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 | a5d76b54 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 52 | #include <linux/page-isolation.h> |
Joonsoo Kim | eefa864b | 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 | |
Emese Revfy | 38addce | 2016-06-20 20:41:19 +0200 | [diff] [blame] | 94 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
Emese Revfy | 0766f78 | 2016-06-20 20:42:34 +0200 | [diff] [blame] | 95 | volatile u64 latent_entropy __latent_entropy; |
Emese Revfy | 38addce | 2016-06-20 20:41:19 +0200 | [diff] [blame] | 96 | EXPORT_SYMBOL(latent_entropy); |
| 97 | #endif |
| 98 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | /* |
Christoph Lameter | 1380891 | 2007-10-16 01:25:27 -0700 | [diff] [blame] | 100 | * Array of node states. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 101 | */ |
Christoph Lameter | 1380891 | 2007-10-16 01:25:27 -0700 | [diff] [blame] | 102 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
| 103 | [N_POSSIBLE] = NODE_MASK_ALL, |
| 104 | [N_ONLINE] = { { [0] = 1UL } }, |
| 105 | #ifndef CONFIG_NUMA |
| 106 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, |
| 107 | #ifdef CONFIG_HIGHMEM |
| 108 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, |
| 109 | #endif |
Lai Jiangshan | 20b2f52 | 2012-12-12 13:52:00 -0800 | [diff] [blame] | 110 | #ifdef CONFIG_MOVABLE_NODE |
| 111 | [N_MEMORY] = { { [0] = 1UL } }, |
| 112 | #endif |
Christoph Lameter | 1380891 | 2007-10-16 01:25:27 -0700 | [diff] [blame] | 113 | [N_CPU] = { { [0] = 1UL } }, |
| 114 | #endif /* NUMA */ |
| 115 | }; |
| 116 | EXPORT_SYMBOL(node_states); |
| 117 | |
Jiang Liu | c3d5f5f | 2013-07-03 15:03:14 -0700 | [diff] [blame] | 118 | /* Protect totalram_pages and zone->managed_pages */ |
| 119 | static DEFINE_SPINLOCK(managed_page_count_lock); |
| 120 | |
Ravikiran G Thirumalai | 6c231b7 | 2005-09-06 15:17:45 -0700 | [diff] [blame] | 121 | unsigned long totalram_pages __read_mostly; |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 122 | unsigned long totalreserve_pages __read_mostly; |
Pintu Kumar | e48322a | 2014-12-18 16:17:15 -0800 | [diff] [blame] | 123 | unsigned long totalcma_pages __read_mostly; |
Johannes Weiner | ab8fabd | 2012-01-10 15:07:42 -0800 | [diff] [blame] | 124 | |
Hugh Dickins | 1b76b02 | 2012-05-11 01:00:07 -0700 | [diff] [blame] | 125 | int percpu_pagelist_fraction; |
Benjamin Herrenschmidt | dcce284 | 2009-06-18 13:24:12 +1000 | [diff] [blame] | 126 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 128 | /* |
| 129 | * A cached value of the page's pageblock's migratetype, used when the page is |
| 130 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when |
| 131 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. |
| 132 | * Also the migratetype set in the page does not necessarily match the pcplist |
| 133 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any |
| 134 | * other index - this ensures that it will be put on the correct CMA freelist. |
| 135 | */ |
| 136 | static inline int get_pcppage_migratetype(struct page *page) |
| 137 | { |
| 138 | return page->index; |
| 139 | } |
| 140 | |
| 141 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) |
| 142 | { |
| 143 | page->index = migratetype; |
| 144 | } |
| 145 | |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 146 | #ifdef CONFIG_PM_SLEEP |
| 147 | /* |
| 148 | * The following functions are used by the suspend/hibernate code to temporarily |
| 149 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations |
| 150 | * while devices are suspended. To avoid races with the suspend/hibernate code, |
| 151 | * they should always be called with pm_mutex held (gfp_allowed_mask also should |
| 152 | * only be modified with pm_mutex held, unless the suspend/hibernate code is |
| 153 | * guaranteed not to run in parallel with that modification). |
| 154 | */ |
Rafael J. Wysocki | c9e664f | 2010-12-03 22:57:45 +0100 | [diff] [blame] | 155 | |
| 156 | static gfp_t saved_gfp_mask; |
| 157 | |
| 158 | void pm_restore_gfp_mask(void) |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 159 | { |
| 160 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
Rafael J. Wysocki | c9e664f | 2010-12-03 22:57:45 +0100 | [diff] [blame] | 161 | if (saved_gfp_mask) { |
| 162 | gfp_allowed_mask = saved_gfp_mask; |
| 163 | saved_gfp_mask = 0; |
| 164 | } |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 165 | } |
| 166 | |
Rafael J. Wysocki | c9e664f | 2010-12-03 22:57:45 +0100 | [diff] [blame] | 167 | void pm_restrict_gfp_mask(void) |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 168 | { |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 169 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
Rafael J. Wysocki | c9e664f | 2010-12-03 22:57:45 +0100 | [diff] [blame] | 170 | WARN_ON(saved_gfp_mask); |
| 171 | saved_gfp_mask = gfp_allowed_mask; |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 172 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 173 | } |
Mel Gorman | f90ac39 | 2012-01-10 15:07:15 -0800 | [diff] [blame] | 174 | |
| 175 | bool pm_suspended_storage(void) |
| 176 | { |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 177 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
Mel Gorman | f90ac39 | 2012-01-10 15:07:15 -0800 | [diff] [blame] | 178 | return false; |
| 179 | return true; |
| 180 | } |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 181 | #endif /* CONFIG_PM_SLEEP */ |
| 182 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 183 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 184 | unsigned int pageblock_order __read_mostly; |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 185 | #endif |
| 186 | |
Hugh Dickins | d98c7a0 | 2006-02-14 13:52:59 -0800 | [diff] [blame] | 187 | static void __free_pages_ok(struct page *page, unsigned int order); |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 188 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 189 | /* |
| 190 | * results with 256, 32 in the lowmem_reserve sysctl: |
| 191 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) |
| 192 | * 1G machine -> (16M dma, 784M normal, 224M high) |
| 193 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA |
| 194 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL |
Yaowei Bai | 84109e1 | 2015-02-12 15:00:22 -0800 | [diff] [blame] | 195 | * 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] | 196 | * |
| 197 | * TBD: should special case ZONE_DMA32 machines here - in those we normally |
| 198 | * don't need any ZONE_NORMAL reservation |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | */ |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 200 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 201 | #ifdef CONFIG_ZONE_DMA |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 202 | 256, |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 203 | #endif |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 204 | #ifdef CONFIG_ZONE_DMA32 |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 205 | 256, |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 206 | #endif |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 207 | #ifdef CONFIG_HIGHMEM |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 208 | 32, |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 209 | #endif |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 210 | 32, |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 211 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 212 | |
| 213 | EXPORT_SYMBOL(totalram_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 214 | |
Helge Deller | 15ad7cd | 2006-12-06 20:40:36 -0800 | [diff] [blame] | 215 | static char * const zone_names[MAX_NR_ZONES] = { |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 216 | #ifdef CONFIG_ZONE_DMA |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 217 | "DMA", |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 218 | #endif |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 219 | #ifdef CONFIG_ZONE_DMA32 |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 220 | "DMA32", |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 221 | #endif |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 222 | "Normal", |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 223 | #ifdef CONFIG_HIGHMEM |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 224 | "HighMem", |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 225 | #endif |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 226 | "Movable", |
Dan Williams | 033fbae | 2015-08-09 15:29:06 -0400 | [diff] [blame] | 227 | #ifdef CONFIG_ZONE_DEVICE |
| 228 | "Device", |
| 229 | #endif |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 230 | }; |
| 231 | |
Vlastimil Babka | 60f3035 | 2016-03-15 14:56:08 -0700 | [diff] [blame] | 232 | char * const migratetype_names[MIGRATE_TYPES] = { |
| 233 | "Unmovable", |
| 234 | "Movable", |
| 235 | "Reclaimable", |
| 236 | "HighAtomic", |
| 237 | #ifdef CONFIG_CMA |
| 238 | "CMA", |
| 239 | #endif |
| 240 | #ifdef CONFIG_MEMORY_ISOLATION |
| 241 | "Isolate", |
| 242 | #endif |
| 243 | }; |
| 244 | |
Kirill A. Shutemov | f1e6155 | 2015-11-06 16:29:50 -0800 | [diff] [blame] | 245 | compound_page_dtor * const compound_page_dtors[] = { |
| 246 | NULL, |
| 247 | free_compound_page, |
| 248 | #ifdef CONFIG_HUGETLB_PAGE |
| 249 | free_huge_page, |
| 250 | #endif |
Kirill A. Shutemov | 9a98225 | 2016-01-15 16:54:17 -0800 | [diff] [blame] | 251 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 252 | free_transhuge_page, |
| 253 | #endif |
Kirill A. Shutemov | f1e6155 | 2015-11-06 16:29:50 -0800 | [diff] [blame] | 254 | }; |
| 255 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 256 | int min_free_kbytes = 1024; |
Han Pingtian | 42aa83c | 2014-01-23 15:53:28 -0800 | [diff] [blame] | 257 | int user_min_free_kbytes = -1; |
Johannes Weiner | 795ae7a | 2016-03-17 14:19:14 -0700 | [diff] [blame] | 258 | int watermark_scale_factor = 10; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 259 | |
Jan Beulich | 2c85f51 | 2009-09-21 17:03:07 -0700 | [diff] [blame] | 260 | static unsigned long __meminitdata nr_kernel_pages; |
| 261 | static unsigned long __meminitdata nr_all_pages; |
Yasunori Goto | a3142c8 | 2007-05-08 00:23:07 -0700 | [diff] [blame] | 262 | static unsigned long __meminitdata dma_reserve; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 263 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 264 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
| 265 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; |
| 266 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; |
| 267 | static unsigned long __initdata required_kernelcore; |
| 268 | static unsigned long __initdata required_movablecore; |
| 269 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 270 | static bool mirrored_kernelcore; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 271 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 272 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ |
| 273 | int movable_zone; |
| 274 | EXPORT_SYMBOL(movable_zone); |
| 275 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 276 | |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 277 | #if MAX_NUMNODES > 1 |
| 278 | int nr_node_ids __read_mostly = MAX_NUMNODES; |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 279 | int nr_online_nodes __read_mostly = 1; |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 280 | EXPORT_SYMBOL(nr_node_ids); |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 281 | EXPORT_SYMBOL(nr_online_nodes); |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 282 | #endif |
| 283 | |
Mel Gorman | 9ef9acb | 2007-10-16 01:25:54 -0700 | [diff] [blame] | 284 | int page_group_by_mobility_disabled __read_mostly; |
| 285 | |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 286 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
| 287 | static inline void reset_deferred_meminit(pg_data_t *pgdat) |
| 288 | { |
| 289 | pgdat->first_deferred_pfn = ULONG_MAX; |
| 290 | } |
| 291 | |
| 292 | /* Returns true if the struct page for the pfn is uninitialised */ |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 293 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 294 | { |
Mel Gorman | ef70b6f | 2016-07-14 12:07:23 -0700 | [diff] [blame] | 295 | int nid = early_pfn_to_nid(pfn); |
| 296 | |
| 297 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 298 | return true; |
| 299 | |
| 300 | return false; |
| 301 | } |
| 302 | |
| 303 | /* |
| 304 | * Returns false when the remaining initialisation should be deferred until |
| 305 | * later in the boot cycle when it can be parallelised. |
| 306 | */ |
| 307 | static inline bool update_defer_init(pg_data_t *pgdat, |
| 308 | unsigned long pfn, unsigned long zone_end, |
| 309 | unsigned long *nr_initialised) |
| 310 | { |
Li Zhang | 987b309 | 2016-03-17 14:20:16 -0700 | [diff] [blame] | 311 | unsigned long max_initialise; |
| 312 | |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 313 | /* Always populate low zones for address-contrained allocations */ |
| 314 | if (zone_end < pgdat_end_pfn(pgdat)) |
| 315 | return true; |
Li Zhang | 987b309 | 2016-03-17 14:20:16 -0700 | [diff] [blame] | 316 | /* |
| 317 | * Initialise at least 2G of a node but also take into account that |
| 318 | * two large system hashes that can take up 1GB for 0.25TB/node. |
| 319 | */ |
| 320 | max_initialise = max(2UL << (30 - PAGE_SHIFT), |
| 321 | (pgdat->node_spanned_pages >> 8)); |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 322 | |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 323 | (*nr_initialised)++; |
Li Zhang | 987b309 | 2016-03-17 14:20:16 -0700 | [diff] [blame] | 324 | if ((*nr_initialised > max_initialise) && |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 325 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
| 326 | pgdat->first_deferred_pfn = pfn; |
| 327 | return false; |
| 328 | } |
| 329 | |
| 330 | return true; |
| 331 | } |
| 332 | #else |
| 333 | static inline void reset_deferred_meminit(pg_data_t *pgdat) |
| 334 | { |
| 335 | } |
| 336 | |
| 337 | static inline bool early_page_uninitialised(unsigned long pfn) |
| 338 | { |
| 339 | return false; |
| 340 | } |
| 341 | |
| 342 | static inline bool update_defer_init(pg_data_t *pgdat, |
| 343 | unsigned long pfn, unsigned long zone_end, |
| 344 | unsigned long *nr_initialised) |
| 345 | { |
| 346 | return true; |
| 347 | } |
| 348 | #endif |
| 349 | |
Mel Gorman | 0b423ca | 2016-05-19 17:14:27 -0700 | [diff] [blame] | 350 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
| 351 | static inline unsigned long *get_pageblock_bitmap(struct page *page, |
| 352 | unsigned long pfn) |
| 353 | { |
| 354 | #ifdef CONFIG_SPARSEMEM |
| 355 | return __pfn_to_section(pfn)->pageblock_flags; |
| 356 | #else |
| 357 | return page_zone(page)->pageblock_flags; |
| 358 | #endif /* CONFIG_SPARSEMEM */ |
| 359 | } |
| 360 | |
| 361 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) |
| 362 | { |
| 363 | #ifdef CONFIG_SPARSEMEM |
| 364 | pfn &= (PAGES_PER_SECTION-1); |
| 365 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
| 366 | #else |
| 367 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); |
| 368 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
| 369 | #endif /* CONFIG_SPARSEMEM */ |
| 370 | } |
| 371 | |
| 372 | /** |
| 373 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages |
| 374 | * @page: The page within the block of interest |
| 375 | * @pfn: The target page frame number |
| 376 | * @end_bitidx: The last bit of interest to retrieve |
| 377 | * @mask: mask of bits that the caller is interested in |
| 378 | * |
| 379 | * Return: pageblock_bits flags |
| 380 | */ |
| 381 | static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page, |
| 382 | unsigned long pfn, |
| 383 | unsigned long end_bitidx, |
| 384 | unsigned long mask) |
| 385 | { |
| 386 | unsigned long *bitmap; |
| 387 | unsigned long bitidx, word_bitidx; |
| 388 | unsigned long word; |
| 389 | |
| 390 | bitmap = get_pageblock_bitmap(page, pfn); |
| 391 | bitidx = pfn_to_bitidx(page, pfn); |
| 392 | word_bitidx = bitidx / BITS_PER_LONG; |
| 393 | bitidx &= (BITS_PER_LONG-1); |
| 394 | |
| 395 | word = bitmap[word_bitidx]; |
| 396 | bitidx += end_bitidx; |
| 397 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; |
| 398 | } |
| 399 | |
| 400 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, |
| 401 | unsigned long end_bitidx, |
| 402 | unsigned long mask) |
| 403 | { |
| 404 | return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask); |
| 405 | } |
| 406 | |
| 407 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) |
| 408 | { |
| 409 | return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK); |
| 410 | } |
| 411 | |
| 412 | /** |
| 413 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages |
| 414 | * @page: The page within the block of interest |
| 415 | * @flags: The flags to set |
| 416 | * @pfn: The target page frame number |
| 417 | * @end_bitidx: The last bit of interest |
| 418 | * @mask: mask of bits that the caller is interested in |
| 419 | */ |
| 420 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, |
| 421 | unsigned long pfn, |
| 422 | unsigned long end_bitidx, |
| 423 | unsigned long mask) |
| 424 | { |
| 425 | unsigned long *bitmap; |
| 426 | unsigned long bitidx, word_bitidx; |
| 427 | unsigned long old_word, word; |
| 428 | |
| 429 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); |
| 430 | |
| 431 | bitmap = get_pageblock_bitmap(page, pfn); |
| 432 | bitidx = pfn_to_bitidx(page, pfn); |
| 433 | word_bitidx = bitidx / BITS_PER_LONG; |
| 434 | bitidx &= (BITS_PER_LONG-1); |
| 435 | |
| 436 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); |
| 437 | |
| 438 | bitidx += end_bitidx; |
| 439 | mask <<= (BITS_PER_LONG - bitidx - 1); |
| 440 | flags <<= (BITS_PER_LONG - bitidx - 1); |
| 441 | |
| 442 | word = READ_ONCE(bitmap[word_bitidx]); |
| 443 | for (;;) { |
| 444 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); |
| 445 | if (word == old_word) |
| 446 | break; |
| 447 | word = old_word; |
| 448 | } |
| 449 | } |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 450 | |
Minchan Kim | ee6f509 | 2012-07-31 16:43:50 -0700 | [diff] [blame] | 451 | void set_pageblock_migratetype(struct page *page, int migratetype) |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 452 | { |
KOSAKI Motohiro | 5d0f3f7 | 2013-11-12 15:08:18 -0800 | [diff] [blame] | 453 | if (unlikely(page_group_by_mobility_disabled && |
| 454 | migratetype < MIGRATE_PCPTYPES)) |
Mel Gorman | 49255c6 | 2009-06-16 15:31:58 -0700 | [diff] [blame] | 455 | migratetype = MIGRATE_UNMOVABLE; |
| 456 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 457 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
| 458 | PB_migrate, PB_migrate_end); |
| 459 | } |
| 460 | |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 461 | #ifdef CONFIG_DEBUG_VM |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 462 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 463 | { |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 464 | int ret = 0; |
| 465 | unsigned seq; |
| 466 | unsigned long pfn = page_to_pfn(page); |
Cody P Schafer | b5e6a5a | 2013-02-22 16:35:28 -0800 | [diff] [blame] | 467 | unsigned long sp, start_pfn; |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 468 | |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 469 | do { |
| 470 | seq = zone_span_seqbegin(zone); |
Cody P Schafer | b5e6a5a | 2013-02-22 16:35:28 -0800 | [diff] [blame] | 471 | start_pfn = zone->zone_start_pfn; |
| 472 | sp = zone->spanned_pages; |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 473 | if (!zone_spans_pfn(zone, pfn)) |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 474 | ret = 1; |
| 475 | } while (zone_span_seqretry(zone, seq)); |
| 476 | |
Cody P Schafer | b5e6a5a | 2013-02-22 16:35:28 -0800 | [diff] [blame] | 477 | if (ret) |
Dave Hansen | 613813e | 2014-06-04 16:07:27 -0700 | [diff] [blame] | 478 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
| 479 | pfn, zone_to_nid(zone), zone->name, |
| 480 | start_pfn, start_pfn + sp); |
Cody P Schafer | b5e6a5a | 2013-02-22 16:35:28 -0800 | [diff] [blame] | 481 | |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 482 | return ret; |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 483 | } |
| 484 | |
| 485 | static int page_is_consistent(struct zone *zone, struct page *page) |
| 486 | { |
Andy Whitcroft | 14e0729 | 2007-05-06 14:49:14 -0700 | [diff] [blame] | 487 | if (!pfn_valid_within(page_to_pfn(page))) |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 488 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 489 | if (zone != page_zone(page)) |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 490 | return 0; |
| 491 | |
| 492 | return 1; |
| 493 | } |
| 494 | /* |
| 495 | * Temporary debugging check for pages not lying within a given zone. |
| 496 | */ |
| 497 | static int bad_range(struct zone *zone, struct page *page) |
| 498 | { |
| 499 | if (page_outside_zone_boundaries(zone, page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 500 | return 1; |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 501 | if (!page_is_consistent(zone, page)) |
| 502 | return 1; |
| 503 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 504 | return 0; |
| 505 | } |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 506 | #else |
| 507 | static inline int bad_range(struct zone *zone, struct page *page) |
| 508 | { |
| 509 | return 0; |
| 510 | } |
| 511 | #endif |
| 512 | |
Kirill A. Shutemov | d230dec | 2014-04-07 15:37:38 -0700 | [diff] [blame] | 513 | static void bad_page(struct page *page, const char *reason, |
| 514 | unsigned long bad_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 515 | { |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 516 | static unsigned long resume; |
| 517 | static unsigned long nr_shown; |
| 518 | static unsigned long nr_unshown; |
| 519 | |
| 520 | /* |
| 521 | * Allow a burst of 60 reports, then keep quiet for that minute; |
| 522 | * or allow a steady drip of one report per second. |
| 523 | */ |
| 524 | if (nr_shown == 60) { |
| 525 | if (time_before(jiffies, resume)) { |
| 526 | nr_unshown++; |
| 527 | goto out; |
| 528 | } |
| 529 | if (nr_unshown) { |
Vlastimil Babka | ff8e811 | 2016-03-15 14:56:24 -0700 | [diff] [blame] | 530 | pr_alert( |
Hugh Dickins | 1e9e636 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 531 | "BUG: Bad page state: %lu messages suppressed\n", |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 532 | nr_unshown); |
| 533 | nr_unshown = 0; |
| 534 | } |
| 535 | nr_shown = 0; |
| 536 | } |
| 537 | if (nr_shown++ == 0) |
| 538 | resume = jiffies + 60 * HZ; |
| 539 | |
Vlastimil Babka | ff8e811 | 2016-03-15 14:56:24 -0700 | [diff] [blame] | 540 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 541 | current->comm, page_to_pfn(page)); |
Vlastimil Babka | ff8e811 | 2016-03-15 14:56:24 -0700 | [diff] [blame] | 542 | __dump_page(page, reason); |
| 543 | bad_flags &= page->flags; |
| 544 | if (bad_flags) |
| 545 | pr_alert("bad because of flags: %#lx(%pGp)\n", |
| 546 | bad_flags, &bad_flags); |
Vlastimil Babka | 4e46211 | 2016-03-15 14:56:21 -0700 | [diff] [blame] | 547 | dump_page_owner(page); |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 548 | |
Dave Jones | 4f31888 | 2011-10-31 17:07:24 -0700 | [diff] [blame] | 549 | print_modules(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 550 | dump_stack(); |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 551 | out: |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 552 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
Mel Gorman | 22b751c | 2013-02-22 16:34:59 -0800 | [diff] [blame] | 553 | page_mapcount_reset(page); /* remove PageBuddy */ |
Rusty Russell | 373d4d0 | 2013-01-21 17:17:39 +1030 | [diff] [blame] | 554 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 555 | } |
| 556 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 557 | /* |
| 558 | * Higher-order pages are called "compound pages". They are structured thusly: |
| 559 | * |
Kirill A. Shutemov | 1d798ca | 2015-11-06 16:29:54 -0800 | [diff] [blame] | 560 | * 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] | 561 | * |
Kirill A. Shutemov | 1d798ca | 2015-11-06 16:29:54 -0800 | [diff] [blame] | 562 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
| 563 | * 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] | 564 | * |
Kirill A. Shutemov | 1d798ca | 2015-11-06 16:29:54 -0800 | [diff] [blame] | 565 | * The first tail page's ->compound_dtor holds the offset in array of compound |
| 566 | * page destructors. See compound_page_dtors. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 567 | * |
Kirill A. Shutemov | 1d798ca | 2015-11-06 16:29:54 -0800 | [diff] [blame] | 568 | * The first tail page's ->compound_order holds the order of allocation. |
Hugh Dickins | 41d78ba | 2006-02-14 13:52:58 -0800 | [diff] [blame] | 569 | * This usage means that zero-order pages may not be compound. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 570 | */ |
Hugh Dickins | d98c7a0 | 2006-02-14 13:52:59 -0800 | [diff] [blame] | 571 | |
Kirill A. Shutemov | 9a98225 | 2016-01-15 16:54:17 -0800 | [diff] [blame] | 572 | void free_compound_page(struct page *page) |
Hugh Dickins | d98c7a0 | 2006-02-14 13:52:59 -0800 | [diff] [blame] | 573 | { |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 574 | __free_pages_ok(page, compound_order(page)); |
Hugh Dickins | d98c7a0 | 2006-02-14 13:52:59 -0800 | [diff] [blame] | 575 | } |
| 576 | |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 577 | void prep_compound_page(struct page *page, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 578 | { |
| 579 | int i; |
| 580 | int nr_pages = 1 << order; |
| 581 | |
Kirill A. Shutemov | f1e6155 | 2015-11-06 16:29:50 -0800 | [diff] [blame] | 582 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 583 | set_compound_order(page, order); |
Christoph Lameter | 6d77795 | 2007-05-06 14:49:40 -0700 | [diff] [blame] | 584 | __SetPageHead(page); |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 585 | for (i = 1; i < nr_pages; i++) { |
| 586 | struct page *p = page + i; |
Youquan Song | 58a84aa | 2011-12-08 14:34:18 -0800 | [diff] [blame] | 587 | set_page_count(p, 0); |
Kirill A. Shutemov | 1c290f6 | 2016-01-15 16:52:07 -0800 | [diff] [blame] | 588 | p->mapping = TAIL_MAPPING; |
Kirill A. Shutemov | 1d798ca | 2015-11-06 16:29:54 -0800 | [diff] [blame] | 589 | set_compound_head(p, page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 590 | } |
Kirill A. Shutemov | 53f9263 | 2016-01-15 16:53:42 -0800 | [diff] [blame] | 591 | atomic_set(compound_mapcount_ptr(page), -1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 592 | } |
| 593 | |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 594 | #ifdef CONFIG_DEBUG_PAGEALLOC |
| 595 | unsigned int _debug_guardpage_minorder; |
Christian Borntraeger | ea6eabb | 2016-03-15 14:55:30 -0700 | [diff] [blame] | 596 | bool _debug_pagealloc_enabled __read_mostly |
| 597 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); |
Joonsoo Kim | 505f6d2 | 2016-03-17 14:17:56 -0700 | [diff] [blame] | 598 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 599 | bool _debug_guardpage_enabled __read_mostly; |
| 600 | |
Joonsoo Kim | 031bc57 | 2014-12-12 16:55:52 -0800 | [diff] [blame] | 601 | static int __init early_debug_pagealloc(char *buf) |
| 602 | { |
| 603 | if (!buf) |
| 604 | return -EINVAL; |
Minfei Huang | 2a138dc | 2016-05-20 16:58:13 -0700 | [diff] [blame] | 605 | return kstrtobool(buf, &_debug_pagealloc_enabled); |
Joonsoo Kim | 031bc57 | 2014-12-12 16:55:52 -0800 | [diff] [blame] | 606 | } |
| 607 | early_param("debug_pagealloc", early_debug_pagealloc); |
| 608 | |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 609 | static bool need_debug_guardpage(void) |
| 610 | { |
Joonsoo Kim | 031bc57 | 2014-12-12 16:55:52 -0800 | [diff] [blame] | 611 | /* If we don't use debug_pagealloc, we don't need guard page */ |
| 612 | if (!debug_pagealloc_enabled()) |
| 613 | return false; |
| 614 | |
Joonsoo Kim | f1c1e9f | 2016-10-07 16:58:18 -0700 | [diff] [blame] | 615 | if (!debug_guardpage_minorder()) |
| 616 | return false; |
| 617 | |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 618 | return true; |
| 619 | } |
| 620 | |
| 621 | static void init_debug_guardpage(void) |
| 622 | { |
Joonsoo Kim | 031bc57 | 2014-12-12 16:55:52 -0800 | [diff] [blame] | 623 | if (!debug_pagealloc_enabled()) |
| 624 | return; |
| 625 | |
Joonsoo Kim | f1c1e9f | 2016-10-07 16:58:18 -0700 | [diff] [blame] | 626 | if (!debug_guardpage_minorder()) |
| 627 | return; |
| 628 | |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 629 | _debug_guardpage_enabled = true; |
| 630 | } |
| 631 | |
| 632 | struct page_ext_operations debug_guardpage_ops = { |
| 633 | .need = need_debug_guardpage, |
| 634 | .init = init_debug_guardpage, |
| 635 | }; |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 636 | |
| 637 | static int __init debug_guardpage_minorder_setup(char *buf) |
| 638 | { |
| 639 | unsigned long res; |
| 640 | |
| 641 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 642 | pr_err("Bad debug_guardpage_minorder value\n"); |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 643 | return 0; |
| 644 | } |
