [diff] [blame] | 2796 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
| 2797 | unsigned long mark, int classzone_idx, unsigned int alloc_flags) |
| 2798 | { |
| 2799 | long free_pages = zone_page_state(z, NR_FREE_PAGES); |
| 2800 | long cma_pages = 0; |
| 2801 | |
| 2802 | #ifdef CONFIG_CMA |
| 2803 | /* If allocation can't use CMA areas don't use free CMA pages */ |
| 2804 | if (!(alloc_flags & ALLOC_CMA)) |
| 2805 | cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES); |
| 2806 | #endif |
| 2807 | |
| 2808 | /* |
| 2809 | * Fast check for order-0 only. If this fails then the reserves |
| 2810 | * need to be calculated. There is a corner case where the check |
| 2811 | * passes but only the high-order atomic reserve are free. If |
| 2812 | * the caller is !atomic then it'll uselessly search the free |
| 2813 | * list. That corner case is then slower but it is harmless. |
| 2814 | */ |
| 2815 | if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx]) |
| 2816 | return true; |
| 2817 | |
| 2818 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, |
| 2819 | free_pages); |
| 2820 | } |
| 2821 | |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 2822 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
Mel Gorman | e2b1919 | 2015-11-06 16:28:09 -0800 | [diff] [blame] | 2823 | unsigned long mark, int classzone_idx) |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2824 | { |
| 2825 | long free_pages = zone_page_state(z, NR_FREE_PAGES); |
| 2826 | |
| 2827 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) |
| 2828 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); |
| 2829 | |
Mel Gorman | e2b1919 | 2015-11-06 16:28:09 -0800 | [diff] [blame] | 2830 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2831 | free_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2832 | } |
| 2833 | |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2834 | #ifdef CONFIG_NUMA |
Johannes Weiner | 81c0a2b | 2013-09-11 14:20:47 -0700 | [diff] [blame] | 2835 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
| 2836 | { |
Johannes Weiner | fff4068 | 2013-12-20 14:54:12 +0000 | [diff] [blame] | 2837 | return local_zone->node == zone->node; |
Johannes Weiner | 81c0a2b | 2013-09-11 14:20:47 -0700 | [diff] [blame] | 2838 | } |
| 2839 | |
David Rientjes | 957f822 | 2012-10-08 16:33:24 -0700 | [diff] [blame] | 2840 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
| 2841 | { |
Mel Gorman | 5f7a75a | 2014-06-04 16:07:15 -0700 | [diff] [blame] | 2842 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) < |
| 2843 | RECLAIM_DISTANCE; |
David Rientjes | 957f822 | 2012-10-08 16:33:24 -0700 | [diff] [blame] | 2844 | } |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2845 | #else /* CONFIG_NUMA */ |
Johannes Weiner | 81c0a2b | 2013-09-11 14:20:47 -0700 | [diff] [blame] | 2846 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
| 2847 | { |
| 2848 | return true; |
| 2849 | } |
| 2850 | |
David Rientjes | 957f822 | 2012-10-08 16:33:24 -0700 | [diff] [blame] | 2851 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
| 2852 | { |
| 2853 | return true; |
| 2854 | } |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2855 | #endif /* CONFIG_NUMA */ |
| 2856 | |
Mel Gorman | 4ffeaf3 | 2014-08-06 16:07:22 -0700 | [diff] [blame] | 2857 | static void reset_alloc_batches(struct zone *preferred_zone) |
| 2858 | { |
| 2859 | struct zone *zone = preferred_zone->zone_pgdat->node_zones; |
| 2860 | |
| 2861 | do { |
| 2862 | mod_zone_page_state(zone, NR_ALLOC_BATCH, |
| 2863 | high_wmark_pages(zone) - low_wmark_pages(zone) - |
| 2864 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); |
Johannes Weiner | 5705465 | 2014-10-09 15:28:17 -0700 | [diff] [blame] | 2865 | clear_bit(ZONE_FAIR_DEPLETED, &zone->flags); |
Mel Gorman | 4ffeaf3 | 2014-08-06 16:07:22 -0700 | [diff] [blame] | 2866 | } while (zone++ != preferred_zone); |
| 2867 | } |
| 2868 | |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2869 | /* |
Paul Jackson | 0798e51 | 2006-12-06 20:31:38 -0800 | [diff] [blame] | 2870 | * get_page_from_freelist goes through the zonelist trying to allocate |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2871 | * a page. |
| 2872 | */ |
| 2873 | static struct page * |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 2874 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
| 2875 | const struct alloc_context *ac) |
Martin Hicks | 753ee72 | 2005-06-21 17:14:41 -0700 | [diff] [blame] | 2876 | { |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 2877 | struct zoneref *z = ac->preferred_zoneref; |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2878 | struct zone *zone; |
Mel Gorman | 3053475 | 2016-05-19 17:14:04 -0700 | [diff] [blame] | 2879 | bool fair_skipped = false; |
| 2880 | bool apply_fair = (alloc_flags & ALLOC_FAIR); |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2881 | |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2882 | zonelist_scan: |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2883 | /* |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2884 | * Scan zonelist, looking for a zone with enough free. |
Vladimir Davydov | 344736f | 2014-10-20 15:50:30 +0400 | [diff] [blame] | 2885 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2886 | */ |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 2887 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 2888 | ac->nodemask) { |
Mel Gorman | be06af0 | 2016-05-19 17:13:47 -0700 | [diff] [blame] | 2889 | struct page *page; |
Johannes Weiner | e085dbc | 2013-09-11 14:20:46 -0700 | [diff] [blame] | 2890 | unsigned long mark; |
| 2891 | |
Mel Gorman | 664eedd | 2014-06-04 16:10:08 -0700 | [diff] [blame] | 2892 | if (cpusets_enabled() && |
| 2893 | (alloc_flags & ALLOC_CPUSET) && |
Vlastimil Babka | 002f290 | 2016-05-19 17:14:30 -0700 | [diff] [blame] | 2894 | !__cpuset_zone_allowed(zone, gfp_mask)) |
Mel Gorman | cd38b11 | 2011-07-25 17:12:29 -0700 | [diff] [blame] | 2895 | continue; |
Johannes Weiner | a756cf5 | 2012-01-10 15:07:49 -0800 | [diff] [blame] | 2896 | /* |
Johannes Weiner | 81c0a2b | 2013-09-11 14:20:47 -0700 | [diff] [blame] | 2897 | * Distribute pages in proportion to the individual |
| 2898 | * zone size to ensure fair page aging. The zone a |
| 2899 | * page was allocated in should have no effect on the |
| 2900 | * time the page has in memory before being reclaimed. |
Johannes Weiner | 81c0a2b | 2013-09-11 14:20:47 -0700 | [diff] [blame] | 2901 | */ |
Mel Gorman | 3053475 | 2016-05-19 17:14:04 -0700 | [diff] [blame] | 2902 | if (apply_fair) { |
Johannes Weiner | 5705465 | 2014-10-09 15:28:17 -0700 | [diff] [blame] | 2903 | if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) { |
Mel Gorman | fa379b9 | 2016-05-19 17:13:41 -0700 | [diff] [blame] | 2904 | fair_skipped = true; |
Johannes Weiner | 3a02576 | 2014-04-07 15:37:48 -0700 | [diff] [blame] | 2905 | continue; |
Mel Gorman | 4ffeaf3 | 2014-08-06 16:07:22 -0700 | [diff] [blame] | 2906 | } |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 2907 | if (!zone_local(ac->preferred_zoneref->zone, zone)) { |
Mel Gorman | 3053475 | 2016-05-19 17:14:04 -0700 | [diff] [blame] | 2908 | if (fair_skipped) |
| 2909 | goto reset_fair; |
| 2910 | apply_fair = false; |
| 2911 | } |
Johannes Weiner | 81c0a2b | 2013-09-11 14:20:47 -0700 | [diff] [blame] | 2912 | } |
| 2913 | /* |
Johannes Weiner | a756cf5 | 2012-01-10 15:07:49 -0800 | [diff] [blame] | 2914 | * When allocating a page cache page for writing, we |
| 2915 | * want to get it from a zone that is within its dirty |
| 2916 | * limit, such that no single zone holds more than its |
| 2917 | * proportional share of globally allowed dirty pages. |
| 2918 | * The dirty limits take into account the zone's |
| 2919 | * lowmem reserves and high watermark so that kswapd |
| 2920 | * should be able to balance it without having to |
| 2921 | * write pages from its LRU list. |
| 2922 | * |
| 2923 | * This may look like it could increase pressure on |
| 2924 | * lower zones by failing allocations in higher zones |
| 2925 | * before they are full. But the pages that do spill |
| 2926 | * over are limited as the lower zones are protected |
| 2927 | * by this very same mechanism. It should not become |
| 2928 | * a practical burden to them. |
| 2929 | * |
| 2930 | * XXX: For now, allow allocations to potentially |
| 2931 | * exceed the per-zone dirty limit in the slowpath |
Mel Gorman | c9ab0c4 | 2015-11-06 16:28:12 -0800 | [diff] [blame] | 2932 | * (spread_dirty_pages unset) before going into reclaim, |
Johannes Weiner | a756cf5 | 2012-01-10 15:07:49 -0800 | [diff] [blame] | 2933 | * which is important when on a NUMA setup the allowed |
| 2934 | * zones are together not big enough to reach the |
| 2935 | * global limit. The proper fix for these situations |
| 2936 | * will require awareness of zones in the |
| 2937 | * dirty-throttling and the flusher threads. |
| 2938 | */ |
Mel Gorman | c9ab0c4 | 2015-11-06 16:28:12 -0800 | [diff] [blame] | 2939 | if (ac->spread_dirty_pages && !zone_dirty_ok(zone)) |
Mel Gorman | 800a1e7 | 2014-06-04 16:10:06 -0700 | [diff] [blame] | 2940 | continue; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2941 | |
Johannes Weiner | e085dbc | 2013-09-11 14:20:46 -0700 | [diff] [blame] | 2942 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
Mel Gorman | 48ee5f3 | 2016-05-19 17:14:07 -0700 | [diff] [blame] | 2943 | if (!zone_watermark_fast(zone, order, mark, |
Mel Gorman | 93ea996 | 2016-05-19 17:14:13 -0700 | [diff] [blame] | 2944 | ac_classzone_idx(ac), alloc_flags)) { |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 2945 | int ret; |
| 2946 | |
Mel Gorman | 5dab291 | 2014-06-04 16:10:14 -0700 | [diff] [blame] | 2947 | /* Checked here to keep the fast path fast */ |
| 2948 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
| 2949 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
| 2950 | goto try_this_zone; |
| 2951 | |
David Rientjes | 957f822 | 2012-10-08 16:33:24 -0700 | [diff] [blame] | 2952 | if (zone_reclaim_mode == 0 || |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 2953 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
Mel Gorman | cd38b11 | 2011-07-25 17:12:29 -0700 | [diff] [blame] | 2954 | continue; |
| 2955 | |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 2956 | ret = zone_reclaim(zone, gfp_mask, order); |
| 2957 | switch (ret) { |
| 2958 | case ZONE_RECLAIM_NOSCAN: |
| 2959 | /* did not scan */ |
Mel Gorman | cd38b11 | 2011-07-25 17:12:29 -0700 | [diff] [blame] | 2960 | continue; |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 2961 | case ZONE_RECLAIM_FULL: |
| 2962 | /* scanned but unreclaimable */ |
Mel Gorman | cd38b11 | 2011-07-25 17:12:29 -0700 | [diff] [blame] | 2963 | continue; |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 2964 | default: |
| 2965 | /* did we reclaim enough */ |
Mel Gorman | fed2719 | 2013-04-29 15:07:57 -0700 | [diff] [blame] | 2966 | if (zone_watermark_ok(zone, order, mark, |
Mel Gorman | 93ea996 | 2016-05-19 17:14:13 -0700 | [diff] [blame] | 2967 | ac_classzone_idx(ac), alloc_flags)) |
Mel Gorman | fed2719 | 2013-04-29 15:07:57 -0700 | [diff] [blame] | 2968 | goto try_this_zone; |
| 2969 | |
Mel Gorman | fed2719 | 2013-04-29 15:07:57 -0700 | [diff] [blame] | 2970 | continue; |
Paul Jackson | 0798e51 | 2006-12-06 20:31:38 -0800 | [diff] [blame] | 2971 | } |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2972 | } |
| 2973 | |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 2974 | try_this_zone: |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 2975 | page = buffered_rmqueue(ac->preferred_zoneref->zone, zone, order, |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2976 | gfp_mask, alloc_flags, ac->migratetype); |
Vlastimil Babka | 7537919 | 2015-02-11 15:25:38 -0800 | [diff] [blame] | 2977 | if (page) { |
Mel Gorman | 479f854 | 2016-05-19 17:14:35 -0700 | [diff] [blame] | 2978 | prep_new_page(page, order, gfp_mask, alloc_flags); |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2979 | |
| 2980 | /* |
| 2981 | * If this is a high-order atomic allocation then check |
| 2982 | * if the pageblock should be reserved for the future |
| 2983 | */ |
| 2984 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) |
| 2985 | reserve_highatomic_pageblock(page, zone, order); |
| 2986 | |
Vlastimil Babka | 7537919 | 2015-02-11 15:25:38 -0800 | [diff] [blame] | 2987 | return page; |
| 2988 | } |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2989 | } |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2990 | |
Mel Gorman | 4ffeaf3 | 2014-08-06 16:07:22 -0700 | [diff] [blame] | 2991 | /* |
| 2992 | * The first pass makes sure allocations are spread fairly within the |
| 2993 | * local node. However, the local node might have free pages left |
| 2994 | * after the fairness batches are exhausted, and remote zones haven't |
| 2995 | * even been considered yet. Try once more without fairness, and |
| 2996 | * include remote zones now, before entering the slowpath and waking |
| 2997 | * kswapd: prefer spilling to a remote zone over swapping locally. |
| 2998 | */ |
Mel Gorman | 3053475 | 2016-05-19 17:14:04 -0700 | [diff] [blame] | 2999 | if (fair_skipped) { |
| 3000 | reset_fair: |
| 3001 | apply_fair = false; |
| 3002 | fair_skipped = false; |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 3003 | reset_alloc_batches(ac->preferred_zoneref->zone); |
Mel Gorman | 0d0bd89 | 2016-06-03 14:55:58 -0700 | [diff] [blame] | 3004 | z = ac->preferred_zoneref; |
Mel Gorman | 4ffeaf3 | 2014-08-06 16:07:22 -0700 | [diff] [blame] | 3005 | goto zonelist_scan; |
Mel Gorman | 3053475 | 2016-05-19 17:14:04 -0700 | [diff] [blame] | 3006 | } |
Mel Gorman | 4ffeaf3 | 2014-08-06 16:07:22 -0700 | [diff] [blame] | 3007 | |
| 3008 | return NULL; |
Martin Hicks | 753ee72 | 2005-06-21 17:14:41 -0700 | [diff] [blame] | 3009 | } |
| 3010 | |
David Rientjes | 29423e77 | 2011-03-22 16:30:47 -0700 | [diff] [blame] | 3011 | /* |
| 3012 | * Large machines with many possible nodes should not always dump per-node |
| 3013 | * meminfo in irq context. |
| 3014 | */ |
| 3015 | static inline bool should_suppress_show_mem(void) |
| 3016 | { |
| 3017 | bool ret = false; |
| 3018 | |
| 3019 | #if NODES_SHIFT > 8 |
| 3020 | ret = in_interrupt(); |
| 3021 | #endif |
| 3022 | return ret; |
| 3023 | } |
| 3024 | |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 3025 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
| 3026 | DEFAULT_RATELIMIT_INTERVAL, |
| 3027 | DEFAULT_RATELIMIT_BURST); |
| 3028 | |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 3029 | void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...) |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 3030 | { |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 3031 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
| 3032 | |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 3033 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
| 3034 | debug_guardpage_minorder() > 0) |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 3035 | return; |
| 3036 | |
| 3037 | /* |
| 3038 | * This documents exceptions given to allocations in certain |
| 3039 | * contexts that are allowed to allocate outside current's set |
| 3040 | * of allowed nodes. |
| 3041 | */ |
| 3042 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
| 3043 | if (test_thread_flag(TIF_MEMDIE) || |
| 3044 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
| 3045 | filter &= ~SHOW_MEM_FILTER_NODES; |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3046 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 3047 | filter &= ~SHOW_MEM_FILTER_NODES; |
| 3048 | |
| 3049 | if (fmt) { |
Joe Perches | 3ee9a4f | 2011-10-31 17:08:35 -0700 | [diff] [blame] | 3050 | struct va_format vaf; |
| 3051 | va_list args; |
| 3052 | |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 3053 | va_start(args, fmt); |
Joe Perches | 3ee9a4f | 2011-10-31 17:08:35 -0700 | [diff] [blame] | 3054 | |
| 3055 | vaf.fmt = fmt; |
| 3056 | vaf.va = &args; |
| 3057 | |
| 3058 | pr_warn("%pV", &vaf); |
| 3059 | |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 3060 | va_end(args); |
| 3061 | } |
| 3062 | |
Vlastimil Babka | c5c990e | 2016-03-15 14:56:02 -0700 | [diff] [blame] | 3063 | pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n", |
| 3064 | current->comm, order, gfp_mask, &gfp_mask); |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 3065 | dump_stack(); |
| 3066 | if (!should_suppress_show_mem()) |
| 3067 | show_mem(filter); |
| 3068 | } |
| 3069 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3070 | static inline struct page * |
| 3071 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3072 | const struct alloc_context *ac, unsigned long *did_some_progress) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3073 | { |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 3074 | struct oom_control oc = { |
| 3075 | .zonelist = ac->zonelist, |
| 3076 | .nodemask = ac->nodemask, |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 3077 | .memcg = NULL, |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 3078 | .gfp_mask = gfp_mask, |
| 3079 | .order = order, |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 3080 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3081 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3082 | |
Johannes Weiner | 9879de7 | 2015-01-26 12:58:32 -0800 | [diff] [blame] | 3083 | *did_some_progress = 0; |
| 3084 | |
Johannes Weiner | 9879de7 | 2015-01-26 12:58:32 -0800 | [diff] [blame] | 3085 | /* |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 3086 | * Acquire the oom lock. If that fails, somebody else is |
| 3087 | * making progress for us. |
Johannes Weiner | 9879de7 | 2015-01-26 12:58:32 -0800 | [diff] [blame] | 3088 | */ |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 3089 | if (!mutex_trylock(&oom_lock)) { |
Johannes Weiner | 9879de7 | 2015-01-26 12:58:32 -0800 | [diff] [blame] | 3090 | *did_some_progress = 1; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3091 | schedule_timeout_uninterruptible(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3092 | return NULL; |
| 3093 | } |
Jens Axboe | 6b1de91 | 2005-11-17 21:35:02 +0100 | [diff] [blame] | 3094 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3095 | /* |
| 3096 | * Go through the zonelist yet one more time, keep very high watermark |
| 3097 | * here, this is only to catch a parallel oom killing, we must fail if |
| 3098 | * we're still under heavy pressure. |
| 3099 | */ |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3100 | page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order, |
| 3101 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 3102 | if (page) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3103 | goto out; |
| 3104 | |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 3105 | if (!(gfp_mask & __GFP_NOFAIL)) { |
Johannes Weiner | 9879de7 | 2015-01-26 12:58:32 -0800 | [diff] [blame] | 3106 | /* Coredumps can quickly deplete all memory reserves */ |
| 3107 | if (current->flags & PF_DUMPCORE) |
| 3108 | goto out; |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 3109 | /* The OOM killer will not help higher order allocs */ |
| 3110 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
| 3111 | goto out; |
David Rientjes | 03668b3 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 3112 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3113 | if (ac->high_zoneidx < ZONE_NORMAL) |
David Rientjes | 03668b3 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 3114 | goto out; |
Johannes Weiner | 9083905 | 2015-06-24 16:57:21 -0700 | [diff] [blame] | 3115 | if (pm_suspended_storage()) |
| 3116 | goto out; |
Michal Hocko | 3da88fb | 2016-05-19 17:13:09 -0700 | [diff] [blame] | 3117 | /* |
| 3118 | * XXX: GFP_NOFS allocations should rather fail than rely on |
| 3119 | * other request to make a forward progress. |
| 3120 | * We are in an unfortunate situation where out_of_memory cannot |
| 3121 | * do much for this context but let's try it to at least get |
| 3122 | * access to memory reserved if the current task is killed (see |
| 3123 | * out_of_memory). Once filesystems are ready to handle allocation |
| 3124 | * failures more gracefully we should just bail out here. |
| 3125 | */ |
| 3126 | |
David Rientjes | 4167e9b | 2015-04-14 15:46:55 -0700 | [diff] [blame] | 3127 | /* The OOM killer may not free memory on a specific node */ |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 3128 | if (gfp_mask & __GFP_THISNODE) |
| 3129 | goto out; |
| 3130 | } |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3131 | /* Exhausted what can be done so it's blamo time */ |
Michal Hocko | 5020e28 | 2016-01-14 15:20:36 -0800 | [diff] [blame] | 3132 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 3133 | *did_some_progress = 1; |
Michal Hocko | 5020e28 | 2016-01-14 15:20:36 -0800 | [diff] [blame] | 3134 | |
| 3135 | if (gfp_mask & __GFP_NOFAIL) { |
| 3136 | page = get_page_from_freelist(gfp_mask, order, |
| 3137 | ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac); |
| 3138 | /* |
| 3139 | * fallback to ignore cpuset restriction if our nodes |
| 3140 | * are depleted |
| 3141 | */ |
| 3142 | if (!page) |
| 3143 | page = get_page_from_freelist(gfp_mask, order, |
| 3144 | ALLOC_NO_WATERMARKS, ac); |
| 3145 | } |
| 3146 | } |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3147 | out: |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 3148 | mutex_unlock(&oom_lock); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3149 | return page; |
| 3150 | } |
| 3151 | |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3152 | |
| 3153 | /* |
| 3154 | * Maximum number of compaction retries wit a progress before OOM |
| 3155 | * killer is consider as the only way to move forward. |
| 3156 | */ |
| 3157 | #define MAX_COMPACT_RETRIES 16 |
| 3158 | |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3159 | #ifdef CONFIG_COMPACTION |
| 3160 | /* Try memory compaction for high-order allocations before reclaim */ |
| 3161 | static struct page * |
| 3162 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 3163 | unsigned int alloc_flags, const struct alloc_context *ac, |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3164 | enum migrate_mode mode, enum compact_result *compact_result) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3165 | { |
Vlastimil Babka | 98dd3b4 | 2014-10-09 15:27:04 -0700 | [diff] [blame] | 3166 | struct page *page; |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3167 | int contended_compaction; |
Vlastimil Babka | 53853e2 | 2014-10-09 15:27:02 -0700 | [diff] [blame] | 3168 | |
Mel Gorman | 6619971 | 2012-01-12 17:19:41 -0800 | [diff] [blame] | 3169 | if (!order) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3170 | return NULL; |
| 3171 | |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 3172 | current->flags |= PF_MEMALLOC; |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3173 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
| 3174 | mode, &contended_compaction); |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 3175 | current->flags &= ~PF_MEMALLOC; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3176 | |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3177 | if (*compact_result <= COMPACT_INACTIVE) |
Vlastimil Babka | 98dd3b4 | 2014-10-09 15:27:04 -0700 | [diff] [blame] | 3178 | return NULL; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3179 | |
Vlastimil Babka | 98dd3b4 | 2014-10-09 15:27:04 -0700 | [diff] [blame] | 3180 | /* |
| 3181 | * At least in one zone compaction wasn't deferred or skipped, so let's |
| 3182 | * count a compaction stall |
| 3183 | */ |
| 3184 | count_vm_event(COMPACTSTALL); |
| 3185 | |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3186 | page = get_page_from_freelist(gfp_mask, order, |
| 3187 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); |
Vlastimil Babka | 98dd3b4 | 2014-10-09 15:27:04 -0700 | [diff] [blame] | 3188 | |
| 3189 | if (page) { |
| 3190 | struct zone *zone = page_zone(page); |
| 3191 | |
| 3192 | zone->compact_blockskip_flush = false; |
| 3193 | compaction_defer_reset(zone, order, true); |
| 3194 | count_vm_event(COMPACTSUCCESS); |
| 3195 | return page; |
| 3196 | } |
| 3197 | |
| 3198 | /* |
Vlastimil Babka | 98dd3b4 | 2014-10-09 15:27:04 -0700 | [diff] [blame] | 3199 | * It's bad if compaction run occurs and fails. The most likely reason |
| 3200 | * is that pages exist, but not enough to satisfy watermarks. |
| 3201 | */ |
| 3202 | count_vm_event(COMPACTFAIL); |
| 3203 | |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3204 | /* |
| 3205 | * In all zones where compaction was attempted (and not |
| 3206 | * deferred or skipped), lock contention has been detected. |
| 3207 | * For THP allocation we do not want to disrupt the others |
| 3208 | * so we fallback to base pages instead. |
| 3209 | */ |
| 3210 | if (contended_compaction == COMPACT_CONTENDED_LOCK) |
| 3211 | *compact_result = COMPACT_CONTENDED; |
| 3212 | |
| 3213 | /* |
| 3214 | * If compaction was aborted due to need_resched(), we do not |
| 3215 | * want to further increase allocation latency, unless it is |
| 3216 | * khugepaged trying to collapse. |
| 3217 | */ |
| 3218 | if (contended_compaction == COMPACT_CONTENDED_SCHED |
| 3219 | && !(current->flags & PF_KTHREAD)) |
| 3220 | *compact_result = COMPACT_CONTENDED; |
| 3221 | |
Vlastimil Babka | 98dd3b4 | 2014-10-09 15:27:04 -0700 | [diff] [blame] | 3222 | cond_resched(); |
| 3223 | |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3224 | return NULL; |
| 3225 | } |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3226 | |
| 3227 | static inline bool |
Michal Hocko | 86a294a | 2016-05-20 16:57:12 -0700 | [diff] [blame] | 3228 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, |
| 3229 | enum compact_result compact_result, enum migrate_mode *migrate_mode, |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3230 | int compaction_retries) |
| 3231 | { |
Michal Hocko | 7854ea6 | 2016-05-20 16:57:09 -0700 | [diff] [blame] | 3232 | int max_retries = MAX_COMPACT_RETRIES; |
| 3233 | |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3234 | if (!order) |
| 3235 | return false; |
| 3236 | |
| 3237 | /* |
| 3238 | * compaction considers all the zone as desperately out of memory |
| 3239 | * so it doesn't really make much sense to retry except when the |
| 3240 | * failure could be caused by weak migration mode. |
| 3241 | */ |
| 3242 | if (compaction_failed(compact_result)) { |
| 3243 | if (*migrate_mode == MIGRATE_ASYNC) { |
| 3244 | *migrate_mode = MIGRATE_SYNC_LIGHT; |
| 3245 | return true; |
| 3246 | } |
| 3247 | return false; |
| 3248 | } |
| 3249 | |
| 3250 | /* |
Michal Hocko | 7854ea6 | 2016-05-20 16:57:09 -0700 | [diff] [blame] | 3251 | * make sure the compaction wasn't deferred or didn't bail out early |
| 3252 | * due to locks contention before we declare that we should give up. |
Michal Hocko | 86a294a | 2016-05-20 16:57:12 -0700 | [diff] [blame] | 3253 | * But do not retry if the given zonelist is not suitable for |
| 3254 | * compaction. |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3255 | */ |
Michal Hocko | 7854ea6 | 2016-05-20 16:57:09 -0700 | [diff] [blame] | 3256 | if (compaction_withdrawn(compact_result)) |
Michal Hocko | 86a294a | 2016-05-20 16:57:12 -0700 | [diff] [blame] | 3257 | return compaction_zonelist_suitable(ac, order, alloc_flags); |
Michal Hocko | 7854ea6 | 2016-05-20 16:57:09 -0700 | [diff] [blame] | 3258 | |
| 3259 | /* |
| 3260 | * !costly requests are much more important than __GFP_REPEAT |
| 3261 | * costly ones because they are de facto nofail and invoke OOM |
| 3262 | * killer to move on while costly can fail and users are ready |
| 3263 | * to cope with that. 1/4 retries is rather arbitrary but we |
| 3264 | * would need much more detailed feedback from compaction to |
| 3265 | * make a better decision. |
| 3266 | */ |
| 3267 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
| 3268 | max_retries /= 4; |
| 3269 | if (compaction_retries <= max_retries) |
| 3270 | return true; |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3271 | |
| 3272 | return false; |
| 3273 | } |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3274 | #else |
| 3275 | static inline struct page * |
| 3276 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 3277 | unsigned int alloc_flags, const struct alloc_context *ac, |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3278 | enum migrate_mode mode, enum compact_result *compact_result) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3279 | { |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3280 | *compact_result = COMPACT_SKIPPED; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3281 | return NULL; |
| 3282 | } |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3283 | |
| 3284 | static inline bool |
Michal Hocko | 86a294a | 2016-05-20 16:57:12 -0700 | [diff] [blame] | 3285 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
| 3286 | enum compact_result compact_result, |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3287 | enum migrate_mode *migrate_mode, |
| 3288 | int compaction_retries) |
| 3289 | { |
Michal Hocko | 31e49bf | 2016-05-20 16:57:15 -0700 | [diff] [blame] | 3290 | struct zone *zone; |
| 3291 | struct zoneref *z; |
| 3292 | |
| 3293 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) |
| 3294 | return false; |
| 3295 | |
| 3296 | /* |
| 3297 | * There are setups with compaction disabled which would prefer to loop |
| 3298 | * inside the allocator rather than hit the oom killer prematurely. |
| 3299 | * Let's give them a good hope and keep retrying while the order-0 |
| 3300 | * watermarks are OK. |
| 3301 | */ |
| 3302 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
| 3303 | ac->nodemask) { |
| 3304 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), |
| 3305 | ac_classzone_idx(ac), alloc_flags)) |
| 3306 | return true; |
| 3307 | } |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3308 | return false; |
| 3309 | } |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3310 | #endif /* CONFIG_COMPACTION */ |
| 3311 | |
Marek Szyprowski | bba9071 | 2012-01-25 12:09:52 +0100 | [diff] [blame] | 3312 | /* Perform direct synchronous page reclaim */ |
| 3313 | static int |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3314 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
| 3315 | const struct alloc_context *ac) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3316 | { |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3317 | struct reclaim_state reclaim_state; |
Marek Szyprowski | bba9071 | 2012-01-25 12:09:52 +0100 | [diff] [blame] | 3318 | int progress; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3319 | |
| 3320 | cond_resched(); |
| 3321 | |
| 3322 | /* We now go into synchronous reclaim */ |
| 3323 | cpuset_memory_pressure_bump(); |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 3324 | current->flags |= PF_MEMALLOC; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3325 | lockdep_set_current_reclaim_state(gfp_mask); |
| 3326 | reclaim_state.reclaimed_slab = 0; |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 3327 | current->reclaim_state = &reclaim_state; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3328 | |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3329 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
| 3330 | ac->nodemask); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3331 | |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 3332 | current->reclaim_state = NULL; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3333 | lockdep_clear_current_reclaim_state(); |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 3334 | current->flags &= ~PF_MEMALLOC; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3335 | |
| 3336 | cond_resched(); |
| 3337 | |
Marek Szyprowski | bba9071 | 2012-01-25 12:09:52 +0100 | [diff] [blame] | 3338 | return progress; |
| 3339 | } |
| 3340 | |
| 3341 | /* The really slow allocator path where we enter direct reclaim */ |
| 3342 | static inline struct page * |
| 3343 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 3344 | unsigned int alloc_flags, const struct alloc_context *ac, |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3345 | unsigned long *did_some_progress) |
Marek Szyprowski | bba9071 | 2012-01-25 12:09:52 +0100 | [diff] [blame] | 3346 | { |
| 3347 | struct page *page = NULL; |
| 3348 | bool drained = false; |
| 3349 | |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3350 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
Mel Gorman | 9ee493c | 2010-09-09 16:38:18 -0700 | [diff] [blame] | 3351 | if (unlikely(!(*did_some_progress))) |
| 3352 | return NULL; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3353 | |
Mel Gorman | 9ee493c | 2010-09-09 16:38:18 -0700 | [diff] [blame] | 3354 | retry: |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3355 | page = get_page_from_freelist(gfp_mask, order, |
| 3356 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); |
Mel Gorman | 9ee493c | 2010-09-09 16:38:18 -0700 | [diff] [blame] | 3357 | |
| 3358 | /* |
| 3359 | * If an allocation failed after direct reclaim, it could be because |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 3360 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
| 3361 | * Shrink them them and try again |
Mel Gorman | 9ee493c | 2010-09-09 16:38:18 -0700 | [diff] [blame] | 3362 | */ |
| 3363 | if (!page && !drained) { |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 3364 | unreserve_highatomic_pageblock(ac); |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 3365 | drain_all_pages(NULL); |