| 645 | _debug_guardpage_minorder = res; |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 646 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 647 | return 0; |
| 648 | } |
Joonsoo Kim | f1c1e9f | 2016-10-07 16:58:18 -0700 | [diff] [blame] | 649 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 650 | |
Joonsoo Kim | acbc15a | 2016-10-07 16:58:15 -0700 | [diff] [blame] | 651 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 652 | unsigned int order, int migratetype) |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 653 | { |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 654 | struct page_ext *page_ext; |
| 655 | |
| 656 | if (!debug_guardpage_enabled()) |
Joonsoo Kim | acbc15a | 2016-10-07 16:58:15 -0700 | [diff] [blame] | 657 | return false; |
| 658 | |
| 659 | if (order >= debug_guardpage_minorder()) |
| 660 | return false; |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 661 | |
| 662 | page_ext = lookup_page_ext(page); |
Yang Shi | f86e427 | 2016-06-03 14:55:38 -0700 | [diff] [blame] | 663 | if (unlikely(!page_ext)) |
Joonsoo Kim | acbc15a | 2016-10-07 16:58:15 -0700 | [diff] [blame] | 664 | return false; |
Yang Shi | f86e427 | 2016-06-03 14:55:38 -0700 | [diff] [blame] | 665 | |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 666 | __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
| 667 | |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 668 | INIT_LIST_HEAD(&page->lru); |
| 669 | set_page_private(page, order); |
| 670 | /* Guard pages are not available for any usage */ |
| 671 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); |
Joonsoo Kim | acbc15a | 2016-10-07 16:58:15 -0700 | [diff] [blame] | 672 | |
| 673 | return true; |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 674 | } |
| 675 | |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 676 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
| 677 | unsigned int order, int migratetype) |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 678 | { |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 679 | struct page_ext *page_ext; |
| 680 | |
| 681 | if (!debug_guardpage_enabled()) |
| 682 | return; |
| 683 | |
| 684 | page_ext = lookup_page_ext(page); |
Yang Shi | f86e427 | 2016-06-03 14:55:38 -0700 | [diff] [blame] | 685 | if (unlikely(!page_ext)) |
| 686 | return; |
| 687 | |
Joonsoo Kim | e30825f | 2014-12-12 16:55:49 -0800 | [diff] [blame] | 688 | __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
| 689 | |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 690 | set_page_private(page, 0); |
| 691 | if (!is_migrate_isolate(migratetype)) |
| 692 | __mod_zone_freepage_state(zone, (1 << order), migratetype); |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 693 | } |
| 694 | #else |
Joonsoo Kim | 980ac16 | 2016-10-07 16:58:27 -0700 | [diff] [blame] | 695 | struct page_ext_operations debug_guardpage_ops; |
Joonsoo Kim | acbc15a | 2016-10-07 16:58:15 -0700 | [diff] [blame] | 696 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
| 697 | unsigned int order, int migratetype) { return false; } |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 698 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
| 699 | unsigned int order, int migratetype) {} |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 700 | #endif |
| 701 | |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 702 | static inline void set_page_order(struct page *page, unsigned int order) |
Andrew Morton | 6aa3001b2 | 2006-04-18 22:20:52 -0700 | [diff] [blame] | 703 | { |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 704 | set_page_private(page, order); |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 705 | __SetPageBuddy(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 706 | } |
| 707 | |
| 708 | static inline void rmv_page_order(struct page *page) |
| 709 | { |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 710 | __ClearPageBuddy(page); |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 711 | set_page_private(page, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 712 | } |
| 713 | |
| 714 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 715 | * This function checks whether a page is free && is the buddy |
| 716 | * we can do coalesce a page and its buddy if |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 717 | * (a) the buddy is not in a hole && |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 718 | * (b) the buddy is in the buddy system && |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 719 | * (c) a page and its buddy have the same order && |
| 720 | * (d) a page and its buddy are in the same zone. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 721 | * |
Wang Sheng-Hui | cf6fe94 | 2013-09-11 14:22:48 -0700 | [diff] [blame] | 722 | * For recording whether a page is in the buddy system, we set ->_mapcount |
| 723 | * PAGE_BUDDY_MAPCOUNT_VALUE. |
| 724 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is |
| 725 | * serialized by zone->lock. |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 726 | * |
| 727 | * For recording page's order, we use page_private(page). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 728 | */ |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 729 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 730 | unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 731 | { |
Andy Whitcroft | 14e0729 | 2007-05-06 14:49:14 -0700 | [diff] [blame] | 732 | if (!pfn_valid_within(page_to_pfn(buddy))) |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 733 | return 0; |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 734 | |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 735 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
Mel Gorman | d34c5fa | 2014-06-04 16:10:10 -0700 | [diff] [blame] | 736 | if (page_zone_id(page) != page_zone_id(buddy)) |
| 737 | return 0; |
| 738 | |
Weijie Yang | 4c5018c | 2015-02-10 14:11:39 -0800 | [diff] [blame] | 739 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
| 740 | |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 741 | return 1; |
| 742 | } |
| 743 | |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 744 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
Mel Gorman | d34c5fa | 2014-06-04 16:10:10 -0700 | [diff] [blame] | 745 | /* |
| 746 | * zone check is done late to avoid uselessly |
| 747 | * calculating zone/node ids for pages that could |
| 748 | * never merge. |
| 749 | */ |
| 750 | if (page_zone_id(page) != page_zone_id(buddy)) |
| 751 | return 0; |
| 752 | |
Weijie Yang | 4c5018c | 2015-02-10 14:11:39 -0800 | [diff] [blame] | 753 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
| 754 | |
Andrew Morton | 6aa3001b2 | 2006-04-18 22:20:52 -0700 | [diff] [blame] | 755 | return 1; |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 756 | } |
Andrew Morton | 6aa3001b2 | 2006-04-18 22:20:52 -0700 | [diff] [blame] | 757 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 758 | } |
| 759 | |
| 760 | /* |
| 761 | * Freeing function for a buddy system allocator. |
| 762 | * |
| 763 | * The concept of a buddy system is to maintain direct-mapped table |
| 764 | * (containing bit values) for memory blocks of various "orders". |
| 765 | * The bottom level table contains the map for the smallest allocatable |
| 766 | * units of memory (here, pages), and each level above it describes |
| 767 | * pairs of units from the levels below, hence, "buddies". |
| 768 | * At a high level, all that happens here is marking the table entry |
| 769 | * at the bottom level available, and propagating the changes upward |
| 770 | * as necessary, plus some accounting needed to play nicely with other |
| 771 | * parts of the VM system. |
| 772 | * 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] | 773 | * free pages of length of (1 << order) and marked with _mapcount |
| 774 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) |
| 775 | * field. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 776 | * 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] | 777 | * other. That is, if we allocate a small block, and both were |
| 778 | * free, the remainder of the region must be split into blocks. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 779 | * 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] | 780 | * triggers coalescing into a block of larger size. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 781 | * |
Nadia Yvette Chambers | 6d49e35 | 2012-12-06 10:39:54 +0100 | [diff] [blame] | 782 | * -- nyc |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 783 | */ |
| 784 | |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 785 | static inline void __free_one_page(struct page *page, |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 786 | unsigned long pfn, |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 787 | struct zone *zone, unsigned int order, |
| 788 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 789 | { |
| 790 | unsigned long page_idx; |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 791 | unsigned long combined_idx; |
KyongHo Cho | 43506fa | 2011-01-13 15:47:24 -0800 | [diff] [blame] | 792 | unsigned long uninitialized_var(buddy_idx); |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 793 | struct page *buddy; |
Vlastimil Babka | d9dddbf | 2016-03-25 14:21:50 -0700 | [diff] [blame] | 794 | unsigned int max_order; |
| 795 | |
| 796 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | |
Cody P Schafer | d29bb97 | 2013-02-22 16:35:25 -0800 | [diff] [blame] | 798 | VM_BUG_ON(!zone_is_initialized(zone)); |
Kirill A. Shutemov | 6e9f0d5 | 2015-02-11 15:25:50 -0800 | [diff] [blame] | 799 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 800 | |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 801 | VM_BUG_ON(migratetype == -1); |
Vlastimil Babka | d9dddbf | 2016-03-25 14:21:50 -0700 | [diff] [blame] | 802 | if (likely(!is_migrate_isolate(migratetype))) |
Joonsoo Kim | 8f82b55 | 2014-11-13 15:19:18 -0800 | [diff] [blame] | 803 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 804 | |
Vlastimil Babka | d9dddbf | 2016-03-25 14:21:50 -0700 | [diff] [blame] | 805 | page_idx = pfn & ((1 << MAX_ORDER) - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 806 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 807 | VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page); |
| 808 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 809 | |
Vlastimil Babka | d9dddbf | 2016-03-25 14:21:50 -0700 | [diff] [blame] | 810 | continue_merging: |
Joonsoo Kim | 3c60509 | 2014-11-13 15:19:21 -0800 | [diff] [blame] | 811 | while (order < max_order - 1) { |
KyongHo Cho | 43506fa | 2011-01-13 15:47:24 -0800 | [diff] [blame] | 812 | buddy_idx = __find_buddy_index(page_idx, order); |
| 813 | buddy = page + (buddy_idx - page_idx); |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 814 | if (!page_is_buddy(page, buddy, order)) |
Vlastimil Babka | d9dddbf | 2016-03-25 14:21:50 -0700 | [diff] [blame] | 815 | goto done_merging; |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 816 | /* |
| 817 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, |
| 818 | * merge with it and move up one order. |
| 819 | */ |
| 820 | if (page_is_guard(buddy)) { |
Joonsoo Kim | 2847cf9 | 2014-12-12 16:55:01 -0800 | [diff] [blame] | 821 | clear_page_guard(zone, buddy, order, migratetype); |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 822 | } else { |
| 823 | list_del(&buddy->lru); |
| 824 | zone->free_area[order].nr_free--; |
| 825 | rmv_page_order(buddy); |
| 826 | } |
KyongHo Cho | 43506fa | 2011-01-13 15:47:24 -0800 | [diff] [blame] | 827 | combined_idx = buddy_idx & page_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 828 | page = page + (combined_idx - page_idx); |
| 829 | page_idx = combined_idx; |
| 830 | order++; |
| 831 | } |
Vlastimil Babka | d9dddbf | 2016-03-25 14:21:50 -0700 | [diff] [blame] | 832 | if (max_order < MAX_ORDER) { |
| 833 | /* If we are here, it means order is >= pageblock_order. |
| 834 | * We want to prevent merge between freepages on isolate |
| 835 | * pageblock and normal pageblock. Without this, pageblock |
| 836 | * isolation could cause incorrect freepage or CMA accounting. |
| 837 | * |
| 838 | * We don't want to hit this code for the more frequent |
| 839 | * low-order merging. |
| 840 | */ |
| 841 | if (unlikely(has_isolate_pageblock(zone))) { |
| 842 | int buddy_mt; |
| 843 | |
| 844 | buddy_idx = __find_buddy_index(page_idx, order); |
| 845 | buddy = page + (buddy_idx - page_idx); |
| 846 | buddy_mt = get_pageblock_migratetype(buddy); |
| 847 | |
| 848 | if (migratetype != buddy_mt |
| 849 | && (is_migrate_isolate(migratetype) || |
| 850 | is_migrate_isolate(buddy_mt))) |
| 851 | goto done_merging; |
| 852 | } |
| 853 | max_order++; |
| 854 | goto continue_merging; |
| 855 | } |
| 856 | |
| 857 | done_merging: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 858 | set_page_order(page, order); |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 859 | |
| 860 | /* |
| 861 | * If this is not the largest possible page, check if the buddy |
| 862 | * of the next-highest order is free. If it is, it's possible |
| 863 | * that pages are being freed that will coalesce soon. In case, |
| 864 | * that is happening, add the free page to the tail of the list |
| 865 | * so it's less likely to be used soon and more likely to be merged |
| 866 | * as a higher order page |
| 867 | */ |
Mel Gorman | b7f50cf | 2010-10-26 14:21:11 -0700 | [diff] [blame] | 868 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 869 | struct page *higher_page, *higher_buddy; |
KyongHo Cho | 43506fa | 2011-01-13 15:47:24 -0800 | [diff] [blame] | 870 | combined_idx = buddy_idx & page_idx; |
| 871 | higher_page = page + (combined_idx - page_idx); |
| 872 | buddy_idx = __find_buddy_index(combined_idx, order + 1); |
Li Haifeng | 0ba8f2d | 2012-09-17 14:09:21 -0700 | [diff] [blame] | 873 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 874 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
| 875 | list_add_tail(&page->lru, |
| 876 | &zone->free_area[order].free_list[migratetype]); |
| 877 | goto out; |
| 878 | } |
| 879 | } |
| 880 | |
| 881 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); |
| 882 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 883 | zone->free_area[order].nr_free++; |
| 884 | } |
| 885 | |
Mel Gorman | 7bfec6f | 2016-05-19 17:14:15 -0700 | [diff] [blame] | 886 | /* |
| 887 | * A bad page could be due to a number of fields. Instead of multiple branches, |
| 888 | * try and check multiple fields with one check. The caller must do a detailed |
| 889 | * check if necessary. |
| 890 | */ |
| 891 | static inline bool page_expected_state(struct page *page, |
| 892 | unsigned long check_flags) |
| 893 | { |
| 894 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
| 895 | return false; |
| 896 | |
| 897 | if (unlikely((unsigned long)page->mapping | |
| 898 | page_ref_count(page) | |
| 899 | #ifdef CONFIG_MEMCG |
| 900 | (unsigned long)page->mem_cgroup | |
| 901 | #endif |
| 902 | (page->flags & check_flags))) |
| 903 | return false; |
| 904 | |
| 905 | return true; |
| 906 | } |
| 907 | |
Mel Gorman | bb552ac | 2016-05-19 17:14:18 -0700 | [diff] [blame] | 908 | static void free_pages_check_bad(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 909 | { |
Mel Gorman | 7bfec6f | 2016-05-19 17:14:15 -0700 | [diff] [blame] | 910 | const char *bad_reason; |
| 911 | unsigned long bad_flags; |
| 912 | |
Mel Gorman | 7bfec6f | 2016-05-19 17:14:15 -0700 | [diff] [blame] | 913 | bad_reason = NULL; |
| 914 | bad_flags = 0; |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 915 | |
Kirill A. Shutemov | 53f9263 | 2016-01-15 16:53:42 -0800 | [diff] [blame] | 916 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 917 | bad_reason = "nonzero mapcount"; |
| 918 | if (unlikely(page->mapping != NULL)) |
| 919 | bad_reason = "non-NULL mapping"; |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 920 | if (unlikely(page_ref_count(page) != 0)) |
Joonsoo Kim | 0139aa7 | 2016-05-19 17:10:49 -0700 | [diff] [blame] | 921 | bad_reason = "nonzero _refcount"; |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 922 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { |
| 923 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; |
| 924 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; |
| 925 | } |
Johannes Weiner | 9edad6e | 2014-12-10 15:44:58 -0800 | [diff] [blame] | 926 | #ifdef CONFIG_MEMCG |
| 927 | if (unlikely(page->mem_cgroup)) |
| 928 | bad_reason = "page still charged to cgroup"; |
| 929 | #endif |
Mel Gorman | 7bfec6f | 2016-05-19 17:14:15 -0700 | [diff] [blame] | 930 | bad_page(page, bad_reason, bad_flags); |
Mel Gorman | bb552ac | 2016-05-19 17:14:18 -0700 | [diff] [blame] | 931 | } |
| 932 | |
| 933 | static inline int free_pages_check(struct page *page) |
| 934 | { |
Mel Gorman | da838d4 | 2016-05-19 17:14:21 -0700 | [diff] [blame] | 935 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
Mel Gorman | bb552ac | 2016-05-19 17:14:18 -0700 | [diff] [blame] | 936 | return 0; |
Mel Gorman | bb552ac | 2016-05-19 17:14:18 -0700 | [diff] [blame] | 937 | |
| 938 | /* Something has gone sideways, find it */ |
| 939 | free_pages_check_bad(page); |
Mel Gorman | 7bfec6f | 2016-05-19 17:14:15 -0700 | [diff] [blame] | 940 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 941 | } |
| 942 | |
Mel Gorman | 4db7548 | 2016-05-19 17:14:32 -0700 | [diff] [blame] | 943 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
| 944 | { |
| 945 | int ret = 1; |
| 946 | |
| 947 | /* |
| 948 | * We rely page->lru.next never has bit 0 set, unless the page |
| 949 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. |
| 950 | */ |
| 951 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); |
| 952 | |
| 953 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { |
| 954 | ret = 0; |
| 955 | goto out; |
| 956 | } |
| 957 | switch (page - head_page) { |
| 958 | case 1: |
| 959 | /* the first tail page: ->mapping is compound_mapcount() */ |
| 960 | if (unlikely(compound_mapcount(page))) { |
| 961 | bad_page(page, "nonzero compound_mapcount", 0); |
| 962 | goto out; |
| 963 | } |
| 964 | break; |
| 965 | case 2: |
| 966 | /* |
| 967 | * the second tail page: ->mapping is |
| 968 | * page_deferred_list().next -- ignore value. |
| 969 | */ |
| 970 | break; |
| 971 | default: |
| 972 | if (page->mapping != TAIL_MAPPING) { |
| 973 | bad_page(page, "corrupted mapping in tail page", 0); |
| 974 | goto out; |
| 975 | } |
| 976 | break; |
| 977 | } |
| 978 | if (unlikely(!PageTail(page))) { |
| 979 | bad_page(page, "PageTail not set", 0); |
| 980 | goto out; |
| 981 | } |
| 982 | if (unlikely(compound_head(page) != head_page)) { |
| 983 | bad_page(page, "compound_head not consistent", 0); |
| 984 | goto out; |
| 985 | } |
| 986 | ret = 0; |
| 987 | out: |
| 988 | page->mapping = NULL; |
| 989 | clear_compound_head(page); |
| 990 | return ret; |
| 991 | } |
| 992 | |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 993 | static __always_inline bool free_pages_prepare(struct page *page, |
| 994 | unsigned int order, bool check_free) |
| 995 | { |
| 996 | int bad = 0; |
| 997 | |
| 998 | VM_BUG_ON_PAGE(PageTail(page), page); |
| 999 | |
| 1000 | trace_mm_page_free(page, order); |
| 1001 | kmemcheck_free_shadow(page, order); |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 1002 | |
| 1003 | /* |
| 1004 | * Check tail pages before head page information is cleared to |
| 1005 | * avoid checking PageCompound for order-0 pages. |
| 1006 | */ |
| 1007 | if (unlikely(order)) { |
| 1008 | bool compound = PageCompound(page); |
| 1009 | int i; |
| 1010 | |
| 1011 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); |
| 1012 | |
Kirill A. Shutemov | 9a73f61 | 2016-07-26 15:25:53 -0700 | [diff] [blame] | 1013 | if (compound) |
| 1014 | ClearPageDoubleMap(page); |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 1015 | for (i = 1; i < (1 << order); i++) { |
| 1016 | if (compound) |
| 1017 | bad += free_tail_pages_check(page, page + i); |
| 1018 | if (unlikely(free_pages_check(page + i))) { |
| 1019 | bad++; |
| 1020 | continue; |
| 1021 | } |
| 1022 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; |
| 1023 | } |
| 1024 | } |
Minchan Kim | bda807d | 2016-07-26 15:23:05 -0700 | [diff] [blame] | 1025 | if (PageMappingFlags(page)) |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 1026 | page->mapping = NULL; |
Vladimir Davydov | c4159a7 | 2016-08-08 23:03:12 +0300 | [diff] [blame] | 1027 | if (memcg_kmem_enabled() && PageKmemcg(page)) |
Vladimir Davydov | 4949148 | 2016-07-26 15:24:24 -0700 | [diff] [blame] | 1028 | memcg_kmem_uncharge(page, order); |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 1029 | if (check_free) |
| 1030 | bad += free_pages_check(page); |
| 1031 | if (bad) |
| 1032 | return false; |
| 1033 | |
| 1034 | page_cpupid_reset_last(page); |
| 1035 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; |
| 1036 | reset_page_owner(page, order); |
| 1037 | |
| 1038 | if (!PageHighMem(page)) { |
| 1039 | debug_check_no_locks_freed(page_address(page), |
| 1040 | PAGE_SIZE << order); |
| 1041 | debug_check_no_obj_freed(page_address(page), |
| 1042 | PAGE_SIZE << order); |
| 1043 | } |
| 1044 | arch_free_page(page, order); |
| 1045 | kernel_poison_pages(page, 1 << order, 0); |
| 1046 | kernel_map_pages(page, 1 << order, 0); |
seokhoon.yoon | 29b52de | 2016-05-20 16:58:47 -0700 | [diff] [blame] | 1047 | kasan_free_pages(page, order); |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 1048 | |
| 1049 | return true; |
| 1050 | } |
Mel Gorman | 4db7548 | 2016-05-19 17:14:32 -0700 | [diff] [blame] | 1051 | |
| 1052 | #ifdef CONFIG_DEBUG_VM |
| 1053 | static inline bool free_pcp_prepare(struct page *page) |
| 1054 | { |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 1055 | return free_pages_prepare(page, 0, true); |
Mel Gorman | 4db7548 | 2016-05-19 17:14:32 -0700 | [diff] [blame] | 1056 | } |
| 1057 | |
| 1058 | static inline bool bulkfree_pcp_prepare(struct page *page) |
| 1059 | { |
| 1060 | return false; |
| 1061 | } |
| 1062 | #else |
| 1063 | static bool free_pcp_prepare(struct page *page) |
| 1064 | { |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 1065 | return free_pages_prepare(page, 0, false); |
Mel Gorman | 4db7548 | 2016-05-19 17:14:32 -0700 | [diff] [blame] | 1066 | } |
| 1067 | |
| 1068 | static bool bulkfree_pcp_prepare(struct page *page) |
| 1069 | { |
| 1070 | return free_pages_check(page); |
| 1071 | } |
| 1072 | #endif /* CONFIG_DEBUG_VM */ |
| 1073 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1074 | /* |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1075 | * Frees a number of pages from the PCP lists |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1076 | * 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] | 1077 | * count is the number of pages to free. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1078 | * |
| 1079 | * If the zone was previously in an "all pages pinned" state then look to |
| 1080 | * see if this freeing clears that state. |
| 1081 | * |
| 1082 | * And clear the zone's pages_scanned counter, to hold off the "all pages are |
| 1083 | * pinned" detection logic. |
| 1084 | */ |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1085 | static void free_pcppages_bulk(struct zone *zone, int count, |
| 1086 | struct per_cpu_pages *pcp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1087 | { |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1088 | int migratetype = 0; |
Mel Gorman | a6f9edd6 | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 1089 | int batch_free = 0; |
Mel Gorman | 0d5d823 | 2014-08-06 16:07:16 -0700 | [diff] [blame] | 1090 | unsigned long nr_scanned; |
Mel Gorman | 3777999 | 2016-05-19 17:13:58 -0700 | [diff] [blame] | 1091 | bool isolated_pageblocks; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1092 | |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 1093 | spin_lock(&zone->lock); |
Mel Gorman | 3777999 | 2016-05-19 17:13:58 -0700 | [diff] [blame] | 1094 | isolated_pageblocks = has_isolate_pageblock(zone); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1095 | nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED); |
Mel Gorman | 0d5d823 | 2014-08-06 16:07:16 -0700 | [diff] [blame] | 1096 | if (nr_scanned) |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1097 | __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned); |
Mel Gorman | f2260e6 | 2009-06-16 15:32:13 -0700 | [diff] [blame] | 1098 | |
Mel Gorman | e5b31ac | 2016-05-19 17:14:24 -0700 | [diff] [blame] | 1099 | while (count) { |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 1100 | struct page *page; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1101 | struct list_head *list; |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 1102 | |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1103 | /* |
Mel Gorman | a6f9edd6 | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 1104 | * Remove pages from lists in a round-robin fashion. A |
| 1105 | * batch_free count is maintained that is incremented when an |
| 1106 | * empty list is encountered. This is so more pages are freed |
| 1107 | * off fuller lists instead of spinning excessively around empty |
| 1108 | * lists |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1109 | */ |
| 1110 | do { |
Mel Gorman | a6f9edd6 | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 1111 | batch_free++; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1112 | if (++migratetype == MIGRATE_PCPTYPES) |
| 1113 | migratetype = 0; |
| 1114 | list = &pcp->lists[migratetype]; |
| 1115 | } while (list_empty(list)); |
| 1116 | |
Namhyung Kim | 1d16871 | 2011-03-22 16:32:45 -0700 | [diff] [blame] | 1117 | /* This is the only non-empty list. Free them all. */ |
| 1118 | if (batch_free == MIGRATE_PCPTYPES) |
Mel Gorman | e5b31ac | 2016-05-19 17:14:24 -0700 | [diff] [blame] | 1119 | batch_free = count; |
Namhyung Kim | 1d16871 | 2011-03-22 16:32:45 -0700 | [diff] [blame] | 1120 | |
Mel Gorman | a6f9edd6 | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 1121 | do { |
Bartlomiej Zolnierkiewicz | 770c8aa | 2012-10-08 16:31:57 -0700 | [diff] [blame] | 1122 | int mt; /* migratetype of the to-be-freed page */ |
| 1123 | |
Geliang Tang | a16601c | 2016-01-14 15:20:30 -0800 | [diff] [blame] | 1124 | page = list_last_entry(list, struct page, lru); |
Mel Gorman | a6f9edd6 | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 1125 | /* must delete as __free_one_page list manipulates */ |
| 1126 | list_del(&page->lru); |
Vlastimil Babka | aa016d1 | 2015-09-08 15:01:22 -0700 | [diff] [blame] | 1127 | |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 1128 | mt = get_pcppage_migratetype(page); |
Vlastimil Babka | aa016d1 | 2015-09-08 15:01:22 -0700 | [diff] [blame] | 1129 | /* MIGRATE_ISOLATE page should not go to pcplists */ |
| 1130 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); |
| 1131 | /* Pageblock could have been isolated meanwhile */ |
Mel Gorman | 3777999 | 2016-05-19 17:13:58 -0700 | [diff] [blame] | 1132 | if (unlikely(isolated_pageblocks)) |
Joonsoo Kim | 51bb1a4 | 2014-11-13 15:19:14 -0800 | [diff] [blame] | 1133 | mt = get_pageblock_migratetype(page); |
Joonsoo Kim | 51bb1a4 | 2014-11-13 15:19:14 -0800 | [diff] [blame] | 1134 | |
Mel Gorman | 4db7548 | 2016-05-19 17:14:32 -0700 | [diff] [blame] | 1135 | if (bulkfree_pcp_prepare(page)) |
| 1136 | continue; |
| 1137 | |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 1138 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); |
Bartlomiej Zolnierkiewicz | 770c8aa | 2012-10-08 16:31:57 -0700 | [diff] [blame] | 1139 | trace_mm_page_pcpu_drain(page, 0, mt); |
Mel Gorman | e5b31ac | 2016-05-19 17:14:24 -0700 | [diff] [blame] | 1140 | } while (--count && --batch_free && !list_empty(list)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1141 | } |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 1142 | spin_unlock(&zone->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1143 | } |
| 1144 | |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 1145 | static void free_one_page(struct zone *zone, |
| 1146 | struct page *page, unsigned long pfn, |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 1147 | unsigned int order, |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 1148 | int migratetype) |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 1149 | { |
Mel Gorman | 0d5d823 | 2014-08-06 16:07:16 -0700 | [diff] [blame] | 1150 | unsigned long nr_scanned; |
Christoph Lameter | 006d22d | 2006-09-25 23:31:48 -0700 | [diff] [blame] | 1151 | spin_lock(&zone->lock); |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1152 | nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED); |
Mel Gorman | 0d5d823 | 2014-08-06 16:07:16 -0700 | [diff] [blame] | 1153 | if (nr_scanned) |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 1154 | __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned); |
Mel Gorman | f2260e6 | 2009-06-16 15:32:13 -0700 | [diff] [blame] | 1155 | |
Joonsoo Kim | ad53f92 | 2014-11-13 15:19:11 -0800 | [diff] [blame] | 1156 | if (unlikely(has_isolate_pageblock(zone) || |
| 1157 | is_migrate_isolate(migratetype))) { |
| 1158 | migratetype = get_pfnblock_migratetype(page, pfn); |
Joonsoo Kim | ad53f92 | 2014-11-13 15:19:11 -0800 | [diff] [blame] | 1159 | } |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 1160 | __free_one_page(page, pfn, zone, order, migratetype); |
Christoph Lameter | 006d22d | 2006-09-25 23:31:48 -0700 | [diff] [blame] | 1161 | spin_unlock(&zone->lock); |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 1162 | } |
| 1163 | |
Robin Holt | 1e8ce83 | 2015-06-30 14:56:45 -0700 | [diff] [blame] | 1164 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
| 1165 | unsigned long zone, int nid) |
| 1166 | { |
Robin Holt | 1e8ce83 | 2015-06-30 14:56:45 -0700 | [diff] [blame] | 1167 | set_page_links(page, zone, nid, pfn); |
Robin Holt | 1e8ce83 | 2015-06-30 14:56:45 -0700 | [diff] [blame] | 1168 | init_page_count(page); |
| 1169 | page_mapcount_reset(page); |
| 1170 | page_cpupid_reset_last(page); |
Robin Holt | 1e8ce83 | 2015-06-30 14:56:45 -0700 | [diff] [blame] | 1171 | |
Robin Holt | 1e8ce83 | 2015-06-30 14:56:45 -0700 | [diff] [blame] | 1172 | INIT_LIST_HEAD(&page->lru); |
| 1173 | #ifdef WANT_PAGE_VIRTUAL |
| 1174 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ |
| 1175 | if (!is_highmem_idx(zone)) |
| 1176 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
| 1177 | #endif |
| 1178 | } |
| 1179 | |
| 1180 | static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone, |
| 1181 | int nid) |
| 1182 | { |
| 1183 | return __init_single_page(pfn_to_page(pfn), pfn, zone, nid); |
| 1184 | } |
| 1185 | |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1186 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
| 1187 | static void init_reserved_page(unsigned long pfn) |
| 1188 | { |
| 1189 | pg_data_t *pgdat; |
| 1190 | int nid, zid; |
| 1191 | |
| 1192 | if (!early_page_uninitialised(pfn)) |
| 1193 | return; |
| 1194 | |
| 1195 | nid = early_pfn_to_nid(pfn); |
| 1196 | pgdat = NODE_DATA(nid); |
| 1197 | |
| 1198 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 1199 | struct zone *zone = &pgdat->node_zones[zid]; |
| 1200 | |
| 1201 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) |
| 1202 | break; |
| 1203 | } |
| 1204 | __init_single_pfn(pfn, zid, nid); |
| 1205 | } |
| 1206 | #else |
| 1207 | static inline void init_reserved_page(unsigned long pfn) |
| 1208 | { |
| 1209 | } |
| 1210 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
| 1211 | |
Nathan Zimmer | 92923ca | 2015-06-30 14:56:48 -0700 | [diff] [blame] | 1212 | /* |
| 1213 | * Initialised pages do not have PageReserved set. This function is |
| 1214 | * called for each range allocated by the bootmem allocator and |
| 1215 | * marks the pages PageReserved. The remaining valid pages are later |
| 1216 | * sent to the buddy page allocator. |
| 1217 | */ |
Stefan Bader | 4b50bcc | 2016-05-20 16:58:38 -0700 | [diff] [blame] | 1218 | 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] | 1219 | { |
| 1220 | unsigned long start_pfn = PFN_DOWN(start); |
| 1221 | unsigned long end_pfn = PFN_UP(end); |
| 1222 | |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1223 | for (; start_pfn < end_pfn; start_pfn++) { |
| 1224 | if (pfn_valid(start_pfn)) { |
| 1225 | struct page *page = pfn_to_page(start_pfn); |
| 1226 | |
| 1227 | init_reserved_page(start_pfn); |
Kirill A. Shutemov | 1d798ca | 2015-11-06 16:29:54 -0800 | [diff] [blame] | 1228 | |
| 1229 | /* Avoid false-positive PageTail() */ |
| 1230 | INIT_LIST_HEAD(&page->lru); |
| 1231 | |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1232 | SetPageReserved(page); |
| 1233 | } |
| 1234 | } |
Nathan Zimmer | 92923ca | 2015-06-30 14:56:48 -0700 | [diff] [blame] | 1235 | } |
| 1236 | |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 1237 | static void __free_pages_ok(struct page *page, unsigned int order) |
| 1238 | { |
| 1239 | unsigned long flags; |
Minchan Kim | 95e3441 | 2012-10-08 16:32:11 -0700 | [diff] [blame] | 1240 | int migratetype; |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 1241 | unsigned long pfn = page_to_pfn(page); |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 1242 | |
Mel Gorman | e2769db | 2016-05-19 17:14:38 -0700 | [diff] [blame] | 1243 | if (!free_pages_prepare(page, order, true)) |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 1244 | return; |
| 1245 | |
Mel Gorman | cfc47a2 | 2014-06-04 16:10:19 -0700 | [diff] [blame] | 1246 | migratetype = get_pfnblock_migratetype(page, pfn); |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 1247 | local_irq_save(flags); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1248 | __count_vm_events(PGFREE, 1 << order); |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 1249 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 1250 | local_irq_restore(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1251 | } |
| 1252 | |
Li Zhang | 949698a | 2016-05-19 17:11:37 -0700 | [diff] [blame] | 1253 | 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] | 1254 | { |
Johannes Weiner | c399307 | 2012-01-10 15:08:10 -0800 | [diff] [blame] | 1255 | unsigned int nr_pages = 1 << order; |
Yinghai Lu | e2d0bd2 | 2013-09-11 14:20:37 -0700 | [diff] [blame] | 1256 | struct page *p = page; |
Johannes Weiner | c399307 | 2012-01-10 15:08:10 -0800 | [diff] [blame] | 1257 | unsigned int loop; |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 1258 | |
Yinghai Lu | e2d0bd2 | 2013-09-11 14:20:37 -0700 | [diff] [blame] | 1259 | prefetchw(p); |
| 1260 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { |
| 1261 | prefetchw(p + 1); |
Johannes Weiner | c399307 | 2012-01-10 15:08:10 -0800 | [diff] [blame] | 1262 | __ClearPageReserved(p); |
| 1263 | set_page_count(p, 0); |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 1264 | } |
Yinghai Lu | e2d0bd2 | 2013-09-11 14:20:37 -0700 | [diff] [blame] | 1265 | __ClearPageReserved(p); |
| 1266 | set_page_count(p, 0); |
Johannes Weiner | c399307 | 2012-01-10 15:08:10 -0800 | [diff] [blame] | 1267 | |
Yinghai Lu | e2d0bd2 | 2013-09-11 14:20:37 -0700 | [diff] [blame] | 1268 | page_zone(page)->managed_pages += nr_pages; |
Johannes Weiner | c399307 | 2012-01-10 15:08:10 -0800 | [diff] [blame] | 1269 | set_page_refcounted(page); |
| 1270 | __free_pages(page, order); |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 1271 | } |
| 1272 | |
Mel Gorman | 75a592a | 2015-06-30 14:56:59 -0700 | [diff] [blame] | 1273 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
| 1274 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) |
Mel Gorman | 7ace991 | 2015-08-06 15:46:13 -0700 | [diff] [blame] | 1275 | |
Mel Gorman | 75a592a | 2015-06-30 14:56:59 -0700 | [diff] [blame] | 1276 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
| 1277 | |
| 1278 | int __meminit early_pfn_to_nid(unsigned long pfn) |
| 1279 | { |
Mel Gorman | 7ace991 | 2015-08-06 15:46:13 -0700 | [diff] [blame] | 1280 | static DEFINE_SPINLOCK(early_pfn_lock); |
Mel Gorman | 75a592a | 2015-06-30 14:56:59 -0700 | [diff] [blame] | 1281 | int nid; |
| 1282 | |
Mel Gorman | 7ace991 | 2015-08-06 15:46:13 -0700 | [diff] [blame] | 1283 | spin_lock(&early_pfn_lock); |
Mel Gorman | 75a592a | 2015-06-30 14:56:59 -0700 | [diff] [blame] | 1284 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
Mel Gorman | 7ace991 | 2015-08-06 15:46:13 -0700 | [diff] [blame] | 1285 | if (nid < 0) |
Mel Gorman | e4568d3 | 2016-07-14 12:07:20 -0700 | [diff] [blame] | 1286 | nid = first_online_node; |
Mel Gorman | 7ace991 | 2015-08-06 15:46:13 -0700 | [diff] [blame] | 1287 | spin_unlock(&early_pfn_lock); |
| 1288 | |
| 1289 | return nid; |
Mel Gorman | 75a592a | 2015-06-30 14:56:59 -0700 | [diff] [blame] | 1290 | } |
| 1291 | #endif |
| 1292 | |
| 1293 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
| 1294 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, |
| 1295 | struct mminit_pfnnid_cache *state) |
| 1296 | { |
| 1297 | int nid; |
| 1298 | |
| 1299 | nid = __early_pfn_to_nid(pfn, state); |
| 1300 | if (nid >= 0 && nid != node) |
| 1301 | return false; |
| 1302 | return true; |
| 1303 | } |
| 1304 | |
| 1305 | /* Only safe to use early in boot when initialisation is single-threaded */ |
| 1306 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) |
| 1307 | { |
| 1308 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); |
| 1309 | } |
| 1310 | |
| 1311 | #else |
| 1312 | |
| 1313 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) |
| 1314 | { |
| 1315 | return true; |
| 1316 | } |
| 1317 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, |
| 1318 | struct mminit_pfnnid_cache *state) |
| 1319 | { |
| 1320 | return true; |
| 1321 | } |
| 1322 | #endif |
| 1323 | |
| 1324 | |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1325 | void __init __free_pages_bootmem(struct page *page, unsigned long pfn, |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 1326 | unsigned int order) |
| 1327 | { |
| 1328 | if (early_page_uninitialised(pfn)) |
| 1329 | return; |
Li Zhang | 949698a | 2016-05-19 17:11:37 -0700 | [diff] [blame] | 1330 | return __free_pages_boot_core(page, order); |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 1331 | } |
| 1332 | |
Joonsoo Kim | 7cf91a9 | 2016-03-15 14:57:51 -0700 | [diff] [blame] | 1333 | /* |
| 1334 | * Check that the whole (or subset of) a pageblock given by the interval of |
| 1335 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it |
| 1336 | * with the migration of free compaction scanner. The scanners then need to |
| 1337 | * use only pfn_valid_within() check for arches that allow holes within |
| 1338 | * pageblocks. |
| 1339 | * |
| 1340 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. |
| 1341 | * |
| 1342 | * It's possible on some configurations to have a setup like node0 node1 node0 |
| 1343 | * i.e. it's possible that all pages within a zones range of pages do not |
| 1344 | * belong to a single zone. We assume that a border between node0 and node1 |
| 1345 | * can occur within a single pageblock, but not a node0 node1 node0 |
| 1346 | * interleaving within a single pageblock. It is therefore sufficient to check |
| 1347 | * the first and last page of a pageblock and avoid checking each individual |
| 1348 | * page in a pageblock. |
| 1349 | */ |
| 1350 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, |
| 1351 | unsigned long end_pfn, struct zone *zone) |
| 1352 | { |
| 1353 | struct page *start_page; |
| 1354 | struct page *end_page; |
| 1355 | |
| 1356 | /* end_pfn is one past the range we are checking */ |
| 1357 | end_pfn--; |
| 1358 | |
| 1359 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) |
| 1360 | return NULL; |
| 1361 | |
| 1362 | start_page = pfn_to_page(start_pfn); |
| 1363 | |
| 1364 | if (page_zone(start_page) != zone) |
| 1365 | return NULL; |
| 1366 | |
| 1367 | end_page = pfn_to_page(end_pfn); |
| 1368 | |
| 1369 | /* This gives a shorter code than deriving page_zone(end_page) */ |
| 1370 | if (page_zone_id(start_page) != page_zone_id(end_page)) |
| 1371 | return NULL; |
| 1372 | |
| 1373 | return start_page; |
| 1374 | } |
| 1375 | |
| 1376 | void set_zone_contiguous(struct zone *zone) |
| 1377 | { |
| 1378 | unsigned long block_start_pfn = zone->zone_start_pfn; |
| 1379 | unsigned long block_end_pfn; |
| 1380 | |
| 1381 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); |
| 1382 | for (; block_start_pfn < zone_end_pfn(zone); |
| 1383 | block_start_pfn = block_end_pfn, |
| 1384 | block_end_pfn += pageblock_nr_pages) { |
| 1385 | |
| 1386 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); |
| 1387 | |
| 1388 | if (!__pageblock_pfn_to_page(block_start_pfn, |
| 1389 | block_end_pfn, zone)) |
| 1390 | return; |
| 1391 | } |
| 1392 | |
| 1393 | /* We confirm that there is no hole */ |
| 1394 | zone->contiguous = true; |
| 1395 | } |
| 1396 | |
| 1397 | void clear_zone_contiguous(struct zone *zone) |
| 1398 | { |
| 1399 | zone->contiguous = false; |
| 1400 | } |
| 1401 | |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1402 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1403 | static void __init deferred_free_range(struct page *page, |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1404 | unsigned long pfn, int nr_pages) |
| 1405 | { |
| 1406 | int i; |
| 1407 | |
| 1408 | if (!page) |
| 1409 | return; |
| 1410 | |
| 1411 | /* Free a large naturally-aligned chunk if possible */ |
Xishi Qiu | e780149 | 2016-10-07 16:58:09 -0700 | [diff] [blame] | 1412 | if (nr_pages == pageblock_nr_pages && |
| 1413 | (pfn & (pageblock_nr_pages - 1)) == 0) { |
Mel Gorman | ac5d253 | 2015-06-30 14:57:20 -0700 | [diff] [blame] | 1414 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
Xishi Qiu | e780149 | 2016-10-07 16:58:09 -0700 | [diff] [blame] | 1415 | __free_pages_boot_core(page, pageblock_order); |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1416 | return; |
| 1417 | } |
| 1418 | |
Xishi Qiu | e780149 | 2016-10-07 16:58:09 -0700 | [diff] [blame] | 1419 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
| 1420 | if ((pfn & (pageblock_nr_pages - 1)) == 0) |
| 1421 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
Li Zhang | 949698a | 2016-05-19 17:11:37 -0700 | [diff] [blame] | 1422 | __free_pages_boot_core(page, 0); |
Xishi Qiu | e780149 | 2016-10-07 16:58:09 -0700 | [diff] [blame] | 1423 | } |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1424 | } |
| 1425 | |
Nicolai Stange | d3cd131 | 2015-08-06 15:46:16 -0700 | [diff] [blame] | 1426 | /* Completion tracking for deferred_init_memmap() threads */ |
| 1427 | static atomic_t pgdat_init_n_undone __initdata; |
| 1428 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); |
| 1429 | |
| 1430 | static inline void __init pgdat_init_report_one_done(void) |
| 1431 | { |
| 1432 | if (atomic_dec_and_test(&pgdat_init_n_undone)) |
| 1433 | complete(&pgdat_init_all_done_comp); |
| 1434 | } |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1435 | |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1436 | /* Initialise remaining memory on a node */ |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1437 | static int __init deferred_init_memmap(void *data) |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1438 | { |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1439 | pg_data_t *pgdat = data; |
| 1440 | int nid = pgdat->node_id; |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1441 | struct mminit_pfnnid_cache nid_init_state = { }; |
| 1442 | unsigned long start = jiffies; |
| 1443 | unsigned long nr_pages = 0; |
| 1444 | unsigned long walk_start, walk_end; |
| 1445 | int i, zid; |
| 1446 | struct zone *zone; |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1447 | unsigned long first_init_pfn = pgdat->first_deferred_pfn; |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1448 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1449 | |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1450 | if (first_init_pfn == ULONG_MAX) { |
Nicolai Stange | d3cd131 | 2015-08-06 15:46:16 -0700 | [diff] [blame] | 1451 | pgdat_init_report_one_done(); |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1452 | return 0; |
| 1453 | } |
| 1454 | |
| 1455 | /* Bind memory initialisation thread to a local node if possible */ |
| 1456 | if (!cpumask_empty(cpumask)) |
| 1457 | set_cpus_allowed_ptr(current, cpumask); |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1458 | |
| 1459 | /* Sanity check boundaries */ |
| 1460 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); |
| 1461 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); |
| 1462 | pgdat->first_deferred_pfn = ULONG_MAX; |
| 1463 | |
| 1464 | /* Only the highest zone is deferred so find it */ |
| 1465 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 1466 | zone = pgdat->node_zones + zid; |
| 1467 | if (first_init_pfn < zone_end_pfn(zone)) |
| 1468 | break; |
| 1469 | } |
| 1470 | |
| 1471 | for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) { |
| 1472 | unsigned long pfn, end_pfn; |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1473 | struct page *page = NULL; |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1474 | struct page *free_base_page = NULL; |
| 1475 | unsigned long free_base_pfn = 0; |
| 1476 | int nr_to_free = 0; |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1477 | |
| 1478 | end_pfn = min(walk_end, zone_end_pfn(zone)); |
| 1479 | pfn = first_init_pfn; |
| 1480 | if (pfn < walk_start) |
| 1481 | pfn = walk_start; |
| 1482 | if (pfn < zone->zone_start_pfn) |
| 1483 | pfn = zone->zone_start_pfn; |
| 1484 | |
| 1485 | for (; pfn < end_pfn; pfn++) { |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1486 | if (!pfn_valid_within(pfn)) |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1487 | goto free_range; |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1488 | |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1489 | /* |
| 1490 | * Ensure pfn_valid is checked every |
Xishi Qiu | e780149 | 2016-10-07 16:58:09 -0700 | [diff] [blame] | 1491 | * pageblock_nr_pages for memory holes |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1492 | */ |
Xishi Qiu | e780149 | 2016-10-07 16:58:09 -0700 | [diff] [blame] | 1493 | if ((pfn & (pageblock_nr_pages - 1)) == 0) { |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1494 | if (!pfn_valid(pfn)) { |
| 1495 | page = NULL; |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1496 | goto free_range; |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1497 | } |
| 1498 | } |
| 1499 | |
| 1500 | if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) { |
| 1501 | page = NULL; |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1502 | goto free_range; |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1503 | } |
| 1504 | |
| 1505 | /* Minimise pfn page lookups and scheduler checks */ |
Xishi Qiu | e780149 | 2016-10-07 16:58:09 -0700 | [diff] [blame] | 1506 | if (page && (pfn & (pageblock_nr_pages - 1)) != 0) { |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1507 | page++; |
| 1508 | } else { |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1509 | nr_pages += nr_to_free; |
| 1510 | deferred_free_range(free_base_page, |
| 1511 | free_base_pfn, nr_to_free); |
| 1512 | free_base_page = NULL; |
| 1513 | free_base_pfn = nr_to_free = 0; |
| 1514 | |
Mel Gorman | 54608c3 | 2015-06-30 14:57:09 -0700 | [diff] [blame] | 1515 | page = pfn_to_page(pfn); |
| 1516 | cond_resched(); |
| 1517 | } |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1518 | |
| 1519 | if (page->flags) { |
| 1520 | VM_BUG_ON(page_zone(page) != zone); |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1521 | goto free_range; |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1522 | } |
| 1523 | |
| 1524 | __init_single_page(page, pfn, zid, nid); |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1525 | if (!free_base_page) { |
| 1526 | free_base_page = page; |
| 1527 | free_base_pfn = pfn; |
| 1528 | nr_to_free = 0; |
| 1529 | } |
| 1530 | nr_to_free++; |
| 1531 | |
| 1532 | /* Where possible, batch up pages for a single free */ |
| 1533 | continue; |
| 1534 | free_range: |
| 1535 | /* Free the current block of pages to allocator */ |
| 1536 | nr_pages += nr_to_free; |
| 1537 | deferred_free_range(free_base_page, free_base_pfn, |
| 1538 | nr_to_free); |
| 1539 | free_base_page = NULL; |
| 1540 | free_base_pfn = nr_to_free = 0; |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1541 | } |
Xishi Qiu | e780149 | 2016-10-07 16:58:09 -0700 | [diff] [blame] | 1542 | /* Free the last block of pages to allocator */ |
| 1543 | nr_pages += nr_to_free; |
| 1544 | deferred_free_range(free_base_page, free_base_pfn, nr_to_free); |
Mel Gorman | a4de83d | 2015-06-30 14:57:16 -0700 | [diff] [blame] | 1545 | |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1546 | first_init_pfn = max(end_pfn, first_init_pfn); |
| 1547 | } |
| 1548 | |
| 1549 | /* Sanity check that the next zone really is unpopulated */ |
| 1550 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); |
| 1551 | |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1552 | 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] | 1553 | jiffies_to_msecs(jiffies - start)); |
Nicolai Stange | d3cd131 | 2015-08-06 15:46:16 -0700 | [diff] [blame] | 1554 | |
| 1555 | pgdat_init_report_one_done(); |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1556 | return 0; |
| 1557 | } |
Joonsoo Kim | 7cf91a9 | 2016-03-15 14:57:51 -0700 | [diff] [blame] | 1558 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1559 | |
| 1560 | void __init page_alloc_init_late(void) |
| 1561 | { |
Joonsoo Kim | 7cf91a9 | 2016-03-15 14:57:51 -0700 | [diff] [blame] | 1562 | struct zone *zone; |
| 1563 | |
| 1564 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1565 | int nid; |
| 1566 | |
Nicolai Stange | d3cd131 | 2015-08-06 15:46:16 -0700 | [diff] [blame] | 1567 | /* There will be num_node_state(N_MEMORY) threads */ |
| 1568 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1569 | for_each_node_state(nid, N_MEMORY) { |
Mel Gorman | 0e1cc95 | 2015-06-30 14:57:27 -0700 | [diff] [blame] | 1570 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
| 1571 | } |
| 1572 | |
| 1573 | /* Block until all are initialised */ |
Nicolai Stange | d3cd131 | 2015-08-06 15:46:16 -0700 | [diff] [blame] | 1574 | wait_for_completion(&pgdat_init_all_done_comp); |
Mel Gorman | 4248b0d | 2015-08-06 15:46:20 -0700 | [diff] [blame] | 1575 | |
| 1576 | /* Reinit limits that are based on free pages after the kernel is up */ |
| 1577 | files_maxfiles_init(); |
Joonsoo Kim | 7cf91a9 | 2016-03-15 14:57:51 -0700 | [diff] [blame] | 1578 | #endif |
| 1579 | |
| 1580 | for_each_populated_zone(zone) |
| 1581 | set_zone_contiguous(zone); |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1582 | } |
Mel Gorman | 7e18adb | 2015-06-30 14:57:05 -0700 | [diff] [blame] | 1583 | |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1584 | #ifdef CONFIG_CMA |
Li Zhong | 9cf510a | 2013-08-23 13:52:52 +0800 | [diff] [blame] | 1585 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1586 | void __init init_cma_reserved_pageblock(struct page *page) |
| 1587 | { |
| 1588 | unsigned i = pageblock_nr_pages; |
| 1589 | struct page *p = page; |
| 1590 | |
| 1591 | do { |
| 1592 | __ClearPageReserved(p); |
| 1593 | set_page_count(p, 0); |
| 1594 | } while (++p, --i); |
| 1595 | |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1596 | set_pageblock_migratetype(page, MIGRATE_CMA); |
Michal Nazarewicz | dc78327 | 2014-07-02 15:22:35 -0700 | [diff] [blame] | 1597 | |
| 1598 | if (pageblock_order >= MAX_ORDER) { |
| 1599 | i = pageblock_nr_pages; |
| 1600 | p = page; |
| 1601 | do { |
| 1602 | set_page_refcounted(p); |
| 1603 | __free_pages(p, MAX_ORDER - 1); |
| 1604 | p += MAX_ORDER_NR_PAGES; |
| 1605 | } while (i -= MAX_ORDER_NR_PAGES); |
| 1606 | } else { |
| 1607 | set_page_refcounted(page); |
| 1608 | __free_pages(page, pageblock_order); |
| 1609 | } |
| 1610 | |
Jiang Liu | 3dcc057 | 2013-07-03 15:03:21 -0700 | [diff] [blame] | 1611 | adjust_managed_page_count(page, pageblock_nr_pages); |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1612 | } |
| 1613 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1614 | |
| 1615 | /* |
| 1616 | * The order of subdivision here is critical for the IO subsystem. |
| 1617 | * Please do not alter this order without good reasons and regression |
| 1618 | * testing. Specifically, as large blocks of memory are subdivided, |
| 1619 | * the order in which smaller blocks are delivered depends on the order |
| 1620 | * they're subdivided in this function. This is the primary factor |
| 1621 | * influencing the order in which pages are delivered to the IO |
| 1622 | * subsystem according to empirical testing, and this is also justified |
| 1623 | * by considering the behavior of a buddy system containing a single |
| 1624 | * large block of memory acted on by a series of small allocations. |
| 1625 | * This behavior is a critical factor in sglist merging's success. |
| 1626 | * |
Nadia Yvette Chambers | 6d49e35 | 2012-12-06 10:39:54 +0100 | [diff] [blame] | 1627 | * -- nyc |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1628 | */ |
Nick Piggin | 085cc7d | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 1629 | static inline void expand(struct zone *zone, struct page *page, |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1630 | int low, int high, struct free_area *area, |
| 1631 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1632 | { |
| 1633 | unsigned long size = 1 << high; |
| 1634 | |
| 1635 | while (high > low) { |
| 1636 | area--; |
| 1637 | high--; |
| 1638 | size >>= 1; |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 1639 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 1640 | |
Joonsoo Kim | acbc15a | 2016-10-07 16:58:15 -0700 | [diff] [blame] | 1641 | /* |
| 1642 | * Mark as guard pages (or page), that will allow to |
| 1643 | * merge back to allocator when buddy will be freed. |
| 1644 | * Corresponding page table entries will not be touched, |
| 1645 | * pages will stay not present in virtual address space |
| 1646 | */ |
| 1647 | if (set_page_guard(zone, &page[size], high, migratetype)) |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 1648 | continue; |
Joonsoo Kim | acbc15a | 2016-10-07 16:58:15 -0700 | [diff] [blame] | 1649 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1650 | list_add(&page[size].lru, &area->free_list[migratetype]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1651 | area->nr_free++; |
| 1652 | set_page_order(&page[size], high); |
| 1653 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1654 | } |
| 1655 | |
Vlastimil Babka | 4e61180 | 2016-05-19 17:14:41 -0700 | [diff] [blame] | 1656 | static void check_new_page_bad(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1657 | { |
Vlastimil Babka | 4e61180 | 2016-05-19 17:14:41 -0700 | [diff] [blame] | 1658 | const char *bad_reason = NULL; |
| 1659 | unsigned long bad_flags = 0; |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 1660 | |
Kirill A. Shutemov | 53f9263 | 2016-01-15 16:53:42 -0800 | [diff] [blame] | 1661 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 1662 | bad_reason = "nonzero mapcount"; |
| 1663 | if (unlikely(page->mapping != NULL)) |
| 1664 | bad_reason = "non-NULL mapping"; |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 1665 | if (unlikely(page_ref_count(page) != 0)) |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 1666 | bad_reason = "nonzero _count"; |
Naoya Horiguchi | f4c18e6 | 2015-08-06 15:47:08 -0700 | [diff] [blame] | 1667 | if (unlikely(page->flags & __PG_HWPOISON)) { |
| 1668 | bad_reason = "HWPoisoned (hardware-corrupted)"; |
| 1669 | bad_flags = __PG_HWPOISON; |
Naoya Horiguchi | e570f56 | 2016-05-20 16:58:50 -0700 | [diff] [blame] | 1670 | /* Don't complain about hwpoisoned pages */ |
| 1671 | page_mapcount_reset(page); /* remove PageBuddy */ |
| 1672 | return; |
Naoya Horiguchi | f4c18e6 | 2015-08-06 15:47:08 -0700 | [diff] [blame] | 1673 | } |
Dave Hansen | f0b791a | 2014-01-23 15:52:49 -0800 | [diff] [blame] | 1674 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
| 1675 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; |
| 1676 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; |
| 1677 | } |
Johannes Weiner | 9edad6e | 2014-12-10 15:44:58 -0800 | [diff] [blame] | 1678 | #ifdef CONFIG_MEMCG |
| 1679 | if (unlikely(page->mem_cgroup)) |
| 1680 | bad_reason = "page still charged to cgroup"; |
| 1681 | #endif |
Vlastimil Babka | 4e61180 | 2016-05-19 17:14:41 -0700 | [diff] [blame] | 1682 | bad_page(page, bad_reason, bad_flags); |
| 1683 | } |
| 1684 | |
| 1685 | /* |
| 1686 | * This page is about to be returned from the page allocator |
| 1687 | */ |
| 1688 | static inline int check_new_page(struct page *page) |
| 1689 | { |
| 1690 | if (likely(page_expected_state(page, |
| 1691 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) |
| 1692 | return 0; |
| 1693 | |
| 1694 | check_new_page_bad(page); |
| 1695 | return 1; |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 1696 | } |
| 1697 | |
Laura Abbott | 1414c7f | 2016-03-15 14:56:30 -0700 | [diff] [blame] | 1698 | static inline bool free_pages_prezeroed(bool poisoned) |
| 1699 | { |
| 1700 | return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && |
| 1701 | page_poisoning_enabled() && poisoned; |
| 1702 | } |
| 1703 | |
Mel Gorman | 479f854 | 2016-05-19 17:14:35 -0700 | [diff] [blame] | 1704 | #ifdef CONFIG_DEBUG_VM |
| 1705 | static bool check_pcp_refill(struct page *page) |
| 1706 | { |
| 1707 | return false; |
| 1708 | } |
| 1709 | |
| 1710 | static bool check_new_pcp(struct page *page) |
| 1711 | { |
| 1712 | return check_new_page(page); |
| 1713 | } |
| 1714 | #else |
| 1715 | static bool check_pcp_refill(struct page *page) |
| 1716 | { |
| 1717 | return check_new_page(page); |
| 1718 | } |
| 1719 | static bool check_new_pcp(struct page *page) |
| 1720 | { |
| 1721 | return false; |
| 1722 | } |
| 1723 | #endif /* CONFIG_DEBUG_VM */ |
| 1724 | |
| 1725 | static bool check_new_pages(struct page *page, unsigned int order) |
| 1726 | { |
| 1727 | int i; |
| 1728 | for (i = 0; i < (1 << order); i++) { |
| 1729 | struct page *p = page + i; |
| 1730 | |
| 1731 | if (unlikely(check_new_page(p))) |
| 1732 | return true; |
| 1733 | } |
| 1734 | |
| 1735 | return false; |
| 1736 | } |
| 1737 | |
Joonsoo Kim | 46f24fd | 2016-07-26 15:23:58 -0700 | [diff] [blame] | 1738 | inline void post_alloc_hook(struct page *page, unsigned int order, |
| 1739 | gfp_t gfp_flags) |
| 1740 | { |
| 1741 | set_page_private(page, 0); |
| 1742 | set_page_refcounted(page); |