Mel Gorman | 9ee493c | 2010-09-09 16:38:18 -0700 | [diff] [blame] | 3366 | drained = true; |
| 3367 | goto retry; |
| 3368 | } |
| 3369 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3370 | return page; |
| 3371 | } |
| 3372 | |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3373 | static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3374 | { |
| 3375 | struct zoneref *z; |
| 3376 | struct zone *zone; |
| 3377 | |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3378 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
| 3379 | ac->high_zoneidx, ac->nodemask) |
Mel Gorman | 93ea996 | 2016-05-19 17:14:13 -0700 | [diff] [blame] | 3380 | wakeup_kswapd(zone, order, ac_classzone_idx(ac)); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3381 | } |
| 3382 | |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 3383 | static inline unsigned int |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3384 | gfp_to_alloc_flags(gfp_t gfp_mask) |
| 3385 | { |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 3386 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3387 | |
Mel Gorman | a56f57f | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3388 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
Namhyung Kim | e6223a3 | 2010-10-26 14:21:59 -0700 | [diff] [blame] | 3389 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
Mel Gorman | a56f57f | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3390 | |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3391 | /* |
| 3392 | * The caller may dip into page reserves a bit more if the caller |
| 3393 | * cannot run direct reclaim, or if the caller has realtime scheduling |
| 3394 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3395 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3396 | */ |
Namhyung Kim | e6223a3 | 2010-10-26 14:21:59 -0700 | [diff] [blame] | 3397 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3398 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3399 | if (gfp_mask & __GFP_ATOMIC) { |
Andrea Arcangeli | 5c3240d | 2011-01-13 15:46:49 -0800 | [diff] [blame] | 3400 | /* |
David Rientjes | b104a35 | 2014-07-30 16:08:24 -0700 | [diff] [blame] | 3401 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
| 3402 | * if it can't schedule. |
Andrea Arcangeli | 5c3240d | 2011-01-13 15:46:49 -0800 | [diff] [blame] | 3403 | */ |
David Rientjes | b104a35 | 2014-07-30 16:08:24 -0700 | [diff] [blame] | 3404 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
Andrea Arcangeli | 5c3240d | 2011-01-13 15:46:49 -0800 | [diff] [blame] | 3405 | alloc_flags |= ALLOC_HARDER; |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3406 | /* |
David Rientjes | b104a35 | 2014-07-30 16:08:24 -0700 | [diff] [blame] | 3407 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
Vladimir Davydov | 344736f | 2014-10-20 15:50:30 +0400 | [diff] [blame] | 3408 | * comment for __cpuset_node_allowed(). |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3409 | */ |
| 3410 | alloc_flags &= ~ALLOC_CPUSET; |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 3411 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3412 | alloc_flags |= ALLOC_HARDER; |
| 3413 | |
Mel Gorman | b37f1dd | 2012-07-31 16:44:03 -0700 | [diff] [blame] | 3414 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
| 3415 | if (gfp_mask & __GFP_MEMALLOC) |
| 3416 | alloc_flags |= ALLOC_NO_WATERMARKS; |
Mel Gorman | 907aed4 | 2012-07-31 16:44:07 -0700 | [diff] [blame] | 3417 | else if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
| 3418 | alloc_flags |= ALLOC_NO_WATERMARKS; |
| 3419 | else if (!in_interrupt() && |
| 3420 | ((current->flags & PF_MEMALLOC) || |
| 3421 | unlikely(test_thread_flag(TIF_MEMDIE)))) |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3422 | alloc_flags |= ALLOC_NO_WATERMARKS; |
| 3423 | } |
Bartlomiej Zolnierkiewicz | d95ea5d | 2012-10-08 16:32:05 -0700 | [diff] [blame] | 3424 | #ifdef CONFIG_CMA |
David Rientjes | 43e7a34 | 2014-10-09 15:27:25 -0700 | [diff] [blame] | 3425 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
Bartlomiej Zolnierkiewicz | d95ea5d | 2012-10-08 16:32:05 -0700 | [diff] [blame] | 3426 | alloc_flags |= ALLOC_CMA; |
| 3427 | #endif |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3428 | return alloc_flags; |
| 3429 | } |
| 3430 | |
Mel Gorman | 072bb0a | 2012-07-31 16:43:58 -0700 | [diff] [blame] | 3431 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
| 3432 | { |
Mel Gorman | b37f1dd | 2012-07-31 16:44:03 -0700 | [diff] [blame] | 3433 | return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS); |
Mel Gorman | 072bb0a | 2012-07-31 16:43:58 -0700 | [diff] [blame] | 3434 | } |
| 3435 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3436 | static inline bool is_thp_gfp_mask(gfp_t gfp_mask) |
| 3437 | { |
| 3438 | return (gfp_mask & (GFP_TRANSHUGE | __GFP_KSWAPD_RECLAIM)) == GFP_TRANSHUGE; |
| 3439 | } |
| 3440 | |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3441 | /* |
| 3442 | * Maximum number of reclaim retries without any progress before OOM killer |
| 3443 | * is consider as the only way to move forward. |
| 3444 | */ |
| 3445 | #define MAX_RECLAIM_RETRIES 16 |
| 3446 | |
| 3447 | /* |
| 3448 | * Checks whether it makes sense to retry the reclaim to make a forward progress |
| 3449 | * for the given allocation request. |
| 3450 | * The reclaim feedback represented by did_some_progress (any progress during |
Michal Hocko | 7854ea6 | 2016-05-20 16:57:09 -0700 | [diff] [blame] | 3451 | * the last reclaim round) and no_progress_loops (number of reclaim rounds without |
| 3452 | * any progress in a row) is considered as well as the reclaimable pages on the |
| 3453 | * applicable zone list (with a backoff mechanism which is a function of |
| 3454 | * no_progress_loops). |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3455 | * |
| 3456 | * Returns true if a retry is viable or false to enter the oom path. |
| 3457 | */ |
| 3458 | static inline bool |
| 3459 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, |
| 3460 | struct alloc_context *ac, int alloc_flags, |
Michal Hocko | 7854ea6 | 2016-05-20 16:57:09 -0700 | [diff] [blame] | 3461 | bool did_some_progress, int no_progress_loops) |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3462 | { |
| 3463 | struct zone *zone; |
| 3464 | struct zoneref *z; |
| 3465 | |
| 3466 | /* |
| 3467 | * Make sure we converge to OOM if we cannot make any progress |
| 3468 | * several times in the row. |
| 3469 | */ |
| 3470 | if (no_progress_loops > MAX_RECLAIM_RETRIES) |
| 3471 | return false; |
| 3472 | |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3473 | /* |
| 3474 | * Keep reclaiming pages while there is a chance this will lead somewhere. |
| 3475 | * If none of the target zones can satisfy our allocation request even |
| 3476 | * if all reclaimable pages are considered then we are screwed and have |
| 3477 | * to go OOM. |
| 3478 | */ |
| 3479 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
| 3480 | ac->nodemask) { |
| 3481 | unsigned long available; |
Michal Hocko | ede3771 | 2016-05-20 16:57:03 -0700 | [diff] [blame] | 3482 | unsigned long reclaimable; |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3483 | |
Michal Hocko | ede3771 | 2016-05-20 16:57:03 -0700 | [diff] [blame] | 3484 | available = reclaimable = zone_reclaimable_pages(zone); |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3485 | available -= DIV_ROUND_UP(no_progress_loops * available, |
| 3486 | MAX_RECLAIM_RETRIES); |
| 3487 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
| 3488 | |
| 3489 | /* |
| 3490 | * Would the allocation succeed if we reclaimed the whole |
| 3491 | * available? |
| 3492 | */ |
| 3493 | if (__zone_watermark_ok(zone, order, min_wmark_pages(zone), |
Michal Hocko | ede3771 | 2016-05-20 16:57:03 -0700 | [diff] [blame] | 3494 | ac_classzone_idx(ac), alloc_flags, available)) { |
| 3495 | /* |
| 3496 | * If we didn't make any progress and have a lot of |
| 3497 | * dirty + writeback pages then we should wait for |
| 3498 | * an IO to complete to slow down the reclaim and |
| 3499 | * prevent from pre mature OOM |
| 3500 | */ |
| 3501 | if (!did_some_progress) { |
| 3502 | unsigned long writeback; |
| 3503 | unsigned long dirty; |
| 3504 | |
| 3505 | writeback = zone_page_state_snapshot(zone, |
| 3506 | NR_WRITEBACK); |
| 3507 | dirty = zone_page_state_snapshot(zone, NR_FILE_DIRTY); |
| 3508 | |
| 3509 | if (2*(writeback + dirty) > reclaimable) { |
| 3510 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
| 3511 | return true; |
| 3512 | } |
| 3513 | } |
| 3514 | |
| 3515 | /* |
| 3516 | * Memory allocation/reclaim might be called from a WQ |
| 3517 | * context and the current implementation of the WQ |
| 3518 | * concurrency control doesn't recognize that |
| 3519 | * a particular WQ is congested if the worker thread is |
| 3520 | * looping without ever sleeping. Therefore we have to |
| 3521 | * do a short sleep here rather than calling |
| 3522 | * cond_resched(). |
| 3523 | */ |
| 3524 | if (current->flags & PF_WQ_WORKER) |
| 3525 | schedule_timeout_uninterruptible(1); |
| 3526 | else |
| 3527 | cond_resched(); |
| 3528 | |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3529 | return true; |
| 3530 | } |
| 3531 | } |
| 3532 | |
| 3533 | return false; |
| 3534 | } |
| 3535 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3536 | static inline struct page * |
| 3537 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3538 | struct alloc_context *ac) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3539 | { |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3540 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3541 | struct page *page = NULL; |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 3542 | unsigned int alloc_flags; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3543 | unsigned long did_some_progress; |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 3544 | enum migrate_mode migration_mode = MIGRATE_ASYNC; |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3545 | enum compact_result compact_result; |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3546 | int compaction_retries = 0; |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3547 | int no_progress_loops = 0; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 3548 | |
Christoph Lameter | 952f3b5 | 2006-12-06 20:33:26 -0800 | [diff] [blame] | 3549 | /* |
Mel Gorman | 72807a7 | 2009-06-16 15:32:18 -0700 | [diff] [blame] | 3550 | * In the slowpath, we sanity check order to avoid ever trying to |
| 3551 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may |
| 3552 | * be using allocators in order of preference for an area that is |
| 3553 | * too large. |
| 3554 | */ |
Mel Gorman | 1fc28b7 | 2009-07-29 15:04:08 -0700 | [diff] [blame] | 3555 | if (order >= MAX_ORDER) { |
| 3556 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); |
Mel Gorman | 72807a7 | 2009-06-16 15:32:18 -0700 | [diff] [blame] | 3557 | return NULL; |
Mel Gorman | 1fc28b7 | 2009-07-29 15:04:08 -0700 | [diff] [blame] | 3558 | } |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 3559 | |
Christoph Lameter | 952f3b5 | 2006-12-06 20:33:26 -0800 | [diff] [blame] | 3560 | /* |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3561 | * We also sanity check to catch abuse of atomic reserves being used by |
| 3562 | * callers that are not in atomic context. |
| 3563 | */ |
| 3564 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == |
| 3565 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) |
| 3566 | gfp_mask &= ~__GFP_ATOMIC; |
| 3567 | |
Johannes Weiner | 9879de7 | 2015-01-26 12:58:32 -0800 | [diff] [blame] | 3568 | retry: |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3569 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3570 | wake_all_kswapds(order, ac); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 3571 | |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 3572 | /* |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 3573 | * OK, we're below the kswapd watermark and have kicked background |
| 3574 | * reclaim. Now things get more complex, so set up alloc_flags according |
| 3575 | * to how we want to proceed. |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 3576 | */ |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3577 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3578 | |
Mel Gorman | e46e7b7 | 2016-06-03 14:56:01 -0700 | [diff] [blame] | 3579 | /* |
| 3580 | * Reset the zonelist iterators if memory policies can be ignored. |
| 3581 | * These allocations are high priority and system rather than user |
| 3582 | * orientated. |
| 3583 | */ |
| 3584 | if ((alloc_flags & ALLOC_NO_WATERMARKS) || !(alloc_flags & ALLOC_CPUSET)) { |
| 3585 | ac->zonelist = node_zonelist(numa_node_id(), gfp_mask); |
| 3586 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
| 3587 | ac->high_zoneidx, ac->nodemask); |
| 3588 | } |
| 3589 | |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3590 | /* This is the last chance, in general, before the goto nopage. */ |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3591 | page = get_page_from_freelist(gfp_mask, order, |
| 3592 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 3593 | if (page) |
| 3594 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3595 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3596 | /* Allocate without watermarks if the context allows */ |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3597 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
Michal Hocko | 33d5310 | 2016-01-14 15:19:05 -0800 | [diff] [blame] | 3598 | page = get_page_from_freelist(gfp_mask, order, |
| 3599 | ALLOC_NO_WATERMARKS, ac); |
| 3600 | if (page) |
| 3601 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3602 | } |
| 3603 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3604 | /* Caller is not willing to reclaim, we can't balance anything */ |
| 3605 | if (!can_direct_reclaim) { |
David Rientjes | aed0a0e | 2014-01-21 15:51:12 -0800 | [diff] [blame] | 3606 | /* |
Michal Hocko | 33d5310 | 2016-01-14 15:19:05 -0800 | [diff] [blame] | 3607 | * All existing users of the __GFP_NOFAIL are blockable, so warn |
| 3608 | * of any new users that actually allow this type of allocation |
| 3609 | * to fail. |
David Rientjes | aed0a0e | 2014-01-21 15:51:12 -0800 | [diff] [blame] | 3610 | */ |
| 3611 | WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3612 | goto nopage; |
David Rientjes | aed0a0e | 2014-01-21 15:51:12 -0800 | [diff] [blame] | 3613 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3614 | |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3615 | /* Avoid recursion of direct reclaim */ |
Michal Hocko | 33d5310 | 2016-01-14 15:19:05 -0800 | [diff] [blame] | 3616 | if (current->flags & PF_MEMALLOC) { |
| 3617 | /* |
| 3618 | * __GFP_NOFAIL request from this context is rather bizarre |
| 3619 | * because we cannot reclaim anything and only can loop waiting |
| 3620 | * for somebody to do a work for us. |
| 3621 | */ |
| 3622 | if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
| 3623 | cond_resched(); |
| 3624 | goto retry; |
| 3625 | } |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3626 | goto nopage; |
Michal Hocko | 33d5310 | 2016-01-14 15:19:05 -0800 | [diff] [blame] | 3627 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3628 | |
David Rientjes | 6583bb6 | 2009-07-29 15:02:06 -0700 | [diff] [blame] | 3629 | /* Avoid allocations with no watermarks from looping endlessly */ |
| 3630 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) |
| 3631 | goto nopage; |
| 3632 | |
Mel Gorman | 77f1fe6 | 2011-01-13 15:45:57 -0800 | [diff] [blame] | 3633 | /* |
| 3634 | * Try direct compaction. The first pass is asynchronous. Subsequent |
| 3635 | * attempts after direct reclaim are synchronous |
| 3636 | */ |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3637 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
| 3638 | migration_mode, |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3639 | &compact_result); |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3640 | if (page) |
| 3641 | goto got_pg; |
David Rientjes | 75f3086 | 2014-06-04 16:08:30 -0700 | [diff] [blame] | 3642 | |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 3643 | /* Checks for THP-specific high-order allocations */ |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3644 | if (is_thp_gfp_mask(gfp_mask)) { |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 3645 | /* |
| 3646 | * If compaction is deferred for high-order allocations, it is |
| 3647 | * because sync compaction recently failed. If this is the case |
| 3648 | * and the caller requested a THP allocation, we do not want |
| 3649 | * to heavily disrupt the system, so we fail the allocation |
| 3650 | * instead of entering direct reclaim. |
| 3651 | */ |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3652 | if (compact_result == COMPACT_DEFERRED) |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 3653 | goto nopage; |
| 3654 | |
| 3655 | /* |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3656 | * Compaction is contended so rather back off than cause |
| 3657 | * excessive stalls. |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 3658 | */ |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3659 | if(compact_result == COMPACT_CONTENDED) |
Vlastimil Babka | 1f9efde | 2014-10-09 15:27:14 -0700 | [diff] [blame] | 3660 | goto nopage; |
| 3661 | } |
Mel Gorman | 6619971 | 2012-01-12 17:19:41 -0800 | [diff] [blame] | 3662 | |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3663 | if (order && compaction_made_progress(compact_result)) |
| 3664 | compaction_retries++; |
David Rientjes | 8fe7804 | 2014-08-06 16:07:54 -0700 | [diff] [blame] | 3665 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3666 | /* Try direct reclaim and then allocating */ |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3667 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, |
| 3668 | &did_some_progress); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3669 | if (page) |
| 3670 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3671 | |
Johannes Weiner | 9083905 | 2015-06-24 16:57:21 -0700 | [diff] [blame] | 3672 | /* Do not loop if specifically requested */ |
| 3673 | if (gfp_mask & __GFP_NORETRY) |
| 3674 | goto noretry; |
| 3675 | |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3676 | /* |
| 3677 | * Do not retry costly high order allocations unless they are |
| 3678 | * __GFP_REPEAT |
| 3679 | */ |
| 3680 | if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_REPEAT)) |
| 3681 | goto noretry; |
| 3682 | |
Michal Hocko | 7854ea6 | 2016-05-20 16:57:09 -0700 | [diff] [blame] | 3683 | /* |
| 3684 | * Costly allocations might have made a progress but this doesn't mean |
| 3685 | * their order will become available due to high fragmentation so |
| 3686 | * always increment the no progress counter for them |
| 3687 | */ |
| 3688 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3689 | no_progress_loops = 0; |
Michal Hocko | 7854ea6 | 2016-05-20 16:57:09 -0700 | [diff] [blame] | 3690 | else |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3691 | no_progress_loops++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3692 | |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3693 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
Michal Hocko | 7854ea6 | 2016-05-20 16:57:09 -0700 | [diff] [blame] | 3694 | did_some_progress > 0, no_progress_loops)) |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3695 | goto retry; |
| 3696 | |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3697 | /* |
| 3698 | * It doesn't make any sense to retry for the compaction if the order-0 |
| 3699 | * reclaim is not able to make any progress because the current |
| 3700 | * implementation of the compaction depends on the sufficient amount |
| 3701 | * of free memory (see __compaction_suitable) |
| 3702 | */ |
| 3703 | if (did_some_progress > 0 && |
Michal Hocko | 86a294a | 2016-05-20 16:57:12 -0700 | [diff] [blame] | 3704 | should_compact_retry(ac, order, alloc_flags, |
| 3705 | compact_result, &migration_mode, |
| 3706 | compaction_retries)) |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3707 | goto retry; |
| 3708 | |
Johannes Weiner | 9083905 | 2015-06-24 16:57:21 -0700 | [diff] [blame] | 3709 | /* Reclaim has failed us, start killing things */ |
| 3710 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); |
| 3711 | if (page) |
| 3712 | goto got_pg; |
| 3713 | |
| 3714 | /* Retry as long as the OOM killer is making progress */ |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3715 | if (did_some_progress) { |
| 3716 | no_progress_loops = 0; |
Johannes Weiner | 9083905 | 2015-06-24 16:57:21 -0700 | [diff] [blame] | 3717 | goto retry; |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3718 | } |
Johannes Weiner | 9083905 | 2015-06-24 16:57:21 -0700 | [diff] [blame] | 3719 | |
| 3720 | noretry: |
| 3721 | /* |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3722 | * High-order allocations do not necessarily loop after direct reclaim |
| 3723 | * and reclaim/compaction depends on compaction being called after |
| 3724 | * reclaim so call directly if necessary. |
| 3725 | * It can become very expensive to allocate transparent hugepages at |
| 3726 | * fault, so use asynchronous memory compaction for THP unless it is |
| 3727 | * khugepaged trying to collapse. All other requests should tolerate |
| 3728 | * at least light sync migration. |
Johannes Weiner | 9083905 | 2015-06-24 16:57:21 -0700 | [diff] [blame] | 3729 | */ |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3730 | if (is_thp_gfp_mask(gfp_mask) && !(current->flags & PF_KTHREAD)) |
| 3731 | migration_mode = MIGRATE_ASYNC; |
| 3732 | else |
| 3733 | migration_mode = MIGRATE_SYNC_LIGHT; |
Johannes Weiner | 9083905 | 2015-06-24 16:57:21 -0700 | [diff] [blame] | 3734 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, |
| 3735 | ac, migration_mode, |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3736 | &compact_result); |
Johannes Weiner | 9083905 | 2015-06-24 16:57:21 -0700 | [diff] [blame] | 3737 | if (page) |
| 3738 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3739 | nopage: |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 3740 | warn_alloc_failed(gfp_mask, order, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3741 | got_pg: |
Mel Gorman | 072bb0a | 2012-07-31 16:43:58 -0700 | [diff] [blame] | 3742 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3743 | } |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3744 | |
| 3745 | /* |
| 3746 | * This is the 'heart' of the zoned buddy allocator. |
| 3747 | */ |
| 3748 | struct page * |
| 3749 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, |
| 3750 | struct zonelist *zonelist, nodemask_t *nodemask) |
| 3751 | { |
Mel Gorman | 5bb1b16 | 2016-05-19 17:13:50 -0700 | [diff] [blame] | 3752 | struct page *page; |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 3753 | unsigned int cpuset_mems_cookie; |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 3754 | unsigned int alloc_flags = ALLOC_WMARK_LOW|ALLOC_FAIR; |
Mel Gorman | 83d4ca8 | 2016-05-19 17:13:56 -0700 | [diff] [blame] | 3755 | gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */ |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3756 | struct alloc_context ac = { |
| 3757 | .high_zoneidx = gfp_zone(gfp_mask), |
Mel Gorman | 682a338 | 2016-05-19 17:13:30 -0700 | [diff] [blame] | 3758 | .zonelist = zonelist, |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3759 | .nodemask = nodemask, |
| 3760 | .migratetype = gfpflags_to_migratetype(gfp_mask), |
| 3761 | }; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3762 | |
Mel Gorman | 682a338 | 2016-05-19 17:13:30 -0700 | [diff] [blame] | 3763 | if (cpusets_enabled()) { |
Mel Gorman | 83d4ca8 | 2016-05-19 17:13:56 -0700 | [diff] [blame] | 3764 | alloc_mask |= __GFP_HARDWALL; |
Mel Gorman | 682a338 | 2016-05-19 17:13:30 -0700 | [diff] [blame] | 3765 | alloc_flags |= ALLOC_CPUSET; |
| 3766 | if (!ac.nodemask) |
| 3767 | ac.nodemask = &cpuset_current_mems_allowed; |
| 3768 | } |
| 3769 | |
Benjamin Herrenschmidt | dcce284 | 2009-06-18 13:24:12 +1000 | [diff] [blame] | 3770 | gfp_mask &= gfp_allowed_mask; |
| 3771 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3772 | lockdep_trace_alloc(gfp_mask); |
| 3773 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3774 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3775 | |
| 3776 | if (should_fail_alloc_page(gfp_mask, order)) |
| 3777 | return NULL; |
| 3778 | |
| 3779 | /* |
| 3780 | * Check the zones suitable for the gfp_mask contain at least one |
| 3781 | * valid zone. It's possible to have an empty zonelist as a result |
David Rientjes | 4167e9b | 2015-04-14 15:46:55 -0700 | [diff] [blame] | 3782 | * of __GFP_THISNODE and a memoryless node |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3783 | */ |
| 3784 | if (unlikely(!zonelist->_zonerefs->zone)) |
| 3785 | return NULL; |
| 3786 | |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3787 | if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE) |
Vlastimil Babka | 21bb9bd | 2014-10-09 15:26:51 -0700 | [diff] [blame] | 3788 | alloc_flags |= ALLOC_CMA; |
| 3789 | |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 3790 | retry_cpuset: |
Mel Gorman | d26914d | 2014-04-03 14:47:24 -0700 | [diff] [blame] | 3791 | cpuset_mems_cookie = read_mems_allowed_begin(); |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 3792 | |
Mel Gorman | c9ab0c4 | 2015-11-06 16:28:12 -0800 | [diff] [blame] | 3793 | /* Dirty zone balancing only done in the fast path */ |
| 3794 | ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
| 3795 | |
Mel Gorman | e46e7b7 | 2016-06-03 14:56:01 -0700 | [diff] [blame] | 3796 | /* |
| 3797 | * The preferred zone is used for statistics but crucially it is |
| 3798 | * also used as the starting point for the zonelist iterator. It |
| 3799 | * may get reset for allocations that ignore memory policies. |
| 3800 | */ |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 3801 | ac.preferred_zoneref = first_zones_zonelist(ac.zonelist, |
| 3802 | ac.high_zoneidx, ac.nodemask); |
| 3803 | if (!ac.preferred_zoneref) { |
Mel Gorman | 5bb1b16 | 2016-05-19 17:13:50 -0700 | [diff] [blame] | 3804 | page = NULL; |
Mel Gorman | 4fcb097 | 2016-05-19 17:14:01 -0700 | [diff] [blame] | 3805 | goto no_zone; |
Mel Gorman | 5bb1b16 | 2016-05-19 17:13:50 -0700 | [diff] [blame] | 3806 | } |
| 3807 | |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 3808 | /* First allocation attempt */ |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3809 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
Mel Gorman | 4fcb097 | 2016-05-19 17:14:01 -0700 | [diff] [blame] | 3810 | if (likely(page)) |
| 3811 | goto out; |
Andrew Morton | 91fbdc0 | 2015-02-11 15:25:04 -0800 | [diff] [blame] | 3812 | |
Mel Gorman | 4fcb097 | 2016-05-19 17:14:01 -0700 | [diff] [blame] | 3813 | /* |
| 3814 | * Runtime PM, block IO and its error handling path can deadlock |
| 3815 | * because I/O on the device might not complete. |
| 3816 | */ |
| 3817 | alloc_mask = memalloc_noio_flags(gfp_mask); |
| 3818 | ac.spread_dirty_pages = false; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3819 | |
Mel Gorman | 4741526 | 2016-05-19 17:14:44 -0700 | [diff] [blame] | 3820 | /* |
| 3821 | * Restore the original nodemask if it was potentially replaced with |
| 3822 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. |
| 3823 | */ |
| 3824 | if (cpusets_enabled()) |
| 3825 | ac.nodemask = nodemask; |
Mel Gorman | 4fcb097 | 2016-05-19 17:14:01 -0700 | [diff] [blame] | 3826 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
Xishi Qiu | 23f086f | 2015-02-11 15:25:07 -0800 | [diff] [blame] | 3827 | |
Mel Gorman | 4fcb097 | 2016-05-19 17:14:01 -0700 | [diff] [blame] | 3828 | no_zone: |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 3829 | /* |
| 3830 | * When updating a task's mems_allowed, it is possible to race with |
| 3831 | * parallel threads in such a way that an allocation can fail while |
| 3832 | * the mask is being updated. If a page allocation is about to fail, |
| 3833 | * check if the cpuset changed during allocation and if so, retry. |
| 3834 | */ |
Mel Gorman | 83d4ca8 | 2016-05-19 17:13:56 -0700 | [diff] [blame] | 3835 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) { |
| 3836 | alloc_mask = gfp_mask; |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 3837 | goto retry_cpuset; |
Mel Gorman | 83d4ca8 | 2016-05-19 17:13:56 -0700 | [diff] [blame] | 3838 | } |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 3839 | |
Mel Gorman | 4fcb097 | 2016-05-19 17:14:01 -0700 | [diff] [blame] | 3840 | out: |
Vladimir Davydov | 4949148 | 2016-07-26 15:24:24 -0700 | [diff] [blame] | 3841 | if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page) { |
| 3842 | if (unlikely(memcg_kmem_charge(page, gfp_mask, order))) { |
| 3843 | __free_pages(page, order); |
| 3844 | page = NULL; |
| 3845 | } else |
| 3846 | __SetPageKmemcg(page); |
| 3847 | } |
| 3848 | |
Mel Gorman | 4fcb097 | 2016-05-19 17:14:01 -0700 | [diff] [blame] | 3849 | if (kmemcheck_enabled && page) |
| 3850 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); |
| 3851 | |
| 3852 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); |
| 3853 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3854 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3855 | } |
Mel Gorman | d239171 | 2009-06-16 15:31:52 -0700 | [diff] [blame] | 3856 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3857 | |
| 3858 | /* |
| 3859 | * Common helper functions. |
| 3860 | */ |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 3861 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3862 | { |
Akinobu Mita | 945a111 | 2009-09-21 17:01:47 -0700 | [diff] [blame] | 3863 | struct page *page; |
| 3864 | |
| 3865 | /* |
| 3866 | * __get_free_pages() returns a 32-bit address, which cannot represent |
| 3867 | * a highmem page |
| 3868 | */ |
| 3869 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
| 3870 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3871 | page = alloc_pages(gfp_mask, order); |
| 3872 | if (!page) |
| 3873 | return 0; |
| 3874 | return (unsigned long) page_address(page); |
| 3875 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3876 | EXPORT_SYMBOL(__get_free_pages); |
| 3877 | |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 3878 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3879 | { |
Akinobu Mita | 945a111 | 2009-09-21 17:01:47 -0700 | [diff] [blame] | 3880 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3881 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3882 | EXPORT_SYMBOL(get_zeroed_page); |
| 3883 | |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 3884 | void __free_pages(struct page *page, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3885 | { |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 3886 | if (put_page_testzero(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3887 | if (order == 0) |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 3888 | free_hot_cold_page(page, false); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3889 | else |
| 3890 | __free_pages_ok(page, order); |
| 3891 | } |
| 3892 | } |
| 3893 | |
| 3894 | EXPORT_SYMBOL(__free_pages); |
| 3895 | |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 3896 | void free_pages(unsigned long addr, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3897 | { |
| 3898 | if (addr != 0) { |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 3899 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3900 | __free_pages(virt_to_page((void *)addr), order); |
| 3901 | } |
| 3902 | } |
| 3903 | |
| 3904 | EXPORT_SYMBOL(free_pages); |
| 3905 | |
Glauber Costa | 6a1a0d3 | 2012-12-18 14:22:00 -0800 | [diff] [blame] | 3906 | /* |
Alexander Duyck | b63ae8c | 2015-05-06 21:11:57 -0700 | [diff] [blame] | 3907 | * Page Fragment: |
| 3908 | * An arbitrary-length arbitrary-offset area of memory which resides |
| 3909 | * within a 0 or higher order page. Multiple fragments within that page |
| 3910 | * are individually refcounted, in the page's reference counter. |
| 3911 | * |
| 3912 | * The page_frag functions below provide a simple allocation framework for |
| 3913 | * page fragments. This is used by the network stack and network device |
| 3914 | * drivers to provide a backing region of memory for use as either an |
| 3915 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. |
| 3916 | */ |
| 3917 | static struct page *__page_frag_refill(struct page_frag_cache *nc, |
| 3918 | gfp_t gfp_mask) |
| 3919 | { |
| 3920 | struct page *page = NULL; |
| 3921 | gfp_t gfp = gfp_mask; |
| 3922 | |
| 3923 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
| 3924 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | |
| 3925 | __GFP_NOMEMALLOC; |
| 3926 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, |
| 3927 | PAGE_FRAG_CACHE_MAX_ORDER); |
| 3928 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; |
| 3929 | #endif |
| 3930 | if (unlikely(!page)) |
| 3931 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); |
| 3932 | |
| 3933 | nc->va = page ? page_address(page) : NULL; |
| 3934 | |
| 3935 | return page; |
| 3936 | } |
| 3937 | |
| 3938 | void *__alloc_page_frag(struct page_frag_cache *nc, |
| 3939 | unsigned int fragsz, gfp_t gfp_mask) |
| 3940 | { |
| 3941 | unsigned int size = PAGE_SIZE; |
| 3942 | struct page *page; |
| 3943 | int offset; |
| 3944 | |
| 3945 | if (unlikely(!nc->va)) { |
| 3946 | refill: |
| 3947 | page = __page_frag_refill(nc, gfp_mask); |