| 1743 | |
| 1744 | arch_alloc_page(page, order); |
| 1745 | kernel_map_pages(page, 1 << order, 1); |
| 1746 | kernel_poison_pages(page, 1 << order, 1); |
| 1747 | kasan_alloc_pages(page, order); |
| 1748 | set_page_owner(page, order, gfp_flags); |
| 1749 | } |
| 1750 | |
Mel Gorman | 479f854 | 2016-05-19 17:14:35 -0700 | [diff] [blame] | 1751 | 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] | 1752 | unsigned int alloc_flags) |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 1753 | { |
| 1754 | int i; |
Laura Abbott | 1414c7f | 2016-03-15 14:56:30 -0700 | [diff] [blame] | 1755 | bool poisoned = true; |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 1756 | |
| 1757 | for (i = 0; i < (1 << order); i++) { |
| 1758 | struct page *p = page + i; |
Laura Abbott | 1414c7f | 2016-03-15 14:56:30 -0700 | [diff] [blame] | 1759 | if (poisoned) |
| 1760 | poisoned &= page_is_poisoned(p); |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 1761 | } |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 1762 | |
Joonsoo Kim | 46f24fd | 2016-07-26 15:23:58 -0700 | [diff] [blame] | 1763 | post_alloc_hook(page, order, gfp_flags); |
Nick Piggin | 17cf440 | 2006-03-22 00:08:41 -0800 | [diff] [blame] | 1764 | |
Laura Abbott | 1414c7f | 2016-03-15 14:56:30 -0700 | [diff] [blame] | 1765 | if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO)) |
Anisse Astier | f4d2897 | 2015-06-24 16:56:36 -0700 | [diff] [blame] | 1766 | for (i = 0; i < (1 << order); i++) |
| 1767 | clear_highpage(page + i); |
Nick Piggin | 17cf440 | 2006-03-22 00:08:41 -0800 | [diff] [blame] | 1768 | |
| 1769 | if (order && (gfp_flags & __GFP_COMP)) |
| 1770 | prep_compound_page(page, order); |
| 1771 | |
Vlastimil Babka | 7537919 | 2015-02-11 15:25:38 -0800 | [diff] [blame] | 1772 | /* |
Michal Hocko | 2f064f3 | 2015-08-21 14:11:51 -0700 | [diff] [blame] | 1773 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
Vlastimil Babka | 7537919 | 2015-02-11 15:25:38 -0800 | [diff] [blame] | 1774 | * allocate the page. The expectation is that the caller is taking |
| 1775 | * steps that will free more memory. The caller should avoid the page |
| 1776 | * being used for !PFMEMALLOC purposes. |
| 1777 | */ |
Michal Hocko | 2f064f3 | 2015-08-21 14:11:51 -0700 | [diff] [blame] | 1778 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
| 1779 | set_page_pfmemalloc(page); |
| 1780 | else |
| 1781 | clear_page_pfmemalloc(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1782 | } |
| 1783 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1784 | /* |
| 1785 | * Go through the free lists for the given migratetype and remove |
| 1786 | * the smallest available page from the freelists |
| 1787 | */ |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 1788 | static inline |
| 1789 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1790 | int migratetype) |
| 1791 | { |
| 1792 | unsigned int current_order; |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 1793 | struct free_area *area; |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1794 | struct page *page; |
| 1795 | |
| 1796 | /* Find a page of the appropriate size in the preferred list */ |
| 1797 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { |
| 1798 | area = &(zone->free_area[current_order]); |
Geliang Tang | a16601c | 2016-01-14 15:20:30 -0800 | [diff] [blame] | 1799 | page = list_first_entry_or_null(&area->free_list[migratetype], |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1800 | struct page, lru); |
Geliang Tang | a16601c | 2016-01-14 15:20:30 -0800 | [diff] [blame] | 1801 | if (!page) |
| 1802 | continue; |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1803 | list_del(&page->lru); |
| 1804 | rmv_page_order(page); |
| 1805 | area->nr_free--; |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1806 | expand(zone, page, order, current_order, area, migratetype); |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 1807 | set_pcppage_migratetype(page, migratetype); |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1808 | return page; |
| 1809 | } |
| 1810 | |
| 1811 | return NULL; |
| 1812 | } |
| 1813 | |
| 1814 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1815 | /* |
| 1816 | * This array describes the order lists are fallen back to when |
| 1817 | * the free lists for the desirable migrate type are depleted |
| 1818 | */ |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1819 | static int fallbacks[MIGRATE_TYPES][4] = { |
Mel Gorman | 974a786 | 2015-11-06 16:28:34 -0800 | [diff] [blame] | 1820 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
| 1821 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
| 1822 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, |
Joonsoo Kim | dc67647 | 2015-04-14 15:45:15 -0700 | [diff] [blame] | 1823 | #ifdef CONFIG_CMA |
Mel Gorman | 974a786 | 2015-11-06 16:28:34 -0800 | [diff] [blame] | 1824 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 1825 | #endif |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 1826 | #ifdef CONFIG_MEMORY_ISOLATION |
Mel Gorman | 974a786 | 2015-11-06 16:28:34 -0800 | [diff] [blame] | 1827 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 1828 | #endif |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1829 | }; |
| 1830 | |
Joonsoo Kim | dc67647 | 2015-04-14 15:45:15 -0700 | [diff] [blame] | 1831 | #ifdef CONFIG_CMA |
| 1832 | static struct page *__rmqueue_cma_fallback(struct zone *zone, |
| 1833 | unsigned int order) |
| 1834 | { |
| 1835 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); |
| 1836 | } |
| 1837 | #else |
| 1838 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
| 1839 | unsigned int order) { return NULL; } |
| 1840 | #endif |
| 1841 | |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1842 | /* |
| 1843 | * 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] | 1844 | * 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] | 1845 | * boundary. If alignment is required, use move_freepages_block() |
| 1846 | */ |
Minchan Kim | 435b405 | 2012-10-08 16:32:16 -0700 | [diff] [blame] | 1847 | int move_freepages(struct zone *zone, |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 1848 | struct page *start_page, struct page *end_page, |
| 1849 | int migratetype) |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1850 | { |
| 1851 | struct page *page; |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 1852 | unsigned int order; |
Mel Gorman | d100313 | 2007-10-16 01:26:00 -0700 | [diff] [blame] | 1853 | int pages_moved = 0; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1854 | |
| 1855 | #ifndef CONFIG_HOLES_IN_ZONE |
| 1856 | /* |
| 1857 | * page_zone is not safe to call in this context when |
| 1858 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant |
| 1859 | * anyway as we check zone boundaries in move_freepages_block(). |
| 1860 | * Remove at a later date when no bug reports exist related to |
Mel Gorman | ac0e5b7 | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1861 | * grouping pages by mobility |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1862 | */ |
Mel Gorman | 97ee4ba | 2014-10-09 15:28:28 -0700 | [diff] [blame] | 1863 | VM_BUG_ON(page_zone(start_page) != page_zone(end_page)); |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1864 | #endif |
| 1865 | |
| 1866 | for (page = start_page; page <= end_page;) { |
Adam Litke | 344c790 | 2008-09-02 14:35:38 -0700 | [diff] [blame] | 1867 | /* Make sure we are not inadvertently changing nodes */ |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 1868 | 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] | 1869 | |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1870 | if (!pfn_valid_within(page_to_pfn(page))) { |
| 1871 | page++; |
| 1872 | continue; |
| 1873 | } |
| 1874 | |
| 1875 | if (!PageBuddy(page)) { |
| 1876 | page++; |
| 1877 | continue; |
| 1878 | } |
| 1879 | |
| 1880 | order = page_order(page); |
Kirill A. Shutemov | 84be48d | 2011-03-22 16:33:41 -0700 | [diff] [blame] | 1881 | list_move(&page->lru, |
| 1882 | &zone->free_area[order].free_list[migratetype]); |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1883 | page += 1 << order; |
Mel Gorman | d100313 | 2007-10-16 01:26:00 -0700 | [diff] [blame] | 1884 | pages_moved += 1 << order; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1885 | } |
| 1886 | |
Mel Gorman | d100313 | 2007-10-16 01:26:00 -0700 | [diff] [blame] | 1887 | return pages_moved; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1888 | } |
| 1889 | |
Minchan Kim | ee6f509 | 2012-07-31 16:43:50 -0700 | [diff] [blame] | 1890 | int move_freepages_block(struct zone *zone, struct page *page, |
Linus Torvalds | 68e3e92 | 2012-06-03 20:05:57 -0700 | [diff] [blame] | 1891 | int migratetype) |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1892 | { |
| 1893 | unsigned long start_pfn, end_pfn; |
| 1894 | struct page *start_page, *end_page; |
| 1895 | |
| 1896 | start_pfn = page_to_pfn(page); |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 1897 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1898 | start_page = pfn_to_page(start_pfn); |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 1899 | end_page = start_page + pageblock_nr_pages - 1; |
| 1900 | end_pfn = start_pfn + pageblock_nr_pages - 1; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1901 | |
| 1902 | /* Do not cross zone boundaries */ |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 1903 | if (!zone_spans_pfn(zone, start_pfn)) |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1904 | start_page = page; |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 1905 | if (!zone_spans_pfn(zone, end_pfn)) |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1906 | return 0; |
| 1907 | |
| 1908 | return move_freepages(zone, start_page, end_page, migratetype); |
| 1909 | } |
| 1910 | |
Mel Gorman | 2f66a68 | 2009-09-21 17:02:31 -0700 | [diff] [blame] | 1911 | static void change_pageblock_range(struct page *pageblock_page, |
| 1912 | int start_order, int migratetype) |
| 1913 | { |
| 1914 | int nr_pageblocks = 1 << (start_order - pageblock_order); |
| 1915 | |
| 1916 | while (nr_pageblocks--) { |
| 1917 | set_pageblock_migratetype(pageblock_page, migratetype); |
| 1918 | pageblock_page += pageblock_nr_pages; |
| 1919 | } |
| 1920 | } |
| 1921 | |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1922 | /* |
Vlastimil Babka | 9c0415e | 2015-02-11 15:28:21 -0800 | [diff] [blame] | 1923 | * When we are falling back to another migratetype during allocation, try to |
| 1924 | * steal extra free pages from the same pageblocks to satisfy further |
| 1925 | * allocations, instead of polluting multiple pageblocks. |
| 1926 | * |
| 1927 | * If we are stealing a relatively large buddy page, it is likely there will |
| 1928 | * be more free pages in the pageblock, so try to steal them all. For |
| 1929 | * reclaimable and unmovable allocations, we steal regardless of page size, |
| 1930 | * as fragmentation caused by those allocations polluting movable pageblocks |
| 1931 | * is worse than movable allocations stealing from unmovable and reclaimable |
| 1932 | * pageblocks. |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1933 | */ |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 1934 | static bool can_steal_fallback(unsigned int order, int start_mt) |
| 1935 | { |
| 1936 | /* |
| 1937 | * Leaving this order check is intended, although there is |
| 1938 | * relaxed order check in next check. The reason is that |
| 1939 | * we can actually steal whole pageblock if this condition met, |
| 1940 | * but, below check doesn't guarantee it and that is just heuristic |
| 1941 | * so could be changed anytime. |
| 1942 | */ |
| 1943 | if (order >= pageblock_order) |
| 1944 | return true; |
| 1945 | |
| 1946 | if (order >= pageblock_order / 2 || |
| 1947 | start_mt == MIGRATE_RECLAIMABLE || |
| 1948 | start_mt == MIGRATE_UNMOVABLE || |
| 1949 | page_group_by_mobility_disabled) |
| 1950 | return true; |
| 1951 | |
| 1952 | return false; |
| 1953 | } |
| 1954 | |
| 1955 | /* |
| 1956 | * This function implements actual steal behaviour. If order is large enough, |
| 1957 | * we can steal whole pageblock. If not, we first move freepages in this |
| 1958 | * pageblock and check whether half of pages are moved or not. If half of |
| 1959 | * pages are moved, we can change migratetype of pageblock and permanently |
| 1960 | * use it's pages as requested migratetype in the future. |
| 1961 | */ |
| 1962 | static void steal_suitable_fallback(struct zone *zone, struct page *page, |
| 1963 | int start_type) |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1964 | { |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 1965 | unsigned int current_order = page_order(page); |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 1966 | int pages; |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1967 | |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1968 | /* Take ownership for orders >= pageblock_order */ |
| 1969 | if (current_order >= pageblock_order) { |
| 1970 | change_pageblock_range(page, current_order, start_type); |
Vlastimil Babka | 3a1086f | 2015-02-11 15:28:18 -0800 | [diff] [blame] | 1971 | return; |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1972 | } |
| 1973 | |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 1974 | pages = move_freepages_block(zone, page, start_type); |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1975 | |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 1976 | /* Claim the whole block if over half of it is free */ |
| 1977 | if (pages >= (1 << (pageblock_order-1)) || |
| 1978 | page_group_by_mobility_disabled) |
| 1979 | set_pageblock_migratetype(page, start_type); |
| 1980 | } |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 1981 | |
Joonsoo Kim | 2149cda | 2015-04-14 15:45:21 -0700 | [diff] [blame] | 1982 | /* |
| 1983 | * Check whether there is a suitable fallback freepage with requested order. |
| 1984 | * If only_stealable is true, this function returns fallback_mt only if |
| 1985 | * we can steal other freepages all together. This would help to reduce |
| 1986 | * fragmentation due to mixed migratetype pages in one pageblock. |
| 1987 | */ |
| 1988 | int find_suitable_fallback(struct free_area *area, unsigned int order, |
| 1989 | int migratetype, bool only_stealable, bool *can_steal) |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 1990 | { |
| 1991 | int i; |
| 1992 | int fallback_mt; |
| 1993 | |
| 1994 | if (area->nr_free == 0) |
| 1995 | return -1; |
| 1996 | |
| 1997 | *can_steal = false; |
| 1998 | for (i = 0;; i++) { |
| 1999 | fallback_mt = fallbacks[migratetype][i]; |
Mel Gorman | 974a786 | 2015-11-06 16:28:34 -0800 | [diff] [blame] | 2000 | if (fallback_mt == MIGRATE_TYPES) |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2001 | break; |
| 2002 | |
| 2003 | if (list_empty(&area->free_list[fallback_mt])) |
| 2004 | continue; |
| 2005 | |
| 2006 | if (can_steal_fallback(order, migratetype)) |
| 2007 | *can_steal = true; |
| 2008 | |
Joonsoo Kim | 2149cda | 2015-04-14 15:45:21 -0700 | [diff] [blame] | 2009 | if (!only_stealable) |
| 2010 | return fallback_mt; |
| 2011 | |
| 2012 | if (*can_steal) |
| 2013 | return fallback_mt; |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 2014 | } |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2015 | |
| 2016 | return -1; |
Srivatsa S. Bhat | fef903e | 2013-09-11 14:20:35 -0700 | [diff] [blame] | 2017 | } |
| 2018 | |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2019 | /* |
| 2020 | * Reserve a pageblock for exclusive use of high-order atomic allocations if |
| 2021 | * there are no empty page blocks that contain a page with a suitable order |
| 2022 | */ |
| 2023 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, |
| 2024 | unsigned int alloc_order) |
| 2025 | { |
| 2026 | int mt; |
| 2027 | unsigned long max_managed, flags; |
| 2028 | |
| 2029 | /* |
| 2030 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. |
| 2031 | * Check is race-prone but harmless. |
| 2032 | */ |
| 2033 | max_managed = (zone->managed_pages / 100) + pageblock_nr_pages; |
| 2034 | if (zone->nr_reserved_highatomic >= max_managed) |
| 2035 | return; |
| 2036 | |
| 2037 | spin_lock_irqsave(&zone->lock, flags); |
| 2038 | |
| 2039 | /* Recheck the nr_reserved_highatomic limit under the lock */ |
| 2040 | if (zone->nr_reserved_highatomic >= max_managed) |
| 2041 | goto out_unlock; |
| 2042 | |
| 2043 | /* Yoink! */ |
| 2044 | mt = get_pageblock_migratetype(page); |
| 2045 | if (mt != MIGRATE_HIGHATOMIC && |
| 2046 | !is_migrate_isolate(mt) && !is_migrate_cma(mt)) { |
| 2047 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
| 2048 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); |
| 2049 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC); |
| 2050 | } |
| 2051 | |
| 2052 | out_unlock: |
| 2053 | spin_unlock_irqrestore(&zone->lock, flags); |
| 2054 | } |
| 2055 | |
| 2056 | /* |
| 2057 | * Used when an allocation is about to fail under memory pressure. This |
| 2058 | * potentially hurts the reliability of high-order allocations when under |
| 2059 | * intense memory pressure but failed atomic allocations should be easier |
| 2060 | * to recover from than an OOM. |
| 2061 | */ |
| 2062 | static void unreserve_highatomic_pageblock(const struct alloc_context *ac) |
| 2063 | { |
| 2064 | struct zonelist *zonelist = ac->zonelist; |
| 2065 | unsigned long flags; |
| 2066 | struct zoneref *z; |
| 2067 | struct zone *zone; |
| 2068 | struct page *page; |
| 2069 | int order; |
| 2070 | |
| 2071 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, |
| 2072 | ac->nodemask) { |
| 2073 | /* Preserve at least one pageblock */ |
| 2074 | if (zone->nr_reserved_highatomic <= pageblock_nr_pages) |
| 2075 | continue; |
| 2076 | |
| 2077 | spin_lock_irqsave(&zone->lock, flags); |
| 2078 | for (order = 0; order < MAX_ORDER; order++) { |
| 2079 | struct free_area *area = &(zone->free_area[order]); |
| 2080 | |
Geliang Tang | a16601c | 2016-01-14 15:20:30 -0800 | [diff] [blame] | 2081 | page = list_first_entry_or_null( |
| 2082 | &area->free_list[MIGRATE_HIGHATOMIC], |
| 2083 | struct page, lru); |
| 2084 | if (!page) |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2085 | continue; |
| 2086 | |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2087 | /* |
| 2088 | * It should never happen but changes to locking could |
| 2089 | * inadvertently allow a per-cpu drain to add pages |
| 2090 | * to MIGRATE_HIGHATOMIC while unreserving so be safe |
| 2091 | * and watch for underflows. |
| 2092 | */ |
| 2093 | zone->nr_reserved_highatomic -= min(pageblock_nr_pages, |
| 2094 | zone->nr_reserved_highatomic); |
| 2095 | |
| 2096 | /* |
| 2097 | * Convert to ac->migratetype and avoid the normal |
| 2098 | * pageblock stealing heuristics. Minimally, the caller |
| 2099 | * is doing the work and needs the pages. More |
| 2100 | * importantly, if the block was always converted to |
| 2101 | * MIGRATE_UNMOVABLE or another type then the number |
| 2102 | * of pageblocks that cannot be completely freed |
| 2103 | * may increase. |
| 2104 | */ |
| 2105 | set_pageblock_migratetype(page, ac->migratetype); |
| 2106 | move_freepages_block(zone, page, ac->migratetype); |
| 2107 | spin_unlock_irqrestore(&zone->lock, flags); |
| 2108 | return; |
| 2109 | } |
| 2110 | spin_unlock_irqrestore(&zone->lock, flags); |
| 2111 | } |
| 2112 | } |
| 2113 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2114 | /* Remove an element from the buddy allocator from the fallback list */ |
Mel Gorman | 0ac3a40 | 2009-06-16 15:32:06 -0700 | [diff] [blame] | 2115 | static inline struct page * |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 2116 | __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2117 | { |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 2118 | struct free_area *area; |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 2119 | unsigned int current_order; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2120 | struct page *page; |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2121 | int fallback_mt; |
| 2122 | bool can_steal; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2123 | |
| 2124 | /* Find the largest possible block of pages in the other list */ |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 2125 | for (current_order = MAX_ORDER-1; |
| 2126 | current_order >= order && current_order <= MAX_ORDER-1; |
| 2127 | --current_order) { |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2128 | area = &(zone->free_area[current_order]); |
| 2129 | fallback_mt = find_suitable_fallback(area, current_order, |
Joonsoo Kim | 2149cda | 2015-04-14 15:45:21 -0700 | [diff] [blame] | 2130 | start_migratetype, false, &can_steal); |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2131 | if (fallback_mt == -1) |
| 2132 | continue; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2133 | |
Geliang Tang | a16601c | 2016-01-14 15:20:30 -0800 | [diff] [blame] | 2134 | page = list_first_entry(&area->free_list[fallback_mt], |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2135 | struct page, lru); |
| 2136 | if (can_steal) |
| 2137 | steal_suitable_fallback(zone, page, start_migratetype); |
Mel Gorman | e010487 | 2007-10-16 01:25:53 -0700 | [diff] [blame] | 2138 | |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2139 | /* Remove the page from the freelists */ |
| 2140 | area->nr_free--; |
| 2141 | list_del(&page->lru); |
| 2142 | rmv_page_order(page); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2143 | |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2144 | expand(zone, page, order, current_order, area, |
| 2145 | start_migratetype); |
| 2146 | /* |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 2147 | * The pcppage_migratetype may differ from pageblock's |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2148 | * migratetype depending on the decisions in |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 2149 | * find_suitable_fallback(). This is OK as long as it does not |
| 2150 | * differ for MIGRATE_CMA pageblocks. Those can be used as |
| 2151 | * fallback only via special __rmqueue_cma_fallback() function |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2152 | */ |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 2153 | set_pcppage_migratetype(page, start_migratetype); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2154 | |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2155 | trace_mm_page_alloc_extfrag(page, order, current_order, |
| 2156 | start_migratetype, fallback_mt); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2157 | |
Joonsoo Kim | 4eb7dce | 2015-04-14 15:45:18 -0700 | [diff] [blame] | 2158 | return page; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2159 | } |
| 2160 | |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 2161 | return NULL; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2162 | } |
| 2163 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 2164 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2165 | * Do the hard work of removing an element from the buddy allocator. |
| 2166 | * Call me with the zone->lock already held. |
| 2167 | */ |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2168 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
Mel Gorman | 6ac0206 | 2016-01-14 15:20:28 -0800 | [diff] [blame] | 2169 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2170 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2171 | struct page *page; |
| 2172 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 2173 | page = __rmqueue_smallest(zone, order, migratetype); |
Mel Gorman | 974a786 | 2015-11-06 16:28:34 -0800 | [diff] [blame] | 2174 | if (unlikely(!page)) { |
Joonsoo Kim | dc67647 | 2015-04-14 15:45:15 -0700 | [diff] [blame] | 2175 | if (migratetype == MIGRATE_MOVABLE) |
| 2176 | page = __rmqueue_cma_fallback(zone, order); |
| 2177 | |
| 2178 | if (!page) |
| 2179 | page = __rmqueue_fallback(zone, order, migratetype); |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 2180 | } |
| 2181 | |
Mel Gorman | 0d3d062 | 2009-09-21 17:02:44 -0700 | [diff] [blame] | 2182 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2183 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2184 | } |
| 2185 | |
Michal Nazarewicz | 5f63b72 | 2012-01-11 15:16:11 +0100 | [diff] [blame] | 2186 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2187 | * Obtain a specified number of elements from the buddy allocator, all under |
| 2188 | * a single hold of the lock, for efficiency. Add them to the supplied list. |
| 2189 | * Returns the number of new pages which were placed at *list. |
| 2190 | */ |
Michal Nazarewicz | 5f63b72 | 2012-01-11 15:16:11 +0100 | [diff] [blame] | 2191 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2192 | unsigned long count, struct list_head *list, |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2193 | int migratetype, bool cold) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2194 | { |
Vlastimil Babka | 5bcc9f8 | 2014-06-04 16:07:22 -0700 | [diff] [blame] | 2195 | int i; |
Michal Nazarewicz | 5f63b72 | 2012-01-11 15:16:11 +0100 | [diff] [blame] | 2196 | |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 2197 | spin_lock(&zone->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2198 | for (i = 0; i < count; ++i) { |
Mel Gorman | 6ac0206 | 2016-01-14 15:20:28 -0800 | [diff] [blame] | 2199 | struct page *page = __rmqueue(zone, order, migratetype); |
Nick Piggin | 085cc7d | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 2200 | if (unlikely(page == NULL)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2201 | break; |
Mel Gorman | 81eabcb | 2007-12-17 16:20:05 -0800 | [diff] [blame] | 2202 | |
Mel Gorman | 479f854 | 2016-05-19 17:14:35 -0700 | [diff] [blame] | 2203 | if (unlikely(check_pcp_refill(page))) |
| 2204 | continue; |
| 2205 | |
Mel Gorman | 81eabcb | 2007-12-17 16:20:05 -0800 | [diff] [blame] | 2206 | /* |
| 2207 | * Split buddy pages returned by expand() are received here |
| 2208 | * in physical page order. The page is added to the callers and |
| 2209 | * list and the list head then moves forward. From the callers |
| 2210 | * perspective, the linked list is ordered by page number in |
| 2211 | * some conditions. This is useful for IO devices that can |
| 2212 | * merge IO requests if the physical pages are ordered |
| 2213 | * properly. |
| 2214 | */ |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2215 | if (likely(!cold)) |
Mel Gorman | e084b2d | 2009-07-29 15:02:04 -0700 | [diff] [blame] | 2216 | list_add(&page->lru, list); |
| 2217 | else |
| 2218 | list_add_tail(&page->lru, list); |
Mel Gorman | 81eabcb | 2007-12-17 16:20:05 -0800 | [diff] [blame] | 2219 | list = &page->lru; |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 2220 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
Bartlomiej Zolnierkiewicz | d1ce749 | 2012-10-08 16:32:02 -0700 | [diff] [blame] | 2221 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
| 2222 | -(1 << order)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2223 | } |
Mel Gorman | f2260e6 | 2009-06-16 15:32:13 -0700 | [diff] [blame] | 2224 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 2225 | spin_unlock(&zone->lock); |
Nick Piggin | 085cc7d | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 2226 | return i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2227 | } |
| 2228 | |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 2229 | #ifdef CONFIG_NUMA |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 2230 | /* |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 2231 | * Called from the vmstat counter updater to drain pagesets of this |
| 2232 | * currently executing processor on remote nodes after they have |
| 2233 | * expired. |
| 2234 | * |
Christoph Lameter | 879336c | 2006-03-22 00:09:08 -0800 | [diff] [blame] | 2235 | * Note that this function must be called with the thread pinned to |
| 2236 | * a single processor. |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 2237 | */ |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 2238 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 2239 | { |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 2240 | unsigned long flags; |
Michal Nazarewicz | 7be12fc | 2014-08-06 16:05:15 -0700 | [diff] [blame] | 2241 | int to_drain, batch; |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 2242 | |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 2243 | local_irq_save(flags); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 2244 | batch = READ_ONCE(pcp->batch); |
Michal Nazarewicz | 7be12fc | 2014-08-06 16:05:15 -0700 | [diff] [blame] | 2245 | to_drain = min(pcp->count, batch); |
KOSAKI Motohiro | 2a13515 | 2012-07-31 16:42:53 -0700 | [diff] [blame] | 2246 | if (to_drain > 0) { |
| 2247 | free_pcppages_bulk(zone, to_drain, pcp); |
| 2248 | pcp->count -= to_drain; |
| 2249 | } |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 2250 | local_irq_restore(flags); |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 2251 | } |
| 2252 | #endif |
| 2253 | |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2254 | /* |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2255 | * Drain pcplists of the indicated processor and zone. |
| 2256 | * |
| 2257 | * The processor must either be the current processor and the |