| 3948 | if (!page) |
| 3949 | return NULL; |
| 3950 | |
| 3951 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
| 3952 | /* if size can vary use size else just use PAGE_SIZE */ |
| 3953 | size = nc->size; |
| 3954 | #endif |
| 3955 | /* Even if we own the page, we do not use atomic_set(). |
| 3956 | * This would break get_page_unless_zero() users. |
| 3957 | */ |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 3958 | page_ref_add(page, size - 1); |
Alexander Duyck | b63ae8c | 2015-05-06 21:11:57 -0700 | [diff] [blame] | 3959 | |
| 3960 | /* reset page count bias and offset to start of new frag */ |
Michal Hocko | 2f064f3 | 2015-08-21 14:11:51 -0700 | [diff] [blame] | 3961 | nc->pfmemalloc = page_is_pfmemalloc(page); |
Alexander Duyck | b63ae8c | 2015-05-06 21:11:57 -0700 | [diff] [blame] | 3962 | nc->pagecnt_bias = size; |
| 3963 | nc->offset = size; |
| 3964 | } |
| 3965 | |
| 3966 | offset = nc->offset - fragsz; |
| 3967 | if (unlikely(offset < 0)) { |
| 3968 | page = virt_to_page(nc->va); |
| 3969 | |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 3970 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
Alexander Duyck | b63ae8c | 2015-05-06 21:11:57 -0700 | [diff] [blame] | 3971 | goto refill; |
| 3972 | |
| 3973 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
| 3974 | /* if size can vary use size else just use PAGE_SIZE */ |
| 3975 | size = nc->size; |
| 3976 | #endif |
| 3977 | /* OK, page count is 0, we can safely set it */ |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 3978 | set_page_count(page, size); |
Alexander Duyck | b63ae8c | 2015-05-06 21:11:57 -0700 | [diff] [blame] | 3979 | |
| 3980 | /* reset page count bias and offset to start of new frag */ |
| 3981 | nc->pagecnt_bias = size; |
| 3982 | offset = size - fragsz; |
| 3983 | } |
| 3984 | |
| 3985 | nc->pagecnt_bias--; |
| 3986 | nc->offset = offset; |
| 3987 | |
| 3988 | return nc->va + offset; |
| 3989 | } |
| 3990 | EXPORT_SYMBOL(__alloc_page_frag); |
| 3991 | |
| 3992 | /* |
| 3993 | * Frees a page fragment allocated out of either a compound or order 0 page. |
| 3994 | */ |
| 3995 | void __free_page_frag(void *addr) |
| 3996 | { |
| 3997 | struct page *page = virt_to_head_page(addr); |
| 3998 | |
| 3999 | if (unlikely(put_page_testzero(page))) |
| 4000 | __free_pages_ok(page, compound_order(page)); |
| 4001 | } |
| 4002 | EXPORT_SYMBOL(__free_page_frag); |
| 4003 | |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 4004 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
| 4005 | size_t size) |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 4006 | { |
| 4007 | if (addr) { |
| 4008 | unsigned long alloc_end = addr + (PAGE_SIZE << order); |
| 4009 | unsigned long used = addr + PAGE_ALIGN(size); |
| 4010 | |
| 4011 | split_page(virt_to_page((void *)addr), order); |
| 4012 | while (used < alloc_end) { |
| 4013 | free_page(used); |
| 4014 | used += PAGE_SIZE; |
| 4015 | } |
| 4016 | } |
| 4017 | return (void *)addr; |
| 4018 | } |
| 4019 | |
Timur Tabi | 2be0ffe | 2008-07-23 21:28:11 -0700 | [diff] [blame] | 4020 | /** |
| 4021 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. |
| 4022 | * @size: the number of bytes to allocate |
| 4023 | * @gfp_mask: GFP flags for the allocation |
| 4024 | * |
| 4025 | * This function is similar to alloc_pages(), except that it allocates the |
| 4026 | * minimum number of pages to satisfy the request. alloc_pages() can only |
| 4027 | * allocate memory in power-of-two pages. |
| 4028 | * |
| 4029 | * This function is also limited by MAX_ORDER. |
| 4030 | * |
| 4031 | * Memory allocated by this function must be released by free_pages_exact(). |
| 4032 | */ |
| 4033 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) |
| 4034 | { |
| 4035 | unsigned int order = get_order(size); |
| 4036 | unsigned long addr; |
| 4037 | |
| 4038 | addr = __get_free_pages(gfp_mask, order); |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 4039 | return make_alloc_exact(addr, order, size); |
Timur Tabi | 2be0ffe | 2008-07-23 21:28:11 -0700 | [diff] [blame] | 4040 | } |
| 4041 | EXPORT_SYMBOL(alloc_pages_exact); |
| 4042 | |
| 4043 | /** |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 4044 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous |
| 4045 | * pages on a node. |
Randy Dunlap | b5e6ab5 | 2011-05-16 13:16:54 -0700 | [diff] [blame] | 4046 | * @nid: the preferred node ID where memory should be allocated |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 4047 | * @size: the number of bytes to allocate |
| 4048 | * @gfp_mask: GFP flags for the allocation |
| 4049 | * |
| 4050 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling |
| 4051 | * back. |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 4052 | */ |
Fabian Frederick | e193181 | 2014-08-06 16:04:59 -0700 | [diff] [blame] | 4053 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 4054 | { |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 4055 | unsigned int order = get_order(size); |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 4056 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
| 4057 | if (!p) |
| 4058 | return NULL; |
| 4059 | return make_alloc_exact((unsigned long)page_address(p), order, size); |
| 4060 | } |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 4061 | |
| 4062 | /** |
Timur Tabi | 2be0ffe | 2008-07-23 21:28:11 -0700 | [diff] [blame] | 4063 | * free_pages_exact - release memory allocated via alloc_pages_exact() |
| 4064 | * @virt: the value returned by alloc_pages_exact. |
| 4065 | * @size: size of allocation, same value as passed to alloc_pages_exact(). |
| 4066 | * |
| 4067 | * Release the memory allocated by a previous call to alloc_pages_exact. |
| 4068 | */ |
| 4069 | void free_pages_exact(void *virt, size_t size) |
| 4070 | { |
| 4071 | unsigned long addr = (unsigned long)virt; |
| 4072 | unsigned long end = addr + PAGE_ALIGN(size); |
| 4073 | |
| 4074 | while (addr < end) { |
| 4075 | free_page(addr); |
| 4076 | addr += PAGE_SIZE; |
| 4077 | } |
| 4078 | } |
| 4079 | EXPORT_SYMBOL(free_pages_exact); |
| 4080 | |
Zhang Yanfei | e0fb581 | 2013-02-22 16:35:54 -0800 | [diff] [blame] | 4081 | /** |
| 4082 | * nr_free_zone_pages - count number of pages beyond high watermark |
| 4083 | * @offset: The zone index of the highest zone |
| 4084 | * |
| 4085 | * nr_free_zone_pages() counts the number of counts pages which are beyond the |
| 4086 | * high watermark within all zones at or below a given zone index. For each |
| 4087 | * zone, the number of pages is calculated as: |
Jiang Liu | 834405c | 2013-07-03 15:03:04 -0700 | [diff] [blame] | 4088 | * managed_pages - high_pages |
Zhang Yanfei | e0fb581 | 2013-02-22 16:35:54 -0800 | [diff] [blame] | 4089 | */ |
Zhang Yanfei | ebec386 | 2013-02-22 16:35:43 -0800 | [diff] [blame] | 4090 | static unsigned long nr_free_zone_pages(int offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4091 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4092 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4093 | struct zone *zone; |
| 4094 | |
Martin J. Bligh | e310fd4 | 2005-07-29 22:59:18 -0700 | [diff] [blame] | 4095 | /* Just pick one node, since fallback list is circular */ |
Zhang Yanfei | ebec386 | 2013-02-22 16:35:43 -0800 | [diff] [blame] | 4096 | unsigned long sum = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4097 | |
Mel Gorman | 0e88460 | 2008-04-28 02:12:14 -0700 | [diff] [blame] | 4098 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4099 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4100 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 4101 | unsigned long size = zone->managed_pages; |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 4102 | unsigned long high = high_wmark_pages(zone); |
Martin J. Bligh | e310fd4 | 2005-07-29 22:59:18 -0700 | [diff] [blame] | 4103 | if (size > high) |
| 4104 | sum += size - high; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4105 | } |
| 4106 | |
| 4107 | return sum; |
| 4108 | } |
| 4109 | |
Zhang Yanfei | e0fb581 | 2013-02-22 16:35:54 -0800 | [diff] [blame] | 4110 | /** |
| 4111 | * nr_free_buffer_pages - count number of pages beyond high watermark |
| 4112 | * |
| 4113 | * nr_free_buffer_pages() counts the number of pages which are beyond the high |
| 4114 | * watermark within ZONE_DMA and ZONE_NORMAL. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4115 | */ |
Zhang Yanfei | ebec386 | 2013-02-22 16:35:43 -0800 | [diff] [blame] | 4116 | unsigned long nr_free_buffer_pages(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4117 | { |
Al Viro | af4ca45 | 2005-10-21 02:55:38 -0400 | [diff] [blame] | 4118 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4119 | } |
Meelap Shah | c2f1a55 | 2007-07-17 04:04:39 -0700 | [diff] [blame] | 4120 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4121 | |
Zhang Yanfei | e0fb581 | 2013-02-22 16:35:54 -0800 | [diff] [blame] | 4122 | /** |
| 4123 | * nr_free_pagecache_pages - count number of pages beyond high watermark |
| 4124 | * |
| 4125 | * nr_free_pagecache_pages() counts the number of pages which are beyond the |
| 4126 | * high watermark within all zones. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4127 | */ |
Zhang Yanfei | ebec386 | 2013-02-22 16:35:43 -0800 | [diff] [blame] | 4128 | unsigned long nr_free_pagecache_pages(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4129 | { |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4130 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4131 | } |
Christoph Lameter | 08e0f6a | 2006-09-27 01:50:06 -0700 | [diff] [blame] | 4132 | |
| 4133 | static inline void show_node(struct zone *zone) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4134 | { |
Kirill A. Shutemov | e5adfff | 2012-12-11 16:00:29 -0800 | [diff] [blame] | 4135 | if (IS_ENABLED(CONFIG_NUMA)) |
Andy Whitcroft | 25ba77c | 2006-12-06 20:33:03 -0800 | [diff] [blame] | 4136 | printk("Node %d ", zone_to_nid(zone)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4137 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4138 | |
Igor Redko | d02bd27 | 2016-03-17 14:19:05 -0700 | [diff] [blame] | 4139 | long si_mem_available(void) |
| 4140 | { |
| 4141 | long available; |
| 4142 | unsigned long pagecache; |
| 4143 | unsigned long wmark_low = 0; |
| 4144 | unsigned long pages[NR_LRU_LISTS]; |
| 4145 | struct zone *zone; |
| 4146 | int lru; |
| 4147 | |
| 4148 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) |
| 4149 | pages[lru] = global_page_state(NR_LRU_BASE + lru); |
| 4150 | |
| 4151 | for_each_zone(zone) |
| 4152 | wmark_low += zone->watermark[WMARK_LOW]; |
| 4153 | |
| 4154 | /* |
| 4155 | * Estimate the amount of memory available for userspace allocations, |
| 4156 | * without causing swapping. |
| 4157 | */ |
| 4158 | available = global_page_state(NR_FREE_PAGES) - totalreserve_pages; |
| 4159 | |
| 4160 | /* |
| 4161 | * Not all the page cache can be freed, otherwise the system will |
| 4162 | * start swapping. Assume at least half of the page cache, or the |
| 4163 | * low watermark worth of cache, needs to stay. |
| 4164 | */ |
| 4165 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; |
| 4166 | pagecache -= min(pagecache / 2, wmark_low); |
| 4167 | available += pagecache; |
| 4168 | |
| 4169 | /* |
| 4170 | * Part of the reclaimable slab consists of items that are in use, |
| 4171 | * and cannot be freed. Cap this estimate at the low watermark. |
| 4172 | */ |
| 4173 | available += global_page_state(NR_SLAB_RECLAIMABLE) - |
| 4174 | min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low); |
| 4175 | |
| 4176 | if (available < 0) |
| 4177 | available = 0; |
| 4178 | return available; |
| 4179 | } |
| 4180 | EXPORT_SYMBOL_GPL(si_mem_available); |
| 4181 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4182 | void si_meminfo(struct sysinfo *val) |
| 4183 | { |
| 4184 | val->totalram = totalram_pages; |
Rafael Aquini | cc7452b | 2014-08-06 16:06:38 -0700 | [diff] [blame] | 4185 | val->sharedram = global_page_state(NR_SHMEM); |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 4186 | val->freeram = global_page_state(NR_FREE_PAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4187 | val->bufferram = nr_blockdev_pages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4188 | val->totalhigh = totalhigh_pages; |
| 4189 | val->freehigh = nr_free_highpages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4190 | val->mem_unit = PAGE_SIZE; |
| 4191 | } |
| 4192 | |
| 4193 | EXPORT_SYMBOL(si_meminfo); |
| 4194 | |
| 4195 | #ifdef CONFIG_NUMA |
| 4196 | void si_meminfo_node(struct sysinfo *val, int nid) |
| 4197 | { |
Jiang Liu | cdd91a7 | 2013-07-03 15:03:27 -0700 | [diff] [blame] | 4198 | int zone_type; /* needs to be signed */ |
| 4199 | unsigned long managed_pages = 0; |
Joonsoo Kim | fc2bd79 | 2016-05-19 17:12:23 -0700 | [diff] [blame] | 4200 | unsigned long managed_highpages = 0; |
| 4201 | unsigned long free_highpages = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4202 | pg_data_t *pgdat = NODE_DATA(nid); |
| 4203 | |
Jiang Liu | cdd91a7 | 2013-07-03 15:03:27 -0700 | [diff] [blame] | 4204 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
| 4205 | managed_pages += pgdat->node_zones[zone_type].managed_pages; |
| 4206 | val->totalram = managed_pages; |
Mel Gorman | 75ef718 | 2016-07-28 15:45:24 -0700 | [diff] [blame] | 4207 | val->sharedram = sum_zone_node_page_state(nid, NR_SHMEM); |
| 4208 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 4209 | #ifdef CONFIG_HIGHMEM |
Joonsoo Kim | fc2bd79 | 2016-05-19 17:12:23 -0700 | [diff] [blame] | 4210 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
| 4211 | struct zone *zone = &pgdat->node_zones[zone_type]; |
| 4212 | |
| 4213 | if (is_highmem(zone)) { |
| 4214 | managed_highpages += zone->managed_pages; |
| 4215 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); |
| 4216 | } |
| 4217 | } |
| 4218 | val->totalhigh = managed_highpages; |
| 4219 | val->freehigh = free_highpages; |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 4220 | #else |
Joonsoo Kim | fc2bd79 | 2016-05-19 17:12:23 -0700 | [diff] [blame] | 4221 | val->totalhigh = managed_highpages; |
| 4222 | val->freehigh = free_highpages; |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 4223 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4224 | val->mem_unit = PAGE_SIZE; |
| 4225 | } |
| 4226 | #endif |
| 4227 | |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4228 | /* |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 4229 | * Determine whether the node should be displayed or not, depending on whether |
| 4230 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4231 | */ |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 4232 | bool skip_free_areas_node(unsigned int flags, int nid) |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4233 | { |
| 4234 | bool ret = false; |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 4235 | unsigned int cpuset_mems_cookie; |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4236 | |
| 4237 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
| 4238 | goto out; |
| 4239 | |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 4240 | do { |
Mel Gorman | d26914d | 2014-04-03 14:47:24 -0700 | [diff] [blame] | 4241 | cpuset_mems_cookie = read_mems_allowed_begin(); |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 4242 | ret = !node_isset(nid, cpuset_current_mems_allowed); |
Mel Gorman | d26914d | 2014-04-03 14:47:24 -0700 | [diff] [blame] | 4243 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4244 | out: |
| 4245 | return ret; |
| 4246 | } |
| 4247 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4248 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
| 4249 | |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4250 | static void show_migration_types(unsigned char type) |
| 4251 | { |
| 4252 | static const char types[MIGRATE_TYPES] = { |
| 4253 | [MIGRATE_UNMOVABLE] = 'U', |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4254 | [MIGRATE_MOVABLE] = 'M', |
Vlastimil Babka | 475a2f9 | 2015-12-11 13:40:29 -0800 | [diff] [blame] | 4255 | [MIGRATE_RECLAIMABLE] = 'E', |
| 4256 | [MIGRATE_HIGHATOMIC] = 'H', |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4257 | #ifdef CONFIG_CMA |
| 4258 | [MIGRATE_CMA] = 'C', |
| 4259 | #endif |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 4260 | #ifdef CONFIG_MEMORY_ISOLATION |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4261 | [MIGRATE_ISOLATE] = 'I', |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 4262 | #endif |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4263 | }; |
| 4264 | char tmp[MIGRATE_TYPES + 1]; |
| 4265 | char *p = tmp; |
| 4266 | int i; |
| 4267 | |
| 4268 | for (i = 0; i < MIGRATE_TYPES; i++) { |
| 4269 | if (type & (1 << i)) |
| 4270 | *p++ = types[i]; |
| 4271 | } |
| 4272 | |
| 4273 | *p = '\0'; |
| 4274 | printk("(%s) ", tmp); |
| 4275 | } |
| 4276 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4277 | /* |
| 4278 | * Show free area list (used inside shift_scroll-lock stuff) |
| 4279 | * We also calculate the percentage fragmentation. We do this by counting the |
| 4280 | * memory on each free list with the exception of the first item on the list. |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4281 | * |
| 4282 | * Bits in @filter: |
| 4283 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's |
| 4284 | * cpuset. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4285 | */ |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 4286 | void show_free_areas(unsigned int filter) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4287 | { |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4288 | unsigned long free_pcp = 0; |
Jes Sorensen | c724191 | 2006-09-27 01:50:05 -0700 | [diff] [blame] | 4289 | int cpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4290 | struct zone *zone; |
| 4291 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 4292 | for_each_populated_zone(zone) { |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 4293 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4294 | continue; |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4295 | |
Konstantin Khlebnikov | 761b067 | 2015-04-14 15:45:32 -0700 | [diff] [blame] | 4296 | for_each_online_cpu(cpu) |
| 4297 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4298 | } |
| 4299 | |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 4300 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
| 4301 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4302 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
| 4303 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
Bartlomiej Zolnierkiewicz | d1ce749 | 2012-10-08 16:32:02 -0700 | [diff] [blame] | 4304 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
Kirill A. Shutemov | 65c4537 | 2016-07-26 15:26:10 -0700 | [diff] [blame] | 4305 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 4306 | " anon_thp: %lu shmem_thp: %lu shmem_pmdmapped: %lu\n" |
| 4307 | #endif |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4308 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 4309 | global_page_state(NR_ACTIVE_ANON), |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 4310 | global_page_state(NR_INACTIVE_ANON), |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 4311 | global_page_state(NR_ISOLATED_ANON), |
| 4312 | global_page_state(NR_ACTIVE_FILE), |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 4313 | global_page_state(NR_INACTIVE_FILE), |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 4314 | global_page_state(NR_ISOLATED_FILE), |
Lee Schermerhorn | 7b85412 | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 4315 | global_page_state(NR_UNEVICTABLE), |
Christoph Lameter | b1e7a8f | 2006-06-30 01:55:39 -0700 | [diff] [blame] | 4316 | global_page_state(NR_FILE_DIRTY), |
Christoph Lameter | ce866b3 | 2006-06-30 01:55:40 -0700 | [diff] [blame] | 4317 | global_page_state(NR_WRITEBACK), |
Christoph Lameter | fd39fc8 | 2006-06-30 01:55:40 -0700 | [diff] [blame] | 4318 | global_page_state(NR_UNSTABLE_NFS), |
KOSAKI Motohiro | 3701b03 | 2009-09-21 17:01:29 -0700 | [diff] [blame] | 4319 | global_page_state(NR_SLAB_RECLAIMABLE), |
| 4320 | global_page_state(NR_SLAB_UNRECLAIMABLE), |
Christoph Lameter | 65ba55f | 2006-06-30 01:55:34 -0700 | [diff] [blame] | 4321 | global_page_state(NR_FILE_MAPPED), |
KOSAKI Motohiro | 4b02108 | 2009-09-21 17:01:33 -0700 | [diff] [blame] | 4322 | global_page_state(NR_SHMEM), |
Andrew Morton | a25700a | 2007-02-08 14:20:40 -0800 | [diff] [blame] | 4323 | global_page_state(NR_PAGETABLE), |
Bartlomiej Zolnierkiewicz | d1ce749 | 2012-10-08 16:32:02 -0700 | [diff] [blame] | 4324 | global_page_state(NR_BOUNCE), |
Kirill A. Shutemov | 65c4537 | 2016-07-26 15:26:10 -0700 | [diff] [blame] | 4325 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 4326 | global_page_state(NR_ANON_THPS) * HPAGE_PMD_NR, |
| 4327 | global_page_state(NR_SHMEM_THPS) * HPAGE_PMD_NR, |
| 4328 | global_page_state(NR_SHMEM_PMDMAPPED) * HPAGE_PMD_NR, |
| 4329 | #endif |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4330 | global_page_state(NR_FREE_PAGES), |
| 4331 | free_pcp, |
Bartlomiej Zolnierkiewicz | d1ce749 | 2012-10-08 16:32:02 -0700 | [diff] [blame] | 4332 | global_page_state(NR_FREE_CMA_PAGES)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4333 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 4334 | for_each_populated_zone(zone) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4335 | int i; |
| 4336 | |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 4337 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4338 | continue; |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4339 | |
| 4340 | free_pcp = 0; |
| 4341 | for_each_online_cpu(cpu) |
| 4342 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; |
| 4343 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4344 | show_node(zone); |
| 4345 | printk("%s" |
| 4346 | " free:%lukB" |
| 4347 | " min:%lukB" |
| 4348 | " low:%lukB" |
| 4349 | " high:%lukB" |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 4350 | " active_anon:%lukB" |
| 4351 | " inactive_anon:%lukB" |
| 4352 | " active_file:%lukB" |
| 4353 | " inactive_file:%lukB" |
Lee Schermerhorn | 7b85412 | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 4354 | " unevictable:%lukB" |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 4355 | " isolated(anon):%lukB" |
| 4356 | " isolated(file):%lukB" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4357 | " present:%lukB" |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 4358 | " managed:%lukB" |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4359 | " mlocked:%lukB" |
| 4360 | " dirty:%lukB" |
| 4361 | " writeback:%lukB" |
| 4362 | " mapped:%lukB" |
KOSAKI Motohiro | 4b02108 | 2009-09-21 17:01:33 -0700 | [diff] [blame] | 4363 | " shmem:%lukB" |
Kirill A. Shutemov | 65c4537 | 2016-07-26 15:26:10 -0700 | [diff] [blame] | 4364 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 4365 | " shmem_thp: %lukB" |
| 4366 | " shmem_pmdmapped: %lukB" |
| 4367 | " anon_thp: %lukB" |
| 4368 | #endif |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4369 | " slab_reclaimable:%lukB" |
| 4370 | " slab_unreclaimable:%lukB" |
KOSAKI Motohiro | c6a7f57 | 2009-09-21 17:01:32 -0700 | [diff] [blame] | 4371 | " kernel_stack:%lukB" |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4372 | " pagetables:%lukB" |
| 4373 | " unstable:%lukB" |
| 4374 | " bounce:%lukB" |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4375 | " free_pcp:%lukB" |
| 4376 | " local_pcp:%ukB" |
Bartlomiej Zolnierkiewicz | d1ce749 | 2012-10-08 16:32:02 -0700 | [diff] [blame] | 4377 | " free_cma:%lukB" |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4378 | " writeback_tmp:%lukB" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4379 | " pages_scanned:%lu" |
| 4380 | " all_unreclaimable? %s" |
| 4381 | "\n", |
| 4382 | zone->name, |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 4383 | K(zone_page_state(zone, NR_FREE_PAGES)), |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 4384 | K(min_wmark_pages(zone)), |
| 4385 | K(low_wmark_pages(zone)), |
| 4386 | K(high_wmark_pages(zone)), |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 4387 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
| 4388 | K(zone_page_state(zone, NR_INACTIVE_ANON)), |
| 4389 | K(zone_page_state(zone, NR_ACTIVE_FILE)), |
| 4390 | K(zone_page_state(zone, NR_INACTIVE_FILE)), |
Lee Schermerhorn | 7b85412 | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 4391 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 4392 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
| 4393 | K(zone_page_state(zone, NR_ISOLATED_FILE)), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4394 | K(zone->present_pages), |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 4395 | K(zone->managed_pages), |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4396 | K(zone_page_state(zone, NR_MLOCK)), |
| 4397 | K(zone_page_state(zone, NR_FILE_DIRTY)), |
| 4398 | K(zone_page_state(zone, NR_WRITEBACK)), |
| 4399 | K(zone_page_state(zone, NR_FILE_MAPPED)), |
KOSAKI Motohiro | 4b02108 | 2009-09-21 17:01:33 -0700 | [diff] [blame] | 4400 | K(zone_page_state(zone, NR_SHMEM)), |
Kirill A. Shutemov | 65c4537 | 2016-07-26 15:26:10 -0700 | [diff] [blame] | 4401 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 4402 | K(zone_page_state(zone, NR_SHMEM_THPS) * HPAGE_PMD_NR), |
| 4403 | K(zone_page_state(zone, NR_SHMEM_PMDMAPPED) |
| 4404 | * HPAGE_PMD_NR), |
| 4405 | K(zone_page_state(zone, NR_ANON_THPS) * HPAGE_PMD_NR), |
| 4406 | #endif |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4407 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
| 4408 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), |
KOSAKI Motohiro | c6a7f57 | 2009-09-21 17:01:32 -0700 | [diff] [blame] | 4409 | zone_page_state(zone, NR_KERNEL_STACK) * |
| 4410 | THREAD_SIZE / 1024, |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4411 | K(zone_page_state(zone, NR_PAGETABLE)), |
| 4412 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), |
| 4413 | K(zone_page_state(zone, NR_BOUNCE)), |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4414 | K(free_pcp), |
| 4415 | K(this_cpu_read(zone->pageset->pcp.count)), |
Bartlomiej Zolnierkiewicz | d1ce749 | 2012-10-08 16:32:02 -0700 | [diff] [blame] | 4416 | K(zone_page_state(zone, NR_FREE_CMA_PAGES)), |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4417 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
Mel Gorman | 0d5d823 | 2014-08-06 16:07:16 -0700 | [diff] [blame] | 4418 | K(zone_page_state(zone, NR_PAGES_SCANNED)), |
Lisa Du | 6e543d5 | 2013-09-11 14:22:36 -0700 | [diff] [blame] | 4419 | (!zone_reclaimable(zone) ? "yes" : "no") |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4420 | ); |
| 4421 | printk("lowmem_reserve[]:"); |
| 4422 | for (i = 0; i < MAX_NR_ZONES; i++) |
Mel Gorman | 3484b2d | 2014-08-06 16:07:14 -0700 | [diff] [blame] | 4423 | printk(" %ld", zone->lowmem_reserve[i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4424 | printk("\n"); |
| 4425 | } |
| 4426 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 4427 | for_each_populated_zone(zone) { |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 4428 | unsigned int order; |
| 4429 | unsigned long nr[MAX_ORDER], flags, total = 0; |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4430 | unsigned char types[MAX_ORDER]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4431 | |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 4432 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4433 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4434 | show_node(zone); |
| 4435 | printk("%s: ", zone->name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4436 | |
| 4437 | spin_lock_irqsave(&zone->lock, flags); |
| 4438 | for (order = 0; order < MAX_ORDER; order++) { |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4439 | struct free_area *area = &zone->free_area[order]; |
| 4440 | int type; |
| 4441 | |
| 4442 | nr[order] = area->nr_free; |
Kirill Korotaev | 8f9de51 | 2006-06-23 02:03:50 -0700 | [diff] [blame] | 4443 | total += nr[order] << order; |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4444 | |
| 4445 | types[order] = 0; |
| 4446 | for (type = 0; type < MIGRATE_TYPES; type++) { |
| 4447 | if (!list_empty(&area->free_list[type])) |
| 4448 | types[order] |= 1 << type; |
| 4449 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4450 | } |
| 4451 | spin_unlock_irqrestore(&zone->lock, flags); |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4452 | for (order = 0; order < MAX_ORDER; order++) { |
Kirill Korotaev | 8f9de51 | 2006-06-23 02:03:50 -0700 | [diff] [blame] | 4453 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4454 | if (nr[order]) |
| 4455 | show_migration_types(types[order]); |
| 4456 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4457 | printk("= %lukB\n", K(total)); |
| 4458 | } |
| 4459 | |
David Rientjes | 949f7ec | 2013-04-29 15:07:48 -0700 | [diff] [blame] | 4460 | hugetlb_show_meminfo(); |
| 4461 | |
Larry Woodman | e6f3602 | 2008-02-04 22:29:30 -0800 | [diff] [blame] | 4462 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
| 4463 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4464 | show_swap_cache_info(); |
| 4465 | } |
| 4466 | |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 4467 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
| 4468 | { |
| 4469 | zoneref->zone = zone; |
| 4470 | zoneref->zone_idx = zone_idx(zone); |
| 4471 | } |
| 4472 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4473 | /* |
| 4474 | * Builds allocation fallback zone lists. |
Christoph Lameter | 1a93205 | 2006-01-06 00:11:16 -0800 | [diff] [blame] | 4475 | * |
| 4476 | * Add all populated zones of a node to the zonelist. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4477 | */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4478 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4479 | int nr_zones) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4480 | { |
Christoph Lameter | 1a93205 | 2006-01-06 00:11:16 -0800 | [diff] [blame] | 4481 | struct zone *zone; |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4482 | enum zone_type zone_type = MAX_NR_ZONES; |
Christoph Lameter | 02a68a5 | 2006-01-06 00:11:18 -0800 | [diff] [blame] | 4483 | |
| 4484 | do { |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 4485 | zone_type--; |
Christoph Lameter | 070f803 | 2006-01-06 00:11:19 -0800 | [diff] [blame] | 4486 | zone = pgdat->node_zones + zone_type; |
Christoph Lameter | 1a93205 | 2006-01-06 00:11:16 -0800 | [diff] [blame] | 4487 | if (populated_zone(zone)) { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4488 | zoneref_set_zone(zone, |
| 4489 | &zonelist->_zonerefs[nr_zones++]); |
Christoph Lameter | 070f803 | 2006-01-06 00:11:19 -0800 | [diff] [blame] | 4490 | check_highest_zone(zone_type); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4491 | } |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 4492 | } while (zone_type); |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4493 | |
Christoph Lameter | 070f803 | 2006-01-06 00:11:19 -0800 | [diff] [blame] | 4494 | return nr_zones; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4495 | } |
| 4496 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4497 | |
| 4498 | /* |
| 4499 | * zonelist_order: |
| 4500 | * 0 = automatic detection of better ordering. |
| 4501 | * 1 = order by ([node] distance, -zonetype) |
| 4502 | * 2 = order by (-zonetype, [node] distance) |
| 4503 | * |
| 4504 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create |
| 4505 | * the same zonelist. So only NUMA can configure this param. |
| 4506 | */ |
| 4507 | #define ZONELIST_ORDER_DEFAULT 0 |
| 4508 | #define ZONELIST_ORDER_NODE 1 |
| 4509 | #define ZONELIST_ORDER_ZONE 2 |
| 4510 | |
| 4511 | /* zonelist order in the kernel. |
| 4512 | * set_zonelist_order() will set this to NODE or ZONE. |
| 4513 | */ |
| 4514 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; |
| 4515 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; |
| 4516 | |
| 4517 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4518 | #ifdef CONFIG_NUMA |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4519 | /* The value user specified ....changed by config */ |
| 4520 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; |
| 4521 | /* string for sysctl */ |
| 4522 | #define NUMA_ZONELIST_ORDER_LEN 16 |
| 4523 | char numa_zonelist_order[16] = "default"; |
| 4524 | |
| 4525 | /* |
| 4526 | * interface for configure zonelist ordering. |
| 4527 | * command line option "numa_zonelist_order" |
| 4528 | * = "[dD]efault - default, automatic configuration. |
| 4529 | * = "[nN]ode - order by node locality, then by zone within node |
| 4530 | * = "[zZ]one - order by zone, then by locality within zone |
| 4531 | */ |
| 4532 | |
| 4533 | static int __parse_numa_zonelist_order(char *s) |
| 4534 | { |
| 4535 | if (*s == 'd' || *s == 'D') { |
| 4536 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; |
| 4537 | } else if (*s == 'n' || *s == 'N') { |
| 4538 | user_zonelist_order = ZONELIST_ORDER_NODE; |
| 4539 | } else if (*s == 'z' || *s == 'Z') { |
| 4540 | user_zonelist_order = ZONELIST_ORDER_ZONE; |
| 4541 | } else { |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 4542 | pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4543 | return -EINVAL; |
| 4544 | } |
| 4545 | return 0; |
| 4546 | } |
| 4547 | |
| 4548 | static __init int setup_numa_zonelist_order(char *s) |
| 4549 | { |
Volodymyr G. Lukiianyk | ecb256f | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 4550 | int ret; |
| 4551 | |
| 4552 | if (!s) |
| 4553 | return 0; |
| 4554 | |
| 4555 | ret = __parse_numa_zonelist_order(s); |
| 4556 | if (ret == 0) |
| 4557 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); |
| 4558 | |
| 4559 | return ret; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4560 | } |
| 4561 | early_param("numa_zonelist_order", setup_numa_zonelist_order); |
| 4562 | |
| 4563 | /* |
| 4564 | * sysctl handler for numa_zonelist_order |
| 4565 | */ |
Joe Perches | cccad5b | 2014-06-06 14:38:09 -0700 | [diff] [blame] | 4566 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 4567 | void __user *buffer, size_t *length, |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4568 | loff_t *ppos) |
| 4569 | { |
| 4570 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; |
| 4571 | int ret; |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 4572 | static DEFINE_MUTEX(zl_order_mutex); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4573 | |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 4574 | mutex_lock(&zl_order_mutex); |
Chen Gang | dacbde0 | 2013-07-03 15:02:35 -0700 | [diff] [blame] | 4575 | if (write) { |
| 4576 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { |
| 4577 | ret = -EINVAL; |
| 4578 | goto out; |
| 4579 | } |
| 4580 | strcpy(saved_string, (char *)table->data); |
| 4581 | } |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 4582 | ret = proc_dostring(table, write, buffer, length, ppos); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4583 | if (ret) |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 4584 | goto out; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4585 | if (write) { |
| 4586 | int oldval = user_zonelist_order; |
Chen Gang | dacbde0 | 2013-07-03 15:02:35 -0700 | [diff] [blame] | 4587 | |
| 4588 | ret = __parse_numa_zonelist_order((char *)table->data); |
| 4589 | if (ret) { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4590 | /* |
| 4591 | * bogus value. restore saved string |
| 4592 | */ |
Chen Gang | dacbde0 | 2013-07-03 15:02:35 -0700 | [diff] [blame] | 4593 | strncpy((char *)table->data, saved_string, |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4594 | NUMA_ZONELIST_ORDER_LEN); |
| 4595 | user_zonelist_order = oldval; |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 4596 | } else if (oldval != user_zonelist_order) { |
| 4597 | mutex_lock(&zonelists_mutex); |
Jiang Liu | 9adb62a | 2012-07-31 16:43:28 -0700 | [diff] [blame] | 4598 | build_all_zonelists(NULL, NULL); |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 4599 | mutex_unlock(&zonelists_mutex); |
| 4600 | } |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4601 | } |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 4602 | out: |
| 4603 | mutex_unlock(&zl_order_mutex); |
| 4604 | return ret; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4605 | } |
| 4606 | |
| 4607 | |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 4608 | #define MAX_NODE_LOAD (nr_online_nodes) |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4609 | static int node_load[MAX_NUMNODES]; |
| 4610 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4611 | /** |
Pavel Pisa | 4dc3b16 | 2005-05-01 08:59:25 -0700 | [diff] [blame] | 4612 | * 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] | 4613 | * @node: node whose fallback list we're appending |
| 4614 | * @used_node_mask: nodemask_t of already used nodes |
| 4615 | * |
| 4616 | * We use a number of factors to determine which is the next node that should |
| 4617 | * appear on a given node's fallback list. The node should not have appeared |
| 4618 | * already in @node's fallback list, and it should be the next closest node |
| 4619 | * according to the distance array (which contains arbitrary distance values |
| 4620 | * from each node to each node in the system), and should also prefer nodes |
| 4621 | * with no CPUs, since presumably they'll have very little allocation pressure |
| 4622 | * on them otherwise. |
| 4623 | * It returns -1 if no node is found. |
| 4624 | */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4625 | 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] | 4626 | { |
Linus Torvalds | 4cf808e | 2006-02-17 20:38:21 +0100 | [diff] [blame] | 4627 | int n, val; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4628 | int min_val = INT_MAX; |
David Rientjes | 00ef2d2 | 2013-02-22 16:35:36 -0800 | [diff] [blame] | 4629 | int best_node = NUMA_NO_NODE; |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 4630 | const struct cpumask *tmp = cpumask_of_node(0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4631 | |
Linus Torvalds | 4cf808e | 2006-02-17 20:38:21 +0100 | [diff] [blame] | 4632 | /* Use the local node if we haven't already */ |
| 4633 | if (!node_isset(node, *used_node_mask)) { |
| 4634 | node_set(node, *used_node_mask); |
| 4635 | return node; |
| 4636 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4637 | |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 4638 | for_each_node_state(n, N_MEMORY) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4639 | |
| 4640 | /* Don't want a node to appear more than once */ |
| 4641 | if (node_isset(n, *used_node_mask)) |
| 4642 | continue; |
| 4643 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4644 | /* Use the distance array to find the distance */ |
| 4645 | val = node_distance(node, n); |
| 4646 | |
Linus Torvalds | 4cf808e | 2006-02-17 20:38:21 +0100 | [diff] [blame] | 4647 | /* Penalize nodes under us ("prefer the next node") */ |
| 4648 | val += (n < node); |
| 4649 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4650 | /* Give preference to headless and unused nodes */ |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 4651 | tmp = cpumask_of_node(n); |
| 4652 | if (!cpumask_empty(tmp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4653 | val += PENALTY_FOR_NODE_WITH_CPUS; |
| 4654 | |
| 4655 | /* Slight preference for less loaded node */ |
| 4656 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); |
| 4657 | val += node_load[n]; |
| 4658 | |
| 4659 | if (val < min_val) { |
| 4660 | min_val = val; |
| 4661 | best_node = n; |
| 4662 | } |
| 4663 | } |
| 4664 | |
| 4665 | if (best_node >= 0) |
| 4666 | node_set(best_node, *used_node_mask); |
| 4667 | |
| 4668 | return best_node; |
| 4669 | } |
| 4670 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4671 | |
| 4672 | /* |
| 4673 | * Build zonelists ordered by node and zones within node. |
| 4674 | * This results in maximum locality--normal zone overflows into local |
| 4675 | * DMA zone, if any--but risks exhausting DMA zone. |
| 4676 | */ |
| 4677 | 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] | 4678 | { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4679 | int j; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4680 | struct zonelist *zonelist; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4681 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4682 | zonelist = &pgdat->node_zonelists[0]; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4683 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4684 | ; |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4685 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4686 | zonelist->_zonerefs[j].zone = NULL; |
| 4687 | zonelist->_zonerefs[j].zone_idx = 0; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4688 | } |
| 4689 | |
| 4690 | /* |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 4691 | * Build gfp_thisnode zonelists |
| 4692 | */ |
| 4693 | static void build_thisnode_zonelists(pg_data_t *pgdat) |
| 4694 | { |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 4695 | int j; |
| 4696 | struct zonelist *zonelist; |
| 4697 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4698 | zonelist = &pgdat->node_zonelists[1]; |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4699 | j = build_zonelists_node(pgdat, zonelist, 0); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4700 | zonelist->_zonerefs[j].zone = NULL; |
| 4701 | zonelist->_zonerefs[j].zone_idx = 0; |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 4702 | } |
| 4703 | |
| 4704 | /* |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4705 | * Build zonelists ordered by zone and nodes within zones. |
| 4706 | * This results in conserving DMA zone[s] until all Normal memory is |
| 4707 | * exhausted, but results in overflowing to remote node while memory |
| 4708 | * may still exist in local DMA zone. |
| 4709 | */ |
| 4710 | static int node_order[MAX_NUMNODES]; |
| 4711 | |
| 4712 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) |
| 4713 | { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4714 | int pos, j, node; |
| 4715 | int zone_type; /* needs to be signed */ |
| 4716 | struct zone *z; |
| 4717 | struct zonelist *zonelist; |
| 4718 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4719 | zonelist = &pgdat->node_zonelists[0]; |
| 4720 | pos = 0; |
| 4721 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { |
| 4722 | for (j = 0; j < nr_nodes; j++) { |
| 4723 | node = node_order[j]; |
| 4724 | z = &NODE_DATA(node)->node_zones[zone_type]; |
| 4725 | if (populated_zone(z)) { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4726 | zoneref_set_zone(z, |
| 4727 | &zonelist->_zonerefs[pos++]); |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4728 | check_highest_zone(zone_type); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4729 | } |
| 4730 | } |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4731 | } |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4732 | zonelist->_zonerefs[pos].zone = NULL; |
| 4733 | zonelist->_zonerefs[pos].zone_idx = 0; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4734 | } |
| 4735 | |
Mel Gorman | 3193913 | 2014-10-09 15:28:30 -0700 | [diff] [blame] | 4736 | #if defined(CONFIG_64BIT) |
| 4737 | /* |
| 4738 | * Devices that require DMA32/DMA are relatively rare and do not justify a |
| 4739 | * penalty to every machine in case the specialised case applies. Default |
| 4740 | * to Node-ordering on 64-bit NUMA machines |
| 4741 | */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4742 | static int default_zonelist_order(void) |
| 4743 | { |
Mel Gorman | 3193913 | 2014-10-09 15:28:30 -0700 | [diff] [blame] | 4744 | return ZONELIST_ORDER_NODE; |
| 4745 | } |
| 4746 | #else |
| 4747 | /* |
| 4748 | * On 32-bit, the Normal zone needs to be preserved for allocations accessible |
| 4749 | * by the kernel. If processes running on node 0 deplete the low memory zone |
| 4750 | * then reclaim will occur more frequency increasing stalls and potentially |
| 4751 | * be easier to OOM if a large percentage of the zone is under writeback or |
| 4752 | * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set. |
| 4753 | * Hence, default to zone ordering on 32-bit. |
| 4754 | */ |
| 4755 | static int default_zonelist_order(void) |
| 4756 | { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4757 | return ZONELIST_ORDER_ZONE; |
| 4758 | } |
Mel Gorman | 3193913 | 2014-10-09 15:28:30 -0700 | [diff] [blame] | 4759 | #endif /* CONFIG_64BIT */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4760 | |
| 4761 | static void set_zonelist_order(void) |
| 4762 | { |
| 4763 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) |
| 4764 | current_zonelist_order = default_zonelist_order(); |
| 4765 | else |
| 4766 | current_zonelist_order = user_zonelist_order; |
| 4767 | } |
| 4768 | |
| 4769 | static void build_zonelists(pg_data_t *pgdat) |
| 4770 | { |
Yaowei Bai | c00eb15 | 2016-01-14 15:19:00 -0800 | [diff] [blame] | 4771 | int i, node, load; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4772 | nodemask_t used_mask; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4773 | int local_node, prev_node; |
| 4774 | struct zonelist *zonelist; |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 4775 | unsigned int order = current_zonelist_order; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4776 | |
| 4777 | /* initialize zonelists */ |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 4778 | for (i = 0; i < MAX_ZONELISTS; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4779 | zonelist = pgdat->node_zonelists + i; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4780 | zonelist->_zonerefs[0].zone = NULL; |
| 4781 | zonelist->_zonerefs[0].zone_idx = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4782 | } |
| 4783 | |
| 4784 | /* NUMA-aware ordering of nodes */ |
| 4785 | local_node = pgdat->node_id; |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 4786 | load = nr_online_nodes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4787 | prev_node = local_node; |
| 4788 | nodes_clear(used_mask); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4789 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4790 | memset(node_order, 0, sizeof(node_order)); |
Yaowei Bai | c00eb15 | 2016-01-14 15:19:00 -0800 | [diff] [blame] | 4791 | i = 0; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4792 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4793 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
| 4794 | /* |
| 4795 | * We don't want to pressure a particular node. |
| 4796 | * So adding penalty to the first node in same |
| 4797 | * distance group to make it round-robin. |
| 4798 | */ |
David Rientjes | 957f822 | 2012-10-08 16:33:24 -0700 | [diff] [blame] | 4799 | if (node_distance(local_node, node) != |
| 4800 | node_distance(local_node, prev_node)) |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4801 | node_load[node] = load; |
| 4802 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4803 | prev_node = node; |
| 4804 | load--; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4805 | if (order == ZONELIST_ORDER_NODE) |
| 4806 | build_zonelists_in_node_order(pgdat, node); |
| 4807 | else |
Yaowei Bai | c00eb15 | 2016-01-14 15:19:00 -0800 | [diff] [blame] | 4808 | node_order[i++] = node; /* remember order */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4809 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4810 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4811 | if (order == ZONELIST_ORDER_ZONE) { |
| 4812 | /* calculate node order -- i.e., DMA last! */ |
Yaowei Bai | c00eb15 | 2016-01-14 15:19:00 -0800 | [diff] [blame] | 4813 | build_zonelists_in_zone_order(pgdat, i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4814 | } |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 4815 | |
| 4816 | build_thisnode_zonelists(pgdat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4817 | } |
| 4818 | |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 4819 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
| 4820 | /* |
| 4821 | * Return node id of node used for "local" allocations. |
| 4822 | * I.e., first node id of first zone in arg node's generic zonelist. |
| 4823 | * Used for initializing percpu 'numa_mem', which is used primarily |
| 4824 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. |
| 4825 | */ |
| 4826 | int local_memory_node(int node) |
| 4827 | { |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 4828 | struct zoneref *z; |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 4829 | |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 4830 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 4831 | gfp_zone(GFP_KERNEL), |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 4832 | NULL); |
| 4833 | return z->zone->node; |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 4834 | } |
| 4835 | #endif |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4836 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4837 | #else /* CONFIG_NUMA */ |
| 4838 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4839 | static void set_zonelist_order(void) |
| 4840 | { |
| 4841 | current_zonelist_order = ZONELIST_ORDER_ZONE; |
| 4842 | } |
| 4843 | |
| 4844 | static void build_zonelists(pg_data_t *pgdat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4845 | { |
Christoph Lameter | 19655d3 | 2006-09-25 23:31:19 -0700 | [diff] [blame] | 4846 | int node, local_node; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4847 | enum zone_type j; |
| 4848 | struct zonelist *zonelist; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4849 | |
| 4850 | local_node = pgdat->node_id; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4851 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4852 | zonelist = &pgdat->node_zonelists[0]; |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4853 | j = build_zonelists_node(pgdat, zonelist, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4854 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4855 | /* |
| 4856 | * Now we build the zonelist so that it contains the zones |
| 4857 | * of all the other nodes. |
| 4858 | * We don't want to pressure a particular node, so when |
| 4859 | * building the zones for node N, we make sure that the |
| 4860 | * zones coming right after the local ones are those from |
| 4861 | * node N+1 (modulo N) |
| 4862 | */ |
| 4863 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { |
| 4864 | if (!node_online(node)) |
| 4865 | continue; |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4866 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4867 | } |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4868 | for (node = 0; node < local_node; node++) { |
| 4869 | if (!node_online(node)) |
| 4870 | continue; |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4871 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4872 | } |
| 4873 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4874 | zonelist->_zonerefs[j].zone = NULL; |
| 4875 | zonelist->_zonerefs[j].zone_idx = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4876 | } |
| 4877 | |
| 4878 | #endif /* CONFIG_NUMA */ |
| 4879 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 4880 | /* |
| 4881 | * Boot pageset table. One per cpu which is going to be used for all |
| 4882 | * zones and all nodes. The parameters will be set in such a way |
| 4883 | * that an item put on a list will immediately be handed over to |
| 4884 | * the buddy list. This is safe since pageset manipulation is done |
| 4885 | * with interrupts disabled. |
| 4886 | * |
| 4887 | * The boot_pagesets must be kept even after bootup is complete for |
| 4888 | * unused processors and/or zones. They do play a role for bootstrapping |
| 4889 | * hotplugged processors. |
| 4890 | * |
| 4891 | * zoneinfo_show() and maybe other functions do |
| 4892 | * not check if the processor is online before following the pageset pointer. |
| 4893 | * Other parts of the kernel may not check if the zone is available. |
| 4894 | */ |
| 4895 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); |
| 4896 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); |
Haicheng Li | 1f52250 | 2010-05-24 14:32:51 -0700 | [diff] [blame] | 4897 | static void setup_zone_pageset(struct zone *zone); |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 4898 | |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 4899 | /* |
| 4900 | * Global mutex to protect against size modification of zonelists |
| 4901 | * as well as to serialize pageset setup for the new populated zone. |
| 4902 | */ |
| 4903 | DEFINE_MUTEX(zonelists_mutex); |
| 4904 | |
Rusty Russell | 9b1a4d3 | 2008-07-28 12:16:30 -0500 | [diff] [blame] | 4905 | /* return values int ....just for stop_machine() */ |
Jiang Liu | 4ed7e02 | 2012-07-31 16:43:35 -0700 | [diff] [blame] | 4906 | static int __build_all_zonelists(void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4907 | { |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 4908 | int nid; |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 4909 | int cpu; |
Jiang Liu | 9adb62a | 2012-07-31 16:43:28 -0700 | [diff] [blame] | 4910 | pg_data_t *self = data; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 4911 | |
Bo Liu | 7f9cfb3 | 2009-08-18 14:11:19 -0700 | [diff] [blame] | 4912 | #ifdef CONFIG_NUMA |
| 4913 | memset(node_load, 0, sizeof(node_load)); |
| 4914 | #endif |
Jiang Liu | 9adb62a | 2012-07-31 16:43:28 -0700 | [diff] [blame] | 4915 | |
| 4916 | if (self && !node_online(self->node_id)) { |
| 4917 | build_zonelists(self); |
Jiang Liu | 9adb62a | 2012-07-31 16:43:28 -0700 | [diff] [blame] | 4918 | } |
| 4919 | |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 4920 | for_each_online_node(nid) { |
Christoph Lameter | 7ea1530 | 2007-10-16 01:25:29 -0700 | [diff] [blame] | 4921 | pg_data_t *pgdat = NODE_DATA(nid); |
| 4922 | |
| 4923 | build_zonelists(pgdat); |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 4924 | } |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 4925 | |
| 4926 | /* |
| 4927 | * Initialize the boot_pagesets that are going to be used |
| 4928 | * for bootstrapping processors. The real pagesets for |
| 4929 | * each zone will be allocated later when the per cpu |
| 4930 | * allocator is available. |
| 4931 | * |
| 4932 | * boot_pagesets are used also for bootstrapping offline |
| 4933 | * cpus if the system is already booted because the pagesets |
| 4934 | * are needed to initialize allocators on a specific cpu too. |
| 4935 | * F.e. the percpu allocator needs the page allocator which |
| 4936 | * needs the percpu allocator in order to allocate its pagesets |
| 4937 | * (a chicken-egg dilemma). |
| 4938 | */ |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 4939 | for_each_possible_cpu(cpu) { |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 4940 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
| 4941 | |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 4942 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
| 4943 | /* |
| 4944 | * We now know the "local memory node" for each node-- |
| 4945 | * i.e., the node of the first zone in the generic zonelist. |
| 4946 | * Set up numa_mem percpu variable for on-line cpus. During |
| 4947 | * boot, only the boot cpu should be on-line; we'll init the |
| 4948 | * secondary cpus' numa_mem as they come on-line. During |
| 4949 | * node/memory hotplug, we'll fixup all on-line cpus. |
| 4950 | */ |
| 4951 | if (cpu_online(cpu)) |
| 4952 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
| 4953 | #endif |
| 4954 | } |
| 4955 | |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 4956 | return 0; |
| 4957 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4958 | |
Rasmus Villemoes | 061f67b | 2015-02-12 15:00:06 -0800 | [diff] [blame] | 4959 | static noinline void __init |
| 4960 | build_all_zonelists_init(void) |
| 4961 | { |
| 4962 | __build_all_zonelists(NULL); |
| 4963 | mminit_verify_zonelist(); |
| 4964 | cpuset_init_current_mems_allowed(); |
| 4965 | } |
| 4966 | |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 4967 | /* |
| 4968 | * Called with zonelists_mutex held always |
| 4969 | * unless system_state == SYSTEM_BOOTING. |
Rasmus Villemoes | 061f67b | 2015-02-12 15:00:06 -0800 | [diff] [blame] | 4970 | * |
| 4971 | * __ref due to (1) call of __meminit annotated setup_zone_pageset |
| 4972 | * [we're only called with non-NULL zone through __meminit paths] and |
| 4973 | * (2) call of __init annotated helper build_all_zonelists_init |
| 4974 | * [protected by SYSTEM_BOOTING]. |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 4975 | */ |
Jiang Liu | 9adb62a | 2012-07-31 16:43:28 -0700 | [diff] [blame] | 4976 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 4977 | { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4978 | set_zonelist_order(); |
| 4979 | |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 4980 | if (system_state == SYSTEM_BOOTING) { |
Rasmus Villemoes | 061f67b | 2015-02-12 15:00:06 -0800 | [diff] [blame] | 4981 | build_all_zonelists_init(); |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 4982 | } else { |
KAMEZAWA Hiroyuki | e9959f0 | 2010-11-24 12:57:09 -0800 | [diff] [blame] | 4983 | #ifdef CONFIG_MEMORY_HOTPLUG |
Jiang Liu | 9adb62a | 2012-07-31 16:43:28 -0700 | [diff] [blame] | 4984 | if (zone) |
| 4985 | setup_zone_pageset(zone); |
KAMEZAWA Hiroyuki | e9959f0 | 2010-11-24 12:57:09 -0800 | [diff] [blame] | 4986 | #endif |
Cody P Schafer | dd1895e | 2013-07-03 15:01:36 -0700 | [diff] [blame] | 4987 | /* we have to stop all cpus to guarantee there is no user |
| 4988 | of zonelist */ |
Jiang Liu | 9adb62a | 2012-07-31 16:43:28 -0700 | [diff] [blame] | 4989 | stop_machine(__build_all_zonelists, pgdat, NULL); |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 4990 | /* cpuset refresh routine should be here */ |
| 4991 | } |
Andrew Morton | bd1e22b | 2006-06-23 02:03:47 -0700 | [diff] [blame] | 4992 | vm_total_pages = nr_free_pagecache_pages(); |
Mel Gorman | 9ef9acb | 2007-10-16 01:25:54 -0700 | [diff] [blame] | 4993 | /* |
| 4994 | * Disable grouping by mobility if the number of pages in the |
| 4995 | * system is too low to allow the mechanism to work. It would be |
| 4996 | * more accurate, but expensive to check per-zone. This check is |
| 4997 | * made on memory-hotadd so a system can start with mobility |
| 4998 | * disabled and enable it later |
| 4999 | */ |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5000 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
Mel Gorman | 9ef9acb | 2007-10-16 01:25:54 -0700 | [diff] [blame] | 5001 | page_group_by_mobility_disabled = 1; |
| 5002 | else |
| 5003 | page_group_by_mobility_disabled = 0; |
| 5004 | |
Joe Perches | 756a025 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 5005 | pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n", |
| 5006 | nr_online_nodes, |
| 5007 | zonelist_order_name[current_zonelist_order], |
| 5008 | page_group_by_mobility_disabled ? "off" : "on", |
| 5009 | vm_total_pages); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 5010 | #ifdef CONFIG_NUMA |
Anton Blanchard | f88dfff | 2014-12-10 15:42:53 -0800 | [diff] [blame] | 5011 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 5012 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5013 | } |
| 5014 | |
| 5015 | /* |
| 5016 | * Helper functions to size the waitqueue hash table. |
| 5017 | * Essentially these want to choose hash table sizes sufficiently |
| 5018 | * large so that collisions trying to wait on pages are rare. |
| 5019 | * But in fact, the number of active page waitqueues on typical |
| 5020 | * systems is ridiculously low, less than 200. So this is even |
| 5021 | * conservative, even though it seems large. |
| 5022 | * |
| 5023 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to |
| 5024 | * waitqueues, i.e. the size of the waitq table given the number of pages. |
| 5025 | */ |
| 5026 | #define PAGES_PER_WAITQUEUE 256 |
| 5027 | |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5028 | #ifndef CONFIG_MEMORY_HOTPLUG |
Yasunori Goto | 02b694d | 2006-06-23 02:03:08 -0700 | [diff] [blame] | 5029 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5030 | { |
| 5031 | unsigned long size = 1; |
| 5032 | |
| 5033 | pages /= PAGES_PER_WAITQUEUE; |
| 5034 | |
| 5035 | while (size < pages) |
| 5036 | size <<= 1; |
| 5037 | |
| 5038 | /* |
| 5039 | * Once we have dozens or even hundreds of threads sleeping |
| 5040 | * on IO we've got bigger problems than wait queue collision. |
| 5041 | * Limit the size of the wait table to a reasonable size. |
| 5042 | */ |
| 5043 | size = min(size, 4096UL); |
| 5044 | |
| 5045 | return max(size, 4UL); |
| 5046 | } |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5047 | #else |
| 5048 | /* |
| 5049 | * A zone's size might be changed by hot-add, so it is not possible to determine |
| 5050 | * a suitable size for its wait_table. So we use the maximum size now. |
| 5051 | * |
| 5052 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: |
| 5053 | * |
| 5054 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. |
| 5055 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. |
| 5056 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. |
| 5057 | * |
| 5058 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages |
| 5059 | * or more by the traditional way. (See above). It equals: |
| 5060 | * |
| 5061 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. |
| 5062 | * ia64(16K page size) : = ( 8G + 4M)byte. |
| 5063 | * powerpc (64K page size) : = (32G +16M)byte. |
| 5064 | */ |
| 5065 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
| 5066 | { |
| 5067 | return 4096UL; |
| 5068 | } |
| 5069 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5070 | |
| 5071 | /* |
| 5072 | * This is an integer logarithm so that shifts can be used later |
| 5073 | * to extract the more random high bits from the multiplicative |
| 5074 | * hash function before the remainder is taken. |
| 5075 | */ |
| 5076 | static inline unsigned long wait_table_bits(unsigned long size) |
| 5077 | { |
| 5078 | return ffz(~size); |
| 5079 | } |
| 5080 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 5081 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5082 | * Initially all pages are reserved - free ones are freed |
| 5083 | * up by free_all_bootmem() once the early boot process is |
| 5084 | * done. Non-atomic initialization, single-pass. |
| 5085 | */ |
Matt Tolentino | c09b424 | 2006-01-17 07:03:44 +0100 | [diff] [blame] | 5086 | 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] | 5087 | unsigned long start_pfn, enum memmap_context context) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5088 | { |
Dan Williams | 4b94ffd | 2016-01-15 16:56:22 -0800 | [diff] [blame] | 5089 | struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn)); |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 5090 | unsigned long end_pfn = start_pfn + size; |
Dan Williams | 4b94ffd | 2016-01-15 16:56:22 -0800 | [diff] [blame] | 5091 | pg_data_t *pgdat = NODE_DATA(nid); |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 5092 | unsigned long pfn; |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 5093 | unsigned long nr_initialised = 0; |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5094 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
| 5095 | struct memblock_region *r = NULL, *tmp; |
| 5096 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5097 | |
Hugh Dickins | 22b31ee | 2009-01-06 14:40:09 -0800 | [diff] [blame] | 5098 | if (highest_memmap_pfn < end_pfn - 1) |
| 5099 | highest_memmap_pfn = end_pfn - 1; |
| 5100 | |
Dan Williams | 4b94ffd | 2016-01-15 16:56:22 -0800 | [diff] [blame] | 5101 | /* |
| 5102 | * Honor reservation requested by the driver for this ZONE_DEVICE |
| 5103 | * memory |
| 5104 | */ |
| 5105 | if (altmap && start_pfn == altmap->base_pfn) |
| 5106 | start_pfn += altmap->reserve; |
| 5107 | |
Greg Ungerer | cbe8dd4 | 2006-01-12 01:05:24 -0800 | [diff] [blame] | 5108 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 5109 | /* |
Andrew Morton | b72d0ff | 2016-03-15 14:55:25 -0700 | [diff] [blame] | 5110 | * There can be holes in boot-time mem_map[]s handed to this |
| 5111 | * function. They do not exist on hotplugged memory. |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 5112 | */ |
Andrew Morton | b72d0ff | 2016-03-15 14:55:25 -0700 | [diff] [blame] | 5113 | if (context != MEMMAP_EARLY) |
| 5114 | goto not_early; |
| 5115 | |
| 5116 | if (!early_pfn_valid(pfn)) |
| 5117 | continue; |
| 5118 | if (!early_pfn_in_nid(pfn, nid)) |
| 5119 | continue; |
| 5120 | if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised)) |
| 5121 | break; |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5122 | |
| 5123 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
Andrew Morton | b72d0ff | 2016-03-15 14:55:25 -0700 | [diff] [blame] | 5124 | /* |
| 5125 | * If not mirrored_kernelcore and ZONE_MOVABLE exists, range |
| 5126 | * from zone_movable_pfn[nid] to end of each node should be |
| 5127 | * ZONE_MOVABLE not ZONE_NORMAL. skip it. |
| 5128 | */ |
| 5129 | if (!mirrored_kernelcore && zone_movable_pfn[nid]) |
| 5130 | if (zone == ZONE_NORMAL && pfn >= zone_movable_pfn[nid]) |
| 5131 | continue; |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5132 | |
Andrew Morton | b72d0ff | 2016-03-15 14:55:25 -0700 | [diff] [blame] | 5133 | /* |
| 5134 | * Check given memblock attribute by firmware which can affect |
| 5135 | * kernel memory layout. If zone==ZONE_MOVABLE but memory is |
| 5136 | * mirrored, it's an overlapped memmap init. skip it. |
| 5137 | */ |
| 5138 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { |
| 5139 | if (!r || pfn >= memblock_region_memory_end_pfn(r)) { |
| 5140 | for_each_memblock(memory, tmp) |
| 5141 | if (pfn < memblock_region_memory_end_pfn(tmp)) |
| 5142 | break; |
| 5143 | r = tmp; |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5144 | } |
Andrew Morton | b72d0ff | 2016-03-15 14:55:25 -0700 | [diff] [blame] | 5145 | if (pfn >= memblock_region_memory_base_pfn(r) && |
| 5146 | memblock_is_mirror(r)) { |
| 5147 | /* already initialized as NORMAL */ |
| 5148 | pfn = memblock_region_memory_end_pfn(r); |
| 5149 | continue; |
| 5150 | } |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 5151 | } |
Andrew Morton | b72d0ff | 2016-03-15 14:55:25 -0700 | [diff] [blame] | 5152 | #endif |
Mel Gorman | ac5d253 | 2015-06-30 14:57:20 -0700 | [diff] [blame] | 5153 | |
Andrew Morton | b72d0ff | 2016-03-15 14:55:25 -0700 | [diff] [blame] | 5154 | not_early: |
Mel Gorman | ac5d253 | 2015-06-30 14:57:20 -0700 | [diff] [blame] | 5155 | /* |
| 5156 | * Mark the block movable so that blocks are reserved for |
| 5157 | * movable at startup. This will force kernel allocations |
| 5158 | * to reserve their blocks rather than leaking throughout |
| 5159 | * the address space during boot when many long-lived |
Mel Gorman | 974a786 | 2015-11-06 16:28:34 -0800 | [diff] [blame] | 5160 | * kernel allocations are made. |