| 2258 | * thread pinned to the current processor or a processor that |
| 2259 | * is not online. |
| 2260 | */ |
| 2261 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
| 2262 | { |
| 2263 | unsigned long flags; |
| 2264 | struct per_cpu_pageset *pset; |
| 2265 | struct per_cpu_pages *pcp; |
| 2266 | |
| 2267 | local_irq_save(flags); |
| 2268 | pset = per_cpu_ptr(zone->pageset, cpu); |
| 2269 | |
| 2270 | pcp = &pset->pcp; |
| 2271 | if (pcp->count) { |
| 2272 | free_pcppages_bulk(zone, pcp->count, pcp); |
| 2273 | pcp->count = 0; |
| 2274 | } |
| 2275 | local_irq_restore(flags); |
| 2276 | } |
| 2277 | |
| 2278 | /* |
| 2279 | * Drain pcplists of all zones on the indicated processor. |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2280 | * |
| 2281 | * The processor must either be the current processor and the |
| 2282 | * thread pinned to the current processor or a processor that |
| 2283 | * is not online. |
| 2284 | */ |
| 2285 | static void drain_pages(unsigned int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2286 | { |
| 2287 | struct zone *zone; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2288 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 2289 | for_each_populated_zone(zone) { |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2290 | drain_pages_zone(cpu, zone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2291 | } |
| 2292 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2293 | |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2294 | /* |
| 2295 | * 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] | 2296 | * |
| 2297 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just |
| 2298 | * the single zone's pages. |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2299 | */ |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2300 | void drain_local_pages(struct zone *zone) |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2301 | { |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2302 | int cpu = smp_processor_id(); |
| 2303 | |
| 2304 | if (zone) |
| 2305 | drain_pages_zone(cpu, zone); |
| 2306 | else |
| 2307 | drain_pages(cpu); |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2308 | } |
| 2309 | |
| 2310 | /* |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2311 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
| 2312 | * |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2313 | * When zone parameter is non-NULL, spill just the single zone's pages. |
| 2314 | * |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2315 | * Note that this code is protected against sending an IPI to an offline |
| 2316 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: |
| 2317 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but |
| 2318 | * nothing keeps CPUs from showing up after we populated the cpumask and |
| 2319 | * before the call to on_each_cpu_mask(). |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2320 | */ |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2321 | void drain_all_pages(struct zone *zone) |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2322 | { |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2323 | int cpu; |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2324 | |
| 2325 | /* |
| 2326 | * Allocate in the BSS so we wont require allocation in |
| 2327 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y |
| 2328 | */ |
| 2329 | static cpumask_t cpus_with_pcps; |
| 2330 | |
| 2331 | /* |
| 2332 | * We don't care about racing with CPU hotplug event |
| 2333 | * as offline notification will cause the notified |
| 2334 | * cpu to drain that CPU pcps and on_each_cpu_mask |
| 2335 | * disables preemption as part of its processing |
| 2336 | */ |
| 2337 | for_each_online_cpu(cpu) { |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2338 | struct per_cpu_pageset *pcp; |
| 2339 | struct zone *z; |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2340 | bool has_pcps = false; |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2341 | |
| 2342 | if (zone) { |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2343 | pcp = per_cpu_ptr(zone->pageset, cpu); |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2344 | if (pcp->pcp.count) |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2345 | has_pcps = true; |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2346 | } else { |
| 2347 | for_each_populated_zone(z) { |
| 2348 | pcp = per_cpu_ptr(z->pageset, cpu); |
| 2349 | if (pcp->pcp.count) { |
| 2350 | has_pcps = true; |
| 2351 | break; |
| 2352 | } |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2353 | } |
| 2354 | } |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2355 | |
Gilad Ben-Yossef | 7404649 | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 2356 | if (has_pcps) |
| 2357 | cpumask_set_cpu(cpu, &cpus_with_pcps); |
| 2358 | else |
| 2359 | cpumask_clear_cpu(cpu, &cpus_with_pcps); |
| 2360 | } |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 2361 | on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages, |
| 2362 | zone, 1); |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 2363 | } |
| 2364 | |
Rafael J. Wysocki | 296699d | 2007-07-29 23:27:18 +0200 | [diff] [blame] | 2365 | #ifdef CONFIG_HIBERNATION |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2366 | |
| 2367 | void mark_free_pages(struct zone *zone) |
| 2368 | { |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 2369 | unsigned long pfn, max_zone_pfn; |
| 2370 | unsigned long flags; |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 2371 | unsigned int order, t; |
Geliang Tang | 86760a2 | 2016-01-14 15:20:33 -0800 | [diff] [blame] | 2372 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2373 | |
Xishi Qiu | 8080fc0 | 2013-09-11 14:21:45 -0700 | [diff] [blame] | 2374 | if (zone_is_empty(zone)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2375 | return; |
| 2376 | |
| 2377 | spin_lock_irqsave(&zone->lock, flags); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 2378 | |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 2379 | max_zone_pfn = zone_end_pfn(zone); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 2380 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
| 2381 | if (pfn_valid(pfn)) { |
Geliang Tang | 86760a2 | 2016-01-14 15:20:33 -0800 | [diff] [blame] | 2382 | page = pfn_to_page(pfn); |
Joonsoo Kim | ba6b097 | 2016-05-19 17:12:16 -0700 | [diff] [blame] | 2383 | |
| 2384 | if (page_zone(page) != zone) |
| 2385 | continue; |
| 2386 | |
Rafael J. Wysocki | 7be9823 | 2007-05-06 14:50:42 -0700 | [diff] [blame] | 2387 | if (!swsusp_page_is_forbidden(page)) |
| 2388 | swsusp_unset_page_free(page); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 2389 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2390 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2391 | for_each_migratetype_order(order, t) { |
Geliang Tang | 86760a2 | 2016-01-14 15:20:33 -0800 | [diff] [blame] | 2392 | list_for_each_entry(page, |
| 2393 | &zone->free_area[order].free_list[t], lru) { |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 2394 | unsigned long i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2395 | |
Geliang Tang | 86760a2 | 2016-01-14 15:20:33 -0800 | [diff] [blame] | 2396 | pfn = page_to_pfn(page); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 2397 | for (i = 0; i < (1UL << order); i++) |
Rafael J. Wysocki | 7be9823 | 2007-05-06 14:50:42 -0700 | [diff] [blame] | 2398 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 2399 | } |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 2400 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2401 | spin_unlock_irqrestore(&zone->lock, flags); |
| 2402 | } |
Mel Gorman | e2c55dc | 2007-10-16 01:25:50 -0700 | [diff] [blame] | 2403 | #endif /* CONFIG_PM */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2404 | |
| 2405 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2406 | * Free a 0-order page |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2407 | * cold == true ? free a cold page : free a hot page |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2408 | */ |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2409 | void free_hot_cold_page(struct page *page, bool cold) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2410 | { |
| 2411 | struct zone *zone = page_zone(page); |
| 2412 | struct per_cpu_pages *pcp; |
| 2413 | unsigned long flags; |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 2414 | unsigned long pfn = page_to_pfn(page); |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 2415 | int migratetype; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2416 | |
Mel Gorman | 4db7548 | 2016-05-19 17:14:32 -0700 | [diff] [blame] | 2417 | if (!free_pcp_prepare(page)) |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 2418 | return; |
| 2419 | |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 2420 | migratetype = get_pfnblock_migratetype(page, pfn); |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 2421 | set_pcppage_migratetype(page, migratetype); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2422 | local_irq_save(flags); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 2423 | __count_vm_event(PGFREE); |
Mel Gorman | da456f1 | 2009-06-16 15:32:08 -0700 | [diff] [blame] | 2424 | |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 2425 | /* |
| 2426 | * We only track unmovable, reclaimable and movable on pcp lists. |
| 2427 | * Free ISOLATE pages back to the allocator because they are being |
| 2428 | * offlined but treat RESERVE as movable pages so we can get those |
| 2429 | * areas back if necessary. Otherwise, we may have to free |
| 2430 | * excessively into the page allocator |
| 2431 | */ |
| 2432 | if (migratetype >= MIGRATE_PCPTYPES) { |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 2433 | if (unlikely(is_migrate_isolate(migratetype))) { |
Mel Gorman | dc4b0ca | 2014-06-04 16:10:17 -0700 | [diff] [blame] | 2434 | free_one_page(zone, page, pfn, 0, migratetype); |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 2435 | goto out; |
| 2436 | } |
| 2437 | migratetype = MIGRATE_MOVABLE; |
| 2438 | } |
| 2439 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 2440 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2441 | if (!cold) |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 2442 | list_add(&page->lru, &pcp->lists[migratetype]); |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2443 | else |
| 2444 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2445 | pcp->count++; |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 2446 | if (pcp->count >= pcp->high) { |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 2447 | unsigned long batch = READ_ONCE(pcp->batch); |
Cody P Schafer | 998d39c | 2013-07-03 15:01:32 -0700 | [diff] [blame] | 2448 | free_pcppages_bulk(zone, batch, pcp); |
| 2449 | pcp->count -= batch; |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 2450 | } |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 2451 | |
| 2452 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2453 | local_irq_restore(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2454 | } |
| 2455 | |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 2456 | /* |
Konstantin Khlebnikov | cc59850 | 2012-01-10 15:07:04 -0800 | [diff] [blame] | 2457 | * Free a list of 0-order pages |
| 2458 | */ |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2459 | void free_hot_cold_page_list(struct list_head *list, bool cold) |
Konstantin Khlebnikov | cc59850 | 2012-01-10 15:07:04 -0800 | [diff] [blame] | 2460 | { |
| 2461 | struct page *page, *next; |
| 2462 | |
| 2463 | list_for_each_entry_safe(page, next, list, lru) { |
Konstantin Khlebnikov | b413d48 | 2012-01-10 15:07:09 -0800 | [diff] [blame] | 2464 | trace_mm_page_free_batched(page, cold); |
Konstantin Khlebnikov | cc59850 | 2012-01-10 15:07:04 -0800 | [diff] [blame] | 2465 | free_hot_cold_page(page, cold); |
| 2466 | } |
| 2467 | } |
| 2468 | |
| 2469 | /* |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 2470 | * split_page takes a non-compound higher-order page, and splits it into |
| 2471 | * n (1<<order) sub-pages: page[0..n] |
| 2472 | * Each sub-page must be freed individually. |
| 2473 | * |
| 2474 | * Note: this is probably too low level an operation for use in drivers. |
| 2475 | * Please consult with lkml before using this in your driver. |
| 2476 | */ |
| 2477 | void split_page(struct page *page, unsigned int order) |
| 2478 | { |
| 2479 | int i; |
| 2480 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 2481 | VM_BUG_ON_PAGE(PageCompound(page), page); |
| 2482 | VM_BUG_ON_PAGE(!page_count(page), page); |
Vegard Nossum | b1eeab6 | 2008-11-25 16:55:53 +0100 | [diff] [blame] | 2483 | |
| 2484 | #ifdef CONFIG_KMEMCHECK |
| 2485 | /* |
| 2486 | * Split shadow pages too, because free(page[0]) would |
| 2487 | * otherwise free the whole shadow. |
| 2488 | */ |
| 2489 | if (kmemcheck_page_is_tracked(page)) |
| 2490 | split_page(virt_to_page(page[0].shadow), order); |
| 2491 | #endif |
| 2492 | |
Joonsoo Kim | a9627bc | 2016-07-26 15:23:49 -0700 | [diff] [blame] | 2493 | for (i = 1; i < (1 << order); i++) |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 2494 | set_page_refcounted(page + i); |
Joonsoo Kim | a9627bc | 2016-07-26 15:23:49 -0700 | [diff] [blame] | 2495 | split_page_owner(page, order); |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 2496 | } |
K. Y. Srinivasan | 5853ff2 | 2013-03-25 15:47:38 -0700 | [diff] [blame] | 2497 | EXPORT_SYMBOL_GPL(split_page); |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 2498 | |
Joonsoo Kim | 3c60509 | 2014-11-13 15:19:21 -0800 | [diff] [blame] | 2499 | int __isolate_free_page(struct page *page, unsigned int order) |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 2500 | { |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 2501 | unsigned long watermark; |
| 2502 | struct zone *zone; |
Bartlomiej Zolnierkiewicz | 2139cbe | 2012-10-08 16:32:00 -0700 | [diff] [blame] | 2503 | int mt; |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 2504 | |
| 2505 | BUG_ON(!PageBuddy(page)); |
| 2506 | |
| 2507 | zone = page_zone(page); |
Marek Szyprowski | 2e30abd | 2012-12-11 16:02:57 -0800 | [diff] [blame] | 2508 | mt = get_pageblock_migratetype(page); |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 2509 | |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 2510 | if (!is_migrate_isolate(mt)) { |
Vlastimil Babka | 8348faf | 2016-10-07 16:58:00 -0700 | [diff] [blame] | 2511 | /* |
| 2512 | * Obey watermarks as if the page was being allocated. We can |
| 2513 | * emulate a high-order watermark check with a raised order-0 |
| 2514 | * watermark, because we already know our high-order page |
| 2515 | * exists. |
| 2516 | */ |
| 2517 | watermark = min_wmark_pages(zone) + (1UL << order); |
Vlastimil Babka | 984fdba | 2016-10-07 16:57:57 -0700 | [diff] [blame] | 2518 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
Marek Szyprowski | 2e30abd | 2012-12-11 16:02:57 -0800 | [diff] [blame] | 2519 | return 0; |
| 2520 | |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 2521 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
Marek Szyprowski | 2e30abd | 2012-12-11 16:02:57 -0800 | [diff] [blame] | 2522 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 2523 | |
| 2524 | /* Remove page from free list */ |
| 2525 | list_del(&page->lru); |
| 2526 | zone->free_area[order].nr_free--; |
| 2527 | rmv_page_order(page); |
Bartlomiej Zolnierkiewicz | 2139cbe | 2012-10-08 16:32:00 -0700 | [diff] [blame] | 2528 | |
zhong jiang | 400bc7f | 2016-07-28 15:45:07 -0700 | [diff] [blame] | 2529 | /* |
| 2530 | * Set the pageblock if the isolated page is at least half of a |
| 2531 | * pageblock |
| 2532 | */ |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 2533 | if (order >= pageblock_order - 1) { |
| 2534 | struct page *endpage = page + (1 << order) - 1; |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 2535 | for (; page < endpage; page += pageblock_nr_pages) { |
| 2536 | int mt = get_pageblock_migratetype(page); |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 2537 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)) |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 2538 | set_pageblock_migratetype(page, |
| 2539 | MIGRATE_MOVABLE); |
| 2540 | } |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 2541 | } |
| 2542 | |
Joonsoo Kim | f3a14ce | 2015-07-17 16:24:15 -0700 | [diff] [blame] | 2543 | |
Mel Gorman | 8fb74b9 | 2013-01-11 14:32:16 -0800 | [diff] [blame] | 2544 | return 1UL << order; |
Mel Gorman | 1fb3f8c | 2012-10-08 16:29:12 -0700 | [diff] [blame] | 2545 | } |
| 2546 | |
| 2547 | /* |
Mel Gorman | 060e741 | 2016-05-19 17:13:27 -0700 | [diff] [blame] | 2548 | * Update NUMA hit/miss statistics |
| 2549 | * |
| 2550 | * Must be called with interrupts disabled. |
| 2551 | * |
| 2552 | * When __GFP_OTHER_NODE is set assume the node of the preferred |
| 2553 | * zone is the local node. This is useful for daemons who allocate |
| 2554 | * memory on behalf of other processes. |
| 2555 | */ |
| 2556 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z, |
| 2557 | gfp_t flags) |
| 2558 | { |
| 2559 | #ifdef CONFIG_NUMA |
| 2560 | int local_nid = numa_node_id(); |
| 2561 | enum zone_stat_item local_stat = NUMA_LOCAL; |
| 2562 | |
| 2563 | if (unlikely(flags & __GFP_OTHER_NODE)) { |
| 2564 | local_stat = NUMA_OTHER; |
| 2565 | local_nid = preferred_zone->node; |
| 2566 | } |
| 2567 | |
| 2568 | if (z->node == local_nid) { |
| 2569 | __inc_zone_state(z, NUMA_HIT); |
| 2570 | __inc_zone_state(z, local_stat); |
| 2571 | } else { |
| 2572 | __inc_zone_state(z, NUMA_MISS); |
| 2573 | __inc_zone_state(preferred_zone, NUMA_FOREIGN); |
| 2574 | } |
| 2575 | #endif |
| 2576 | } |
| 2577 | |
| 2578 | /* |
Vlastimil Babka | 7537919 | 2015-02-11 15:25:38 -0800 | [diff] [blame] | 2579 | * 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] | 2580 | */ |
Mel Gorman | 0a15c3e | 2009-06-16 15:32:05 -0700 | [diff] [blame] | 2581 | static inline |
| 2582 | struct page *buffered_rmqueue(struct zone *preferred_zone, |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 2583 | struct zone *zone, unsigned int order, |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 2584 | gfp_t gfp_flags, unsigned int alloc_flags, |
| 2585 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2586 | { |
| 2587 | unsigned long flags; |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 2588 | struct page *page; |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 2589 | bool cold = ((gfp_flags & __GFP_COLD) != 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2590 | |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 2591 | if (likely(order == 0)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2592 | struct per_cpu_pages *pcp; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 2593 | struct list_head *list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2594 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2595 | local_irq_save(flags); |
Mel Gorman | 479f854 | 2016-05-19 17:14:35 -0700 | [diff] [blame] | 2596 | do { |
| 2597 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
| 2598 | list = &pcp->lists[migratetype]; |
| 2599 | if (list_empty(list)) { |
| 2600 | pcp->count += rmqueue_bulk(zone, 0, |
| 2601 | pcp->batch, list, |
| 2602 | migratetype, cold); |
| 2603 | if (unlikely(list_empty(list))) |
| 2604 | goto failed; |
| 2605 | } |
Mel Gorman | b92a6ed | 2007-10-16 01:25:50 -0700 | [diff] [blame] | 2606 | |
Mel Gorman | 479f854 | 2016-05-19 17:14:35 -0700 | [diff] [blame] | 2607 | if (cold) |
| 2608 | page = list_last_entry(list, struct page, lru); |
| 2609 | else |
| 2610 | page = list_first_entry(list, struct page, lru); |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 2611 | |
Vlastimil Babka | 83b9355 | 2016-06-03 14:55:52 -0700 | [diff] [blame] | 2612 | list_del(&page->lru); |
| 2613 | pcp->count--; |
| 2614 | |
| 2615 | } while (check_new_pcp(page)); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2616 | } else { |
Michal Hocko | 0f352e5 | 2016-03-17 14:19:32 -0700 | [diff] [blame] | 2617 | /* |
| 2618 | * We most definitely don't want callers attempting to |
| 2619 | * allocate greater than order-1 page units with __GFP_NOFAIL. |
| 2620 | */ |
| 2621 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2622 | spin_lock_irqsave(&zone->lock, flags); |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2623 | |
Mel Gorman | 479f854 | 2016-05-19 17:14:35 -0700 | [diff] [blame] | 2624 | do { |
| 2625 | page = NULL; |
| 2626 | if (alloc_flags & ALLOC_HARDER) { |
| 2627 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); |
| 2628 | if (page) |
| 2629 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
| 2630 | } |
| 2631 | if (!page) |
| 2632 | page = __rmqueue(zone, order, migratetype); |
| 2633 | } while (page && check_new_pages(page, order)); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 2634 | spin_unlock(&zone->lock); |
| 2635 | if (!page) |
| 2636 | goto failed; |
Bartlomiej Zolnierkiewicz | d1ce749 | 2012-10-08 16:32:02 -0700 | [diff] [blame] | 2637 | __mod_zone_freepage_state(zone, -(1 << order), |
Vlastimil Babka | bb14c2c | 2015-09-08 15:01:25 -0700 | [diff] [blame] | 2638 | get_pcppage_migratetype(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2639 | } |
| 2640 | |
Mel Gorman | 16709d1 | 2016-07-28 15:46:56 -0700 | [diff] [blame] | 2641 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
Andi Kleen | 78afd56 | 2011-03-22 16:33:12 -0700 | [diff] [blame] | 2642 | zone_statistics(preferred_zone, zone, gfp_flags); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 2643 | local_irq_restore(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2644 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 2645 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2646 | return page; |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 2647 | |
| 2648 | failed: |
| 2649 | local_irq_restore(flags); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 2650 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2651 | } |
| 2652 | |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2653 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
| 2654 | |
Akinobu Mita | b2588c4 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 2655 | static struct { |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2656 | struct fault_attr attr; |
| 2657 | |
Viresh Kumar | 621a5f7 | 2015-09-26 15:04:07 -0700 | [diff] [blame] | 2658 | bool ignore_gfp_highmem; |
Mel Gorman | 71baba4 | 2015-11-06 16:28:28 -0800 | [diff] [blame] | 2659 | bool ignore_gfp_reclaim; |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 2660 | u32 min_order; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2661 | } fail_page_alloc = { |
| 2662 | .attr = FAULT_ATTR_INITIALIZER, |
Mel Gorman | 71baba4 | 2015-11-06 16:28:28 -0800 | [diff] [blame] | 2663 | .ignore_gfp_reclaim = true, |
Viresh Kumar | 621a5f7 | 2015-09-26 15:04:07 -0700 | [diff] [blame] | 2664 | .ignore_gfp_highmem = true, |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 2665 | .min_order = 1, |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2666 | }; |
| 2667 | |
| 2668 | static int __init setup_fail_page_alloc(char *str) |
| 2669 | { |
| 2670 | return setup_fault_attr(&fail_page_alloc.attr, str); |
| 2671 | } |
| 2672 | __setup("fail_page_alloc=", setup_fail_page_alloc); |
| 2673 | |
Gavin Shan | deaf386 | 2012-07-31 16:41:51 -0700 | [diff] [blame] | 2674 | 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] | 2675 | { |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 2676 | if (order < fail_page_alloc.min_order) |
Gavin Shan | deaf386 | 2012-07-31 16:41:51 -0700 | [diff] [blame] | 2677 | return false; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2678 | if (gfp_mask & __GFP_NOFAIL) |
Gavin Shan | deaf386 | 2012-07-31 16:41:51 -0700 | [diff] [blame] | 2679 | return false; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2680 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
Gavin Shan | deaf386 | 2012-07-31 16:41:51 -0700 | [diff] [blame] | 2681 | return false; |
Mel Gorman | 71baba4 | 2015-11-06 16:28:28 -0800 | [diff] [blame] | 2682 | if (fail_page_alloc.ignore_gfp_reclaim && |
| 2683 | (gfp_mask & __GFP_DIRECT_RECLAIM)) |
Gavin Shan | deaf386 | 2012-07-31 16:41:51 -0700 | [diff] [blame] | 2684 | return false; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2685 | |
| 2686 | return should_fail(&fail_page_alloc.attr, 1 << order); |
| 2687 | } |
| 2688 | |
| 2689 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS |
| 2690 | |
| 2691 | static int __init fail_page_alloc_debugfs(void) |
| 2692 | { |
Al Viro | f4ae40a | 2011-07-24 04:33:43 -0400 | [diff] [blame] | 2693 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2694 | struct dentry *dir; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2695 | |
Akinobu Mita | dd48c08 | 2011-08-03 16:21:01 -0700 | [diff] [blame] | 2696 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
| 2697 | &fail_page_alloc.attr); |
| 2698 | if (IS_ERR(dir)) |
| 2699 | return PTR_ERR(dir); |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2700 | |
Akinobu Mita | b2588c4 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 2701 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
Mel Gorman | 71baba4 | 2015-11-06 16:28:28 -0800 | [diff] [blame] | 2702 | &fail_page_alloc.ignore_gfp_reclaim)) |
Akinobu Mita | b2588c4 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 2703 | goto fail; |
| 2704 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, |
| 2705 | &fail_page_alloc.ignore_gfp_highmem)) |
| 2706 | goto fail; |
| 2707 | if (!debugfs_create_u32("min-order", mode, dir, |
| 2708 | &fail_page_alloc.min_order)) |
| 2709 | goto fail; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2710 | |
Akinobu Mita | b2588c4 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 2711 | return 0; |
| 2712 | fail: |
Akinobu Mita | dd48c08 | 2011-08-03 16:21:01 -0700 | [diff] [blame] | 2713 | debugfs_remove_recursive(dir); |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2714 | |
Akinobu Mita | b2588c4 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 2715 | return -ENOMEM; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2716 | } |
| 2717 | |
| 2718 | late_initcall(fail_page_alloc_debugfs); |
| 2719 | |
| 2720 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ |
| 2721 | |
| 2722 | #else /* CONFIG_FAIL_PAGE_ALLOC */ |
| 2723 | |
Gavin Shan | deaf386 | 2012-07-31 16:41:51 -0700 | [diff] [blame] | 2724 | 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] | 2725 | { |
Gavin Shan | deaf386 | 2012-07-31 16:41:51 -0700 | [diff] [blame] | 2726 | return false; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 2727 | } |
| 2728 | |
| 2729 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ |
| 2730 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2731 | /* |
Mel Gorman | 97a16fc | 2015-11-06 16:28:40 -0800 | [diff] [blame] | 2732 | * Return true if free base pages are above 'mark'. For high-order checks it |
| 2733 | * will return true of the order-0 watermark is reached and there is at least |
| 2734 | * one free page of a suitable size. Checking now avoids taking the zone lock |
| 2735 | * to check in the allocation paths if no pages are free. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2736 | */ |
Michal Hocko | 86a294a | 2016-05-20 16:57:12 -0700 | [diff] [blame] | 2737 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
| 2738 | int classzone_idx, unsigned int alloc_flags, |
| 2739 | long free_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2740 | { |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 2741 | long min = mark; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2742 | int o; |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 2743 | const bool alloc_harder = (alloc_flags & ALLOC_HARDER); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2744 | |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2745 | /* free_pages may go negative - that's OK */ |
Michal Hocko | df0a6da | 2012-01-10 15:08:02 -0800 | [diff] [blame] | 2746 | free_pages -= (1 << order) - 1; |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2747 | |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2748 | if (alloc_flags & ALLOC_HIGH) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2749 | min -= min / 2; |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2750 | |
| 2751 | /* |
| 2752 | * If the caller does not have rights to ALLOC_HARDER then subtract |
| 2753 | * the high-atomic reserves. This will over-estimate the size of the |
| 2754 | * atomic reserve but it avoids a search. |
| 2755 | */ |
Mel Gorman | 97a16fc | 2015-11-06 16:28:40 -0800 | [diff] [blame] | 2756 | if (likely(!alloc_harder)) |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2757 | free_pages -= z->nr_reserved_highatomic; |
| 2758 | else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2759 | min -= min / 4; |
Mel Gorman | e2b1919 | 2015-11-06 16:28:09 -0800 | [diff] [blame] | 2760 | |
Bartlomiej Zolnierkiewicz | d95ea5d | 2012-10-08 16:32:05 -0700 | [diff] [blame] | 2761 | #ifdef CONFIG_CMA |
| 2762 | /* If allocation can't use CMA areas don't use free CMA pages */ |
| 2763 | if (!(alloc_flags & ALLOC_CMA)) |