Mel Gorman | ac5d253 | 2015-06-30 14:57:20 -0700 | [diff] [blame] | 5161 | * |
| 5162 | * bitmap is created for zone's valid pfn range. but memmap |
| 5163 | * can be created for invalid pages (for alignment) |
| 5164 | * check here not to call set_pageblock_migratetype() against |
| 5165 | * pfn out of zone. |
| 5166 | */ |
| 5167 | if (!(pfn & (pageblock_nr_pages - 1))) { |
| 5168 | struct page *page = pfn_to_page(pfn); |
| 5169 | |
| 5170 | __init_single_page(page, pfn, zone, nid); |
| 5171 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
| 5172 | } else { |
| 5173 | __init_single_pfn(pfn, zone, nid); |
| 5174 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5175 | } |
| 5176 | } |
| 5177 | |
Andi Kleen | 1e548de | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 5178 | static void __meminit zone_init_free_lists(struct zone *zone) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5179 | { |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 5180 | unsigned int order, t; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 5181 | for_each_migratetype_order(order, t) { |
| 5182 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5183 | zone->free_area[order].nr_free = 0; |
| 5184 | } |
| 5185 | } |
| 5186 | |
| 5187 | #ifndef __HAVE_ARCH_MEMMAP_INIT |
| 5188 | #define memmap_init(size, nid, zone, start_pfn) \ |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 5189 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5190 | #endif |
| 5191 | |
David Rientjes | 7cd2b0a | 2014-06-23 13:22:04 -0700 | [diff] [blame] | 5192 | static int zone_batchsize(struct zone *zone) |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5193 | { |
David Howells | 3a6be87 | 2009-05-06 16:03:03 -0700 | [diff] [blame] | 5194 | #ifdef CONFIG_MMU |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5195 | int batch; |
| 5196 | |
| 5197 | /* |
| 5198 | * The per-cpu-pages pools are set to around 1000th of the |
Seth, Rohit | ba56e91 | 2005-10-29 18:15:47 -0700 | [diff] [blame] | 5199 | * 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] | 5200 | * |
| 5201 | * OK, so we don't know how big the cache is. So guess. |
| 5202 | */ |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 5203 | batch = zone->managed_pages / 1024; |
Seth, Rohit | ba56e91 | 2005-10-29 18:15:47 -0700 | [diff] [blame] | 5204 | if (batch * PAGE_SIZE > 512 * 1024) |
| 5205 | batch = (512 * 1024) / PAGE_SIZE; |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5206 | batch /= 4; /* We effectively *= 4 below */ |
| 5207 | if (batch < 1) |
| 5208 | batch = 1; |
| 5209 | |
| 5210 | /* |
Nick Piggin | 0ceaacc | 2005-12-04 13:55:25 +1100 | [diff] [blame] | 5211 | * Clamp the batch to a 2^n - 1 value. Having a power |
| 5212 | * of 2 value was found to be more likely to have |
| 5213 | * suboptimal cache aliasing properties in some cases. |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5214 | * |
Nick Piggin | 0ceaacc | 2005-12-04 13:55:25 +1100 | [diff] [blame] | 5215 | * For example if 2 tasks are alternately allocating |
| 5216 | * batches of pages, one task can end up with a lot |
| 5217 | * of pages of one half of the possible page colors |
| 5218 | * and the other with pages of the other colors. |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5219 | */ |
David Howells | 9155203 | 2009-05-06 16:03:02 -0700 | [diff] [blame] | 5220 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
Seth, Rohit | ba56e91 | 2005-10-29 18:15:47 -0700 | [diff] [blame] | 5221 | |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5222 | return batch; |
David Howells | 3a6be87 | 2009-05-06 16:03:03 -0700 | [diff] [blame] | 5223 | |
| 5224 | #else |
| 5225 | /* The deferral and batching of frees should be suppressed under NOMMU |
| 5226 | * conditions. |
| 5227 | * |
| 5228 | * The problem is that NOMMU needs to be able to allocate large chunks |
| 5229 | * of contiguous memory as there's no hardware page translation to |
| 5230 | * assemble apparent contiguous memory from discontiguous pages. |
| 5231 | * |
| 5232 | * Queueing large contiguous runs of pages for batching, however, |
| 5233 | * causes the pages to actually be freed in smaller chunks. As there |
| 5234 | * can be a significant delay between the individual batches being |
| 5235 | * recycled, this leads to the once large chunks of space being |
| 5236 | * fragmented and becoming unavailable for high-order allocations. |
| 5237 | */ |
| 5238 | return 0; |
| 5239 | #endif |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5240 | } |
| 5241 | |
Cody P Schafer | 8d7a8fa | 2013-07-03 15:01:31 -0700 | [diff] [blame] | 5242 | /* |
| 5243 | * pcp->high and pcp->batch values are related and dependent on one another: |
| 5244 | * ->batch must never be higher then ->high. |
| 5245 | * The following function updates them in a safe manner without read side |
| 5246 | * locking. |
| 5247 | * |
| 5248 | * Any new users of pcp->batch and pcp->high should ensure they can cope with |
| 5249 | * those fields changing asynchronously (acording the the above rule). |
| 5250 | * |
| 5251 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function |
| 5252 | * outside of boot time (or some other assurance that no concurrent updaters |
| 5253 | * exist). |
| 5254 | */ |
| 5255 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, |
| 5256 | unsigned long batch) |
| 5257 | { |
| 5258 | /* start with a fail safe value for batch */ |
| 5259 | pcp->batch = 1; |
| 5260 | smp_wmb(); |
| 5261 | |
| 5262 | /* Update high, then batch, in order */ |
| 5263 | pcp->high = high; |
| 5264 | smp_wmb(); |
| 5265 | |
| 5266 | pcp->batch = batch; |
| 5267 | } |
| 5268 | |
Cody P Schafer | 3664033 | 2013-07-03 15:01:40 -0700 | [diff] [blame] | 5269 | /* a companion to pageset_set_high() */ |
Cody P Schafer | 4008bab | 2013-07-03 15:01:28 -0700 | [diff] [blame] | 5270 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
| 5271 | { |
Cody P Schafer | 8d7a8fa | 2013-07-03 15:01:31 -0700 | [diff] [blame] | 5272 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
Cody P Schafer | 4008bab | 2013-07-03 15:01:28 -0700 | [diff] [blame] | 5273 | } |
| 5274 | |
Cody P Schafer | 88c90db | 2013-07-03 15:01:35 -0700 | [diff] [blame] | 5275 | static void pageset_init(struct per_cpu_pageset *p) |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 5276 | { |
| 5277 | struct per_cpu_pages *pcp; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 5278 | int migratetype; |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 5279 | |
Magnus Damm | 1c6fe94 | 2005-10-26 01:58:59 -0700 | [diff] [blame] | 5280 | memset(p, 0, sizeof(*p)); |
| 5281 | |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 5282 | pcp = &p->pcp; |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 5283 | pcp->count = 0; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 5284 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
| 5285 | INIT_LIST_HEAD(&pcp->lists[migratetype]); |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 5286 | } |
| 5287 | |
Cody P Schafer | 88c90db | 2013-07-03 15:01:35 -0700 | [diff] [blame] | 5288 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
| 5289 | { |
| 5290 | pageset_init(p); |
| 5291 | pageset_set_batch(p, batch); |
| 5292 | } |
| 5293 | |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5294 | /* |
Cody P Schafer | 3664033 | 2013-07-03 15:01:40 -0700 | [diff] [blame] | 5295 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5296 | * to the value high for the pageset p. |
| 5297 | */ |
Cody P Schafer | 3664033 | 2013-07-03 15:01:40 -0700 | [diff] [blame] | 5298 | static void pageset_set_high(struct per_cpu_pageset *p, |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5299 | unsigned long high) |
| 5300 | { |
Cody P Schafer | 8d7a8fa | 2013-07-03 15:01:31 -0700 | [diff] [blame] | 5301 | unsigned long batch = max(1UL, high / 4); |
| 5302 | if ((high / 4) > (PAGE_SHIFT * 8)) |
| 5303 | batch = PAGE_SHIFT * 8; |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5304 | |
Cody P Schafer | 8d7a8fa | 2013-07-03 15:01:31 -0700 | [diff] [blame] | 5305 | pageset_update(&p->pcp, high, batch); |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5306 | } |
| 5307 | |
David Rientjes | 7cd2b0a | 2014-06-23 13:22:04 -0700 | [diff] [blame] | 5308 | static void pageset_set_high_and_batch(struct zone *zone, |
| 5309 | struct per_cpu_pageset *pcp) |
Cody P Schafer | 56cef2b | 2013-07-03 15:01:38 -0700 | [diff] [blame] | 5310 | { |
Cody P Schafer | 56cef2b | 2013-07-03 15:01:38 -0700 | [diff] [blame] | 5311 | if (percpu_pagelist_fraction) |
Cody P Schafer | 3664033 | 2013-07-03 15:01:40 -0700 | [diff] [blame] | 5312 | pageset_set_high(pcp, |
Cody P Schafer | 56cef2b | 2013-07-03 15:01:38 -0700 | [diff] [blame] | 5313 | (zone->managed_pages / |
| 5314 | percpu_pagelist_fraction)); |
| 5315 | else |
| 5316 | pageset_set_batch(pcp, zone_batchsize(zone)); |
| 5317 | } |
| 5318 | |
Cody P Schafer | 169f6c1 | 2013-07-03 15:01:41 -0700 | [diff] [blame] | 5319 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
| 5320 | { |
| 5321 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); |
| 5322 | |
| 5323 | pageset_init(pcp); |
| 5324 | pageset_set_high_and_batch(zone, pcp); |
| 5325 | } |
| 5326 | |
Jiang Liu | 4ed7e02 | 2012-07-31 16:43:35 -0700 | [diff] [blame] | 5327 | static void __meminit setup_zone_pageset(struct zone *zone) |
Wu Fengguang | 319774e | 2010-05-24 14:32:49 -0700 | [diff] [blame] | 5328 | { |
| 5329 | int cpu; |
Wu Fengguang | 319774e | 2010-05-24 14:32:49 -0700 | [diff] [blame] | 5330 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
Cody P Schafer | 56cef2b | 2013-07-03 15:01:38 -0700 | [diff] [blame] | 5331 | for_each_possible_cpu(cpu) |
| 5332 | zone_pageset_init(zone, cpu); |
Mel Gorman | 75ef718 | 2016-07-28 15:45:24 -0700 | [diff] [blame] | 5333 | |
| 5334 | if (!zone->zone_pgdat->per_cpu_nodestats) { |
| 5335 | zone->zone_pgdat->per_cpu_nodestats = |
| 5336 | alloc_percpu(struct per_cpu_nodestat); |
| 5337 | } |
Wu Fengguang | 319774e | 2010-05-24 14:32:49 -0700 | [diff] [blame] | 5338 | } |
| 5339 | |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5340 | /* |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 5341 | * Allocate per cpu pagesets and initialize them. |
| 5342 | * Before this call only boot pagesets were available. |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 5343 | */ |
Al Viro | 78d9955 | 2005-12-15 09:18:25 +0000 | [diff] [blame] | 5344 | void __init setup_per_cpu_pageset(void) |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5345 | { |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 5346 | struct zone *zone; |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5347 | |
Wu Fengguang | 319774e | 2010-05-24 14:32:49 -0700 | [diff] [blame] | 5348 | for_each_populated_zone(zone) |
| 5349 | setup_zone_pageset(zone); |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5350 | } |
| 5351 | |
Sam Ravnborg | 577a32f | 2007-05-17 23:29:25 +0200 | [diff] [blame] | 5352 | static noinline __init_refok |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5353 | 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] | 5354 | { |
| 5355 | int i; |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5356 | size_t alloc_size; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5357 | |
| 5358 | /* |
| 5359 | * The per-page waitqueue mechanism uses hashed waitqueues |
| 5360 | * per zone. |
| 5361 | */ |
Yasunori Goto | 02b694d | 2006-06-23 02:03:08 -0700 | [diff] [blame] | 5362 | zone->wait_table_hash_nr_entries = |
| 5363 | wait_table_hash_nr_entries(zone_size_pages); |
| 5364 | zone->wait_table_bits = |
| 5365 | wait_table_bits(zone->wait_table_hash_nr_entries); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5366 | alloc_size = zone->wait_table_hash_nr_entries |
| 5367 | * sizeof(wait_queue_head_t); |
| 5368 | |
Heiko Carstens | cd94b9d | 2008-05-23 13:04:52 -0700 | [diff] [blame] | 5369 | if (!slab_is_available()) { |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5370 | zone->wait_table = (wait_queue_head_t *) |
Santosh Shilimkar | 6782832 | 2014-01-21 15:50:25 -0800 | [diff] [blame] | 5371 | memblock_virt_alloc_node_nopanic( |
| 5372 | alloc_size, zone->zone_pgdat->node_id); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5373 | } else { |
| 5374 | /* |
| 5375 | * This case means that a zone whose size was 0 gets new memory |
| 5376 | * via memory hot-add. |
| 5377 | * But it may be the case that a new node was hot-added. In |
| 5378 | * this case vmalloc() will not be able to use this new node's |
| 5379 | * memory - this wait_table must be initialized to use this new |
| 5380 | * node itself as well. |
| 5381 | * To use this new node's memory, further consideration will be |
| 5382 | * necessary. |
| 5383 | */ |
Jesper Juhl | 8691f3a | 2007-10-16 01:24:49 -0700 | [diff] [blame] | 5384 | zone->wait_table = vmalloc(alloc_size); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5385 | } |
| 5386 | if (!zone->wait_table) |
| 5387 | return -ENOMEM; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5388 | |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 5389 | for (i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5390 | init_waitqueue_head(zone->wait_table + i); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5391 | |
| 5392 | return 0; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5393 | } |
| 5394 | |
Matt Tolentino | c09b424 | 2006-01-17 07:03:44 +0100 | [diff] [blame] | 5395 | static __meminit void zone_pcp_init(struct zone *zone) |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5396 | { |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 5397 | /* |
| 5398 | * per cpu subsystem is not up at this point. The following code |
| 5399 | * relies on the ability of the linker to provide the |
| 5400 | * offset of a (static) per cpu variable into the per cpu area. |
| 5401 | */ |
| 5402 | zone->pageset = &boot_pageset; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5403 | |
Xishi Qiu | b38a872 | 2013-11-12 15:07:20 -0800 | [diff] [blame] | 5404 | if (populated_zone(zone)) |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 5405 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
| 5406 | zone->name, zone->present_pages, |
| 5407 | zone_batchsize(zone)); |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5408 | } |
| 5409 | |
Jiang Liu | 4ed7e02 | 2012-07-31 16:43:35 -0700 | [diff] [blame] | 5410 | int __meminit init_currently_empty_zone(struct zone *zone, |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5411 | unsigned long zone_start_pfn, |
Yaowei Bai | b171e40 | 2015-11-05 18:47:06 -0800 | [diff] [blame] | 5412 | unsigned long size) |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5413 | { |
| 5414 | struct pglist_data *pgdat = zone->zone_pgdat; |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5415 | int ret; |
| 5416 | ret = zone_wait_table_init(zone, size); |
| 5417 | if (ret) |
| 5418 | return ret; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5419 | pgdat->nr_zones = zone_idx(zone) + 1; |
| 5420 | |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5421 | zone->zone_start_pfn = zone_start_pfn; |
| 5422 | |
Mel Gorman | 708614e | 2008-07-23 21:26:51 -0700 | [diff] [blame] | 5423 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
| 5424 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", |
| 5425 | pgdat->node_id, |
| 5426 | (unsigned long)zone_idx(zone), |
| 5427 | zone_start_pfn, (zone_start_pfn + size)); |
| 5428 | |
Andi Kleen | 1e548de | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 5429 | zone_init_free_lists(zone); |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5430 | |
| 5431 | return 0; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5432 | } |
| 5433 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 5434 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5435 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
Mel Gorman | 8a942fd | 2015-06-30 14:56:55 -0700 | [diff] [blame] | 5436 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5437 | /* |
| 5438 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5439 | */ |
Mel Gorman | 8a942fd | 2015-06-30 14:56:55 -0700 | [diff] [blame] | 5440 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
| 5441 | struct mminit_pfnnid_cache *state) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5442 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5443 | unsigned long start_pfn, end_pfn; |
Yinghai Lu | e76b63f | 2013-09-11 14:22:17 -0700 | [diff] [blame] | 5444 | int nid; |
Russ Anderson | 7c243c7 | 2013-04-29 15:07:59 -0700 | [diff] [blame] | 5445 | |
Mel Gorman | 8a942fd | 2015-06-30 14:56:55 -0700 | [diff] [blame] | 5446 | if (state->last_start <= pfn && pfn < state->last_end) |
| 5447 | return state->last_nid; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5448 | |
Yinghai Lu | e76b63f | 2013-09-11 14:22:17 -0700 | [diff] [blame] | 5449 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
| 5450 | if (nid != -1) { |
Mel Gorman | 8a942fd | 2015-06-30 14:56:55 -0700 | [diff] [blame] | 5451 | state->last_start = start_pfn; |
| 5452 | state->last_end = end_pfn; |
| 5453 | state->last_nid = nid; |
Yinghai Lu | e76b63f | 2013-09-11 14:22:17 -0700 | [diff] [blame] | 5454 | } |
| 5455 | |
| 5456 | return nid; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5457 | } |
| 5458 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ |
| 5459 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5460 | /** |
Santosh Shilimkar | 6782832 | 2014-01-21 15:50:25 -0800 | [diff] [blame] | 5461 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 5462 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
Santosh Shilimkar | 6782832 | 2014-01-21 15:50:25 -0800 | [diff] [blame] | 5463 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5464 | * |
Zhang Zhen | 7d01817 | 2014-06-04 16:10:53 -0700 | [diff] [blame] | 5465 | * If an architecture guarantees that all ranges registered contain no holes |
| 5466 | * and may be freed, this this function may be used instead of calling |
| 5467 | * memblock_free_early_nid() manually. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5468 | */ |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5469 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5470 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5471 | unsigned long start_pfn, end_pfn; |
| 5472 | int i, this_nid; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5473 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5474 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
| 5475 | start_pfn = min(start_pfn, max_low_pfn); |
| 5476 | end_pfn = min(end_pfn, max_low_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5477 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5478 | if (start_pfn < end_pfn) |
Santosh Shilimkar | 6782832 | 2014-01-21 15:50:25 -0800 | [diff] [blame] | 5479 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
| 5480 | (end_pfn - start_pfn) << PAGE_SHIFT, |
| 5481 | this_nid); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5482 | } |
| 5483 | } |
| 5484 | |
| 5485 | /** |
| 5486 | * 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] | 5487 | * @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] | 5488 | * |
Zhang Zhen | 7d01817 | 2014-06-04 16:10:53 -0700 | [diff] [blame] | 5489 | * If an architecture guarantees that all ranges registered contain no holes and may |
| 5490 | * be freed, this function may be used instead of calling memory_present() manually. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5491 | */ |
| 5492 | void __init sparse_memory_present_with_active_regions(int nid) |
| 5493 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5494 | unsigned long start_pfn, end_pfn; |
| 5495 | int i, this_nid; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5496 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5497 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
| 5498 | memory_present(this_nid, start_pfn, end_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5499 | } |
| 5500 | |
| 5501 | /** |
| 5502 | * 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] | 5503 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
| 5504 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. |
| 5505 | * @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] | 5506 | * |
| 5507 | * It returns the start and end page frame of a node based on information |
Zhang Zhen | 7d01817 | 2014-06-04 16:10:53 -0700 | [diff] [blame] | 5508 | * provided by memblock_set_node(). If called for a node |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5509 | * 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] | 5510 | * PFNs will be 0. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5511 | */ |
Yasunori Goto | a3142c8 | 2007-05-08 00:23:07 -0700 | [diff] [blame] | 5512 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5513 | unsigned long *start_pfn, unsigned long *end_pfn) |
| 5514 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5515 | unsigned long this_start_pfn, this_end_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5516 | int i; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5517 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5518 | *start_pfn = -1UL; |
| 5519 | *end_pfn = 0; |
| 5520 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5521 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
| 5522 | *start_pfn = min(*start_pfn, this_start_pfn); |
| 5523 | *end_pfn = max(*end_pfn, this_end_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5524 | } |
| 5525 | |
Christoph Lameter | 633c066 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 5526 | if (*start_pfn == -1UL) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5527 | *start_pfn = 0; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5528 | } |
| 5529 | |
| 5530 | /* |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 5531 | * This finds a zone that can be used for ZONE_MOVABLE pages. The |
| 5532 | * assumption is made that zones within a node are ordered in monotonic |
| 5533 | * increasing memory addresses so that the "highest" populated zone is used |
| 5534 | */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 5535 | static void __init find_usable_zone_for_movable(void) |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 5536 | { |
| 5537 | int zone_index; |
| 5538 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { |
| 5539 | if (zone_index == ZONE_MOVABLE) |
| 5540 | continue; |
| 5541 | |
| 5542 | if (arch_zone_highest_possible_pfn[zone_index] > |
| 5543 | arch_zone_lowest_possible_pfn[zone_index]) |
| 5544 | break; |
| 5545 | } |
| 5546 | |
| 5547 | VM_BUG_ON(zone_index == -1); |
| 5548 | movable_zone = zone_index; |
| 5549 | } |
| 5550 | |
| 5551 | /* |
| 5552 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 5553 | * because it is sized independent of architecture. Unlike the other zones, |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 5554 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
| 5555 | * in each node depending on the size of each node and how evenly kernelcore |
| 5556 | * is distributed. This helper function adjusts the zone ranges |
| 5557 | * provided by the architecture for a given node by using the end of the |
| 5558 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that |
| 5559 | * zones within a node are in order of monotonic increases memory addresses |
| 5560 | */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 5561 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 5562 | unsigned long zone_type, |
| 5563 | unsigned long node_start_pfn, |
| 5564 | unsigned long node_end_pfn, |
| 5565 | unsigned long *zone_start_pfn, |
| 5566 | unsigned long *zone_end_pfn) |
| 5567 | { |
| 5568 | /* Only adjust if ZONE_MOVABLE is on this node */ |
| 5569 | if (zone_movable_pfn[nid]) { |
| 5570 | /* Size ZONE_MOVABLE */ |
| 5571 | if (zone_type == ZONE_MOVABLE) { |
| 5572 | *zone_start_pfn = zone_movable_pfn[nid]; |
| 5573 | *zone_end_pfn = min(node_end_pfn, |
| 5574 | arch_zone_highest_possible_pfn[movable_zone]); |
| 5575 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 5576 | /* Check if this whole range is within ZONE_MOVABLE */ |
| 5577 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) |
| 5578 | *zone_start_pfn = *zone_end_pfn; |
| 5579 | } |
| 5580 | } |
| 5581 | |
| 5582 | /* |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5583 | * Return the number of pages a zone spans in a node, including holes |
| 5584 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() |
| 5585 | */ |
Paul Mundt | 6ea6e68 | 2007-07-15 23:38:20 -0700 | [diff] [blame] | 5586 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5587 | unsigned long zone_type, |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5588 | unsigned long node_start_pfn, |
| 5589 | unsigned long node_end_pfn, |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5590 | unsigned long *zone_start_pfn, |
| 5591 | unsigned long *zone_end_pfn, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5592 | unsigned long *ignored) |
| 5593 | { |
Xishi Qiu | b5685e9 | 2015-09-08 15:04:16 -0700 | [diff] [blame] | 5594 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
Xishi Qiu | f9126ab | 2015-08-14 15:35:16 -0700 | [diff] [blame] | 5595 | if (!node_start_pfn && !node_end_pfn) |
| 5596 | return 0; |
| 5597 | |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5598 | /* Get the start and end of the zone */ |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5599 | *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
| 5600 | *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 5601 | adjust_zone_range_for_zone_movable(nid, zone_type, |
| 5602 | node_start_pfn, node_end_pfn, |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5603 | zone_start_pfn, zone_end_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5604 | |
| 5605 | /* Check that this node has pages within the zone's required range */ |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5606 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5607 | return 0; |
| 5608 | |
| 5609 | /* Move the zone boundaries inside the node if necessary */ |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5610 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
| 5611 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5612 | |
| 5613 | /* Return the spanned pages */ |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5614 | return *zone_end_pfn - *zone_start_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5615 | } |
| 5616 | |
| 5617 | /* |
| 5618 | * 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] | 5619 | * then all holes in the requested range will be accounted for. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5620 | */ |
Yinghai Lu | 3299625 | 2009-12-15 17:59:02 -0800 | [diff] [blame] | 5621 | unsigned long __meminit __absent_pages_in_range(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5622 | unsigned long range_start_pfn, |
| 5623 | unsigned long range_end_pfn) |
| 5624 | { |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 5625 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
| 5626 | unsigned long start_pfn, end_pfn; |
| 5627 | int i; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5628 | |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 5629 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
| 5630 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); |
| 5631 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); |
| 5632 | nr_absent -= end_pfn - start_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5633 | } |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 5634 | return nr_absent; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5635 | } |
| 5636 | |
| 5637 | /** |
| 5638 | * absent_pages_in_range - Return number of page frames in holes within a range |
| 5639 | * @start_pfn: The start PFN to start searching for holes |
| 5640 | * @end_pfn: The end PFN to stop searching for holes |
| 5641 | * |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 5642 | * 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] | 5643 | */ |
| 5644 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, |
| 5645 | unsigned long end_pfn) |
| 5646 | { |
| 5647 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); |
| 5648 | } |
| 5649 | |
| 5650 | /* 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] | 5651 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5652 | unsigned long zone_type, |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5653 | unsigned long node_start_pfn, |
| 5654 | unsigned long node_end_pfn, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5655 | unsigned long *ignored) |
| 5656 | { |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 5657 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
| 5658 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; |
Mel Gorman | 9c7cd68 | 2006-09-27 01:49:58 -0700 | [diff] [blame] | 5659 | unsigned long zone_start_pfn, zone_end_pfn; |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5660 | unsigned long nr_absent; |
Mel Gorman | 9c7cd68 | 2006-09-27 01:49:58 -0700 | [diff] [blame] | 5661 | |
Xishi Qiu | b5685e9 | 2015-09-08 15:04:16 -0700 | [diff] [blame] | 5662 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
Xishi Qiu | f9126ab | 2015-08-14 15:35:16 -0700 | [diff] [blame] | 5663 | if (!node_start_pfn && !node_end_pfn) |
| 5664 | return 0; |
| 5665 | |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 5666 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
| 5667 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); |
Mel Gorman | 9c7cd68 | 2006-09-27 01:49:58 -0700 | [diff] [blame] | 5668 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 5669 | adjust_zone_range_for_zone_movable(nid, zone_type, |
| 5670 | node_start_pfn, node_end_pfn, |
| 5671 | &zone_start_pfn, &zone_end_pfn); |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5672 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
| 5673 | |
| 5674 | /* |
| 5675 | * ZONE_MOVABLE handling. |
| 5676 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages |
| 5677 | * and vice versa. |
| 5678 | */ |
| 5679 | if (zone_movable_pfn[nid]) { |
| 5680 | if (mirrored_kernelcore) { |
| 5681 | unsigned long start_pfn, end_pfn; |
| 5682 | struct memblock_region *r; |
| 5683 | |
| 5684 | for_each_memblock(memory, r) { |
| 5685 | start_pfn = clamp(memblock_region_memory_base_pfn(r), |
| 5686 | zone_start_pfn, zone_end_pfn); |
| 5687 | end_pfn = clamp(memblock_region_memory_end_pfn(r), |
| 5688 | zone_start_pfn, zone_end_pfn); |
| 5689 | |
| 5690 | if (zone_type == ZONE_MOVABLE && |
| 5691 | memblock_is_mirror(r)) |
| 5692 | nr_absent += end_pfn - start_pfn; |
| 5693 | |
| 5694 | if (zone_type == ZONE_NORMAL && |
| 5695 | !memblock_is_mirror(r)) |
| 5696 | nr_absent += end_pfn - start_pfn; |
| 5697 | } |
| 5698 | } else { |
| 5699 | if (zone_type == ZONE_NORMAL) |
| 5700 | nr_absent += node_end_pfn - zone_movable_pfn[nid]; |
| 5701 | } |
| 5702 | } |
| 5703 | |
| 5704 | return nr_absent; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5705 | } |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 5706 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 5707 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Paul Mundt | 6ea6e68 | 2007-07-15 23:38:20 -0700 | [diff] [blame] | 5708 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5709 | unsigned long zone_type, |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5710 | unsigned long node_start_pfn, |
| 5711 | unsigned long node_end_pfn, |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5712 | unsigned long *zone_start_pfn, |
| 5713 | unsigned long *zone_end_pfn, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5714 | unsigned long *zones_size) |
| 5715 | { |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5716 | unsigned int zone; |
| 5717 | |
| 5718 | *zone_start_pfn = node_start_pfn; |
| 5719 | for (zone = 0; zone < zone_type; zone++) |
| 5720 | *zone_start_pfn += zones_size[zone]; |
| 5721 | |
| 5722 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; |
| 5723 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5724 | return zones_size[zone_type]; |
| 5725 | } |
| 5726 | |
Paul Mundt | 6ea6e68 | 2007-07-15 23:38:20 -0700 | [diff] [blame] | 5727 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5728 | unsigned long zone_type, |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5729 | unsigned long node_start_pfn, |
| 5730 | unsigned long node_end_pfn, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5731 | unsigned long *zholes_size) |
| 5732 | { |
| 5733 | if (!zholes_size) |
| 5734 | return 0; |
| 5735 | |
| 5736 | return zholes_size[zone_type]; |
| 5737 | } |
Yinghai Lu | 20e6926 | 2013-03-01 14:51:27 -0800 | [diff] [blame] | 5738 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 5739 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5740 | |
Yasunori Goto | a3142c8 | 2007-05-08 00:23:07 -0700 | [diff] [blame] | 5741 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5742 | unsigned long node_start_pfn, |
| 5743 | unsigned long node_end_pfn, |
| 5744 | unsigned long *zones_size, |
| 5745 | unsigned long *zholes_size) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5746 | { |
Gu Zheng | febd594 | 2015-06-24 16:57:02 -0700 | [diff] [blame] | 5747 | unsigned long realtotalpages = 0, totalpages = 0; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5748 | enum zone_type i; |
| 5749 | |
Gu Zheng | febd594 | 2015-06-24 16:57:02 -0700 | [diff] [blame] | 5750 | for (i = 0; i < MAX_NR_ZONES; i++) { |
| 5751 | struct zone *zone = pgdat->node_zones + i; |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5752 | unsigned long zone_start_pfn, zone_end_pfn; |