Mel Gorman | 97a16fc | 2015-11-06 16:28:40 -0800 | [diff] [blame] | 2764 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); |
Bartlomiej Zolnierkiewicz | d95ea5d | 2012-10-08 16:32:05 -0700 | [diff] [blame] | 2765 | #endif |
Tomasz Stanislawski | 026b081 | 2013-06-12 14:05:02 -0700 | [diff] [blame] | 2766 | |
Mel Gorman | 97a16fc | 2015-11-06 16:28:40 -0800 | [diff] [blame] | 2767 | /* |
| 2768 | * Check watermarks for an order-0 allocation request. If these |
| 2769 | * are not met, then a high-order request also cannot go ahead |
| 2770 | * even if a suitable page happened to be free. |
| 2771 | */ |
| 2772 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2773 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2774 | |
Mel Gorman | 97a16fc | 2015-11-06 16:28:40 -0800 | [diff] [blame] | 2775 | /* If this is an order-0 request then the watermark is fine */ |
| 2776 | if (!order) |
| 2777 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2778 | |
Mel Gorman | 97a16fc | 2015-11-06 16:28:40 -0800 | [diff] [blame] | 2779 | /* For a high-order request, check at least one suitable page is free */ |
| 2780 | for (o = order; o < MAX_ORDER; o++) { |
| 2781 | struct free_area *area = &z->free_area[o]; |
| 2782 | int mt; |
| 2783 | |
| 2784 | if (!area->nr_free) |
| 2785 | continue; |
| 2786 | |
| 2787 | if (alloc_harder) |
| 2788 | return true; |
| 2789 | |
| 2790 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
| 2791 | if (!list_empty(&area->free_list[mt])) |
| 2792 | return true; |
| 2793 | } |
| 2794 | |
| 2795 | #ifdef CONFIG_CMA |
| 2796 | if ((alloc_flags & ALLOC_CMA) && |
| 2797 | !list_empty(&area->free_list[MIGRATE_CMA])) { |
| 2798 | return true; |
| 2799 | } |
| 2800 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2801 | } |
Mel Gorman | 97a16fc | 2015-11-06 16:28:40 -0800 | [diff] [blame] | 2802 | return false; |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2803 | } |
| 2804 | |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 2805 | 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] | 2806 | int classzone_idx, unsigned int alloc_flags) |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2807 | { |
| 2808 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, |
| 2809 | zone_page_state(z, NR_FREE_PAGES)); |
| 2810 | } |
| 2811 | |
Mel Gorman | 48ee5f3 | 2016-05-19 17:14:07 -0700 | [diff] [blame] | 2812 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
| 2813 | unsigned long mark, int classzone_idx, unsigned int alloc_flags) |
| 2814 | { |
| 2815 | long free_pages = zone_page_state(z, NR_FREE_PAGES); |
| 2816 | long cma_pages = 0; |
| 2817 | |
| 2818 | #ifdef CONFIG_CMA |
| 2819 | /* If allocation can't use CMA areas don't use free CMA pages */ |
| 2820 | if (!(alloc_flags & ALLOC_CMA)) |
| 2821 | cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES); |
| 2822 | #endif |
| 2823 | |
| 2824 | /* |
| 2825 | * Fast check for order-0 only. If this fails then the reserves |
| 2826 | * need to be calculated. There is a corner case where the check |
| 2827 | * passes but only the high-order atomic reserve are free. If |
| 2828 | * the caller is !atomic then it'll uselessly search the free |
| 2829 | * list. That corner case is then slower but it is harmless. |
| 2830 | */ |
| 2831 | if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx]) |
| 2832 | return true; |
| 2833 | |
| 2834 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, |
| 2835 | free_pages); |
| 2836 | } |
| 2837 | |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 2838 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
Mel Gorman | e2b1919 | 2015-11-06 16:28:09 -0800 | [diff] [blame] | 2839 | unsigned long mark, int classzone_idx) |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2840 | { |
| 2841 | long free_pages = zone_page_state(z, NR_FREE_PAGES); |
| 2842 | |
| 2843 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) |
| 2844 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); |
| 2845 | |
Mel Gorman | e2b1919 | 2015-11-06 16:28:09 -0800 | [diff] [blame] | 2846 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2847 | free_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2848 | } |
| 2849 | |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2850 | #ifdef CONFIG_NUMA |
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 | { |
Mel Gorman | 5f7a75a | 2014-06-04 16:07:15 -0700 | [diff] [blame] | 2853 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) < |
| 2854 | RECLAIM_DISTANCE; |
David Rientjes | 957f822 | 2012-10-08 16:33:24 -0700 | [diff] [blame] | 2855 | } |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2856 | #else /* CONFIG_NUMA */ |
David Rientjes | 957f822 | 2012-10-08 16:33:24 -0700 | [diff] [blame] | 2857 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
| 2858 | { |
| 2859 | return true; |
| 2860 | } |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2861 | #endif /* CONFIG_NUMA */ |
| 2862 | |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2863 | /* |
Paul Jackson | 0798e51 | 2006-12-06 20:31:38 -0800 | [diff] [blame] | 2864 | * get_page_from_freelist goes through the zonelist trying to allocate |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2865 | * a page. |
| 2866 | */ |
| 2867 | static struct page * |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 2868 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
| 2869 | const struct alloc_context *ac) |
Martin Hicks | 753ee72 | 2005-06-21 17:14:41 -0700 | [diff] [blame] | 2870 | { |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 2871 | struct zoneref *z = ac->preferred_zoneref; |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2872 | struct zone *zone; |
Mel Gorman | 3b8c0be | 2016-07-28 15:46:53 -0700 | [diff] [blame] | 2873 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
| 2874 | |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2875 | /* |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2876 | * Scan zonelist, looking for a zone with enough free. |
Vladimir Davydov | 344736f | 2014-10-20 15:50:30 +0400 | [diff] [blame] | 2877 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2878 | */ |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 2879 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 2880 | ac->nodemask) { |
Mel Gorman | be06af0 | 2016-05-19 17:13:47 -0700 | [diff] [blame] | 2881 | struct page *page; |
Johannes Weiner | e085dbc | 2013-09-11 14:20:46 -0700 | [diff] [blame] | 2882 | unsigned long mark; |
| 2883 | |
Mel Gorman | 664eedd | 2014-06-04 16:10:08 -0700 | [diff] [blame] | 2884 | if (cpusets_enabled() && |
| 2885 | (alloc_flags & ALLOC_CPUSET) && |
Vlastimil Babka | 002f290 | 2016-05-19 17:14:30 -0700 | [diff] [blame] | 2886 | !__cpuset_zone_allowed(zone, gfp_mask)) |
Mel Gorman | cd38b11 | 2011-07-25 17:12:29 -0700 | [diff] [blame] | 2887 | continue; |
Johannes Weiner | a756cf5 | 2012-01-10 15:07:49 -0800 | [diff] [blame] | 2888 | /* |
| 2889 | * When allocating a page cache page for writing, we |
Mel Gorman | 281e372 | 2016-07-28 15:46:11 -0700 | [diff] [blame] | 2890 | * want to get it from a node that is within its dirty |
| 2891 | * limit, such that no single node holds more than its |
Johannes Weiner | a756cf5 | 2012-01-10 15:07:49 -0800 | [diff] [blame] | 2892 | * proportional share of globally allowed dirty pages. |
Mel Gorman | 281e372 | 2016-07-28 15:46:11 -0700 | [diff] [blame] | 2893 | * The dirty limits take into account the node's |
Johannes Weiner | a756cf5 | 2012-01-10 15:07:49 -0800 | [diff] [blame] | 2894 | * lowmem reserves and high watermark so that kswapd |
| 2895 | * should be able to balance it without having to |
| 2896 | * write pages from its LRU list. |
| 2897 | * |
Johannes Weiner | a756cf5 | 2012-01-10 15:07:49 -0800 | [diff] [blame] | 2898 | * XXX: For now, allow allocations to potentially |
Mel Gorman | 281e372 | 2016-07-28 15:46:11 -0700 | [diff] [blame] | 2899 | * exceed the per-node dirty limit in the slowpath |
Mel Gorman | c9ab0c4 | 2015-11-06 16:28:12 -0800 | [diff] [blame] | 2900 | * (spread_dirty_pages unset) before going into reclaim, |
Johannes Weiner | a756cf5 | 2012-01-10 15:07:49 -0800 | [diff] [blame] | 2901 | * which is important when on a NUMA setup the allowed |
Mel Gorman | 281e372 | 2016-07-28 15:46:11 -0700 | [diff] [blame] | 2902 | * nodes are together not big enough to reach the |
Johannes Weiner | a756cf5 | 2012-01-10 15:07:49 -0800 | [diff] [blame] | 2903 | * global limit. The proper fix for these situations |
Mel Gorman | 281e372 | 2016-07-28 15:46:11 -0700 | [diff] [blame] | 2904 | * will require awareness of nodes in the |
Johannes Weiner | a756cf5 | 2012-01-10 15:07:49 -0800 | [diff] [blame] | 2905 | * dirty-throttling and the flusher threads. |
| 2906 | */ |
Mel Gorman | 3b8c0be | 2016-07-28 15:46:53 -0700 | [diff] [blame] | 2907 | if (ac->spread_dirty_pages) { |
| 2908 | if (last_pgdat_dirty_limit == zone->zone_pgdat) |
| 2909 | continue; |
| 2910 | |
| 2911 | if (!node_dirty_ok(zone->zone_pgdat)) { |
| 2912 | last_pgdat_dirty_limit = zone->zone_pgdat; |
| 2913 | continue; |
| 2914 | } |
| 2915 | } |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2916 | |
Johannes Weiner | e085dbc | 2013-09-11 14:20:46 -0700 | [diff] [blame] | 2917 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
Mel Gorman | 48ee5f3 | 2016-05-19 17:14:07 -0700 | [diff] [blame] | 2918 | if (!zone_watermark_fast(zone, order, mark, |
Mel Gorman | 93ea996 | 2016-05-19 17:14:13 -0700 | [diff] [blame] | 2919 | ac_classzone_idx(ac), alloc_flags)) { |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 2920 | int ret; |
| 2921 | |
Mel Gorman | 5dab291 | 2014-06-04 16:10:14 -0700 | [diff] [blame] | 2922 | /* Checked here to keep the fast path fast */ |
| 2923 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
| 2924 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
| 2925 | goto try_this_zone; |
| 2926 | |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 2927 | if (node_reclaim_mode == 0 || |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 2928 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
Mel Gorman | cd38b11 | 2011-07-25 17:12:29 -0700 | [diff] [blame] | 2929 | continue; |
| 2930 | |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 2931 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 2932 | switch (ret) { |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 2933 | case NODE_RECLAIM_NOSCAN: |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 2934 | /* did not scan */ |
Mel Gorman | cd38b11 | 2011-07-25 17:12:29 -0700 | [diff] [blame] | 2935 | continue; |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 2936 | case NODE_RECLAIM_FULL: |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 2937 | /* scanned but unreclaimable */ |
Mel Gorman | cd38b11 | 2011-07-25 17:12:29 -0700 | [diff] [blame] | 2938 | continue; |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 2939 | default: |
| 2940 | /* did we reclaim enough */ |
Mel Gorman | fed2719 | 2013-04-29 15:07:57 -0700 | [diff] [blame] | 2941 | if (zone_watermark_ok(zone, order, mark, |
Mel Gorman | 93ea996 | 2016-05-19 17:14:13 -0700 | [diff] [blame] | 2942 | ac_classzone_idx(ac), alloc_flags)) |
Mel Gorman | fed2719 | 2013-04-29 15:07:57 -0700 | [diff] [blame] | 2943 | goto try_this_zone; |
| 2944 | |
Mel Gorman | fed2719 | 2013-04-29 15:07:57 -0700 | [diff] [blame] | 2945 | continue; |
Paul Jackson | 0798e51 | 2006-12-06 20:31:38 -0800 | [diff] [blame] | 2946 | } |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2947 | } |
| 2948 | |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 2949 | try_this_zone: |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 2950 | page = buffered_rmqueue(ac->preferred_zoneref->zone, zone, order, |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2951 | gfp_mask, alloc_flags, ac->migratetype); |
Vlastimil Babka | 7537919 | 2015-02-11 15:25:38 -0800 | [diff] [blame] | 2952 | if (page) { |
Mel Gorman | 479f854 | 2016-05-19 17:14:35 -0700 | [diff] [blame] | 2953 | prep_new_page(page, order, gfp_mask, alloc_flags); |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 2954 | |
| 2955 | /* |
| 2956 | * If this is a high-order atomic allocation then check |
| 2957 | * if the pageblock should be reserved for the future |
| 2958 | */ |
| 2959 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) |
| 2960 | reserve_highatomic_pageblock(page, zone, order); |
| 2961 | |
Vlastimil Babka | 7537919 | 2015-02-11 15:25:38 -0800 | [diff] [blame] | 2962 | return page; |
| 2963 | } |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2964 | } |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 2965 | |
Mel Gorman | 4ffeaf3 | 2014-08-06 16:07:22 -0700 | [diff] [blame] | 2966 | return NULL; |
Martin Hicks | 753ee72 | 2005-06-21 17:14:41 -0700 | [diff] [blame] | 2967 | } |
| 2968 | |
David Rientjes | 29423e77 | 2011-03-22 16:30:47 -0700 | [diff] [blame] | 2969 | /* |
| 2970 | * Large machines with many possible nodes should not always dump per-node |
| 2971 | * meminfo in irq context. |
| 2972 | */ |
| 2973 | static inline bool should_suppress_show_mem(void) |
| 2974 | { |
| 2975 | bool ret = false; |
| 2976 | |
| 2977 | #if NODES_SHIFT > 8 |
| 2978 | ret = in_interrupt(); |
| 2979 | #endif |
| 2980 | return ret; |
| 2981 | } |
| 2982 | |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 2983 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
| 2984 | DEFAULT_RATELIMIT_INTERVAL, |
| 2985 | DEFAULT_RATELIMIT_BURST); |
| 2986 | |
Michal Hocko | 7877cdc | 2016-10-07 17:01:55 -0700 | [diff] [blame] | 2987 | void warn_alloc(gfp_t gfp_mask, const char *fmt, ...) |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 2988 | { |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 2989 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
Michal Hocko | 7877cdc | 2016-10-07 17:01:55 -0700 | [diff] [blame] | 2990 | struct va_format vaf; |
| 2991 | va_list args; |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 2992 | |
Stanislaw Gruszka | c0a32fc | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 2993 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
| 2994 | debug_guardpage_minorder() > 0) |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 2995 | return; |
| 2996 | |
| 2997 | /* |
| 2998 | * This documents exceptions given to allocations in certain |
| 2999 | * contexts that are allowed to allocate outside current's set |
| 3000 | * of allowed nodes. |
| 3001 | */ |
| 3002 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
| 3003 | if (test_thread_flag(TIF_MEMDIE) || |
| 3004 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
| 3005 | filter &= ~SHOW_MEM_FILTER_NODES; |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3006 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 3007 | filter &= ~SHOW_MEM_FILTER_NODES; |
| 3008 | |
Michal Hocko | 7877cdc | 2016-10-07 17:01:55 -0700 | [diff] [blame] | 3009 | pr_warn("%s: ", current->comm); |
Joe Perches | 3ee9a4f | 2011-10-31 17:08:35 -0700 | [diff] [blame] | 3010 | |
Michal Hocko | 7877cdc | 2016-10-07 17:01:55 -0700 | [diff] [blame] | 3011 | va_start(args, fmt); |
| 3012 | vaf.fmt = fmt; |
| 3013 | vaf.va = &args; |
| 3014 | pr_cont("%pV", &vaf); |
| 3015 | va_end(args); |
Joe Perches | 3ee9a4f | 2011-10-31 17:08:35 -0700 | [diff] [blame] | 3016 | |
Michal Hocko | 7877cdc | 2016-10-07 17:01:55 -0700 | [diff] [blame] | 3017 | pr_cont(", mode:%#x(%pGg)\n", gfp_mask, &gfp_mask); |
Joe Perches | 3ee9a4f | 2011-10-31 17:08:35 -0700 | [diff] [blame] | 3018 | |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 3019 | dump_stack(); |
| 3020 | if (!should_suppress_show_mem()) |
| 3021 | show_mem(filter); |
| 3022 | } |
| 3023 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3024 | static inline struct page * |
| 3025 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3026 | const struct alloc_context *ac, unsigned long *did_some_progress) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3027 | { |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 3028 | struct oom_control oc = { |
| 3029 | .zonelist = ac->zonelist, |
| 3030 | .nodemask = ac->nodemask, |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 3031 | .memcg = NULL, |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 3032 | .gfp_mask = gfp_mask, |
| 3033 | .order = order, |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 3034 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3035 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3036 | |
Johannes Weiner | 9879de7 | 2015-01-26 12:58:32 -0800 | [diff] [blame] | 3037 | *did_some_progress = 0; |
| 3038 | |
Johannes Weiner | 9879de7 | 2015-01-26 12:58:32 -0800 | [diff] [blame] | 3039 | /* |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 3040 | * Acquire the oom lock. If that fails, somebody else is |
| 3041 | * making progress for us. |
Johannes Weiner | 9879de7 | 2015-01-26 12:58:32 -0800 | [diff] [blame] | 3042 | */ |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 3043 | if (!mutex_trylock(&oom_lock)) { |
Johannes Weiner | 9879de7 | 2015-01-26 12:58:32 -0800 | [diff] [blame] | 3044 | *did_some_progress = 1; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3045 | schedule_timeout_uninterruptible(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3046 | return NULL; |
| 3047 | } |
Jens Axboe | 6b1de91 | 2005-11-17 21:35:02 +0100 | [diff] [blame] | 3048 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3049 | /* |
| 3050 | * Go through the zonelist yet one more time, keep very high watermark |
| 3051 | * here, this is only to catch a parallel oom killing, we must fail if |
| 3052 | * we're still under heavy pressure. |
| 3053 | */ |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3054 | page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order, |
| 3055 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 3056 | if (page) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3057 | goto out; |
| 3058 | |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 3059 | if (!(gfp_mask & __GFP_NOFAIL)) { |
Johannes Weiner | 9879de7 | 2015-01-26 12:58:32 -0800 | [diff] [blame] | 3060 | /* Coredumps can quickly deplete all memory reserves */ |
| 3061 | if (current->flags & PF_DUMPCORE) |
| 3062 | goto out; |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 3063 | /* The OOM killer will not help higher order allocs */ |
| 3064 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
| 3065 | goto out; |
David Rientjes | 03668b3 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 3066 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3067 | if (ac->high_zoneidx < ZONE_NORMAL) |
David Rientjes | 03668b3 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 3068 | goto out; |
Johannes Weiner | 9083905 | 2015-06-24 16:57:21 -0700 | [diff] [blame] | 3069 | if (pm_suspended_storage()) |
| 3070 | goto out; |
Michal Hocko | 3da88fb | 2016-05-19 17:13:09 -0700 | [diff] [blame] | 3071 | /* |
| 3072 | * XXX: GFP_NOFS allocations should rather fail than rely on |
| 3073 | * other request to make a forward progress. |
| 3074 | * We are in an unfortunate situation where out_of_memory cannot |
| 3075 | * do much for this context but let's try it to at least get |
| 3076 | * access to memory reserved if the current task is killed (see |
| 3077 | * out_of_memory). Once filesystems are ready to handle allocation |
| 3078 | * failures more gracefully we should just bail out here. |
| 3079 | */ |
| 3080 | |
David Rientjes | 4167e9b | 2015-04-14 15:46:55 -0700 | [diff] [blame] | 3081 | /* The OOM killer may not free memory on a specific node */ |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 3082 | if (gfp_mask & __GFP_THISNODE) |
| 3083 | goto out; |
| 3084 | } |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3085 | /* Exhausted what can be done so it's blamo time */ |
Michal Hocko | 5020e28 | 2016-01-14 15:20:36 -0800 | [diff] [blame] | 3086 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 3087 | *did_some_progress = 1; |
Michal Hocko | 5020e28 | 2016-01-14 15:20:36 -0800 | [diff] [blame] | 3088 | |
| 3089 | if (gfp_mask & __GFP_NOFAIL) { |
| 3090 | page = get_page_from_freelist(gfp_mask, order, |
| 3091 | ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac); |
| 3092 | /* |
| 3093 | * fallback to ignore cpuset restriction if our nodes |
| 3094 | * are depleted |
| 3095 | */ |
| 3096 | if (!page) |
| 3097 | page = get_page_from_freelist(gfp_mask, order, |
| 3098 | ALLOC_NO_WATERMARKS, ac); |
| 3099 | } |
| 3100 | } |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3101 | out: |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 3102 | mutex_unlock(&oom_lock); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3103 | return page; |
| 3104 | } |
| 3105 | |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3106 | /* |
| 3107 | * Maximum number of compaction retries wit a progress before OOM |
| 3108 | * killer is consider as the only way to move forward. |
| 3109 | */ |
| 3110 | #define MAX_COMPACT_RETRIES 16 |
| 3111 | |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3112 | #ifdef CONFIG_COMPACTION |
| 3113 | /* Try memory compaction for high-order allocations before reclaim */ |
| 3114 | static struct page * |
| 3115 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 3116 | unsigned int alloc_flags, const struct alloc_context *ac, |
Vlastimil Babka | a5508cd | 2016-07-28 15:49:28 -0700 | [diff] [blame] | 3117 | enum compact_priority prio, enum compact_result *compact_result) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3118 | { |
Vlastimil Babka | 98dd3b4 | 2014-10-09 15:27:04 -0700 | [diff] [blame] | 3119 | struct page *page; |
Vlastimil Babka | 53853e2 | 2014-10-09 15:27:02 -0700 | [diff] [blame] | 3120 | |
Mel Gorman | 6619971 | 2012-01-12 17:19:41 -0800 | [diff] [blame] | 3121 | if (!order) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3122 | return NULL; |
| 3123 | |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 3124 | current->flags |= PF_MEMALLOC; |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3125 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
Vlastimil Babka | c3486f5 | 2016-07-28 15:49:30 -0700 | [diff] [blame] | 3126 | prio); |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 3127 | current->flags &= ~PF_MEMALLOC; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3128 | |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3129 | if (*compact_result <= COMPACT_INACTIVE) |
Vlastimil Babka | 98dd3b4 | 2014-10-09 15:27:04 -0700 | [diff] [blame] | 3130 | return NULL; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3131 | |
Vlastimil Babka | 98dd3b4 | 2014-10-09 15:27:04 -0700 | [diff] [blame] | 3132 | /* |
| 3133 | * At least in one zone compaction wasn't deferred or skipped, so let's |
| 3134 | * count a compaction stall |
| 3135 | */ |
| 3136 | count_vm_event(COMPACTSTALL); |
| 3137 | |
Vlastimil Babka | 31a6c19 | 2016-07-28 15:49:13 -0700 | [diff] [blame] | 3138 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
Vlastimil Babka | 98dd3b4 | 2014-10-09 15:27:04 -0700 | [diff] [blame] | 3139 | |
| 3140 | if (page) { |
| 3141 | struct zone *zone = page_zone(page); |
| 3142 | |
| 3143 | zone->compact_blockskip_flush = false; |
| 3144 | compaction_defer_reset(zone, order, true); |
| 3145 | count_vm_event(COMPACTSUCCESS); |
| 3146 | return page; |
| 3147 | } |
| 3148 | |
| 3149 | /* |
Vlastimil Babka | 98dd3b4 | 2014-10-09 15:27:04 -0700 | [diff] [blame] | 3150 | * It's bad if compaction run occurs and fails. The most likely reason |
| 3151 | * is that pages exist, but not enough to satisfy watermarks. |
| 3152 | */ |
| 3153 | count_vm_event(COMPACTFAIL); |
| 3154 | |
| 3155 | cond_resched(); |
| 3156 | |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3157 | return NULL; |
| 3158 | } |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3159 | |
Vlastimil Babka | 3250845 | 2016-10-07 17:00:28 -0700 | [diff] [blame] | 3160 | static inline bool |
| 3161 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, |
| 3162 | enum compact_result compact_result, |
| 3163 | enum compact_priority *compact_priority, |
Vlastimil Babka | d943649 | 2016-10-07 17:00:31 -0700 | [diff] [blame] | 3164 | int *compaction_retries) |
Vlastimil Babka | 3250845 | 2016-10-07 17:00:28 -0700 | [diff] [blame] | 3165 | { |
| 3166 | int max_retries = MAX_COMPACT_RETRIES; |
Vlastimil Babka | c2033b0 | 2016-10-07 17:00:34 -0700 | [diff] [blame] | 3167 | int min_priority; |
Vlastimil Babka | 3250845 | 2016-10-07 17:00:28 -0700 | [diff] [blame] | 3168 | |
| 3169 | if (!order) |
| 3170 | return false; |
| 3171 | |
Vlastimil Babka | d943649 | 2016-10-07 17:00:31 -0700 | [diff] [blame] | 3172 | if (compaction_made_progress(compact_result)) |
| 3173 | (*compaction_retries)++; |
| 3174 | |
Vlastimil Babka | 3250845 | 2016-10-07 17:00:28 -0700 | [diff] [blame] | 3175 | /* |
| 3176 | * compaction considers all the zone as desperately out of memory |
| 3177 | * so it doesn't really make much sense to retry except when the |
| 3178 | * failure could be caused by insufficient priority |
| 3179 | */ |
Vlastimil Babka | d943649 | 2016-10-07 17:00:31 -0700 | [diff] [blame] | 3180 | if (compaction_failed(compact_result)) |
| 3181 | goto check_priority; |
Vlastimil Babka | 3250845 | 2016-10-07 17:00:28 -0700 | [diff] [blame] | 3182 | |
| 3183 | /* |
| 3184 | * make sure the compaction wasn't deferred or didn't bail out early |
| 3185 | * due to locks contention before we declare that we should give up. |
| 3186 | * But do not retry if the given zonelist is not suitable for |
| 3187 | * compaction. |
| 3188 | */ |
| 3189 | if (compaction_withdrawn(compact_result)) |
| 3190 | return compaction_zonelist_suitable(ac, order, alloc_flags); |
| 3191 | |
| 3192 | /* |
| 3193 | * !costly requests are much more important than __GFP_REPEAT |
| 3194 | * costly ones because they are de facto nofail and invoke OOM |
| 3195 | * killer to move on while costly can fail and users are ready |
| 3196 | * to cope with that. 1/4 retries is rather arbitrary but we |
| 3197 | * would need much more detailed feedback from compaction to |
| 3198 | * make a better decision. |
| 3199 | */ |
| 3200 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
| 3201 | max_retries /= 4; |
Vlastimil Babka | d943649 | 2016-10-07 17:00:31 -0700 | [diff] [blame] | 3202 | if (*compaction_retries <= max_retries) |
Vlastimil Babka | 3250845 | 2016-10-07 17:00:28 -0700 | [diff] [blame] | 3203 | return true; |
| 3204 | |
Vlastimil Babka | d943649 | 2016-10-07 17:00:31 -0700 | [diff] [blame] | 3205 | /* |
| 3206 | * Make sure there are attempts at the highest priority if we exhausted |
| 3207 | * all retries or failed at the lower priorities. |
| 3208 | */ |
| 3209 | check_priority: |
Vlastimil Babka | c2033b0 | 2016-10-07 17:00:34 -0700 | [diff] [blame] | 3210 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
| 3211 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; |
| 3212 | if (*compact_priority > min_priority) { |
Vlastimil Babka | d943649 | 2016-10-07 17:00:31 -0700 | [diff] [blame] | 3213 | (*compact_priority)--; |
| 3214 | *compaction_retries = 0; |
| 3215 | return true; |
| 3216 | } |
Vlastimil Babka | 3250845 | 2016-10-07 17:00:28 -0700 | [diff] [blame] | 3217 | return false; |
| 3218 | } |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3219 | #else |
| 3220 | static inline struct page * |
| 3221 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 3222 | unsigned int alloc_flags, const struct alloc_context *ac, |
Vlastimil Babka | a5508cd | 2016-07-28 15:49:28 -0700 | [diff] [blame] | 3223 | enum compact_priority prio, enum compact_result *compact_result) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3224 | { |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3225 | *compact_result = COMPACT_SKIPPED; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3226 | return NULL; |
| 3227 | } |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3228 | |
| 3229 | static inline bool |
Michal Hocko | 86a294a | 2016-05-20 16:57:12 -0700 | [diff] [blame] | 3230 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
| 3231 | enum compact_result compact_result, |
Vlastimil Babka | a5508cd | 2016-07-28 15:49:28 -0700 | [diff] [blame] | 3232 | enum compact_priority *compact_priority, |
Vlastimil Babka | d943649 | 2016-10-07 17:00:31 -0700 | [diff] [blame] | 3233 | int *compaction_retries) |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3234 | { |
Michal Hocko | 31e49bf | 2016-05-20 16:57:15 -0700 | [diff] [blame] | 3235 | struct zone *zone; |
| 3236 | struct zoneref *z; |
| 3237 | |
| 3238 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) |
| 3239 | return false; |
| 3240 | |
| 3241 | /* |
| 3242 | * There are setups with compaction disabled which would prefer to loop |
| 3243 | * inside the allocator rather than hit the oom killer prematurely. |
| 3244 | * Let's give them a good hope and keep retrying while the order-0 |
| 3245 | * watermarks are OK. |
| 3246 | */ |
| 3247 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
| 3248 | ac->nodemask) { |
| 3249 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), |
| 3250 | ac_classzone_idx(ac), alloc_flags)) |
| 3251 | return true; |
| 3252 | } |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3253 | return false; |
| 3254 | } |
Vlastimil Babka | 3250845 | 2016-10-07 17:00:28 -0700 | [diff] [blame] | 3255 | #endif /* CONFIG_COMPACTION */ |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 3256 | |
Marek Szyprowski | bba9071 | 2012-01-25 12:09:52 +0100 | [diff] [blame] | 3257 | /* Perform direct synchronous page reclaim */ |
| 3258 | static int |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3259 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
| 3260 | const struct alloc_context *ac) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3261 | { |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3262 | struct reclaim_state reclaim_state; |
Marek Szyprowski | bba9071 | 2012-01-25 12:09:52 +0100 | [diff] [blame] | 3263 | int progress; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3264 | |
| 3265 | cond_resched(); |
| 3266 | |
| 3267 | /* We now go into synchronous reclaim */ |
| 3268 | cpuset_memory_pressure_bump(); |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 3269 | current->flags |= PF_MEMALLOC; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3270 | lockdep_set_current_reclaim_state(gfp_mask); |
| 3271 | reclaim_state.reclaimed_slab = 0; |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 3272 | current->reclaim_state = &reclaim_state; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3273 | |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3274 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
| 3275 | ac->nodemask); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3276 | |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 3277 | current->reclaim_state = NULL; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3278 | lockdep_clear_current_reclaim_state(); |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 3279 | current->flags &= ~PF_MEMALLOC; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3280 | |
| 3281 | cond_resched(); |
| 3282 | |
Marek Szyprowski | bba9071 | 2012-01-25 12:09:52 +0100 | [diff] [blame] | 3283 | return progress; |
| 3284 | } |
| 3285 | |
| 3286 | /* The really slow allocator path where we enter direct reclaim */ |
| 3287 | static inline struct page * |
| 3288 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 3289 | unsigned int alloc_flags, const struct alloc_context *ac, |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3290 | unsigned long *did_some_progress) |
Marek Szyprowski | bba9071 | 2012-01-25 12:09:52 +0100 | [diff] [blame] | 3291 | { |
| 3292 | struct page *page = NULL; |
| 3293 | bool drained = false; |
| 3294 | |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3295 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
Mel Gorman | 9ee493c | 2010-09-09 16:38:18 -0700 | [diff] [blame] | 3296 | if (unlikely(!(*did_some_progress))) |
| 3297 | return NULL; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3298 | |
Mel Gorman | 9ee493c | 2010-09-09 16:38:18 -0700 | [diff] [blame] | 3299 | retry: |
Vlastimil Babka | 31a6c19 | 2016-07-28 15:49:13 -0700 | [diff] [blame] | 3300 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
Mel Gorman | 9ee493c | 2010-09-09 16:38:18 -0700 | [diff] [blame] | 3301 | |
| 3302 | /* |
| 3303 | * If an allocation failed after direct reclaim, it could be because |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 3304 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
| 3305 | * Shrink them them and try again |
Mel Gorman | 9ee493c | 2010-09-09 16:38:18 -0700 | [diff] [blame] | 3306 | */ |
| 3307 | if (!page && !drained) { |
Mel Gorman | 0aaa29a | 2015-11-06 16:28:37 -0800 | [diff] [blame] | 3308 | unreserve_highatomic_pageblock(ac); |
Vlastimil Babka | 93481ff | 2014-12-10 15:43:01 -0800 | [diff] [blame] | 3309 | drain_all_pages(NULL); |
Mel Gorman | 9ee493c | 2010-09-09 16:38:18 -0700 | [diff] [blame] | 3310 | drained = true; |
| 3311 | goto retry; |
| 3312 | } |
| 3313 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3314 | return page; |
| 3315 | } |
| 3316 | |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3317 | 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] | 3318 | { |
| 3319 | struct zoneref *z; |
| 3320 | struct zone *zone; |
Mel Gorman | e1a5563 | 2016-07-28 15:46:26 -0700 | [diff] [blame] | 3321 | pg_data_t *last_pgdat = NULL; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3322 | |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3323 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
Mel Gorman | e1a5563 | 2016-07-28 15:46:26 -0700 | [diff] [blame] | 3324 | ac->high_zoneidx, ac->nodemask) { |
| 3325 | if (last_pgdat != zone->zone_pgdat) |
Mel Gorman | 52e9f87 | 2016-07-28 15:46:29 -0700 | [diff] [blame] | 3326 | wakeup_kswapd(zone, order, ac->high_zoneidx); |
Mel Gorman | e1a5563 | 2016-07-28 15:46:26 -0700 | [diff] [blame] | 3327 | last_pgdat = zone->zone_pgdat; |
| 3328 | } |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3329 | } |
| 3330 | |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 3331 | static inline unsigned int |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3332 | gfp_to_alloc_flags(gfp_t gfp_mask) |
| 3333 | { |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 3334 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3335 | |
Mel Gorman | a56f57f | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3336 | /* __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] | 3337 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
Mel Gorman | a56f57f | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3338 | |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3339 | /* |
| 3340 | * The caller may dip into page reserves a bit more if the caller |
| 3341 | * cannot run direct reclaim, or if the caller has realtime scheduling |
| 3342 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3343 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3344 | */ |
Namhyung Kim | e6223a3 | 2010-10-26 14:21:59 -0700 | [diff] [blame] | 3345 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3346 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3347 | if (gfp_mask & __GFP_ATOMIC) { |
Andrea Arcangeli | 5c3240d | 2011-01-13 15:46:49 -0800 | [diff] [blame] | 3348 | /* |
David Rientjes | b104a35 | 2014-07-30 16:08:24 -0700 | [diff] [blame] | 3349 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
| 3350 | * if it can't schedule. |
Andrea Arcangeli | 5c3240d | 2011-01-13 15:46:49 -0800 | [diff] [blame] | 3351 | */ |
David Rientjes | b104a35 | 2014-07-30 16:08:24 -0700 | [diff] [blame] | 3352 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
Andrea Arcangeli | 5c3240d | 2011-01-13 15:46:49 -0800 | [diff] [blame] | 3353 | alloc_flags |= ALLOC_HARDER; |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3354 | /* |
David Rientjes | b104a35 | 2014-07-30 16:08:24 -0700 | [diff] [blame] | 3355 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
Vladimir Davydov | 344736f | 2014-10-20 15:50:30 +0400 | [diff] [blame] | 3356 | * comment for __cpuset_node_allowed(). |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3357 | */ |
| 3358 | alloc_flags &= ~ALLOC_CPUSET; |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 3359 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3360 | alloc_flags |= ALLOC_HARDER; |
| 3361 | |
Bartlomiej Zolnierkiewicz | d95ea5d | 2012-10-08 16:32:05 -0700 | [diff] [blame] | 3362 | #ifdef CONFIG_CMA |
David Rientjes | 43e7a34 | 2014-10-09 15:27:25 -0700 | [diff] [blame] | 3363 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
Bartlomiej Zolnierkiewicz | d95ea5d | 2012-10-08 16:32:05 -0700 | [diff] [blame] | 3364 | alloc_flags |= ALLOC_CMA; |
| 3365 | #endif |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3366 | return alloc_flags; |
| 3367 | } |
| 3368 | |
Mel Gorman | 072bb0a | 2012-07-31 16:43:58 -0700 | [diff] [blame] | 3369 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
| 3370 | { |
Vlastimil Babka | 31a6c19 | 2016-07-28 15:49:13 -0700 | [diff] [blame] | 3371 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) |
| 3372 | return false; |
| 3373 | |
| 3374 | if (gfp_mask & __GFP_MEMALLOC) |
| 3375 | return true; |
| 3376 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
| 3377 | return true; |
| 3378 | if (!in_interrupt() && |
| 3379 | ((current->flags & PF_MEMALLOC) || |
| 3380 | unlikely(test_thread_flag(TIF_MEMDIE)))) |
| 3381 | return true; |
| 3382 | |
| 3383 | return false; |
Mel Gorman | 072bb0a | 2012-07-31 16:43:58 -0700 | [diff] [blame] | 3384 | } |
| 3385 | |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3386 | /* |
| 3387 | * Maximum number of reclaim retries without any progress before OOM killer |
| 3388 | * is consider as the only way to move forward. |
| 3389 | */ |
| 3390 | #define MAX_RECLAIM_RETRIES 16 |
| 3391 | |
| 3392 | /* |
| 3393 | * Checks whether it makes sense to retry the reclaim to make a forward progress |
| 3394 | * for the given allocation request. |
| 3395 | * The reclaim feedback represented by did_some_progress (any progress during |
Michal Hocko | 7854ea6 | 2016-05-20 16:57:09 -0700 | [diff] [blame] | 3396 | * the last reclaim round) and no_progress_loops (number of reclaim rounds without |
| 3397 | * any progress in a row) is considered as well as the reclaimable pages on the |
| 3398 | * applicable zone list (with a backoff mechanism which is a function of |
| 3399 | * no_progress_loops). |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3400 | * |
| 3401 | * Returns true if a retry is viable or false to enter the oom path. |
| 3402 | */ |
| 3403 | static inline bool |
| 3404 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, |
| 3405 | struct alloc_context *ac, int alloc_flags, |
Vlastimil Babka | 423b452 | 2016-10-07 17:00:40 -0700 | [diff] [blame] | 3406 | bool did_some_progress, int *no_progress_loops) |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3407 | { |
| 3408 | struct zone *zone; |
| 3409 | struct zoneref *z; |
| 3410 | |
| 3411 | /* |
Vlastimil Babka | 423b452 | 2016-10-07 17:00:40 -0700 | [diff] [blame] | 3412 | * Costly allocations might have made a progress but this doesn't mean |
| 3413 | * their order will become available due to high fragmentation so |
| 3414 | * always increment the no progress counter for them |
| 3415 | */ |
| 3416 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) |
| 3417 | *no_progress_loops = 0; |
| 3418 | else |
| 3419 | (*no_progress_loops)++; |
| 3420 | |
| 3421 | /* |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3422 | * Make sure we converge to OOM if we cannot make any progress |
| 3423 | * several times in the row. |
| 3424 | */ |
Vlastimil Babka | 423b452 | 2016-10-07 17:00:40 -0700 | [diff] [blame] | 3425 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3426 | return false; |
| 3427 | |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3428 | /* |
Mel Gorman | bca6759 | 2016-07-28 15:47:05 -0700 | [diff] [blame] | 3429 | * Keep reclaiming pages while there is a chance this will lead |
| 3430 | * somewhere. If none of the target zones can satisfy our allocation |
| 3431 | * request even if all reclaimable pages are considered then we are |
| 3432 | * screwed and have to go OOM. |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3433 | */ |
| 3434 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
| 3435 | ac->nodemask) { |
| 3436 | unsigned long available; |
Michal Hocko | ede3771 | 2016-05-20 16:57:03 -0700 | [diff] [blame] | 3437 | unsigned long reclaimable; |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3438 | |
Mel Gorman | 5a1c84b | 2016-07-28 15:47:31 -0700 | [diff] [blame] | 3439 | available = reclaimable = zone_reclaimable_pages(zone); |
Vlastimil Babka | 423b452 | 2016-10-07 17:00:40 -0700 | [diff] [blame] | 3440 | available -= DIV_ROUND_UP((*no_progress_loops) * available, |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3441 | MAX_RECLAIM_RETRIES); |
Mel Gorman | 5a1c84b | 2016-07-28 15:47:31 -0700 | [diff] [blame] | 3442 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3443 | |
| 3444 | /* |
| 3445 | * Would the allocation succeed if we reclaimed the whole |
Mel Gorman | 5a1c84b | 2016-07-28 15:47:31 -0700 | [diff] [blame] | 3446 | * available? |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3447 | */ |
Mel Gorman | 5a1c84b | 2016-07-28 15:47:31 -0700 | [diff] [blame] | 3448 | if (__zone_watermark_ok(zone, order, min_wmark_pages(zone), |
| 3449 | ac_classzone_idx(ac), alloc_flags, available)) { |
Michal Hocko | ede3771 | 2016-05-20 16:57:03 -0700 | [diff] [blame] | 3450 | /* |
| 3451 | * If we didn't make any progress and have a lot of |
| 3452 | * dirty + writeback pages then we should wait for |
| 3453 | * an IO to complete to slow down the reclaim and |
| 3454 | * prevent from pre mature OOM |
| 3455 | */ |
| 3456 | if (!did_some_progress) { |
Mel Gorman | 11fb998 | 2016-07-28 15:46:20 -0700 | [diff] [blame] | 3457 | unsigned long write_pending; |
Michal Hocko | ede3771 | 2016-05-20 16:57:03 -0700 | [diff] [blame] | 3458 | |
Mel Gorman | 5a1c84b | 2016-07-28 15:47:31 -0700 | [diff] [blame] | 3459 | write_pending = zone_page_state_snapshot(zone, |
| 3460 | NR_ZONE_WRITE_PENDING); |
Michal Hocko | ede3771 | 2016-05-20 16:57:03 -0700 | [diff] [blame] | 3461 | |
Mel Gorman | 11fb998 | 2016-07-28 15:46:20 -0700 | [diff] [blame] | 3462 | if (2 * write_pending > reclaimable) { |
Michal Hocko | ede3771 | 2016-05-20 16:57:03 -0700 | [diff] [blame] | 3463 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
| 3464 | return true; |
| 3465 | } |
| 3466 | } |
Mel Gorman | 5a1c84b | 2016-07-28 15:47:31 -0700 | [diff] [blame] | 3467 | |
Michal Hocko | ede3771 | 2016-05-20 16:57:03 -0700 | [diff] [blame] | 3468 | /* |
| 3469 | * Memory allocation/reclaim might be called from a WQ |
| 3470 | * context and the current implementation of the WQ |
| 3471 | * concurrency control doesn't recognize that |
| 3472 | * a particular WQ is congested if the worker thread is |
| 3473 | * looping without ever sleeping. Therefore we have to |
| 3474 | * do a short sleep here rather than calling |
| 3475 | * cond_resched(). |
| 3476 | */ |
| 3477 | if (current->flags & PF_WQ_WORKER) |
| 3478 | schedule_timeout_uninterruptible(1); |
| 3479 | else |
| 3480 | cond_resched(); |
| 3481 | |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3482 | return true; |
| 3483 | } |
| 3484 | } |
| 3485 | |
| 3486 | return false; |
| 3487 | } |
| 3488 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3489 | static inline struct page * |
| 3490 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3491 | struct alloc_context *ac) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3492 | { |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3493 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3494 | struct page *page = NULL; |
Mel Gorman | c603844 | 2016-05-19 17:13:38 -0700 | [diff] [blame] | 3495 | unsigned int alloc_flags; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3496 | unsigned long did_some_progress; |
Vlastimil Babka | a5508cd | 2016-07-28 15:49:28 -0700 | [diff] [blame] | 3497 | enum compact_priority compact_priority = DEF_COMPACT_PRIORITY; |
Michal Hocko | c5d01d0 | 2016-05-20 16:56:53 -0700 | [diff] [blame] | 3498 | enum compact_result compact_result; |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3499 | int compaction_retries = 0; |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3500 | int no_progress_loops = 0; |
Michal Hocko | 63f53de | 2016-10-07 17:01:58 -0700 | [diff] [blame] | 3501 | unsigned long alloc_start = jiffies; |
| 3502 | unsigned int stall_timeout = 10 * HZ; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 3503 | |
Christoph Lameter | 952f3b5 | 2006-12-06 20:33:26 -0800 | [diff] [blame] | 3504 | /* |
Mel Gorman | 72807a7 | 2009-06-16 15:32:18 -0700 | [diff] [blame] | 3505 | * In the slowpath, we sanity check order to avoid ever trying to |
| 3506 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may |
| 3507 | * be using allocators in order of preference for an area that is |
| 3508 | * too large. |
| 3509 | */ |
Mel Gorman | 1fc28b7 | 2009-07-29 15:04:08 -0700 | [diff] [blame] | 3510 | if (order >= MAX_ORDER) { |
| 3511 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); |
Mel Gorman | 72807a7 | 2009-06-16 15:32:18 -0700 | [diff] [blame] | 3512 | return NULL; |
Mel Gorman | 1fc28b7 | 2009-07-29 15:04:08 -0700 | [diff] [blame] | 3513 | } |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 3514 | |
Christoph Lameter | 952f3b5 | 2006-12-06 20:33:26 -0800 | [diff] [blame] | 3515 | /* |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3516 | * We also sanity check to catch abuse of atomic reserves being used by |
| 3517 | * callers that are not in atomic context. |
| 3518 | */ |
| 3519 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == |
| 3520 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) |
| 3521 | gfp_mask &= ~__GFP_ATOMIC; |
| 3522 | |
Vlastimil Babka | 31a6c19 | 2016-07-28 15:49:13 -0700 | [diff] [blame] | 3523 | /* |
| 3524 | * The fast path uses conservative alloc_flags to succeed only until |
| 3525 | * kswapd needs to be woken up, and to avoid the cost of setting up |
| 3526 | * alloc_flags precisely. So we do that now. |
| 3527 | */ |
| 3528 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
| 3529 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3530 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3531 | wake_all_kswapds(order, ac); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 3532 | |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 3533 | /* |
Vlastimil Babka | 2377123 | 2016-07-28 15:49:16 -0700 | [diff] [blame] | 3534 | * The adjusted alloc_flags might result in immediate success, so try |
| 3535 | * that first |
| 3536 | */ |
| 3537 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
| 3538 | if (page) |
| 3539 | goto got_pg; |
| 3540 | |
Vlastimil Babka | a8161d1 | 2016-07-28 15:49:19 -0700 | [diff] [blame] | 3541 | /* |
| 3542 | * For costly allocations, try direct compaction first, as it's likely |
| 3543 | * that we have enough base pages and don't need to reclaim. Don't try |
| 3544 | * that for allocations that are allowed to ignore watermarks, as the |
| 3545 | * ALLOC_NO_WATERMARKS attempt didn't yet happen. |
| 3546 | */ |
| 3547 | if (can_direct_reclaim && order > PAGE_ALLOC_COSTLY_ORDER && |
| 3548 | !gfp_pfmemalloc_allowed(gfp_mask)) { |
| 3549 | page = __alloc_pages_direct_compact(gfp_mask, order, |
| 3550 | alloc_flags, ac, |
Vlastimil Babka | a5508cd | 2016-07-28 15:49:28 -0700 | [diff] [blame] | 3551 | INIT_COMPACT_PRIORITY, |
Vlastimil Babka | a8161d1 | 2016-07-28 15:49:19 -0700 | [diff] [blame] | 3552 | &compact_result); |
| 3553 | if (page) |
| 3554 | goto got_pg; |
| 3555 | |
Vlastimil Babka | 3eb2771 | 2016-07-28 15:49:22 -0700 | [diff] [blame] | 3556 | /* |
| 3557 | * Checks for costly allocations with __GFP_NORETRY, which |
| 3558 | * includes THP page fault allocations |
| 3559 | */ |
| 3560 | if (gfp_mask & __GFP_NORETRY) { |
Vlastimil Babka | a8161d1 | 2016-07-28 15:49:19 -0700 | [diff] [blame] | 3561 | /* |
| 3562 | * If compaction is deferred for high-order allocations, |
| 3563 | * it is because sync compaction recently failed. If |
| 3564 | * this is the case and the caller requested a THP |
| 3565 | * allocation, we do not want to heavily disrupt the |
| 3566 | * system, so we fail the allocation instead of entering |
| 3567 | * direct reclaim. |
| 3568 | */ |
| 3569 | if (compact_result == COMPACT_DEFERRED) |
| 3570 | goto nopage; |
| 3571 | |
| 3572 | /* |
Vlastimil Babka | 3eb2771 | 2016-07-28 15:49:22 -0700 | [diff] [blame] | 3573 | * Looks like reclaim/compaction is worth trying, but |
| 3574 | * sync compaction could be very expensive, so keep |
Vlastimil Babka | 2516035 | 2016-07-28 15:49:25 -0700 | [diff] [blame] | 3575 | * using async compaction. |
Vlastimil Babka | a8161d1 | 2016-07-28 15:49:19 -0700 | [diff] [blame] | 3576 | */ |
Vlastimil Babka | a5508cd | 2016-07-28 15:49:28 -0700 | [diff] [blame] | 3577 | compact_priority = INIT_COMPACT_PRIORITY; |
Vlastimil Babka | a8161d1 | 2016-07-28 15:49:19 -0700 | [diff] [blame] | 3578 | } |
| 3579 | } |
Vlastimil Babka | 2377123 | 2016-07-28 15:49:16 -0700 | [diff] [blame] | 3580 | |
| 3581 | retry: |
| 3582 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
| 3583 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
| 3584 | wake_all_kswapds(order, ac); |
| 3585 | |
| 3586 | if (gfp_pfmemalloc_allowed(gfp_mask)) |
| 3587 | alloc_flags = ALLOC_NO_WATERMARKS; |
| 3588 | |
| 3589 | /* |
Mel Gorman | e46e7b7 | 2016-06-03 14:56:01 -0700 | [diff] [blame] | 3590 | * Reset the zonelist iterators if memory policies can be ignored. |
| 3591 | * These allocations are high priority and system rather than user |
| 3592 | * orientated. |
| 3593 | */ |
Vlastimil Babka | 2377123 | 2016-07-28 15:49:16 -0700 | [diff] [blame] | 3594 | if (!(alloc_flags & ALLOC_CPUSET) || (alloc_flags & ALLOC_NO_WATERMARKS)) { |
Mel Gorman | e46e7b7 | 2016-06-03 14:56:01 -0700 | [diff] [blame] | 3595 | ac->zonelist = node_zonelist(numa_node_id(), gfp_mask); |
| 3596 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
| 3597 | ac->high_zoneidx, ac->nodemask); |
| 3598 | } |
| 3599 | |
Vlastimil Babka | 2377123 | 2016-07-28 15:49:16 -0700 | [diff] [blame] | 3600 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
Vlastimil Babka | 31a6c19 | 2016-07-28 15:49:13 -0700 | [diff] [blame] | 3601 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 3602 | if (page) |
| 3603 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3604 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3605 | /* Caller is not willing to reclaim, we can't balance anything */ |
| 3606 | if (!can_direct_reclaim) { |
David Rientjes | aed0a0e | 2014-01-21 15:51:12 -0800 | [diff] [blame] | 3607 | /* |
Michal Hocko | 33d5310 | 2016-01-14 15:19:05 -0800 | [diff] [blame] | 3608 | * All existing users of the __GFP_NOFAIL are blockable, so warn |
| 3609 | * of any new users that actually allow this type of allocation |
| 3610 | * to fail. |
David Rientjes | aed0a0e | 2014-01-21 15:51:12 -0800 | [diff] [blame] | 3611 | */ |
| 3612 | WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3613 | goto nopage; |
David Rientjes | aed0a0e | 2014-01-21 15:51:12 -0800 | [diff] [blame] | 3614 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3615 | |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3616 | /* Avoid recursion of direct reclaim */ |
Michal Hocko | 33d5310 | 2016-01-14 15:19:05 -0800 | [diff] [blame] | 3617 | if (current->flags & PF_MEMALLOC) { |
| 3618 | /* |
| 3619 | * __GFP_NOFAIL request from this context is rather bizarre |
| 3620 | * because we cannot reclaim anything and only can loop waiting |
| 3621 | * for somebody to do a work for us. |
| 3622 | */ |
| 3623 | if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
| 3624 | cond_resched(); |
| 3625 | goto retry; |
| 3626 | } |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 3627 | goto nopage; |
Michal Hocko | 33d5310 | 2016-01-14 15:19:05 -0800 | [diff] [blame] | 3628 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3629 | |
David Rientjes | 6583bb6 | 2009-07-29 15:02:06 -0700 | [diff] [blame] | 3630 | /* Avoid allocations with no watermarks from looping endlessly */ |
| 3631 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) |
| 3632 | goto nopage; |
| 3633 | |
David Rientjes | 8fe7804 | 2014-08-06 16:07:54 -0700 | [diff] [blame] | 3634 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3635 | /* Try direct reclaim and then allocating */ |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3636 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, |
| 3637 | &did_some_progress); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3638 | if (page) |
| 3639 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3640 | |
Vlastimil Babka | a8161d1 | 2016-07-28 15:49:19 -0700 | [diff] [blame] | 3641 | /* Try direct compaction and then allocating */ |
| 3642 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
Vlastimil Babka | a5508cd | 2016-07-28 15:49:28 -0700 | [diff] [blame] | 3643 | compact_priority, &compact_result); |
Vlastimil Babka | a8161d1 | 2016-07-28 15:49:19 -0700 | [diff] [blame] | 3644 | if (page) |
| 3645 | goto got_pg; |
| 3646 | |
Johannes Weiner | 9083905 | 2015-06-24 16:57:21 -0700 | [diff] [blame] | 3647 | /* Do not loop if specifically requested */ |
| 3648 | if (gfp_mask & __GFP_NORETRY) |
Vlastimil Babka | a8161d1 | 2016-07-28 15:49:19 -0700 | [diff] [blame] | 3649 | goto nopage; |
Johannes Weiner | 9083905 | 2015-06-24 16:57:21 -0700 | [diff] [blame] | 3650 | |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3651 | /* |
| 3652 | * Do not retry costly high order allocations unless they are |
| 3653 | * __GFP_REPEAT |
| 3654 | */ |
| 3655 | if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_REPEAT)) |
Vlastimil Babka | a8161d1 | 2016-07-28 15:49:19 -0700 | [diff] [blame] | 3656 | goto nopage; |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3657 | |
Michal Hocko | 63f53de | 2016-10-07 17:01:58 -0700 | [diff] [blame] | 3658 | /* Make sure we know about allocations which stall for too long */ |
| 3659 | if (time_after(jiffies, alloc_start + stall_timeout)) { |
| 3660 | warn_alloc(gfp_mask, |
| 3661 | "page alloction stalls for %ums, order:%u\n", |
| 3662 | jiffies_to_msecs(jiffies-alloc_start), order); |
| 3663 | stall_timeout += 10 * HZ; |
| 3664 | } |
| 3665 | |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3666 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
Vlastimil Babka | 423b452 | 2016-10-07 17:00:40 -0700 | [diff] [blame] | 3667 | did_some_progress > 0, &no_progress_loops)) |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3668 | goto retry; |
| 3669 | |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3670 | /* |
| 3671 | * It doesn't make any sense to retry for the compaction if the order-0 |
| 3672 | * reclaim is not able to make any progress because the current |
| 3673 | * implementation of the compaction depends on the sufficient amount |
| 3674 | * of free memory (see __compaction_suitable) |
| 3675 | */ |
| 3676 | if (did_some_progress > 0 && |
Michal Hocko | 86a294a | 2016-05-20 16:57:12 -0700 | [diff] [blame] | 3677 | should_compact_retry(ac, order, alloc_flags, |
Vlastimil Babka | a5508cd | 2016-07-28 15:49:28 -0700 | [diff] [blame] | 3678 | compact_result, &compact_priority, |
Vlastimil Babka | d943649 | 2016-10-07 17:00:31 -0700 | [diff] [blame] | 3679 | &compaction_retries)) |
Michal Hocko | 33c2d21 | 2016-05-20 16:57:06 -0700 | [diff] [blame] | 3680 | goto retry; |
| 3681 | |
Johannes Weiner | 9083905 | 2015-06-24 16:57:21 -0700 | [diff] [blame] | 3682 | /* Reclaim has failed us, start killing things */ |
| 3683 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); |
| 3684 | if (page) |
| 3685 | goto got_pg; |
| 3686 | |
| 3687 | /* Retry as long as the OOM killer is making progress */ |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3688 | if (did_some_progress) { |
| 3689 | no_progress_loops = 0; |
Johannes Weiner | 9083905 | 2015-06-24 16:57:21 -0700 | [diff] [blame] | 3690 | goto retry; |
Michal Hocko | 0a0337e | 2016-05-20 16:57:00 -0700 | [diff] [blame] | 3691 | } |
Johannes Weiner | 9083905 | 2015-06-24 16:57:21 -0700 | [diff] [blame] | 3692 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3693 | nopage: |
Michal Hocko | 7877cdc | 2016-10-07 17:01:55 -0700 | [diff] [blame] | 3694 | warn_alloc(gfp_mask, |
| 3695 | "page allocation failure: order:%u", order); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3696 | got_pg: |
Mel Gorman | 072bb0a | 2012-07-31 16:43:58 -0700 | [diff] [blame] | 3697 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3698 | } |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3699 | |
| 3700 | /* |
| 3701 | * This is the 'heart' of the zoned buddy allocator. |
| 3702 | */ |
| 3703 | struct page * |
| 3704 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, |
| 3705 | struct zonelist *zonelist, nodemask_t *nodemask) |
| 3706 | { |
Mel Gorman | 5bb1b16 | 2016-05-19 17:13:50 -0700 | [diff] [blame] | 3707 | struct page *page; |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 3708 | unsigned int cpuset_mems_cookie; |
Mel Gorman | e6cbd7f | 2016-07-28 15:46:50 -0700 | [diff] [blame] | 3709 | unsigned int alloc_flags = ALLOC_WMARK_LOW; |
Mel Gorman | 83d4ca8 | 2016-05-19 17:13:56 -0700 | [diff] [blame] | 3710 | 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] | 3711 | struct alloc_context ac = { |
| 3712 | .high_zoneidx = gfp_zone(gfp_mask), |
Mel Gorman | 682a338 | 2016-05-19 17:13:30 -0700 | [diff] [blame] | 3713 | .zonelist = zonelist, |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3714 | .nodemask = nodemask, |
| 3715 | .migratetype = gfpflags_to_migratetype(gfp_mask), |
| 3716 | }; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3717 | |
Mel Gorman | 682a338 | 2016-05-19 17:13:30 -0700 | [diff] [blame] | 3718 | if (cpusets_enabled()) { |
Mel Gorman | 83d4ca8 | 2016-05-19 17:13:56 -0700 | [diff] [blame] | 3719 | alloc_mask |= __GFP_HARDWALL; |
Mel Gorman | 682a338 | 2016-05-19 17:13:30 -0700 | [diff] [blame] | 3720 | alloc_flags |= ALLOC_CPUSET; |
| 3721 | if (!ac.nodemask) |
| 3722 | ac.nodemask = &cpuset_current_mems_allowed; |
| 3723 | } |
| 3724 | |
Benjamin Herrenschmidt | dcce284 | 2009-06-18 13:24:12 +1000 | [diff] [blame] | 3725 | gfp_mask &= gfp_allowed_mask; |
| 3726 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3727 | lockdep_trace_alloc(gfp_mask); |