Gu Zheng | febd594 | 2015-06-24 16:57:02 -0700 | [diff] [blame] | 5753 | unsigned long size, real_size; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5754 | |
Gu Zheng | febd594 | 2015-06-24 16:57:02 -0700 | [diff] [blame] | 5755 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
| 5756 | node_start_pfn, |
| 5757 | node_end_pfn, |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5758 | &zone_start_pfn, |
| 5759 | &zone_end_pfn, |
Gu Zheng | febd594 | 2015-06-24 16:57:02 -0700 | [diff] [blame] | 5760 | zones_size); |
| 5761 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5762 | node_start_pfn, node_end_pfn, |
| 5763 | zholes_size); |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5764 | if (size) |
| 5765 | zone->zone_start_pfn = zone_start_pfn; |
| 5766 | else |
| 5767 | zone->zone_start_pfn = 0; |
Gu Zheng | febd594 | 2015-06-24 16:57:02 -0700 | [diff] [blame] | 5768 | zone->spanned_pages = size; |
| 5769 | zone->present_pages = real_size; |
| 5770 | |
| 5771 | totalpages += size; |
| 5772 | realtotalpages += real_size; |
| 5773 | } |
| 5774 | |
| 5775 | pgdat->node_spanned_pages = totalpages; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5776 | pgdat->node_present_pages = realtotalpages; |
| 5777 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, |
| 5778 | realtotalpages); |
| 5779 | } |
| 5780 | |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5781 | #ifndef CONFIG_SPARSEMEM |
| 5782 | /* |
| 5783 | * Calculate the size of the zone->blockflags rounded to an unsigned long |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5784 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
| 5785 | * 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] | 5786 | * round what is now in bits to nearest long in bits, then return it in |
| 5787 | * bytes. |
| 5788 | */ |
Linus Torvalds | 7c45512 | 2013-02-18 09:58:02 -0800 | [diff] [blame] | 5789 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5790 | { |
| 5791 | unsigned long usemapsize; |
| 5792 | |
Linus Torvalds | 7c45512 | 2013-02-18 09:58:02 -0800 | [diff] [blame] | 5793 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5794 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
| 5795 | usemapsize = usemapsize >> pageblock_order; |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5796 | usemapsize *= NR_PAGEBLOCK_BITS; |
| 5797 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); |
| 5798 | |
| 5799 | return usemapsize / 8; |
| 5800 | } |
| 5801 | |
| 5802 | static void __init setup_usemap(struct pglist_data *pgdat, |
Linus Torvalds | 7c45512 | 2013-02-18 09:58:02 -0800 | [diff] [blame] | 5803 | struct zone *zone, |
| 5804 | unsigned long zone_start_pfn, |
| 5805 | unsigned long zonesize) |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5806 | { |
Linus Torvalds | 7c45512 | 2013-02-18 09:58:02 -0800 | [diff] [blame] | 5807 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5808 | zone->pageblock_flags = NULL; |
Julia Lawall | 58a01a4 | 2009-01-06 14:39:28 -0800 | [diff] [blame] | 5809 | if (usemapsize) |
Santosh Shilimkar | 6782832 | 2014-01-21 15:50:25 -0800 | [diff] [blame] | 5810 | zone->pageblock_flags = |
| 5811 | memblock_virt_alloc_node_nopanic(usemapsize, |
| 5812 | pgdat->node_id); |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5813 | } |
| 5814 | #else |
Linus Torvalds | 7c45512 | 2013-02-18 09:58:02 -0800 | [diff] [blame] | 5815 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
| 5816 | unsigned long zone_start_pfn, unsigned long zonesize) {} |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5817 | #endif /* CONFIG_SPARSEMEM */ |
| 5818 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5819 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 5820 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5821 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
Chen Gang | 15ca220 | 2013-09-11 14:20:27 -0700 | [diff] [blame] | 5822 | void __paginginit set_pageblock_order(void) |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5823 | { |
Andrew Morton | 955c1cd | 2012-05-29 15:06:31 -0700 | [diff] [blame] | 5824 | unsigned int order; |
| 5825 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5826 | /* Check that pageblock_nr_pages has not already been setup */ |
| 5827 | if (pageblock_order) |
| 5828 | return; |
| 5829 | |
Andrew Morton | 955c1cd | 2012-05-29 15:06:31 -0700 | [diff] [blame] | 5830 | if (HPAGE_SHIFT > PAGE_SHIFT) |
| 5831 | order = HUGETLB_PAGE_ORDER; |
| 5832 | else |
| 5833 | order = MAX_ORDER - 1; |
| 5834 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5835 | /* |
| 5836 | * Assume the largest contiguous order of interest is a huge page. |
Andrew Morton | 955c1cd | 2012-05-29 15:06:31 -0700 | [diff] [blame] | 5837 | * This value may be variable depending on boot parameters on IA64 and |
| 5838 | * powerpc. |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5839 | */ |
| 5840 | pageblock_order = order; |
| 5841 | } |
| 5842 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ |
| 5843 | |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 5844 | /* |
| 5845 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() |
Andrew Morton | 955c1cd | 2012-05-29 15:06:31 -0700 | [diff] [blame] | 5846 | * is unused as pageblock_order is set at compile-time. See |
| 5847 | * include/linux/pageblock-flags.h for the values of pageblock_order based on |
| 5848 | * the kernel config |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 5849 | */ |
Chen Gang | 15ca220 | 2013-09-11 14:20:27 -0700 | [diff] [blame] | 5850 | void __paginginit set_pageblock_order(void) |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 5851 | { |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 5852 | } |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5853 | |
| 5854 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ |
| 5855 | |
Jiang Liu | 01cefae | 2012-12-12 13:52:19 -0800 | [diff] [blame] | 5856 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
| 5857 | unsigned long present_pages) |
| 5858 | { |
| 5859 | unsigned long pages = spanned_pages; |
| 5860 | |
| 5861 | /* |
| 5862 | * Provide a more accurate estimation if there are holes within |
| 5863 | * the zone and SPARSEMEM is in use. If there are holes within the |
| 5864 | * zone, each populated memory region may cost us one or two extra |
| 5865 | * memmap pages due to alignment because memmap pages for each |
| 5866 | * populated regions may not naturally algined on page boundary. |
| 5867 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
| 5868 | */ |
| 5869 | if (spanned_pages > present_pages + (present_pages >> 4) && |
| 5870 | IS_ENABLED(CONFIG_SPARSEMEM)) |
| 5871 | pages = present_pages; |
| 5872 | |
| 5873 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; |
| 5874 | } |
| 5875 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5876 | /* |
| 5877 | * Set up the zone data structures: |
| 5878 | * - mark all pages reserved |
| 5879 | * - mark all memory queues empty |
| 5880 | * - clear the memory bitmaps |
Minchan Kim | 6527af5 | 2012-07-31 16:46:16 -0700 | [diff] [blame] | 5881 | * |
| 5882 | * NOTE: pgdat should get zeroed by caller. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5883 | */ |
Wei Yang | 7f3eb55 | 2015-09-08 14:59:50 -0700 | [diff] [blame] | 5884 | static void __paginginit free_area_init_core(struct pglist_data *pgdat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5885 | { |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 5886 | enum zone_type j; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5887 | int nid = pgdat->node_id; |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5888 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5889 | |
Dave Hansen | 208d54e | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 5890 | pgdat_resize_init(pgdat); |
Andrea Arcangeli | 8177a42 | 2012-03-23 20:56:34 +0100 | [diff] [blame] | 5891 | #ifdef CONFIG_NUMA_BALANCING |
| 5892 | spin_lock_init(&pgdat->numabalancing_migrate_lock); |
| 5893 | pgdat->numabalancing_migrate_nr_pages = 0; |
| 5894 | pgdat->numabalancing_migrate_next_window = jiffies; |
| 5895 | #endif |
Kirill A. Shutemov | a3d0a918 | 2016-02-02 16:57:08 -0800 | [diff] [blame] | 5896 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 5897 | spin_lock_init(&pgdat->split_queue_lock); |
| 5898 | INIT_LIST_HEAD(&pgdat->split_queue); |
| 5899 | pgdat->split_queue_len = 0; |
| 5900 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5901 | init_waitqueue_head(&pgdat->kswapd_wait); |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 5902 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
Vlastimil Babka | 698b1b3 | 2016-03-17 14:18:08 -0700 | [diff] [blame] | 5903 | #ifdef CONFIG_COMPACTION |
| 5904 | init_waitqueue_head(&pgdat->kcompactd_wait); |
| 5905 | #endif |
Joonsoo Kim | eefa864 | 2014-12-12 16:55:46 -0800 | [diff] [blame] | 5906 | pgdat_page_ext_init(pgdat); |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame^] | 5907 | spin_lock_init(&pgdat->lru_lock); |
Michal Nazarewicz | 5f63b72 | 2012-01-11 15:16:11 +0100 | [diff] [blame] | 5908 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5909 | for (j = 0; j < MAX_NR_ZONES; j++) { |
| 5910 | struct zone *zone = pgdat->node_zones + j; |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 5911 | unsigned long size, realsize, freesize, memmap_pages; |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5912 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5913 | |
Gu Zheng | febd594 | 2015-06-24 16:57:02 -0700 | [diff] [blame] | 5914 | size = zone->spanned_pages; |
| 5915 | realsize = freesize = zone->present_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5916 | |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 5917 | /* |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 5918 | * Adjust freesize so that it accounts for how much memory |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 5919 | * is used by this zone for memmap. This affects the watermark |
| 5920 | * and per-cpu initialisations |
| 5921 | */ |
Jiang Liu | 01cefae | 2012-12-12 13:52:19 -0800 | [diff] [blame] | 5922 | memmap_pages = calc_memmap_size(size, realsize); |
Zhong Hongbo | ba914f4 | 2014-12-12 16:56:21 -0800 | [diff] [blame] | 5923 | if (!is_highmem_idx(j)) { |
| 5924 | if (freesize >= memmap_pages) { |
| 5925 | freesize -= memmap_pages; |
| 5926 | if (memmap_pages) |
| 5927 | printk(KERN_DEBUG |
| 5928 | " %s zone: %lu pages used for memmap\n", |
| 5929 | zone_names[j], memmap_pages); |
| 5930 | } else |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 5931 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
Zhong Hongbo | ba914f4 | 2014-12-12 16:56:21 -0800 | [diff] [blame] | 5932 | zone_names[j], memmap_pages, freesize); |
| 5933 | } |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 5934 | |
Christoph Lameter | 6267276 | 2007-02-10 01:43:07 -0800 | [diff] [blame] | 5935 | /* Account for reserved pages */ |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 5936 | if (j == 0 && freesize > dma_reserve) { |
| 5937 | freesize -= dma_reserve; |
Yinghai Lu | d903ef9 | 2008-10-18 20:27:06 -0700 | [diff] [blame] | 5938 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
Christoph Lameter | 6267276 | 2007-02-10 01:43:07 -0800 | [diff] [blame] | 5939 | zone_names[0], dma_reserve); |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 5940 | } |
| 5941 | |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 5942 | if (!is_highmem_idx(j)) |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 5943 | nr_kernel_pages += freesize; |
Jiang Liu | 01cefae | 2012-12-12 13:52:19 -0800 | [diff] [blame] | 5944 | /* Charge for highmem memmap if there are enough kernel pages */ |
| 5945 | else if (nr_kernel_pages > memmap_pages * 2) |
| 5946 | nr_kernel_pages -= memmap_pages; |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 5947 | nr_all_pages += freesize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5948 | |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 5949 | /* |
| 5950 | * Set an approximate value for lowmem here, it will be adjusted |
| 5951 | * when the bootmem allocator frees pages into the buddy system. |
| 5952 | * And all highmem pages will be managed by the buddy system. |
| 5953 | */ |
| 5954 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 5955 | #ifdef CONFIG_NUMA |
Christoph Lameter | d5f541e | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 5956 | zone->node = nid; |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 5957 | zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio) |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 5958 | / 100; |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 5959 | zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100; |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 5960 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5961 | zone->name = zone_names[j]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5962 | zone->zone_pgdat = pgdat; |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame^] | 5963 | spin_lock_init(&zone->lock); |
| 5964 | zone_seqlock_init(zone); |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5965 | zone_pcp_init(zone); |
Johannes Weiner | 81c0a2b | 2013-09-11 14:20:47 -0700 | [diff] [blame] | 5966 | |
| 5967 | /* For bootup, initialized properly in watermark setup */ |
| 5968 | mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages); |
| 5969 | |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 5970 | lruvec_init(&zone->lruvec); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5971 | if (!size) |
| 5972 | continue; |
| 5973 | |
Andrew Morton | 955c1cd | 2012-05-29 15:06:31 -0700 | [diff] [blame] | 5974 | set_pageblock_order(); |
Linus Torvalds | 7c45512 | 2013-02-18 09:58:02 -0800 | [diff] [blame] | 5975 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
Yaowei Bai | b171e40 | 2015-11-05 18:47:06 -0800 | [diff] [blame] | 5976 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5977 | BUG_ON(ret); |
Heiko Carstens | 76cdd58 | 2008-05-14 16:05:52 -0700 | [diff] [blame] | 5978 | memmap_init(size, nid, j, zone_start_pfn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5979 | } |
| 5980 | } |
| 5981 | |
Sam Ravnborg | 577a32f | 2007-05-17 23:29:25 +0200 | [diff] [blame] | 5982 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5983 | { |
Tony Luck | b0aeba7 | 2015-11-10 10:09:47 -0800 | [diff] [blame] | 5984 | unsigned long __maybe_unused start = 0; |
Laura Abbott | a1c34a3 | 2015-11-05 18:48:46 -0800 | [diff] [blame] | 5985 | unsigned long __maybe_unused offset = 0; |
| 5986 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5987 | /* Skip empty nodes */ |
| 5988 | if (!pgdat->node_spanned_pages) |
| 5989 | return; |
| 5990 | |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 5991 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
Tony Luck | b0aeba7 | 2015-11-10 10:09:47 -0800 | [diff] [blame] | 5992 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
| 5993 | offset = pgdat->node_start_pfn - start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5994 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
| 5995 | if (!pgdat->node_mem_map) { |
Tony Luck | b0aeba7 | 2015-11-10 10:09:47 -0800 | [diff] [blame] | 5996 | unsigned long size, end; |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 5997 | struct page *map; |
| 5998 | |
Bob Picco | e984bb4 | 2006-05-20 15:00:31 -0700 | [diff] [blame] | 5999 | /* |
| 6000 | * The zone's endpoints aren't required to be MAX_ORDER |
| 6001 | * aligned but the node_mem_map endpoints must be in order |
| 6002 | * for the buddy allocator to function correctly. |
| 6003 | */ |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 6004 | end = pgdat_end_pfn(pgdat); |
Bob Picco | e984bb4 | 2006-05-20 15:00:31 -0700 | [diff] [blame] | 6005 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
| 6006 | size = (end - start) * sizeof(struct page); |
Dave Hansen | 6f167ec | 2005-06-23 00:07:39 -0700 | [diff] [blame] | 6007 | map = alloc_remap(pgdat->node_id, size); |
| 6008 | if (!map) |
Santosh Shilimkar | 6782832 | 2014-01-21 15:50:25 -0800 | [diff] [blame] | 6009 | map = memblock_virt_alloc_node_nopanic(size, |
| 6010 | pgdat->node_id); |
Laura Abbott | a1c34a3 | 2015-11-05 18:48:46 -0800 | [diff] [blame] | 6011 | pgdat->node_mem_map = map + offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6012 | } |
Roman Zippel | 12d810c | 2007-05-31 00:40:54 -0700 | [diff] [blame] | 6013 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6014 | /* |
| 6015 | * With no DISCONTIG, the global mem_map is just set as node 0's |
| 6016 | */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6017 | if (pgdat == NODE_DATA(0)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6018 | mem_map = NODE_DATA(0)->node_mem_map; |
Laura Abbott | a1c34a3 | 2015-11-05 18:48:46 -0800 | [diff] [blame] | 6019 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6020 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
Laura Abbott | a1c34a3 | 2015-11-05 18:48:46 -0800 | [diff] [blame] | 6021 | mem_map -= offset; |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 6022 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6023 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6024 | #endif |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 6025 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6026 | } |
| 6027 | |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 6028 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
| 6029 | unsigned long node_start_pfn, unsigned long *zholes_size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6030 | { |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 6031 | pg_data_t *pgdat = NODE_DATA(nid); |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 6032 | unsigned long start_pfn = 0; |
| 6033 | unsigned long end_pfn = 0; |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 6034 | |
Minchan Kim | 88fdf75 | 2012-07-31 16:46:14 -0700 | [diff] [blame] | 6035 | /* pg_data_t should be reset to zero when it's allocated */ |
Linus Torvalds | 8783b6e | 2012-08-02 10:37:03 -0700 | [diff] [blame] | 6036 | WARN_ON(pgdat->nr_zones || pgdat->classzone_idx); |
Minchan Kim | 88fdf75 | 2012-07-31 16:46:14 -0700 | [diff] [blame] | 6037 | |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 6038 | reset_deferred_meminit(pgdat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6039 | pgdat->node_id = nid; |
| 6040 | pgdat->node_start_pfn = node_start_pfn; |
Mel Gorman | 75ef718 | 2016-07-28 15:45:24 -0700 | [diff] [blame] | 6041 | pgdat->per_cpu_nodestats = NULL; |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 6042 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
| 6043 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
Juergen Gross | 8d29e18 | 2015-02-11 15:26:01 -0800 | [diff] [blame] | 6044 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
Zhen Lei | 4ada0c5 | 2015-09-08 15:04:19 -0700 | [diff] [blame] | 6045 | (u64)start_pfn << PAGE_SHIFT, |
| 6046 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 6047 | #else |
| 6048 | start_pfn = node_start_pfn; |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 6049 | #endif |
| 6050 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, |
| 6051 | zones_size, zholes_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6052 | |
| 6053 | alloc_node_mem_map(pgdat); |
Yinghai Lu | e8c27ac | 2008-06-01 13:15:22 -0700 | [diff] [blame] | 6054 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
| 6055 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", |
| 6056 | nid, (unsigned long)pgdat, |
| 6057 | (unsigned long)pgdat->node_mem_map); |
| 6058 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6059 | |
Wei Yang | 7f3eb55 | 2015-09-08 14:59:50 -0700 | [diff] [blame] | 6060 | free_area_init_core(pgdat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6061 | } |
| 6062 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 6063 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 6064 | |
| 6065 | #if MAX_NUMNODES > 1 |
| 6066 | /* |
| 6067 | * Figure out the number of possible node ids. |
| 6068 | */ |
Cody P Schafer | f9872ca | 2013-04-29 15:08:01 -0700 | [diff] [blame] | 6069 | void __init setup_nr_node_ids(void) |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 6070 | { |
Wei Yang | 904a955 | 2015-09-08 14:59:48 -0700 | [diff] [blame] | 6071 | unsigned int highest; |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 6072 | |
Wei Yang | 904a955 | 2015-09-08 14:59:48 -0700 | [diff] [blame] | 6073 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 6074 | nr_node_ids = highest + 1; |
| 6075 | } |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 6076 | #endif |
| 6077 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6078 | /** |
Tejun Heo | 1e01979 | 2011-07-12 09:45:34 +0200 | [diff] [blame] | 6079 | * node_map_pfn_alignment - determine the maximum internode alignment |
| 6080 | * |
| 6081 | * This function should be called after node map is populated and sorted. |
| 6082 | * It calculates the maximum power of two alignment which can distinguish |
| 6083 | * all the nodes. |
| 6084 | * |
| 6085 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value |
| 6086 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the |
| 6087 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is |
| 6088 | * shifted, 1GiB is enough and this function will indicate so. |
| 6089 | * |
| 6090 | * This is used to test whether pfn -> nid mapping of the chosen memory |
| 6091 | * model has fine enough granularity to avoid incorrect mapping for the |
| 6092 | * populated node map. |
| 6093 | * |
| 6094 | * Returns the determined alignment in pfn's. 0 if there is no alignment |
| 6095 | * requirement (single node). |
| 6096 | */ |
| 6097 | unsigned long __init node_map_pfn_alignment(void) |
| 6098 | { |
| 6099 | unsigned long accl_mask = 0, last_end = 0; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6100 | unsigned long start, end, mask; |
Tejun Heo | 1e01979 | 2011-07-12 09:45:34 +0200 | [diff] [blame] | 6101 | int last_nid = -1; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6102 | int i, nid; |
Tejun Heo | 1e01979 | 2011-07-12 09:45:34 +0200 | [diff] [blame] | 6103 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6104 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
Tejun Heo | 1e01979 | 2011-07-12 09:45:34 +0200 | [diff] [blame] | 6105 | if (!start || last_nid < 0 || last_nid == nid) { |
| 6106 | last_nid = nid; |
| 6107 | last_end = end; |
| 6108 | continue; |
| 6109 | } |
| 6110 | |
| 6111 | /* |
| 6112 | * Start with a mask granular enough to pin-point to the |
| 6113 | * start pfn and tick off bits one-by-one until it becomes |
| 6114 | * too coarse to separate the current node from the last. |
| 6115 | */ |
| 6116 | mask = ~((1 << __ffs(start)) - 1); |
| 6117 | while (mask && last_end <= (start & (mask << 1))) |
| 6118 | mask <<= 1; |
| 6119 | |
| 6120 | /* accumulate all internode masks */ |
| 6121 | accl_mask |= mask; |
| 6122 | } |
| 6123 | |
| 6124 | /* convert mask to number of pages */ |
| 6125 | return ~accl_mask + 1; |
| 6126 | } |
| 6127 | |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 6128 | /* Find the lowest pfn for a node */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 6129 | static unsigned long __init find_min_pfn_for_node(int nid) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6130 | { |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 6131 | unsigned long min_pfn = ULONG_MAX; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6132 | unsigned long start_pfn; |
| 6133 | int i; |
Mel Gorman | 1abbfb4 | 2006-11-23 12:01:41 +0000 | [diff] [blame] | 6134 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6135 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
| 6136 | min_pfn = min(min_pfn, start_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6137 | |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 6138 | if (min_pfn == ULONG_MAX) { |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 6139 | pr_warn("Could not find start_pfn for node %d\n", nid); |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 6140 | return 0; |
| 6141 | } |
| 6142 | |
| 6143 | return min_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6144 | } |
| 6145 | |
| 6146 | /** |
| 6147 | * find_min_pfn_with_active_regions - Find the minimum PFN registered |
| 6148 | * |
| 6149 | * It returns the minimum PFN based on information provided via |
Zhang Zhen | 7d01817 | 2014-06-04 16:10:53 -0700 | [diff] [blame] | 6150 | * memblock_set_node(). |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6151 | */ |
| 6152 | unsigned long __init find_min_pfn_with_active_regions(void) |
| 6153 | { |
| 6154 | return find_min_pfn_for_node(MAX_NUMNODES); |
| 6155 | } |
| 6156 | |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6157 | /* |
| 6158 | * early_calculate_totalpages() |
| 6159 | * Sum pages in active regions for movable zone. |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6160 | * Populate N_MEMORY for calculating usable_nodes. |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6161 | */ |
Adrian Bunk | 484f51f | 2007-10-16 01:26:03 -0700 | [diff] [blame] | 6162 | static unsigned long __init early_calculate_totalpages(void) |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6163 | { |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6164 | unsigned long totalpages = 0; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6165 | unsigned long start_pfn, end_pfn; |
| 6166 | int i, nid; |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6167 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6168 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
| 6169 | unsigned long pages = end_pfn - start_pfn; |
| 6170 | |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6171 | totalpages += pages; |
| 6172 | if (pages) |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6173 | node_set_state(nid, N_MEMORY); |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6174 | } |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 6175 | return totalpages; |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6176 | } |
| 6177 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6178 | /* |
| 6179 | * Find the PFN the Movable zone begins in each node. Kernel memory |
| 6180 | * is spread evenly between nodes as long as the nodes have enough |
| 6181 | * memory. When they don't, some nodes will have more kernelcore than |
| 6182 | * others |
| 6183 | */ |
Kautuk Consul | b224ef8 | 2012-03-21 16:34:15 -0700 | [diff] [blame] | 6184 | static void __init find_zone_movable_pfns_for_nodes(void) |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6185 | { |
| 6186 | int i, nid; |
| 6187 | unsigned long usable_startpfn; |
| 6188 | unsigned long kernelcore_node, kernelcore_remaining; |
Yinghai Lu | 66918dc | 2009-06-30 11:41:37 -0700 | [diff] [blame] | 6189 | /* save the state before borrow the nodemask */ |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6190 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6191 | unsigned long totalpages = early_calculate_totalpages(); |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6192 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
Emil Medve | 136199f | 2014-04-07 15:37:52 -0700 | [diff] [blame] | 6193 | struct memblock_region *r; |
Tang Chen | b2f3eeb | 2014-01-21 15:49:38 -0800 | [diff] [blame] | 6194 | |
| 6195 | /* Need to find movable_zone earlier when movable_node is specified. */ |
| 6196 | find_usable_zone_for_movable(); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6197 | |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6198 | /* |
Tang Chen | b2f3eeb | 2014-01-21 15:49:38 -0800 | [diff] [blame] | 6199 | * If movable_node is specified, ignore kernelcore and movablecore |
| 6200 | * options. |
| 6201 | */ |
| 6202 | if (movable_node_is_enabled()) { |
Emil Medve | 136199f | 2014-04-07 15:37:52 -0700 | [diff] [blame] | 6203 | for_each_memblock(memory, r) { |
| 6204 | if (!memblock_is_hotpluggable(r)) |
Tang Chen | b2f3eeb | 2014-01-21 15:49:38 -0800 | [diff] [blame] | 6205 | continue; |
| 6206 | |
Emil Medve | 136199f | 2014-04-07 15:37:52 -0700 | [diff] [blame] | 6207 | nid = r->nid; |
Tang Chen | b2f3eeb | 2014-01-21 15:49:38 -0800 | [diff] [blame] | 6208 | |
Emil Medve | 136199f | 2014-04-07 15:37:52 -0700 | [diff] [blame] | 6209 | usable_startpfn = PFN_DOWN(r->base); |
Tang Chen | b2f3eeb | 2014-01-21 15:49:38 -0800 | [diff] [blame] | 6210 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
| 6211 | min(usable_startpfn, zone_movable_pfn[nid]) : |
| 6212 | usable_startpfn; |
| 6213 | } |
| 6214 | |
| 6215 | goto out2; |
| 6216 | } |
| 6217 | |
| 6218 | /* |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 6219 | * If kernelcore=mirror is specified, ignore movablecore option |
| 6220 | */ |
| 6221 | if (mirrored_kernelcore) { |
| 6222 | bool mem_below_4gb_not_mirrored = false; |
| 6223 | |
| 6224 | for_each_memblock(memory, r) { |
| 6225 | if (memblock_is_mirror(r)) |
| 6226 | continue; |
| 6227 | |
| 6228 | nid = r->nid; |
| 6229 | |
| 6230 | usable_startpfn = memblock_region_memory_base_pfn(r); |
| 6231 | |
| 6232 | if (usable_startpfn < 0x100000) { |
| 6233 | mem_below_4gb_not_mirrored = true; |
| 6234 | continue; |
| 6235 | } |
| 6236 | |
| 6237 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
| 6238 | min(usable_startpfn, zone_movable_pfn[nid]) : |
| 6239 | usable_startpfn; |
| 6240 | } |
| 6241 | |
| 6242 | if (mem_below_4gb_not_mirrored) |
| 6243 | pr_warn("This configuration results in unmirrored kernel memory."); |
| 6244 | |
| 6245 | goto out2; |
| 6246 | } |
| 6247 | |
| 6248 | /* |
Tang Chen | b2f3eeb | 2014-01-21 15:49:38 -0800 | [diff] [blame] | 6249 | * If movablecore=nn[KMG] was specified, calculate what size of |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6250 | * kernelcore that corresponds so that memory usable for |
| 6251 | * any allocation type is evenly spread. If both kernelcore |
| 6252 | * and movablecore are specified, then the value of kernelcore |
| 6253 | * will be used for required_kernelcore if it's greater than |
| 6254 | * what movablecore would have allowed. |
| 6255 | */ |
| 6256 | if (required_movablecore) { |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6257 | unsigned long corepages; |
| 6258 | |
| 6259 | /* |
| 6260 | * Round-up so that ZONE_MOVABLE is at least as large as what |
| 6261 | * was requested by the user |
| 6262 | */ |
| 6263 | required_movablecore = |
| 6264 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); |
Xishi Qiu | 9fd745d | 2015-11-05 18:48:11 -0800 | [diff] [blame] | 6265 | required_movablecore = min(totalpages, required_movablecore); |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6266 | corepages = totalpages - required_movablecore; |
| 6267 | |
| 6268 | required_kernelcore = max(required_kernelcore, corepages); |
| 6269 | } |
| 6270 | |
Xishi Qiu | bde304b | 2015-11-05 18:48:56 -0800 | [diff] [blame] | 6271 | /* |
| 6272 | * If kernelcore was not specified or kernelcore size is larger |
| 6273 | * than totalpages, there is no ZONE_MOVABLE. |
| 6274 | */ |
| 6275 | if (!required_kernelcore || required_kernelcore >= totalpages) |
Yinghai Lu | 66918dc | 2009-06-30 11:41:37 -0700 | [diff] [blame] | 6276 | goto out; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6277 | |
| 6278 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6279 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
| 6280 | |
| 6281 | restart: |
| 6282 | /* Spread kernelcore memory as evenly as possible throughout nodes */ |
| 6283 | kernelcore_node = required_kernelcore / usable_nodes; |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6284 | for_each_node_state(nid, N_MEMORY) { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6285 | unsigned long start_pfn, end_pfn; |
| 6286 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6287 | /* |
| 6288 | * Recalculate kernelcore_node if the division per node |
| 6289 | * now exceeds what is necessary to satisfy the requested |
| 6290 | * amount of memory for the kernel |
| 6291 | */ |
| 6292 | if (required_kernelcore < kernelcore_node) |
| 6293 | kernelcore_node = required_kernelcore / usable_nodes; |
| 6294 | |
| 6295 | /* |
| 6296 | * As the map is walked, we track how much memory is usable |
| 6297 | * by the kernel using kernelcore_remaining. When it is |
| 6298 | * 0, the rest of the node is usable by ZONE_MOVABLE |
| 6299 | */ |
| 6300 | kernelcore_remaining = kernelcore_node; |
| 6301 | |
| 6302 | /* Go through each range of PFNs within this node */ |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6303 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6304 | unsigned long size_pages; |
| 6305 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6306 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6307 | if (start_pfn >= end_pfn) |
| 6308 | continue; |
| 6309 | |
| 6310 | /* Account for what is only usable for kernelcore */ |
| 6311 | if (start_pfn < usable_startpfn) { |
| 6312 | unsigned long kernel_pages; |
| 6313 | kernel_pages = min(end_pfn, usable_startpfn) |
| 6314 | - start_pfn; |
| 6315 | |
| 6316 | kernelcore_remaining -= min(kernel_pages, |
| 6317 | kernelcore_remaining); |
| 6318 | required_kernelcore -= min(kernel_pages, |
| 6319 | required_kernelcore); |
| 6320 | |
| 6321 | /* Continue if range is now fully accounted */ |
| 6322 | if (end_pfn <= usable_startpfn) { |
| 6323 | |
| 6324 | /* |
| 6325 | * Push zone_movable_pfn to the end so |
| 6326 | * that if we have to rebalance |
| 6327 | * kernelcore across nodes, we will |
| 6328 | * not double account here |
| 6329 | */ |
| 6330 | zone_movable_pfn[nid] = end_pfn; |
| 6331 | continue; |
| 6332 | } |
| 6333 | start_pfn = usable_startpfn; |
| 6334 | } |
| 6335 | |
| 6336 | /* |
| 6337 | * The usable PFN range for ZONE_MOVABLE is from |
| 6338 | * start_pfn->end_pfn. Calculate size_pages as the |
| 6339 | * number of pages used as kernelcore |
| 6340 | */ |
| 6341 | size_pages = end_pfn - start_pfn; |
| 6342 | if (size_pages > kernelcore_remaining) |
| 6343 | size_pages = kernelcore_remaining; |
| 6344 | zone_movable_pfn[nid] = start_pfn + size_pages; |
| 6345 | |
| 6346 | /* |
| 6347 | * Some kernelcore has been met, update counts and |
| 6348 | * break if the kernelcore for this node has been |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 6349 | * satisfied |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6350 | */ |
| 6351 | required_kernelcore -= min(required_kernelcore, |
| 6352 | size_pages); |
| 6353 | kernelcore_remaining -= size_pages; |
| 6354 | if (!kernelcore_remaining) |
| 6355 | break; |
| 6356 | } |
| 6357 | } |
| 6358 | |
| 6359 | /* |
| 6360 | * If there is still required_kernelcore, we do another pass with one |
| 6361 | * less node in the count. This will push zone_movable_pfn[nid] further |
| 6362 | * along on the nodes that still have memory until kernelcore is |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 6363 | * satisfied |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6364 | */ |
| 6365 | usable_nodes--; |
| 6366 | if (usable_nodes && required_kernelcore > usable_nodes) |
| 6367 | goto restart; |
| 6368 | |
Tang Chen | b2f3eeb | 2014-01-21 15:49:38 -0800 | [diff] [blame] | 6369 | out2: |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6370 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
| 6371 | for (nid = 0; nid < MAX_NUMNODES; nid++) |
| 6372 | zone_movable_pfn[nid] = |
| 6373 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); |
Yinghai Lu | 66918dc | 2009-06-30 11:41:37 -0700 | [diff] [blame] | 6374 | |
Yinghai Lu | 20e6926 | 2013-03-01 14:51:27 -0800 | [diff] [blame] | 6375 | out: |
Yinghai Lu | 66918dc | 2009-06-30 11:41:37 -0700 | [diff] [blame] | 6376 | /* restore the node_state */ |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6377 | node_states[N_MEMORY] = saved_node_state; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6378 | } |
| 6379 | |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6380 | /* Any regular or high memory on that node ? */ |
| 6381 | static void check_for_memory(pg_data_t *pgdat, int nid) |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6382 | { |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6383 | enum zone_type zone_type; |
| 6384 | |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6385 | if (N_MEMORY == N_NORMAL_MEMORY) |
| 6386 | return; |
| 6387 | |
| 6388 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6389 | struct zone *zone = &pgdat->node_zones[zone_type]; |
Xishi Qiu | b38a872 | 2013-11-12 15:07:20 -0800 | [diff] [blame] | 6390 | if (populated_zone(zone)) { |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6391 | node_set_state(nid, N_HIGH_MEMORY); |
| 6392 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && |
| 6393 | zone_type <= ZONE_NORMAL) |
| 6394 | node_set_state(nid, N_NORMAL_MEMORY); |
Bob Liu | d0048b0 | 2012-01-12 17:19:07 -0800 | [diff] [blame] | 6395 | break; |
| 6396 | } |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6397 | } |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6398 | } |
| 6399 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6400 | /** |
| 6401 | * free_area_init_nodes - Initialise all pg_data_t and zone data |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 6402 | * @max_zone_pfn: an array of max PFNs for each zone |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6403 | * |
| 6404 | * This will call free_area_init_node() for each active node in the system. |
Zhang Zhen | 7d01817 | 2014-06-04 16:10:53 -0700 | [diff] [blame] | 6405 | * Using the page ranges provided by memblock_set_node(), the size of each |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6406 | * zone in each node and their holes is calculated. If the maximum PFN |
| 6407 | * between two adjacent zones match, it is assumed that the zone is empty. |
| 6408 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed |
| 6409 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone |
| 6410 | * starts where the previous one ended. For example, ZONE_DMA32 starts |
| 6411 | * at arch_max_dma_pfn. |
| 6412 | */ |
| 6413 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) |
| 6414 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6415 | unsigned long start_pfn, end_pfn; |
| 6416 | int i, nid; |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 6417 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6418 | /* Record where the zone boundaries are */ |
| 6419 | memset(arch_zone_lowest_possible_pfn, 0, |
| 6420 | sizeof(arch_zone_lowest_possible_pfn)); |
| 6421 | memset(arch_zone_highest_possible_pfn, 0, |
| 6422 | sizeof(arch_zone_highest_possible_pfn)); |
Oliver O'Halloran | 90cae1f | 2016-07-26 15:22:17 -0700 | [diff] [blame] | 6423 | |
| 6424 | start_pfn = find_min_pfn_with_active_regions(); |
| 6425 | |
| 6426 | for (i = 0; i < MAX_NR_ZONES; i++) { |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6427 | if (i == ZONE_MOVABLE) |
| 6428 | continue; |
Oliver O'Halloran | 90cae1f | 2016-07-26 15:22:17 -0700 | [diff] [blame] | 6429 | |
| 6430 | end_pfn = max(max_zone_pfn[i], start_pfn); |
| 6431 | arch_zone_lowest_possible_pfn[i] = start_pfn; |
| 6432 | arch_zone_highest_possible_pfn[i] = end_pfn; |
| 6433 | |
| 6434 | start_pfn = end_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6435 | } |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6436 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
| 6437 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; |
| 6438 | |
| 6439 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ |
| 6440 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); |
Kautuk Consul | b224ef8 | 2012-03-21 16:34:15 -0700 | [diff] [blame] | 6441 | find_zone_movable_pfns_for_nodes(); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6442 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6443 | /* Print out the zone ranges */ |
Anton Blanchard | f88dfff | 2014-12-10 15:42:53 -0800 | [diff] [blame] | 6444 | pr_info("Zone ranges:\n"); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6445 | for (i = 0; i < MAX_NR_ZONES; i++) { |
| 6446 | if (i == ZONE_MOVABLE) |
| 6447 | continue; |
Anton Blanchard | f88dfff | 2014-12-10 15:42:53 -0800 | [diff] [blame] | 6448 | pr_info(" %-8s ", zone_names[i]); |
David Rientjes | 72f0ba0 | 2010-03-05 13:42:14 -0800 | [diff] [blame] | 6449 | if (arch_zone_lowest_possible_pfn[i] == |
| 6450 | arch_zone_highest_possible_pfn[i]) |
Anton Blanchard | f88dfff | 2014-12-10 15:42:53 -0800 | [diff] [blame] | 6451 | pr_cont("empty\n"); |
David Rientjes | 72f0ba0 | 2010-03-05 13:42:14 -0800 | [diff] [blame] | 6452 | else |
Juergen Gross | 8d29e18 | 2015-02-11 15:26:01 -0800 | [diff] [blame] | 6453 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
| 6454 | (u64)arch_zone_lowest_possible_pfn[i] |
| 6455 | << PAGE_SHIFT, |
| 6456 | ((u64)arch_zone_highest_possible_pfn[i] |
Bjorn Helgaas | a62e2f4 | 2012-05-29 15:06:30 -0700 | [diff] [blame] | 6457 | << PAGE_SHIFT) - 1); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6458 | } |
| 6459 | |
| 6460 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ |
Anton Blanchard | f88dfff | 2014-12-10 15:42:53 -0800 | [diff] [blame] | 6461 | pr_info("Movable zone start for each node\n"); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6462 | for (i = 0; i < MAX_NUMNODES; i++) { |
| 6463 | if (zone_movable_pfn[i]) |
Juergen Gross | 8d29e18 | 2015-02-11 15:26:01 -0800 | [diff] [blame] | 6464 | pr_info(" Node %d: %#018Lx\n", i, |
| 6465 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6466 | } |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6467 | |
Wanpeng Li | f2d52fe | 2012-10-08 16:32:24 -0700 | [diff] [blame] | 6468 | /* Print out the early node map */ |
Anton Blanchard | f88dfff | 2014-12-10 15:42:53 -0800 | [diff] [blame] | 6469 | pr_info("Early memory node ranges\n"); |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6470 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
Juergen Gross | 8d29e18 | 2015-02-11 15:26:01 -0800 | [diff] [blame] | 6471 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
| 6472 | (u64)start_pfn << PAGE_SHIFT, |
| 6473 | ((u64)end_pfn << PAGE_SHIFT) - 1); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6474 | |
| 6475 | /* Initialise every node */ |
Mel Gorman | 708614e | 2008-07-23 21:26:51 -0700 | [diff] [blame] | 6476 | mminit_verify_pageflags_layout(); |
Christoph Lameter | 8ef8286 | 2007-02-20 13:57:52 -0800 | [diff] [blame] | 6477 | setup_nr_node_ids(); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6478 | for_each_online_node(nid) { |
| 6479 | pg_data_t *pgdat = NODE_DATA(nid); |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 6480 | free_area_init_node(nid, NULL, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6481 | find_min_pfn_for_node(nid), NULL); |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6482 | |
| 6483 | /* Any memory on that node */ |
| 6484 | if (pgdat->node_present_pages) |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6485 | node_set_state(nid, N_MEMORY); |
| 6486 | check_for_memory(pgdat, nid); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6487 | } |
| 6488 | } |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6489 | |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6490 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6491 | { |
| 6492 | unsigned long long coremem; |
| 6493 | if (!p) |
| 6494 | return -EINVAL; |
| 6495 | |
| 6496 | coremem = memparse(p, &p); |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6497 | *core = coremem >> PAGE_SHIFT; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6498 | |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6499 | /* Paranoid check that UL is enough for the coremem value */ |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6500 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
| 6501 | |
| 6502 | return 0; |
| 6503 | } |
Mel Gorman | ed7ed36 | 2007-07-17 04:03:14 -0700 | [diff] [blame] | 6504 | |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6505 | /* |
| 6506 | * kernelcore=size sets the amount of memory for use for allocations that |
| 6507 | * cannot be reclaimed or migrated. |
| 6508 | */ |
| 6509 | static int __init cmdline_parse_kernelcore(char *p) |
| 6510 | { |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 6511 | /* parse kernelcore=mirror */ |
| 6512 | if (parse_option_str(p, "mirror")) { |
| 6513 | mirrored_kernelcore = true; |
| 6514 | return 0; |
| 6515 | } |
| 6516 | |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6517 | return cmdline_parse_core(p, &required_kernelcore); |
| 6518 | } |
| 6519 | |
| 6520 | /* |
| 6521 | * movablecore=size sets the amount of memory for use for allocations that |
| 6522 | * can be reclaimed or migrated. |
| 6523 | */ |
| 6524 | static int __init cmdline_parse_movablecore(char *p) |
| 6525 | { |
| 6526 | return cmdline_parse_core(p, &required_movablecore); |
| 6527 | } |
| 6528 | |
Mel Gorman | ed7ed36 | 2007-07-17 04:03:14 -0700 | [diff] [blame] | 6529 | early_param("kernelcore", cmdline_parse_kernelcore); |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6530 | early_param("movablecore", cmdline_parse_movablecore); |
Mel Gorman | ed7ed36 | 2007-07-17 04:03:14 -0700 | [diff] [blame] | 6531 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 6532 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6533 | |
Jiang Liu | c3d5f5f | 2013-07-03 15:03:14 -0700 | [diff] [blame] | 6534 | void adjust_managed_page_count(struct page *page, long count) |
| 6535 | { |
| 6536 | spin_lock(&managed_page_count_lock); |
| 6537 | page_zone(page)->managed_pages += count; |
| 6538 | totalram_pages += count; |
Jiang Liu | 3dcc057 | 2013-07-03 15:03:21 -0700 | [diff] [blame] | 6539 | #ifdef CONFIG_HIGHMEM |
| 6540 | if (PageHighMem(page)) |
| 6541 | totalhigh_pages += count; |
| 6542 | #endif |
Jiang Liu | c3d5f5f | 2013-07-03 15:03:14 -0700 | [diff] [blame] | 6543 | spin_unlock(&managed_page_count_lock); |
| 6544 | } |
Jiang Liu | 3dcc057 | 2013-07-03 15:03:21 -0700 | [diff] [blame] | 6545 | EXPORT_SYMBOL(adjust_managed_page_count); |
Jiang Liu | c3d5f5f | 2013-07-03 15:03:14 -0700 | [diff] [blame] | 6546 | |
Jiang Liu | 1119969 | 2013-07-03 15:02:48 -0700 | [diff] [blame] | 6547 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
Jiang Liu | 69afade | 2013-04-29 15:06:21 -0700 | [diff] [blame] | 6548 | { |
Jiang Liu | 1119969 | 2013-07-03 15:02:48 -0700 | [diff] [blame] | 6549 | void *pos; |
| 6550 | unsigned long pages = 0; |
Jiang Liu | 69afade | 2013-04-29 15:06:21 -0700 | [diff] [blame] | 6551 | |
Jiang Liu | 1119969 | 2013-07-03 15:02:48 -0700 | [diff] [blame] | 6552 | start = (void *)PAGE_ALIGN((unsigned long)start); |
| 6553 | end = (void *)((unsigned long)end & PAGE_MASK); |
| 6554 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { |
Jiang Liu | dbe67df | 2013-07-03 15:02:51 -0700 | [diff] [blame] | 6555 | if ((unsigned int)poison <= 0xFF) |
Jiang Liu | 1119969 | 2013-07-03 15:02:48 -0700 | [diff] [blame] | 6556 | memset(pos, poison, PAGE_SIZE); |
| 6557 | free_reserved_page(virt_to_page(pos)); |
Jiang Liu | 69afade | 2013-04-29 15:06:21 -0700 | [diff] [blame] | 6558 | } |
| 6559 | |
| 6560 | if (pages && s) |
Jiang Liu | 1119969 | 2013-07-03 15:02:48 -0700 | [diff] [blame] | 6561 | pr_info("Freeing %s memory: %ldK (%p - %p)\n", |
Jiang Liu | 69afade | 2013-04-29 15:06:21 -0700 | [diff] [blame] | 6562 | s, pages << (PAGE_SHIFT - 10), start, end); |
| 6563 | |
| 6564 | return pages; |
| 6565 | } |
Jiang Liu | 1119969 | 2013-07-03 15:02:48 -0700 | [diff] [blame] | 6566 | EXPORT_SYMBOL(free_reserved_area); |
Jiang Liu | 69afade | 2013-04-29 15:06:21 -0700 | [diff] [blame] | 6567 | |
Jiang Liu | cfa11e0 | 2013-04-29 15:07:00 -0700 | [diff] [blame] | 6568 | #ifdef CONFIG_HIGHMEM |
| 6569 | void free_highmem_page(struct page *page) |
| 6570 | { |
| 6571 | __free_reserved_page(page); |
| 6572 | totalram_pages++; |
Jiang Liu | 7b4b2a0 | 2013-07-03 15:03:11 -0700 | [diff] [blame] | 6573 | page_zone(page)->managed_pages++; |
Jiang Liu | cfa11e0 | 2013-04-29 15:07:00 -0700 | [diff] [blame] | 6574 | totalhigh_pages++; |
| 6575 | } |
| 6576 | #endif |
| 6577 | |
Jiang Liu | 7ee3d4e | 2013-07-03 15:03:41 -0700 | [diff] [blame] | 6578 | |
| 6579 | void __init mem_init_print_info(const char *str) |
| 6580 | { |
| 6581 | unsigned long physpages, codesize, datasize, rosize, bss_size; |
| 6582 | unsigned long init_code_size, init_data_size; |
| 6583 | |
| 6584 | physpages = get_num_physpages(); |
| 6585 | codesize = _etext - _stext; |
| 6586 | datasize = _edata - _sdata; |
| 6587 | rosize = __end_rodata - __start_rodata; |
| 6588 | bss_size = __bss_stop - __bss_start; |
| 6589 | init_data_size = __init_end - __init_begin; |
| 6590 | init_code_size = _einittext - _sinittext; |
| 6591 | |
| 6592 | /* |
| 6593 | * Detect special cases and adjust section sizes accordingly: |
| 6594 | * 1) .init.* may be embedded into .data sections |
| 6595 | * 2) .init.text.* may be out of [__init_begin, __init_end], |
| 6596 | * please refer to arch/tile/kernel/vmlinux.lds.S. |
| 6597 | * 3) .rodata.* may be embedded into .text or .data sections. |
| 6598 | */ |
| 6599 | #define adj_init_size(start, end, size, pos, adj) \ |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 6600 | do { \ |
| 6601 | if (start <= pos && pos < end && size > adj) \ |
| 6602 | size -= adj; \ |
| 6603 | } while (0) |
Jiang Liu | 7ee3d4e | 2013-07-03 15:03:41 -0700 | [diff] [blame] | 6604 | |
| 6605 | adj_init_size(__init_begin, __init_end, init_data_size, |
| 6606 | _sinittext, init_code_size); |
| 6607 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); |
| 6608 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); |
| 6609 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); |
| 6610 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); |
| 6611 | |
| 6612 | #undef adj_init_size |
| 6613 | |
Joe Perches | 756a025 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 6614 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
Jiang Liu | 7ee3d4e | 2013-07-03 15:03:41 -0700 | [diff] [blame] | 6615 | #ifdef CONFIG_HIGHMEM |
Joe Perches | 756a025 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 6616 | ", %luK highmem" |
Jiang Liu | 7ee3d4e | 2013-07-03 15:03:41 -0700 | [diff] [blame] | 6617 | #endif |
Joe Perches | 756a025 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 6618 | "%s%s)\n", |
| 6619 | nr_free_pages() << (PAGE_SHIFT - 10), |
| 6620 | physpages << (PAGE_SHIFT - 10), |
| 6621 | codesize >> 10, datasize >> 10, rosize >> 10, |
| 6622 | (init_data_size + init_code_size) >> 10, bss_size >> 10, |
| 6623 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10), |
| 6624 | totalcma_pages << (PAGE_SHIFT - 10), |
Jiang Liu | 7ee3d4e | 2013-07-03 15:03:41 -0700 | [diff] [blame] | 6625 | #ifdef CONFIG_HIGHMEM |
Joe Perches | 756a025 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 6626 | totalhigh_pages << (PAGE_SHIFT - 10), |
Jiang Liu | 7ee3d4e | 2013-07-03 15:03:41 -0700 | [diff] [blame] | 6627 | #endif |
Joe Perches | 756a025 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 6628 | str ? ", " : "", str ? str : ""); |
Jiang Liu | 7ee3d4e | 2013-07-03 15:03:41 -0700 | [diff] [blame] | 6629 | } |
| 6630 | |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 6631 | /** |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 6632 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
| 6633 | * @new_dma_reserve: The number of pages to mark reserved |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 6634 | * |
Yaowei Bai | 013110a | 2015-09-08 15:04:10 -0700 | [diff] [blame] | 6635 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 6636 | * In the DMA zone, a significant percentage may be consumed by kernel image |
| 6637 | * and other unfreeable allocations which can skew the watermarks badly. This |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 6638 | * function may optionally be used to account for unfreeable pages in the |
| 6639 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and |
| 6640 | * smaller per-cpu batchsize. |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 6641 | */ |
| 6642 | void __init set_dma_reserve(unsigned long new_dma_reserve) |
| 6643 | { |
| 6644 | dma_reserve = new_dma_reserve; |
| 6645 | } |
| 6646 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6647 | void __init free_area_init(unsigned long *zones_size) |
| 6648 | { |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 6649 | free_area_init_node(0, zones_size, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6650 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
| 6651 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6652 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6653 | static int page_alloc_cpu_notify(struct notifier_block *self, |
| 6654 | unsigned long action, void *hcpu) |
| 6655 | { |
| 6656 | int cpu = (unsigned long)hcpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6657 | |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 6658 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
Konstantin Khlebnikov | f0cb3c7 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 6659 | lru_add_drain_cpu(cpu); |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 6660 | drain_pages(cpu); |
| 6661 | |
| 6662 | /* |
| 6663 | * Spill the event counters of the dead processor |
| 6664 | * into the current processors event counters. |
| 6665 | * This artificially elevates the count of the current |
| 6666 | * processor. |
| 6667 | */ |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 6668 | vm_events_fold_cpu(cpu); |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 6669 | |
| 6670 | /* |
| 6671 | * Zero the differential counters of the dead processor |
| 6672 | * so that the vm statistics are consistent. |
| 6673 | * |
| 6674 | * This is only okay since the processor is dead and cannot |
| 6675 | * race with what we are doing. |
| 6676 | */ |
Christoph Lameter | 2bb921e | 2013-09-11 14:21:30 -0700 | [diff] [blame] | 6677 | cpu_vm_stats_fold(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6678 | } |
| 6679 | return NOTIFY_OK; |
| 6680 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6681 | |
| 6682 | void __init page_alloc_init(void) |
| 6683 | { |
| 6684 | hotcpu_notifier(page_alloc_cpu_notify, 0); |
| 6685 | } |
| 6686 | |
| 6687 | /* |
Yaowei Bai | 34b1006 | 2015-09-08 15:04:13 -0700 | [diff] [blame] | 6688 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6689 | * or min_free_kbytes changes. |
| 6690 | */ |
| 6691 | static void calculate_totalreserve_pages(void) |
| 6692 | { |
| 6693 | struct pglist_data *pgdat; |
| 6694 | unsigned long reserve_pages = 0; |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 6695 | enum zone_type i, j; |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6696 | |
| 6697 | for_each_online_pgdat(pgdat) { |
| 6698 | for (i = 0; i < MAX_NR_ZONES; i++) { |
| 6699 | struct zone *zone = pgdat->node_zones + i; |
Mel Gorman | 3484b2d | 2014-08-06 16:07:14 -0700 | [diff] [blame] | 6700 | long max = 0; |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6701 | |
| 6702 | /* Find valid and maximum lowmem_reserve in the zone */ |
| 6703 | for (j = i; j < MAX_NR_ZONES; j++) { |
| 6704 | if (zone->lowmem_reserve[j] > max) |
| 6705 | max = zone->lowmem_reserve[j]; |
| 6706 | } |
| 6707 | |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 6708 | /* we treat the high watermark as reserved pages. */ |
| 6709 | max += high_wmark_pages(zone); |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6710 | |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6711 | if (max > zone->managed_pages) |
| 6712 | max = zone->managed_pages; |
Johannes Weiner | a8d0143 | 2016-01-14 15:20:15 -0800 | [diff] [blame] | 6713 | |
| 6714 | zone->totalreserve_pages = max; |
| 6715 | |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6716 | reserve_pages += max; |
| 6717 | } |
| 6718 | } |
| 6719 | totalreserve_pages = reserve_pages; |
| 6720 | } |
| 6721 | |
| 6722 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6723 | * setup_per_zone_lowmem_reserve - called whenever |
Yaowei Bai | 34b1006 | 2015-09-08 15:04:13 -0700 | [diff] [blame] | 6724 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6725 | * has a correct pages reserved value, so an adequate number of |
| 6726 | * pages are left in the zone after a successful __alloc_pages(). |
| 6727 | */ |
| 6728 | static void setup_per_zone_lowmem_reserve(void) |
| 6729 | { |
| 6730 | struct pglist_data *pgdat; |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 6731 | enum zone_type j, idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6732 | |
KAMEZAWA Hiroyuki | ec936fc | 2006-03-27 01:15:59 -0800 | [diff] [blame] | 6733 | for_each_online_pgdat(pgdat) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6734 | for (j = 0; j < MAX_NR_ZONES; j++) { |
| 6735 | struct zone *zone = pgdat->node_zones + j; |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6736 | unsigned long managed_pages = zone->managed_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6737 | |
| 6738 | zone->lowmem_reserve[j] = 0; |
| 6739 | |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 6740 | idx = j; |
| 6741 | while (idx) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6742 | struct zone *lower_zone; |
| 6743 | |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 6744 | idx--; |
| 6745 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6746 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
| 6747 | sysctl_lowmem_reserve_ratio[idx] = 1; |
| 6748 | |
| 6749 | lower_zone = pgdat->node_zones + idx; |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6750 | lower_zone->lowmem_reserve[j] = managed_pages / |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6751 | sysctl_lowmem_reserve_ratio[idx]; |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6752 | managed_pages += lower_zone->managed_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6753 | } |
| 6754 | } |
| 6755 | } |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6756 | |
| 6757 | /* update totalreserve_pages */ |
| 6758 | calculate_totalreserve_pages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6759 | } |
| 6760 | |
Mel Gorman | cfd3da1 | 2011-04-25 21:36:42 +0000 | [diff] [blame] | 6761 | static void __setup_per_zone_wmarks(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6762 | { |
| 6763 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); |
| 6764 | unsigned long lowmem_pages = 0; |
| 6765 | struct zone *zone; |
| 6766 | unsigned long flags; |
| 6767 | |
| 6768 | /* Calculate total number of !ZONE_HIGHMEM pages */ |
| 6769 | for_each_zone(zone) { |
| 6770 | if (!is_highmem(zone)) |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6771 | lowmem_pages += zone->managed_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6772 | } |
| 6773 | |
| 6774 | for_each_zone(zone) { |
Andrew Morton | ac924c6 | 2006-05-15 09:43:59 -0700 | [diff] [blame] | 6775 | u64 tmp; |
| 6776 | |
Gerald Schaefer | 1125b4e | 2008-10-18 20:27:11 -0700 | [diff] [blame] | 6777 | spin_lock_irqsave(&zone->lock, flags); |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6778 | tmp = (u64)pages_min * zone->managed_pages; |
Andrew Morton | ac924c6 | 2006-05-15 09:43:59 -0700 | [diff] [blame] | 6779 | do_div(tmp, lowmem_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6780 | if (is_highmem(zone)) { |
| 6781 | /* |
Nick Piggin | 669ed17 | 2005-11-13 16:06:45 -0800 | [diff] [blame] | 6782 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
| 6783 | * need highmem pages, so cap pages_min to a small |
| 6784 | * value here. |
| 6785 | * |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 6786 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
Yaowei Bai | 42ff270 | 2015-04-14 15:47:14 -0700 | [diff] [blame] | 6787 | * deltas control asynch page reclaim, and so should |
Nick Piggin | 669ed17 | 2005-11-13 16:06:45 -0800 | [diff] [blame] | 6788 | * not be capped for highmem. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6789 | */ |
Andrew Morton | 90ae8d6 | 2013-02-22 16:32:22 -0800 | [diff] [blame] | 6790 | unsigned long min_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6791 | |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6792 | min_pages = zone->managed_pages / 1024; |
Andrew Morton | 90ae8d6 | 2013-02-22 16:32:22 -0800 | [diff] [blame] | 6793 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 6794 | zone->watermark[WMARK_MIN] = min_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6795 | } else { |
Nick Piggin | 669ed17 | 2005-11-13 16:06:45 -0800 | [diff] [blame] | 6796 | /* |
| 6797 | * If it's a lowmem zone, reserve a number of pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6798 | * proportionate to the zone's size. |
| 6799 | */ |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 6800 | zone->watermark[WMARK_MIN] = tmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6801 | } |
| 6802 | |
Johannes Weiner | 795ae7a | 2016-03-17 14:19:14 -0700 | [diff] [blame] | 6803 | /* |
| 6804 | * Set the kswapd watermarks distance according to the |
| 6805 | * scale factor in proportion to available memory, but |
| 6806 | * ensure a minimum size on small systems. |
| 6807 | */ |
| 6808 | tmp = max_t(u64, tmp >> 2, |
| 6809 | mult_frac(zone->managed_pages, |
| 6810 | watermark_scale_factor, 10000)); |
| 6811 | |
| 6812 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; |
| 6813 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; |
Marek Szyprowski | 49f223a | 2012-01-25 12:49:24 +0100 | [diff] [blame] | 6814 | |
Johannes Weiner | 81c0a2b | 2013-09-11 14:20:47 -0700 | [diff] [blame] | 6815 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, |
Johannes Weiner | abe5f97 | 2014-10-02 16:21:10 -0700 | [diff] [blame] | 6816 | high_wmark_pages(zone) - low_wmark_pages(zone) - |
| 6817 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); |
Johannes Weiner | 81c0a2b | 2013-09-11 14:20:47 -0700 | [diff] [blame] | 6818 | |
Gerald Schaefer | 1125b4e | 2008-10-18 20:27:11 -0700 | [diff] [blame] | 6819 | spin_unlock_irqrestore(&zone->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6820 | } |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6821 | |
| 6822 | /* update totalreserve_pages */ |
| 6823 | calculate_totalreserve_pages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6824 | } |
| 6825 | |
Mel Gorman | cfd3da1 | 2011-04-25 21:36:42 +0000 | [diff] [blame] | 6826 | /** |
| 6827 | * setup_per_zone_wmarks - called when min_free_kbytes changes |
| 6828 | * or when memory is hot-{added|removed} |
| 6829 | * |
| 6830 | * Ensures that the watermark[min,low,high] values for each zone are set |
| 6831 | * correctly with respect to min_free_kbytes. |
| 6832 | */ |
| 6833 | void setup_per_zone_wmarks(void) |
| 6834 | { |
| 6835 | mutex_lock(&zonelists_mutex); |
| 6836 | __setup_per_zone_wmarks(); |
| 6837 | mutex_unlock(&zonelists_mutex); |
| 6838 | } |
| 6839 | |
Randy Dunlap | 55a4462 | 2009-09-21 17:01:20 -0700 | [diff] [blame] | 6840 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6841 | * Initialise min_free_kbytes. |
| 6842 | * |
| 6843 | * For small machines we want it small (128k min). For large machines |
| 6844 | * we want it large (64MB max). But it is not linear, because network |
| 6845 | * bandwidth does not increase linearly with machine size. We use |
| 6846 | * |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 6847 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6848 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
| 6849 | * |
| 6850 | * which yields |
| 6851 | * |
| 6852 | * 16MB: 512k |
| 6853 | * 32MB: 724k |
| 6854 | * 64MB: 1024k |
| 6855 | * 128MB: 1448k |
| 6856 | * 256MB: 2048k |
| 6857 | * 512MB: 2896k |
| 6858 | * 1024MB: 4096k |
| 6859 | * 2048MB: 5792k |
| 6860 | * 4096MB: 8192k |
| 6861 | * 8192MB: 11584k |
| 6862 | * 16384MB: 16384k |
| 6863 | */ |
KOSAKI Motohiro | 1b79acc | 2011-05-24 17:11:32 -0700 | [diff] [blame] | 6864 | int __meminit init_per_zone_wmark_min(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6865 | { |
| 6866 | unsigned long lowmem_kbytes; |
Michal Hocko | 5f12733 | 2013-07-08 16:00:40 -0700 | [diff] [blame] | 6867 | int new_min_free_kbytes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6868 | |
| 6869 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); |
Michal Hocko | 5f12733 | 2013-07-08 16:00:40 -0700 | [diff] [blame] | 6870 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6871 | |
Michal Hocko | 5f12733 | 2013-07-08 16:00:40 -0700 | [diff] [blame] | 6872 | if (new_min_free_kbytes > user_min_free_kbytes) { |
| 6873 | min_free_kbytes = new_min_free_kbytes; |
| 6874 | if (min_free_kbytes < 128) |
| 6875 | min_free_kbytes = 128; |
| 6876 | if (min_free_kbytes > 65536) |
| 6877 | min_free_kbytes = 65536; |
| 6878 | } else { |
| 6879 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", |
| 6880 | new_min_free_kbytes, user_min_free_kbytes); |
| 6881 | } |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 6882 | setup_per_zone_wmarks(); |
KOSAKI Motohiro | a6cccdc | 2011-05-24 17:11:33 -0700 | [diff] [blame] | 6883 | refresh_zone_stat_thresholds(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6884 | setup_per_zone_lowmem_reserve(); |
| 6885 | return 0; |
| 6886 | } |
Jason Baron | bc22af7 | 2016-05-05 16:22:12 -0700 | [diff] [blame] | 6887 | core_initcall(init_per_zone_wmark_min) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6888 | |
| 6889 | /* |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 6890 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6891 | * that we can call two helper functions whenever min_free_kbytes |
| 6892 | * changes. |
| 6893 | */ |
Joe Perches | cccad5b | 2014-06-06 14:38:09 -0700 | [diff] [blame] | 6894 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 6895 | void __user *buffer, size_t *length, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6896 | { |
Han Pingtian | da8c757 | 2014-01-23 15:53:17 -0800 | [diff] [blame] | 6897 | int rc; |
| 6898 | |
| 6899 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
| 6900 | if (rc) |
| 6901 | return rc; |
| 6902 | |
Michal Hocko | 5f12733 | 2013-07-08 16:00:40 -0700 | [diff] [blame] | 6903 | if (write) { |
| 6904 | user_min_free_kbytes = min_free_kbytes; |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 6905 | setup_per_zone_wmarks(); |
Michal Hocko | 5f12733 | 2013-07-08 16:00:40 -0700 | [diff] [blame] | 6906 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6907 | return 0; |
| 6908 | } |
| 6909 | |
Johannes Weiner | 795ae7a | 2016-03-17 14:19:14 -0700 | [diff] [blame] | 6910 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
| 6911 | void __user *buffer, size_t *length, loff_t *ppos) |
| 6912 | { |
| 6913 | int rc; |
| 6914 | |
| 6915 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
| 6916 | if (rc) |
| 6917 | return rc; |
| 6918 | |
| 6919 | if (write) |
| 6920 | setup_per_zone_wmarks(); |
| 6921 | |
| 6922 | return 0; |
| 6923 | } |
| 6924 | |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 6925 | #ifdef CONFIG_NUMA |
Joe Perches | cccad5b | 2014-06-06 14:38:09 -0700 | [diff] [blame] | 6926 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 6927 | void __user *buffer, size_t *length, loff_t *ppos) |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 6928 | { |
| 6929 | struct zone *zone; |
| 6930 | int rc; |
| 6931 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 6932 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 6933 | if (rc) |
| 6934 | return rc; |
| 6935 | |
| 6936 | for_each_zone(zone) |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6937 | zone->min_unmapped_pages = (zone->managed_pages * |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 6938 | sysctl_min_unmapped_ratio) / 100; |
| 6939 | return 0; |
| 6940 | } |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 6941 | |
Joe Perches | cccad5b | 2014-06-06 14:38:09 -0700 | [diff] [blame] | 6942 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 6943 | void __user *buffer, size_t *length, loff_t *ppos) |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 6944 | { |
| 6945 | struct zone *zone; |
| 6946 | int rc; |
| 6947 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 6948 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 6949 | if (rc) |
| 6950 | return rc; |
| 6951 | |
| 6952 | for_each_zone(zone) |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6953 | zone->min_slab_pages = (zone->managed_pages * |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 6954 | sysctl_min_slab_ratio) / 100; |
| 6955 | return 0; |
| 6956 | } |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 6957 | #endif |
| 6958 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6959 | /* |
| 6960 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around |
| 6961 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() |
| 6962 | * whenever sysctl_lowmem_reserve_ratio changes. |
| 6963 | * |
| 6964 | * The reserve ratio obviously has absolutely no relation with the |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 6965 | * minimum watermarks. The lowmem reserve ratio can only make sense |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6966 | * if in function of the boot time zone sizes. |
| 6967 | */ |
Joe Perches | cccad5b | 2014-06-06 14:38:09 -0700 | [diff] [blame] | 6968 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 6969 | void __user *buffer, size_t *length, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6970 | { |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 6971 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6972 | setup_per_zone_lowmem_reserve(); |
| 6973 | return 0; |
| 6974 | } |
| 6975 | |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 6976 | /* |
| 6977 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 6978 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
| 6979 | * pagelist can have before it gets flushed back to buddy allocator. |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 6980 | */ |
Joe Perches | cccad5b | 2014-06-06 14:38:09 -0700 | [diff] [blame] | 6981 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 6982 | void __user *buffer, size_t *length, loff_t *ppos) |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 6983 | { |
| 6984 | struct zone *zone; |
David Rientjes | 7cd2b0a | 2014-06-23 13:22:04 -0700 | [diff] [blame] | 6985 | int old_percpu_pagelist_fraction; |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 6986 | int ret; |
| 6987 | |
Cody P Schafer | c8e251f | 2013-07-03 15:01:29 -0700 | [diff] [blame] | 6988 | mutex_lock(&pcp_batch_high_lock); |
David Rientjes | 7cd2b0a | 2014-06-23 13:22:04 -0700 | [diff] [blame] | 6989 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; |
| 6990 | |
| 6991 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
| 6992 | if (!write || ret < 0) |
| 6993 | goto out; |
| 6994 | |
| 6995 | /* Sanity checking to avoid pcp imbalance */ |
| 6996 | if (percpu_pagelist_fraction && |
| 6997 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { |
| 6998 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; |
| 6999 | ret = -EINVAL; |
| 7000 | goto out; |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 7001 | } |
David Rientjes | 7cd2b0a | 2014-06-23 13:22:04 -0700 | [diff] [blame] | 7002 | |
| 7003 | /* No change? */ |
| 7004 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) |
| 7005 | goto out; |
| 7006 | |
| 7007 | for_each_populated_zone(zone) { |
| 7008 | unsigned int cpu; |
| 7009 | |
| 7010 | for_each_possible_cpu(cpu) |
| 7011 | pageset_set_high_and_batch(zone, |
| 7012 | per_cpu_ptr(zone->pageset, cpu)); |
| 7013 | } |
| 7014 | out: |
Cody P Schafer | c8e251f | 2013-07-03 15:01:29 -0700 | [diff] [blame] | 7015 | mutex_unlock(&pcp_batch_high_lock); |
David Rientjes | 7cd2b0a | 2014-06-23 13:22:04 -0700 | [diff] [blame] | 7016 | return ret; |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 7017 | } |
| 7018 | |
Rasmus Villemoes | a9919c7 | 2015-06-24 16:56:28 -0700 | [diff] [blame] | 7019 | #ifdef CONFIG_NUMA |
David S. Miller | f034b5d | 2006-08-24 03:08:07 -0700 | [diff] [blame] | 7020 | int hashdist = HASHDIST_DEFAULT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7021 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7022 | static int __init set_hashdist(char *str) |
| 7023 | { |
| 7024 | if (!str) |
| 7025 | return 0; |
| 7026 | hashdist = simple_strtoul(str, &str, 0); |
| 7027 | return 1; |
| 7028 | } |
| 7029 | __setup("hashdist=", set_hashdist); |
| 7030 | #endif |
| 7031 | |
| 7032 | /* |
| 7033 | * allocate a large system hash table from bootmem |
| 7034 | * - it is assumed that the hash table must contain an exact power-of-2 |
| 7035 | * quantity of entries |
| 7036 | * - limit is the number of hash buckets, not the total allocation size |
| 7037 | */ |
| 7038 | void *__init alloc_large_system_hash(const char *tablename, |
| 7039 | unsigned long bucketsize, |
| 7040 | unsigned long numentries, |
| 7041 | int scale, |
| 7042 | int flags, |
| 7043 | unsigned int *_hash_shift, |
| 7044 | unsigned int *_hash_mask, |
Tim Bird | 31fe62b | 2012-05-23 13:33:35 +0000 | [diff] [blame] | 7045 | unsigned long low_limit, |
| 7046 | unsigned long high_limit) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7047 | { |
Tim Bird | 31fe62b | 2012-05-23 13:33:35 +0000 | [diff] [blame] | 7048 | unsigned long long max = high_limit; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7049 | unsigned long log2qty, size; |
| 7050 | void *table = NULL; |
| 7051 | |
| 7052 | /* allow the kernel cmdline to have a say */ |
| 7053 | if (!numentries) { |
| 7054 | /* round applicable memory size up to nearest megabyte */ |
Andrew Morton | 0490366 | 2006-12-06 20:37:33 -0800 | [diff] [blame] | 7055 | numentries = nr_kernel_pages; |
Jerry Zhou | a7e8331 | 2013-09-11 14:20:26 -0700 | [diff] [blame] | 7056 | |
| 7057 | /* It isn't necessary when PAGE_SIZE >= 1MB */ |
| 7058 | if (PAGE_SHIFT < 20) |
| 7059 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7060 | |
| 7061 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
| 7062 | if (scale > PAGE_SHIFT) |
| 7063 | numentries >>= (scale - PAGE_SHIFT); |
| 7064 | else |
| 7065 | numentries <<= (PAGE_SHIFT - scale); |
Paul Mundt | 9ab37b8 | 2007-01-05 16:36:30 -0800 | [diff] [blame] | 7066 | |
| 7067 | /* Make sure we've got at least a 0-order allocation.. */ |
Jan Beulich | 2c85f51 | 2009-09-21 17:03:07 -0700 | [diff] [blame] | 7068 | if (unlikely(flags & HASH_SMALL)) { |
| 7069 | /* Makes no sense without HASH_EARLY */ |
| 7070 | WARN_ON(!(flags & HASH_EARLY)); |
| 7071 | if (!(numentries >> *_hash_shift)) { |
| 7072 | numentries = 1UL << *_hash_shift; |
| 7073 | BUG_ON(!numentries); |
| 7074 | } |
| 7075 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) |
Paul Mundt | 9ab37b8 | 2007-01-05 16:36:30 -0800 | [diff] [blame] | 7076 | numentries = PAGE_SIZE / bucketsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7077 | } |
John Hawkes | 6e692ed | 2006-03-25 03:08:02 -0800 | [diff] [blame] | 7078 | numentries = roundup_pow_of_two(numentries); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7079 | |
| 7080 | /* limit allocation size to 1/16 total memory by default */ |
| 7081 | if (max == 0) { |
| 7082 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; |
| 7083 | do_div(max, bucketsize); |
| 7084 | } |
Dimitri Sivanich | 074b851 | 2012-02-08 12:39:07 -0800 | [diff] [blame] | 7085 | max = min(max, 0x80000000ULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7086 | |
Tim Bird | 31fe62b | 2012-05-23 13:33:35 +0000 | [diff] [blame] | 7087 | if (numentries < low_limit) |
| 7088 | numentries = low_limit; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7089 | if (numentries > max) |
| 7090 | numentries = max; |
| 7091 | |
David Howells | f0d1b0b | 2006-12-08 02:37:49 -0800 | [diff] [blame] | 7092 | log2qty = ilog2(numentries); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7093 | |
| 7094 | do { |
| 7095 | size = bucketsize << log2qty; |
| 7096 | if (flags & HASH_EARLY) |
Santosh Shilimkar | 6782832 | 2014-01-21 15:50:25 -0800 | [diff] [blame] | 7097 | table = memblock_virt_alloc_nopanic(size, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7098 | else if (hashdist) |
| 7099 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); |
| 7100 | else { |
Eric Dumazet | 1037b83 | 2007-07-15 23:38:05 -0700 | [diff] [blame] | 7101 | /* |
| 7102 | * If bucketsize is not a power-of-two, we may free |
Mel Gorman | a1dd268 | 2009-06-16 15:32:19 -0700 | [diff] [blame] | 7103 | * some pages at the end of hash table which |
| 7104 | * alloc_pages_exact() automatically does |
Eric Dumazet | 1037b83 | 2007-07-15 23:38:05 -0700 | [diff] [blame] | 7105 | */ |
Catalin Marinas | 264ef8a | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 7106 | if (get_order(size) < MAX_ORDER) { |
Mel Gorman | a1dd268 | 2009-06-16 15:32:19 -0700 | [diff] [blame] | 7107 | table = alloc_pages_exact(size, GFP_ATOMIC); |
Catalin Marinas | 264ef8a | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 7108 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
| 7109 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7110 | } |
| 7111 | } while (!table && size > PAGE_SIZE && --log2qty); |
| 7112 | |
| 7113 | if (!table) |
| 7114 | panic("Failed to allocate %s hash table\n", tablename); |
| 7115 | |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 7116 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n", |
| 7117 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7118 | |
| 7119 | if (_hash_shift) |
| 7120 | *_hash_shift = log2qty; |
| 7121 | if (_hash_mask) |
| 7122 | *_hash_mask = (1 << log2qty) - 1; |
| 7123 | |
| 7124 | return table; |
| 7125 | } |
KAMEZAWA Hiroyuki | a117e66 | 2006-03-27 01:15:25 -0800 | [diff] [blame] | 7126 | |
KAMEZAWA Hiroyuki | a5d76b5 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 7127 | /* |
Minchan Kim | 8093451 | 2012-07-31 16:43:01 -0700 | [diff] [blame] | 7128 | * This function checks whether pageblock includes unmovable pages or not. |
| 7129 | * If @count is not zero, it is okay to include less @count unmovable pages |
| 7130 | * |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 7131 | * PageLRU check without isolation or lru_lock could race so that |
Minchan Kim | 8093451 | 2012-07-31 16:43:01 -0700 | [diff] [blame] | 7132 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't |
| 7133 | * expect this function should be exact. |
KAMEZAWA Hiroyuki | a5d76b5 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 7134 | */ |
Wen Congyang | b023f46 | 2012-12-11 16:00:45 -0800 | [diff] [blame] | 7135 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
| 7136 | bool skip_hwpoisoned_pages) |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7137 | { |
| 7138 | unsigned long pfn, iter, found; |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7139 | int mt; |
| 7140 | |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7141 | /* |
| 7142 | * For avoiding noise data, lru_add_drain_all() should be called |
Minchan Kim | 8093451 | 2012-07-31 16:43:01 -0700 | [diff] [blame] | 7143 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7144 | */ |
| 7145 | if (zone_idx(zone) == ZONE_MOVABLE) |
Minchan Kim | 8093451 | 2012-07-31 16:43:01 -0700 | [diff] [blame] | 7146 | return false; |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7147 | mt = get_pageblock_migratetype(page); |
| 7148 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) |
Minchan Kim | 8093451 | 2012-07-31 16:43:01 -0700 | [diff] [blame] | 7149 | return false; |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7150 | |
| 7151 | pfn = page_to_pfn(page); |
| 7152 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { |
| 7153 | unsigned long check = pfn + iter; |
| 7154 | |
Namhyung Kim | 29723fc | 2011-02-25 14:44:25 -0800 | [diff] [blame] | 7155 | if (!pfn_valid_within(check)) |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7156 | continue; |
Namhyung Kim | 29723fc | 2011-02-25 14:44:25 -0800 | [diff] [blame] | 7157 | |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7158 | page = pfn_to_page(check); |
Naoya Horiguchi | c8721bb | 2013-09-11 14:22:09 -0700 | [diff] [blame] | 7159 | |
| 7160 | /* |
| 7161 | * Hugepages are not in LRU lists, but they're movable. |
| 7162 | * We need not scan over tail pages bacause we don't |
| 7163 | * handle each tail page individually in migration. |
| 7164 | */ |
| 7165 | if (PageHuge(page)) { |
| 7166 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; |
| 7167 | continue; |
| 7168 | } |
| 7169 | |
Minchan Kim | 97d255c | 2012-07-31 16:42:59 -0700 | [diff] [blame] | 7170 | /* |
| 7171 | * We can't use page_count without pin a page |
| 7172 | * because another CPU can free compound page. |
| 7173 | * This check already skips compound tails of THP |
Joonsoo Kim | 0139aa7 | 2016-05-19 17:10:49 -0700 | [diff] [blame] | 7174 | * because their page->_refcount is zero at all time. |
Minchan Kim | 97d255c | 2012-07-31 16:42:59 -0700 | [diff] [blame] | 7175 | */ |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 7176 | if (!page_ref_count(page)) { |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7177 | if (PageBuddy(page)) |
| 7178 | iter += (1 << page_order(page)) - 1; |
| 7179 | continue; |
| 7180 | } |
Minchan Kim | 97d255c | 2012-07-31 16:42:59 -0700 | [diff] [blame] | 7181 | |
Wen Congyang | b023f46 | 2012-12-11 16:00:45 -0800 | [diff] [blame] | 7182 | /* |
| 7183 | * The HWPoisoned page may be not in buddy system, and |
| 7184 | * page_count() is not 0. |
| 7185 | */ |
| 7186 | if (skip_hwpoisoned_pages && PageHWPoison(page)) |
| 7187 | continue; |
| 7188 | |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7189 | if (!PageLRU(page)) |
| 7190 | found++; |
| 7191 | /* |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 7192 | * If there are RECLAIMABLE pages, we need to check |
| 7193 | * it. But now, memory offline itself doesn't call |
| 7194 | * shrink_node_slabs() and it still to be fixed. |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7195 | */ |
| 7196 | /* |
| 7197 | * If the page is not RAM, page_count()should be 0. |
| 7198 | * we don't need more check. This is an _used_ not-movable page. |
| 7199 | * |
| 7200 | * The problematic thing here is PG_reserved pages. PG_reserved |
| 7201 | * is set to both of a memory hole page and a _used_ kernel |
| 7202 | * page at boot. |
| 7203 | */ |
| 7204 | if (found > count) |
Minchan Kim | 8093451 | 2012-07-31 16:43:01 -0700 | [diff] [blame] | 7205 | return true; |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7206 | } |
Minchan Kim | 8093451 | 2012-07-31 16:43:01 -0700 | [diff] [blame] | 7207 | return false; |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7208 | } |
| 7209 | |
| 7210 | bool is_pageblock_removable_nolock(struct page *page) |
| 7211 | { |
Michal Hocko | 656a070 | 2012-01-20 14:33:58 -0800 | [diff] [blame] | 7212 | struct zone *zone; |
| 7213 | unsigned long pfn; |
Michal Hocko | 687875f | 2012-01-20 14:33:55 -0800 | [diff] [blame] | 7214 | |
| 7215 | /* |
| 7216 | * We have to be careful here because we are iterating over memory |
| 7217 | * sections which are not zone aware so we might end up outside of |
| 7218 | * the zone but still within the section. |
Michal Hocko | 656a070 | 2012-01-20 14:33:58 -0800 | [diff] [blame] | 7219 | * We have to take care about the node as well. If the node is offline |
| 7220 | * its NODE_DATA will be NULL - see page_zone. |
Michal Hocko | 687875f | 2012-01-20 14:33:55 -0800 | [diff] [blame] | 7221 | */ |
Michal Hocko | 656a070 | 2012-01-20 14:33:58 -0800 | [diff] [blame] | 7222 | if (!node_online(page_to_nid(page))) |
| 7223 | return false; |
| 7224 | |
| 7225 | zone = page_zone(page); |
| 7226 | pfn = page_to_pfn(page); |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 7227 | if (!zone_spans_pfn(zone, pfn)) |
Michal Hocko | 687875f | 2012-01-20 14:33:55 -0800 | [diff] [blame] | 7228 | return false; |
| 7229 | |
Wen Congyang | b023f46 | 2012-12-11 16:00:45 -0800 | [diff] [blame] | 7230 | return !has_unmovable_pages(zone, page, 0, true); |
KAMEZAWA Hiroyuki | a5d76b5 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 7231 | } |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 7232 | |
Vlastimil Babka | 080fe20 | 2016-02-05 15:36:41 -0800 | [diff] [blame] | 7233 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7234 | |
| 7235 | static unsigned long pfn_max_align_down(unsigned long pfn) |
| 7236 | { |
| 7237 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, |
| 7238 | pageblock_nr_pages) - 1); |
| 7239 | } |
| 7240 | |
| 7241 | static unsigned long pfn_max_align_up(unsigned long pfn) |
| 7242 | { |
| 7243 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, |
| 7244 | pageblock_nr_pages)); |
| 7245 | } |
| 7246 | |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7247 | /* [start, end) must belong to a single zone. */ |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 7248 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
| 7249 | unsigned long start, unsigned long end) |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7250 | { |
| 7251 | /* This function is based on compact_zone() from compaction.c. */ |
Minchan Kim | beb51ea | 2012-10-08 16:33:51 -0700 | [diff] [blame] | 7252 | unsigned long nr_reclaimed; |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7253 | unsigned long pfn = start; |
| 7254 | unsigned int tries = 0; |
| 7255 | int ret = 0; |
| 7256 | |
Marek Szyprowski | be49a6e | 2012-12-12 13:51:19 -0800 | [diff] [blame] | 7257 | migrate_prep(); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7258 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 7259 | while (pfn < end || !list_empty(&cc->migratepages)) { |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7260 | if (fatal_signal_pending(current)) { |
| 7261 | ret = -EINTR; |
| 7262 | break; |
| 7263 | } |
| 7264 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 7265 | if (list_empty(&cc->migratepages)) { |
| 7266 | cc->nr_migratepages = 0; |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 7267 | pfn = isolate_migratepages_range(cc, pfn, end); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7268 | if (!pfn) { |
| 7269 | ret = -EINTR; |
| 7270 | break; |
| 7271 | } |
| 7272 | tries = 0; |
| 7273 | } else if (++tries == 5) { |
| 7274 | ret = ret < 0 ? ret : -EBUSY; |
| 7275 | break; |
| 7276 | } |
| 7277 | |
Minchan Kim | beb51ea | 2012-10-08 16:33:51 -0700 | [diff] [blame] | 7278 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
| 7279 | &cc->migratepages); |
| 7280 | cc->nr_migratepages -= nr_reclaimed; |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 7281 | |
Hugh Dickins | 9c620e2 | 2013-02-22 16:35:14 -0800 | [diff] [blame] | 7282 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 7283 | NULL, 0, cc->mode, MR_CMA); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7284 | } |
Srinivas Pandruvada | 2a6f512 | 2013-02-22 16:32:09 -0800 | [diff] [blame] | 7285 | if (ret < 0) { |
| 7286 | putback_movable_pages(&cc->migratepages); |
| 7287 | return ret; |
| 7288 | } |
| 7289 | return 0; |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7290 | } |
| 7291 | |
| 7292 | /** |
| 7293 | * alloc_contig_range() -- tries to allocate given range of pages |
| 7294 | * @start: start PFN to allocate |
| 7295 | * @end: one-past-the-last PFN to allocate |
Michal Nazarewicz | 0815f3d | 2012-04-03 15:06:15 +0200 | [diff] [blame] | 7296 | * @migratetype: migratetype of the underlaying pageblocks (either |
| 7297 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks |
| 7298 | * in range must have the same migratetype and it must |
| 7299 | * be either of the two. |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7300 | * |
| 7301 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES |
| 7302 | * aligned, however it's the caller's responsibility to guarantee that |
| 7303 | * we are the only thread that changes migrate type of pageblocks the |
| 7304 | * pages fall in. |
| 7305 | * |
| 7306 | * The PFN range must belong to a single zone. |
| 7307 | * |
| 7308 | * Returns zero on success or negative error code. On success all |
| 7309 | * pages which PFN is in [start, end) are allocated for the caller and |
| 7310 | * need to be freed with free_contig_range(). |
| 7311 | */ |
Michal Nazarewicz | 0815f3d | 2012-04-03 15:06:15 +0200 | [diff] [blame] | 7312 | int alloc_contig_range(unsigned long start, unsigned long end, |
| 7313 | unsigned migratetype) |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7314 | { |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7315 | unsigned long outer_start, outer_end; |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 7316 | unsigned int order; |
| 7317 | int ret = 0; |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7318 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 7319 | struct compact_control cc = { |
| 7320 | .nr_migratepages = 0, |
| 7321 | .order = -1, |
| 7322 | .zone = page_zone(pfn_to_page(start)), |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 7323 | .mode = MIGRATE_SYNC, |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 7324 | .ignore_skip_hint = true, |
| 7325 | }; |
| 7326 | INIT_LIST_HEAD(&cc.migratepages); |
| 7327 | |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7328 | /* |
| 7329 | * What we do here is we mark all pageblocks in range as |
| 7330 | * MIGRATE_ISOLATE. Because pageblock and max order pages may |
| 7331 | * have different sizes, and due to the way page allocator |
| 7332 | * work, we align the range to biggest of the two pages so |
| 7333 | * that page allocator won't try to merge buddies from |
| 7334 | * different pageblocks and change MIGRATE_ISOLATE to some |
| 7335 | * other migration type. |
| 7336 | * |
| 7337 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we |
| 7338 | * migrate the pages from an unaligned range (ie. pages that |
| 7339 | * we are interested in). This will put all the pages in |
| 7340 | * range back to page allocator as MIGRATE_ISOLATE. |
| 7341 | * |
| 7342 | * When this is done, we take the pages in range from page |
| 7343 | * allocator removing them from the buddy system. This way |
| 7344 | * page allocator will never consider using them. |
| 7345 | * |
| 7346 | * This lets us mark the pageblocks back as |
| 7347 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the |
| 7348 | * aligned range but not in the unaligned, original range are |
| 7349 | * put back to page allocator so that buddy can use them. |
| 7350 | */ |
| 7351 | |
| 7352 | ret = start_isolate_page_range(pfn_max_align_down(start), |
Wen Congyang | b023f46 | 2012-12-11 16:00:45 -0800 | [diff] [blame] | 7353 | pfn_max_align_up(end), migratetype, |
| 7354 | false); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7355 | if (ret) |
Bob Liu | 86a595f | 2012-10-25 13:37:56 -0700 | [diff] [blame] | 7356 | return ret; |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7357 | |
Joonsoo Kim | 8ef5849 | 2016-01-14 15:18:45 -0800 | [diff] [blame] | 7358 | /* |
| 7359 | * In case of -EBUSY, we'd like to know which page causes problem. |
| 7360 | * So, just fall through. We will check it in test_pages_isolated(). |
| 7361 | */ |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 7362 | ret = __alloc_contig_migrate_range(&cc, start, end); |
Joonsoo Kim | 8ef5849 | 2016-01-14 15:18:45 -0800 | [diff] [blame] | 7363 | if (ret && ret != -EBUSY) |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7364 | goto done; |
| 7365 | |
| 7366 | /* |
| 7367 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES |
| 7368 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's |
| 7369 | * more, all pages in [start, end) are free in page allocator. |
| 7370 | * What we are going to do is to allocate all pages from |
| 7371 | * [start, end) (that is remove them from page allocator). |
| 7372 | * |
| 7373 | * The only problem is that pages at the beginning and at the |
| 7374 | * end of interesting range may be not aligned with pages that |
| 7375 | * page allocator holds, ie. they can be part of higher order |
| 7376 | * pages. Because of this, we reserve the bigger range and |
| 7377 | * once this is done free the pages we are not interested in. |
| 7378 | * |
| 7379 | * We don't have to hold zone->lock here because the pages are |
| 7380 | * isolated thus they won't get removed from buddy. |
| 7381 | */ |
| 7382 | |
| 7383 | lru_add_drain_all(); |
Vlastimil Babka | 510f550 | 2014-12-10 15:43:07 -0800 | [diff] [blame] | 7384 | drain_all_pages(cc.zone); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7385 | |
| 7386 | order = 0; |
| 7387 | outer_start = start; |
| 7388 | while (!PageBuddy(pfn_to_page(outer_start))) { |
| 7389 | if (++order >= MAX_ORDER) { |
Joonsoo Kim | 8ef5849 | 2016-01-14 15:18:45 -0800 | [diff] [blame] | 7390 | outer_start = start; |
| 7391 | break; |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7392 | } |
| 7393 | outer_start &= ~0UL << order; |
| 7394 | } |
| 7395 | |
Joonsoo Kim | 8ef5849 | 2016-01-14 15:18:45 -0800 | [diff] [blame] | 7396 | if (outer_start != start) { |
| 7397 | order = page_order(pfn_to_page(outer_start)); |
| 7398 | |
| 7399 | /* |
| 7400 | * outer_start page could be small order buddy page and |
| 7401 | * it doesn't include start page. Adjust outer_start |
| 7402 | * in this case to report failed page properly |
| 7403 | * on tracepoint in test_pages_isolated() |
| 7404 | */ |
| 7405 | if (outer_start + (1UL << order) <= start) |
| 7406 | outer_start = start; |
| 7407 | } |
| 7408 | |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7409 | /* Make sure the range is really isolated. */ |
Wen Congyang | b023f46 | 2012-12-11 16:00:45 -0800 | [diff] [blame] | 7410 | if (test_pages_isolated(outer_start, end, false)) { |
Michal Nazarewicz | dae803e | 2014-11-13 15:19:27 -0800 | [diff] [blame] | 7411 | pr_info("%s: [%lx, %lx) PFNs busy\n", |
| 7412 | __func__, outer_start, end); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7413 | ret = -EBUSY; |
| 7414 | goto done; |
| 7415 | } |
| 7416 | |
Marek Szyprowski | 49f223a | 2012-01-25 12:49:24 +0100 | [diff] [blame] | 7417 | /* Grab isolated pages from freelists. */ |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 7418 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7419 | if (!outer_end) { |
| 7420 | ret = -EBUSY; |
| 7421 | goto done; |
| 7422 | } |
| 7423 | |
| 7424 | /* Free head and tail (if any) */ |
| 7425 | if (start != outer_start) |
| 7426 | free_contig_range(outer_start, start - outer_start); |
| 7427 | if (end != outer_end) |
| 7428 | free_contig_range(end, outer_end - end); |
| 7429 | |
| 7430 | done: |
| 7431 | undo_isolate_page_range(pfn_max_align_down(start), |
Michal Nazarewicz | 0815f3d | 2012-04-03 15:06:15 +0200 | [diff] [blame] | 7432 | pfn_max_align_up(end), migratetype); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7433 | return ret; |
| 7434 | } |
| 7435 | |
| 7436 | void free_contig_range(unsigned long pfn, unsigned nr_pages) |
| 7437 | { |
Marek Szyprowski | bcc2b02 | 2012-12-20 15:05:18 -0800 | [diff] [blame] | 7438 | unsigned int count = 0; |
| 7439 | |
| 7440 | for (; nr_pages--; pfn++) { |
| 7441 | struct page *page = pfn_to_page(pfn); |
| 7442 | |
| 7443 | count += page_count(page) != 1; |
| 7444 | __free_page(page); |
| 7445 | } |
| 7446 | WARN(count != 0, "%d pages are still in use!\n", count); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7447 | } |
| 7448 | #endif |
| 7449 | |
Jiang Liu | 4ed7e02 | 2012-07-31 16:43:35 -0700 | [diff] [blame] | 7450 | #ifdef CONFIG_MEMORY_HOTPLUG |
Cody P Schafer | 0a647f3 | 2013-07-03 15:01:33 -0700 | [diff] [blame] | 7451 | /* |
| 7452 | * The zone indicated has a new number of managed_pages; batch sizes and percpu |
| 7453 | * page high values need to be recalulated. |
| 7454 | */ |
Jiang Liu | 4ed7e02 | 2012-07-31 16:43:35 -0700 | [diff] [blame] | 7455 | void __meminit zone_pcp_update(struct zone *zone) |
| 7456 | { |
Cody P Schafer | 0a647f3 | 2013-07-03 15:01:33 -0700 | [diff] [blame] | 7457 | unsigned cpu; |
Cody P Schafer | c8e251f | 2013-07-03 15:01:29 -0700 | [diff] [blame] | 7458 | mutex_lock(&pcp_batch_high_lock); |
Cody P Schafer | 0a647f3 | 2013-07-03 15:01:33 -0700 | [diff] [blame] | 7459 | for_each_possible_cpu(cpu) |
Cody P Schafer | 169f6c1 | 2013-07-03 15:01:41 -0700 | [diff] [blame] | 7460 | pageset_set_high_and_batch(zone, |
| 7461 | per_cpu_ptr(zone->pageset, cpu)); |
Cody P Schafer | c8e251f | 2013-07-03 15:01:29 -0700 | [diff] [blame] | 7462 | mutex_unlock(&pcp_batch_high_lock); |
Jiang Liu | 4ed7e02 | 2012-07-31 16:43:35 -0700 | [diff] [blame] | 7463 | } |
| 7464 | #endif |
| 7465 | |
Jiang Liu | 340175b | 2012-07-31 16:43:32 -0700 | [diff] [blame] | 7466 | void zone_pcp_reset(struct zone *zone) |
| 7467 | { |
| 7468 | unsigned long flags; |
Minchan Kim | 5a88381 | 2012-10-08 16:33:39 -0700 | [diff] [blame] | 7469 | int cpu; |
| 7470 | struct per_cpu_pageset *pset; |
Jiang Liu | 340175b | 2012-07-31 16:43:32 -0700 | [diff] [blame] | 7471 | |
| 7472 | /* avoid races with drain_pages() */ |
| 7473 | local_irq_save(flags); |
| 7474 | if (zone->pageset != &boot_pageset) { |
Minchan Kim | 5a88381 | 2012-10-08 16:33:39 -0700 | [diff] [blame] | 7475 | for_each_online_cpu(cpu) { |
| 7476 | pset = per_cpu_ptr(zone->pageset, cpu); |
| 7477 | drain_zonestat(zone, pset); |
| 7478 | } |
Jiang Liu | 340175b | 2012-07-31 16:43:32 -0700 | [diff] [blame] | 7479 | free_percpu(zone->pageset); |
| 7480 | zone->pageset = &boot_pageset; |
| 7481 | } |
| 7482 | local_irq_restore(flags); |
| 7483 | } |
| 7484 | |
Wen Congyang | 6dcd73d | 2012-12-11 16:01:01 -0800 | [diff] [blame] | 7485 | #ifdef CONFIG_MEMORY_HOTREMOVE |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 7486 | /* |
Joonsoo Kim | b9eb631 | 2016-05-19 17:12:06 -0700 | [diff] [blame] | 7487 | * All pages in the range must be in a single zone and isolated |
| 7488 | * before calling this. |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 7489 | */ |
| 7490 | void |
| 7491 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
| 7492 | { |
| 7493 | struct page *page; |
| 7494 | struct zone *zone; |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 7495 | unsigned int order, i; |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 7496 | unsigned long pfn; |
| 7497 | unsigned long flags; |
| 7498 | /* find the first valid pfn */ |
| 7499 | for (pfn = start_pfn; pfn < end_pfn; pfn++) |
| 7500 | if (pfn_valid(pfn)) |
| 7501 | break; |
| 7502 | if (pfn == end_pfn) |
| 7503 | return; |
| 7504 | zone = page_zone(pfn_to_page(pfn)); |
| 7505 | spin_lock_irqsave(&zone->lock, flags); |
| 7506 | pfn = start_pfn; |
| 7507 | while (pfn < end_pfn) { |
| 7508 | if (!pfn_valid(pfn)) { |
| 7509 | pfn++; |
| 7510 | continue; |
| 7511 | } |
| 7512 | page = pfn_to_page(pfn); |
Wen Congyang | b023f46 | 2012-12-11 16:00:45 -0800 | [diff] [blame] | 7513 | /* |
| 7514 | * The HWPoisoned page may be not in buddy system, and |
| 7515 | * page_count() is not 0. |
| 7516 | */ |
| 7517 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { |
| 7518 | pfn++; |
| 7519 | SetPageReserved(page); |
| 7520 | continue; |
| 7521 | } |
| 7522 | |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 7523 | BUG_ON(page_count(page)); |
| 7524 | BUG_ON(!PageBuddy(page)); |
| 7525 | order = page_order(page); |
| 7526 | #ifdef CONFIG_DEBUG_VM |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 7527 | pr_info("remove from free list %lx %d %lx\n", |
| 7528 | pfn, 1 << order, end_pfn); |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 7529 | #endif |
| 7530 | list_del(&page->lru); |
| 7531 | rmv_page_order(page); |
| 7532 | zone->free_area[order].nr_free--; |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 7533 | for (i = 0; i < (1 << order); i++) |
| 7534 | SetPageReserved((page+i)); |
| 7535 | pfn += (1 << order); |
| 7536 | } |
| 7537 | spin_unlock_irqrestore(&zone->lock, flags); |
| 7538 | } |
| 7539 | #endif |
Wu Fengguang | 8d22ba1 | 2009-12-16 12:19:58 +0100 | [diff] [blame] | 7540 | |
Wu Fengguang | 8d22ba1 | 2009-12-16 12:19:58 +0100 | [diff] [blame] | 7541 | bool is_free_buddy_page(struct page *page) |
| 7542 | { |
| 7543 | struct zone *zone = page_zone(page); |
| 7544 | unsigned long pfn = page_to_pfn(page); |
| 7545 | unsigned long flags; |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 7546 | unsigned int order; |
Wu Fengguang | 8d22ba1 | 2009-12-16 12:19:58 +0100 | [diff] [blame] | 7547 | |
| 7548 | spin_lock_irqsave(&zone->lock, flags); |
| 7549 | for (order = 0; order < MAX_ORDER; order++) { |
| 7550 | struct page *page_head = page - (pfn & ((1 << order) - 1)); |
| 7551 | |
| 7552 | if (PageBuddy(page_head) && page_order(page_head) >= order) |
| 7553 | break; |
| 7554 | } |
| 7555 | spin_unlock_irqrestore(&zone->lock, flags); |
| 7556 | |
| 7557 | return order < MAX_ORDER; |
| 7558 | } |