| 3728 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 3729 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3730 | |
| 3731 | if (should_fail_alloc_page(gfp_mask, order)) |
| 3732 | return NULL; |
| 3733 | |
| 3734 | /* |
| 3735 | * Check the zones suitable for the gfp_mask contain at least one |
| 3736 | * 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] | 3737 | * of __GFP_THISNODE and a memoryless node |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3738 | */ |
| 3739 | if (unlikely(!zonelist->_zonerefs->zone)) |
| 3740 | return NULL; |
| 3741 | |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3742 | if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE) |
Vlastimil Babka | 21bb9bd | 2014-10-09 15:26:51 -0700 | [diff] [blame] | 3743 | alloc_flags |= ALLOC_CMA; |
| 3744 | |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 3745 | retry_cpuset: |
Mel Gorman | d26914d | 2014-04-03 14:47:24 -0700 | [diff] [blame] | 3746 | cpuset_mems_cookie = read_mems_allowed_begin(); |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 3747 | |
Mel Gorman | c9ab0c4 | 2015-11-06 16:28:12 -0800 | [diff] [blame] | 3748 | /* Dirty zone balancing only done in the fast path */ |
| 3749 | ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
| 3750 | |
Mel Gorman | e46e7b7 | 2016-06-03 14:56:01 -0700 | [diff] [blame] | 3751 | /* |
| 3752 | * The preferred zone is used for statistics but crucially it is |
| 3753 | * also used as the starting point for the zonelist iterator. It |
| 3754 | * may get reset for allocations that ignore memory policies. |
| 3755 | */ |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 3756 | ac.preferred_zoneref = first_zones_zonelist(ac.zonelist, |
| 3757 | ac.high_zoneidx, ac.nodemask); |
| 3758 | if (!ac.preferred_zoneref) { |
Mel Gorman | 5bb1b16 | 2016-05-19 17:13:50 -0700 | [diff] [blame] | 3759 | page = NULL; |
Mel Gorman | 4fcb097 | 2016-05-19 17:14:01 -0700 | [diff] [blame] | 3760 | goto no_zone; |
Mel Gorman | 5bb1b16 | 2016-05-19 17:13:50 -0700 | [diff] [blame] | 3761 | } |
| 3762 | |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 3763 | /* First allocation attempt */ |
Vlastimil Babka | a926375 | 2015-02-11 15:25:41 -0800 | [diff] [blame] | 3764 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
Mel Gorman | 4fcb097 | 2016-05-19 17:14:01 -0700 | [diff] [blame] | 3765 | if (likely(page)) |
| 3766 | goto out; |
Andrew Morton | 91fbdc0 | 2015-02-11 15:25:04 -0800 | [diff] [blame] | 3767 | |
Mel Gorman | 4fcb097 | 2016-05-19 17:14:01 -0700 | [diff] [blame] | 3768 | /* |
| 3769 | * Runtime PM, block IO and its error handling path can deadlock |
| 3770 | * because I/O on the device might not complete. |
| 3771 | */ |
| 3772 | alloc_mask = memalloc_noio_flags(gfp_mask); |
| 3773 | ac.spread_dirty_pages = false; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3774 | |
Mel Gorman | 4741526 | 2016-05-19 17:14:44 -0700 | [diff] [blame] | 3775 | /* |
| 3776 | * Restore the original nodemask if it was potentially replaced with |
| 3777 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. |
| 3778 | */ |
| 3779 | if (cpusets_enabled()) |
| 3780 | ac.nodemask = nodemask; |
Mel Gorman | 4fcb097 | 2016-05-19 17:14:01 -0700 | [diff] [blame] | 3781 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
Xishi Qiu | 23f086f | 2015-02-11 15:25:07 -0800 | [diff] [blame] | 3782 | |
Mel Gorman | 4fcb097 | 2016-05-19 17:14:01 -0700 | [diff] [blame] | 3783 | no_zone: |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 3784 | /* |
| 3785 | * When updating a task's mems_allowed, it is possible to race with |
| 3786 | * parallel threads in such a way that an allocation can fail while |
| 3787 | * the mask is being updated. If a page allocation is about to fail, |
| 3788 | * check if the cpuset changed during allocation and if so, retry. |
| 3789 | */ |
Mel Gorman | 83d4ca8 | 2016-05-19 17:13:56 -0700 | [diff] [blame] | 3790 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) { |
| 3791 | alloc_mask = gfp_mask; |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 3792 | goto retry_cpuset; |
Mel Gorman | 83d4ca8 | 2016-05-19 17:13:56 -0700 | [diff] [blame] | 3793 | } |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 3794 | |
Mel Gorman | 4fcb097 | 2016-05-19 17:14:01 -0700 | [diff] [blame] | 3795 | out: |
Vladimir Davydov | c4159a7 | 2016-08-08 23:03:12 +0300 | [diff] [blame] | 3796 | if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page && |
| 3797 | unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) { |
| 3798 | __free_pages(page, order); |
| 3799 | page = NULL; |
Vladimir Davydov | 4949148 | 2016-07-26 15:24:24 -0700 | [diff] [blame] | 3800 | } |
| 3801 | |
Mel Gorman | 4fcb097 | 2016-05-19 17:14:01 -0700 | [diff] [blame] | 3802 | if (kmemcheck_enabled && page) |
| 3803 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); |
| 3804 | |
| 3805 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); |
| 3806 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 3807 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3808 | } |
Mel Gorman | d239171 | 2009-06-16 15:31:52 -0700 | [diff] [blame] | 3809 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3810 | |
| 3811 | /* |
| 3812 | * Common helper functions. |
| 3813 | */ |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 3814 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3815 | { |
Akinobu Mita | 945a111 | 2009-09-21 17:01:47 -0700 | [diff] [blame] | 3816 | struct page *page; |
| 3817 | |
| 3818 | /* |
| 3819 | * __get_free_pages() returns a 32-bit address, which cannot represent |
| 3820 | * a highmem page |
| 3821 | */ |
| 3822 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
| 3823 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3824 | page = alloc_pages(gfp_mask, order); |
| 3825 | if (!page) |
| 3826 | return 0; |
| 3827 | return (unsigned long) page_address(page); |
| 3828 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3829 | EXPORT_SYMBOL(__get_free_pages); |
| 3830 | |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 3831 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3832 | { |
Akinobu Mita | 945a111 | 2009-09-21 17:01:47 -0700 | [diff] [blame] | 3833 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3834 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3835 | EXPORT_SYMBOL(get_zeroed_page); |
| 3836 | |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 3837 | void __free_pages(struct page *page, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3838 | { |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 3839 | if (put_page_testzero(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3840 | if (order == 0) |
Mel Gorman | b745bc8 | 2014-06-04 16:10:22 -0700 | [diff] [blame] | 3841 | free_hot_cold_page(page, false); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3842 | else |
| 3843 | __free_pages_ok(page, order); |
| 3844 | } |
| 3845 | } |
| 3846 | |
| 3847 | EXPORT_SYMBOL(__free_pages); |
| 3848 | |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 3849 | void free_pages(unsigned long addr, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3850 | { |
| 3851 | if (addr != 0) { |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 3852 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3853 | __free_pages(virt_to_page((void *)addr), order); |
| 3854 | } |
| 3855 | } |
| 3856 | |
| 3857 | EXPORT_SYMBOL(free_pages); |
| 3858 | |
Glauber Costa | 6a1a0d3 | 2012-12-18 14:22:00 -0800 | [diff] [blame] | 3859 | /* |
Alexander Duyck | b63ae8c | 2015-05-06 21:11:57 -0700 | [diff] [blame] | 3860 | * Page Fragment: |
| 3861 | * An arbitrary-length arbitrary-offset area of memory which resides |
| 3862 | * within a 0 or higher order page. Multiple fragments within that page |
| 3863 | * are individually refcounted, in the page's reference counter. |
| 3864 | * |
| 3865 | * The page_frag functions below provide a simple allocation framework for |
| 3866 | * page fragments. This is used by the network stack and network device |
| 3867 | * drivers to provide a backing region of memory for use as either an |
| 3868 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. |
| 3869 | */ |
| 3870 | static struct page *__page_frag_refill(struct page_frag_cache *nc, |
| 3871 | gfp_t gfp_mask) |
| 3872 | { |
| 3873 | struct page *page = NULL; |
| 3874 | gfp_t gfp = gfp_mask; |
| 3875 | |
| 3876 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
| 3877 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | |
| 3878 | __GFP_NOMEMALLOC; |
| 3879 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, |
| 3880 | PAGE_FRAG_CACHE_MAX_ORDER); |
| 3881 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; |
| 3882 | #endif |
| 3883 | if (unlikely(!page)) |
| 3884 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); |
| 3885 | |
| 3886 | nc->va = page ? page_address(page) : NULL; |
| 3887 | |
| 3888 | return page; |
| 3889 | } |
| 3890 | |
| 3891 | void *__alloc_page_frag(struct page_frag_cache *nc, |
| 3892 | unsigned int fragsz, gfp_t gfp_mask) |
| 3893 | { |
| 3894 | unsigned int size = PAGE_SIZE; |
| 3895 | struct page *page; |
| 3896 | int offset; |
| 3897 | |
| 3898 | if (unlikely(!nc->va)) { |
| 3899 | refill: |
| 3900 | page = __page_frag_refill(nc, gfp_mask); |
| 3901 | if (!page) |
| 3902 | return NULL; |
| 3903 | |
| 3904 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
| 3905 | /* if size can vary use size else just use PAGE_SIZE */ |
| 3906 | size = nc->size; |
| 3907 | #endif |
| 3908 | /* Even if we own the page, we do not use atomic_set(). |
| 3909 | * This would break get_page_unless_zero() users. |
| 3910 | */ |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 3911 | page_ref_add(page, size - 1); |
Alexander Duyck | b63ae8c | 2015-05-06 21:11:57 -0700 | [diff] [blame] | 3912 | |
| 3913 | /* reset page count bias and offset to start of new frag */ |
Michal Hocko | 2f064f3 | 2015-08-21 14:11:51 -0700 | [diff] [blame] | 3914 | nc->pfmemalloc = page_is_pfmemalloc(page); |
Alexander Duyck | b63ae8c | 2015-05-06 21:11:57 -0700 | [diff] [blame] | 3915 | nc->pagecnt_bias = size; |
| 3916 | nc->offset = size; |
| 3917 | } |
| 3918 | |
| 3919 | offset = nc->offset - fragsz; |
| 3920 | if (unlikely(offset < 0)) { |
| 3921 | page = virt_to_page(nc->va); |
| 3922 | |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 3923 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
Alexander Duyck | b63ae8c | 2015-05-06 21:11:57 -0700 | [diff] [blame] | 3924 | goto refill; |
| 3925 | |
| 3926 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
| 3927 | /* if size can vary use size else just use PAGE_SIZE */ |
| 3928 | size = nc->size; |
| 3929 | #endif |
| 3930 | /* OK, page count is 0, we can safely set it */ |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 3931 | set_page_count(page, size); |
Alexander Duyck | b63ae8c | 2015-05-06 21:11:57 -0700 | [diff] [blame] | 3932 | |
| 3933 | /* reset page count bias and offset to start of new frag */ |
| 3934 | nc->pagecnt_bias = size; |
| 3935 | offset = size - fragsz; |
| 3936 | } |
| 3937 | |
| 3938 | nc->pagecnt_bias--; |
| 3939 | nc->offset = offset; |
| 3940 | |
| 3941 | return nc->va + offset; |
| 3942 | } |
| 3943 | EXPORT_SYMBOL(__alloc_page_frag); |
| 3944 | |
| 3945 | /* |
| 3946 | * Frees a page fragment allocated out of either a compound or order 0 page. |
| 3947 | */ |
| 3948 | void __free_page_frag(void *addr) |
| 3949 | { |
| 3950 | struct page *page = virt_to_head_page(addr); |
| 3951 | |
| 3952 | if (unlikely(put_page_testzero(page))) |
| 3953 | __free_pages_ok(page, compound_order(page)); |
| 3954 | } |
| 3955 | EXPORT_SYMBOL(__free_page_frag); |
| 3956 | |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 3957 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
| 3958 | size_t size) |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 3959 | { |
| 3960 | if (addr) { |
| 3961 | unsigned long alloc_end = addr + (PAGE_SIZE << order); |
| 3962 | unsigned long used = addr + PAGE_ALIGN(size); |
| 3963 | |
| 3964 | split_page(virt_to_page((void *)addr), order); |
| 3965 | while (used < alloc_end) { |
| 3966 | free_page(used); |
| 3967 | used += PAGE_SIZE; |
| 3968 | } |
| 3969 | } |
| 3970 | return (void *)addr; |
| 3971 | } |
| 3972 | |
Timur Tabi | 2be0ffe | 2008-07-23 21:28:11 -0700 | [diff] [blame] | 3973 | /** |
| 3974 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. |
| 3975 | * @size: the number of bytes to allocate |
| 3976 | * @gfp_mask: GFP flags for the allocation |
| 3977 | * |
| 3978 | * This function is similar to alloc_pages(), except that it allocates the |
| 3979 | * minimum number of pages to satisfy the request. alloc_pages() can only |
| 3980 | * allocate memory in power-of-two pages. |
| 3981 | * |
| 3982 | * This function is also limited by MAX_ORDER. |
| 3983 | * |
| 3984 | * Memory allocated by this function must be released by free_pages_exact(). |
| 3985 | */ |
| 3986 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) |
| 3987 | { |
| 3988 | unsigned int order = get_order(size); |
| 3989 | unsigned long addr; |
| 3990 | |
| 3991 | addr = __get_free_pages(gfp_mask, order); |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 3992 | return make_alloc_exact(addr, order, size); |
Timur Tabi | 2be0ffe | 2008-07-23 21:28:11 -0700 | [diff] [blame] | 3993 | } |
| 3994 | EXPORT_SYMBOL(alloc_pages_exact); |
| 3995 | |
| 3996 | /** |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 3997 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous |
| 3998 | * pages on a node. |
Randy Dunlap | b5e6ab5 | 2011-05-16 13:16:54 -0700 | [diff] [blame] | 3999 | * @nid: the preferred node ID where memory should be allocated |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 4000 | * @size: the number of bytes to allocate |
| 4001 | * @gfp_mask: GFP flags for the allocation |
| 4002 | * |
| 4003 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling |
| 4004 | * back. |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 4005 | */ |
Fabian Frederick | e193181 | 2014-08-06 16:04:59 -0700 | [diff] [blame] | 4006 | 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] | 4007 | { |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 4008 | unsigned int order = get_order(size); |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 4009 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
| 4010 | if (!p) |
| 4011 | return NULL; |
| 4012 | return make_alloc_exact((unsigned long)page_address(p), order, size); |
| 4013 | } |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 4014 | |
| 4015 | /** |
Timur Tabi | 2be0ffe | 2008-07-23 21:28:11 -0700 | [diff] [blame] | 4016 | * free_pages_exact - release memory allocated via alloc_pages_exact() |
| 4017 | * @virt: the value returned by alloc_pages_exact. |
| 4018 | * @size: size of allocation, same value as passed to alloc_pages_exact(). |
| 4019 | * |
| 4020 | * Release the memory allocated by a previous call to alloc_pages_exact. |
| 4021 | */ |
| 4022 | void free_pages_exact(void *virt, size_t size) |
| 4023 | { |
| 4024 | unsigned long addr = (unsigned long)virt; |
| 4025 | unsigned long end = addr + PAGE_ALIGN(size); |
| 4026 | |
| 4027 | while (addr < end) { |
| 4028 | free_page(addr); |
| 4029 | addr += PAGE_SIZE; |
| 4030 | } |
| 4031 | } |
| 4032 | EXPORT_SYMBOL(free_pages_exact); |
| 4033 | |
Zhang Yanfei | e0fb581 | 2013-02-22 16:35:54 -0800 | [diff] [blame] | 4034 | /** |
| 4035 | * nr_free_zone_pages - count number of pages beyond high watermark |
| 4036 | * @offset: The zone index of the highest zone |
| 4037 | * |
| 4038 | * nr_free_zone_pages() counts the number of counts pages which are beyond the |
| 4039 | * high watermark within all zones at or below a given zone index. For each |
| 4040 | * zone, the number of pages is calculated as: |
Jiang Liu | 834405c | 2013-07-03 15:03:04 -0700 | [diff] [blame] | 4041 | * managed_pages - high_pages |
Zhang Yanfei | e0fb581 | 2013-02-22 16:35:54 -0800 | [diff] [blame] | 4042 | */ |
Zhang Yanfei | ebec386 | 2013-02-22 16:35:43 -0800 | [diff] [blame] | 4043 | static unsigned long nr_free_zone_pages(int offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4044 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4045 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4046 | struct zone *zone; |
| 4047 | |
Martin J. Bligh | e310fd4 | 2005-07-29 22:59:18 -0700 | [diff] [blame] | 4048 | /* Just pick one node, since fallback list is circular */ |
Zhang Yanfei | ebec386 | 2013-02-22 16:35:43 -0800 | [diff] [blame] | 4049 | unsigned long sum = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4050 | |
Mel Gorman | 0e88460 | 2008-04-28 02:12:14 -0700 | [diff] [blame] | 4051 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4052 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4053 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 4054 | unsigned long size = zone->managed_pages; |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 4055 | unsigned long high = high_wmark_pages(zone); |
Martin J. Bligh | e310fd4 | 2005-07-29 22:59:18 -0700 | [diff] [blame] | 4056 | if (size > high) |
| 4057 | sum += size - high; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4058 | } |
| 4059 | |
| 4060 | return sum; |
| 4061 | } |
| 4062 | |
Zhang Yanfei | e0fb581 | 2013-02-22 16:35:54 -0800 | [diff] [blame] | 4063 | /** |
| 4064 | * nr_free_buffer_pages - count number of pages beyond high watermark |
| 4065 | * |
| 4066 | * nr_free_buffer_pages() counts the number of pages which are beyond the high |
| 4067 | * watermark within ZONE_DMA and ZONE_NORMAL. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4068 | */ |
Zhang Yanfei | ebec386 | 2013-02-22 16:35:43 -0800 | [diff] [blame] | 4069 | unsigned long nr_free_buffer_pages(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4070 | { |
Al Viro | af4ca45 | 2005-10-21 02:55:38 -0400 | [diff] [blame] | 4071 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4072 | } |
Meelap Shah | c2f1a55 | 2007-07-17 04:04:39 -0700 | [diff] [blame] | 4073 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4074 | |
Zhang Yanfei | e0fb581 | 2013-02-22 16:35:54 -0800 | [diff] [blame] | 4075 | /** |
| 4076 | * nr_free_pagecache_pages - count number of pages beyond high watermark |
| 4077 | * |
| 4078 | * nr_free_pagecache_pages() counts the number of pages which are beyond the |
| 4079 | * high watermark within all zones. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4080 | */ |
Zhang Yanfei | ebec386 | 2013-02-22 16:35:43 -0800 | [diff] [blame] | 4081 | unsigned long nr_free_pagecache_pages(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4082 | { |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4083 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4084 | } |
Christoph Lameter | 08e0f6a | 2006-09-27 01:50:06 -0700 | [diff] [blame] | 4085 | |
| 4086 | static inline void show_node(struct zone *zone) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4087 | { |
Kirill A. Shutemov | e5adfff | 2012-12-11 16:00:29 -0800 | [diff] [blame] | 4088 | if (IS_ENABLED(CONFIG_NUMA)) |
Andy Whitcroft | 25ba77c | 2006-12-06 20:33:03 -0800 | [diff] [blame] | 4089 | printk("Node %d ", zone_to_nid(zone)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4090 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4091 | |
Igor Redko | d02bd27 | 2016-03-17 14:19:05 -0700 | [diff] [blame] | 4092 | long si_mem_available(void) |
| 4093 | { |
| 4094 | long available; |
| 4095 | unsigned long pagecache; |
| 4096 | unsigned long wmark_low = 0; |
| 4097 | unsigned long pages[NR_LRU_LISTS]; |
| 4098 | struct zone *zone; |
| 4099 | int lru; |
| 4100 | |
| 4101 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) |
Mel Gorman | 2f95ff9 | 2016-08-11 15:32:57 -0700 | [diff] [blame] | 4102 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
Igor Redko | d02bd27 | 2016-03-17 14:19:05 -0700 | [diff] [blame] | 4103 | |
| 4104 | for_each_zone(zone) |
| 4105 | wmark_low += zone->watermark[WMARK_LOW]; |
| 4106 | |
| 4107 | /* |
| 4108 | * Estimate the amount of memory available for userspace allocations, |
| 4109 | * without causing swapping. |
| 4110 | */ |
| 4111 | available = global_page_state(NR_FREE_PAGES) - totalreserve_pages; |
| 4112 | |
| 4113 | /* |
| 4114 | * Not all the page cache can be freed, otherwise the system will |
| 4115 | * start swapping. Assume at least half of the page cache, or the |
| 4116 | * low watermark worth of cache, needs to stay. |
| 4117 | */ |
| 4118 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; |
| 4119 | pagecache -= min(pagecache / 2, wmark_low); |
| 4120 | available += pagecache; |
| 4121 | |
| 4122 | /* |
| 4123 | * Part of the reclaimable slab consists of items that are in use, |
| 4124 | * and cannot be freed. Cap this estimate at the low watermark. |
| 4125 | */ |
| 4126 | available += global_page_state(NR_SLAB_RECLAIMABLE) - |
| 4127 | min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low); |
| 4128 | |
| 4129 | if (available < 0) |
| 4130 | available = 0; |
| 4131 | return available; |
| 4132 | } |
| 4133 | EXPORT_SYMBOL_GPL(si_mem_available); |
| 4134 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4135 | void si_meminfo(struct sysinfo *val) |
| 4136 | { |
| 4137 | val->totalram = totalram_pages; |
Mel Gorman | 11fb998 | 2016-07-28 15:46:20 -0700 | [diff] [blame] | 4138 | val->sharedram = global_node_page_state(NR_SHMEM); |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 4139 | val->freeram = global_page_state(NR_FREE_PAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4140 | val->bufferram = nr_blockdev_pages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4141 | val->totalhigh = totalhigh_pages; |
| 4142 | val->freehigh = nr_free_highpages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4143 | val->mem_unit = PAGE_SIZE; |
| 4144 | } |
| 4145 | |
| 4146 | EXPORT_SYMBOL(si_meminfo); |
| 4147 | |
| 4148 | #ifdef CONFIG_NUMA |
| 4149 | void si_meminfo_node(struct sysinfo *val, int nid) |
| 4150 | { |
Jiang Liu | cdd91a7 | 2013-07-03 15:03:27 -0700 | [diff] [blame] | 4151 | int zone_type; /* needs to be signed */ |
| 4152 | unsigned long managed_pages = 0; |
Joonsoo Kim | fc2bd79 | 2016-05-19 17:12:23 -0700 | [diff] [blame] | 4153 | unsigned long managed_highpages = 0; |
| 4154 | unsigned long free_highpages = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4155 | pg_data_t *pgdat = NODE_DATA(nid); |
| 4156 | |
Jiang Liu | cdd91a7 | 2013-07-03 15:03:27 -0700 | [diff] [blame] | 4157 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
| 4158 | managed_pages += pgdat->node_zones[zone_type].managed_pages; |
| 4159 | val->totalram = managed_pages; |
Mel Gorman | 11fb998 | 2016-07-28 15:46:20 -0700 | [diff] [blame] | 4160 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
Mel Gorman | 75ef718 | 2016-07-28 15:45:24 -0700 | [diff] [blame] | 4161 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 4162 | #ifdef CONFIG_HIGHMEM |
Joonsoo Kim | fc2bd79 | 2016-05-19 17:12:23 -0700 | [diff] [blame] | 4163 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
| 4164 | struct zone *zone = &pgdat->node_zones[zone_type]; |
| 4165 | |
| 4166 | if (is_highmem(zone)) { |
| 4167 | managed_highpages += zone->managed_pages; |
| 4168 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); |
| 4169 | } |
| 4170 | } |
| 4171 | val->totalhigh = managed_highpages; |
| 4172 | val->freehigh = free_highpages; |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 4173 | #else |
Joonsoo Kim | fc2bd79 | 2016-05-19 17:12:23 -0700 | [diff] [blame] | 4174 | val->totalhigh = managed_highpages; |
| 4175 | val->freehigh = free_highpages; |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 4176 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4177 | val->mem_unit = PAGE_SIZE; |
| 4178 | } |
| 4179 | #endif |
| 4180 | |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4181 | /* |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 4182 | * Determine whether the node should be displayed or not, depending on whether |
| 4183 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4184 | */ |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 4185 | bool skip_free_areas_node(unsigned int flags, int nid) |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4186 | { |
| 4187 | bool ret = false; |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 4188 | unsigned int cpuset_mems_cookie; |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4189 | |
| 4190 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
| 4191 | goto out; |
| 4192 | |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 4193 | do { |
Mel Gorman | d26914d | 2014-04-03 14:47:24 -0700 | [diff] [blame] | 4194 | cpuset_mems_cookie = read_mems_allowed_begin(); |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 4195 | ret = !node_isset(nid, cpuset_current_mems_allowed); |
Mel Gorman | d26914d | 2014-04-03 14:47:24 -0700 | [diff] [blame] | 4196 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4197 | out: |
| 4198 | return ret; |
| 4199 | } |
| 4200 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4201 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
| 4202 | |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4203 | static void show_migration_types(unsigned char type) |
| 4204 | { |
| 4205 | static const char types[MIGRATE_TYPES] = { |
| 4206 | [MIGRATE_UNMOVABLE] = 'U', |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4207 | [MIGRATE_MOVABLE] = 'M', |
Vlastimil Babka | 475a2f9 | 2015-12-11 13:40:29 -0800 | [diff] [blame] | 4208 | [MIGRATE_RECLAIMABLE] = 'E', |
| 4209 | [MIGRATE_HIGHATOMIC] = 'H', |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4210 | #ifdef CONFIG_CMA |
| 4211 | [MIGRATE_CMA] = 'C', |
| 4212 | #endif |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 4213 | #ifdef CONFIG_MEMORY_ISOLATION |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4214 | [MIGRATE_ISOLATE] = 'I', |
Minchan Kim | 194159f | 2013-02-22 16:33:58 -0800 | [diff] [blame] | 4215 | #endif |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4216 | }; |
| 4217 | char tmp[MIGRATE_TYPES + 1]; |
| 4218 | char *p = tmp; |
| 4219 | int i; |
| 4220 | |
| 4221 | for (i = 0; i < MIGRATE_TYPES; i++) { |
| 4222 | if (type & (1 << i)) |
| 4223 | *p++ = types[i]; |
| 4224 | } |
| 4225 | |
| 4226 | *p = '\0'; |
| 4227 | printk("(%s) ", tmp); |
| 4228 | } |
| 4229 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4230 | /* |
| 4231 | * Show free area list (used inside shift_scroll-lock stuff) |
| 4232 | * We also calculate the percentage fragmentation. We do this by counting the |
| 4233 | * 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] | 4234 | * |
| 4235 | * Bits in @filter: |
| 4236 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's |
| 4237 | * cpuset. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4238 | */ |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 4239 | void show_free_areas(unsigned int filter) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4240 | { |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4241 | unsigned long free_pcp = 0; |
Jes Sorensen | c724191 | 2006-09-27 01:50:05 -0700 | [diff] [blame] | 4242 | int cpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4243 | struct zone *zone; |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 4244 | pg_data_t *pgdat; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4245 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 4246 | for_each_populated_zone(zone) { |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 4247 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4248 | continue; |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4249 | |
Konstantin Khlebnikov | 761b067 | 2015-04-14 15:45:32 -0700 | [diff] [blame] | 4250 | for_each_online_cpu(cpu) |
| 4251 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4252 | } |
| 4253 | |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 4254 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
| 4255 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4256 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
| 4257 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
Bartlomiej Zolnierkiewicz | d1ce749 | 2012-10-08 16:32:02 -0700 | [diff] [blame] | 4258 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4259 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 4260 | global_node_page_state(NR_ACTIVE_ANON), |
| 4261 | global_node_page_state(NR_INACTIVE_ANON), |
| 4262 | global_node_page_state(NR_ISOLATED_ANON), |
| 4263 | global_node_page_state(NR_ACTIVE_FILE), |
| 4264 | global_node_page_state(NR_INACTIVE_FILE), |
| 4265 | global_node_page_state(NR_ISOLATED_FILE), |
| 4266 | global_node_page_state(NR_UNEVICTABLE), |
Mel Gorman | 11fb998 | 2016-07-28 15:46:20 -0700 | [diff] [blame] | 4267 | global_node_page_state(NR_FILE_DIRTY), |
| 4268 | global_node_page_state(NR_WRITEBACK), |
| 4269 | global_node_page_state(NR_UNSTABLE_NFS), |
KOSAKI Motohiro | 3701b03 | 2009-09-21 17:01:29 -0700 | [diff] [blame] | 4270 | global_page_state(NR_SLAB_RECLAIMABLE), |
| 4271 | global_page_state(NR_SLAB_UNRECLAIMABLE), |
Mel Gorman | 50658e2 | 2016-07-28 15:46:14 -0700 | [diff] [blame] | 4272 | global_node_page_state(NR_FILE_MAPPED), |
Mel Gorman | 11fb998 | 2016-07-28 15:46:20 -0700 | [diff] [blame] | 4273 | global_node_page_state(NR_SHMEM), |
Andrew Morton | a25700a | 2007-02-08 14:20:40 -0800 | [diff] [blame] | 4274 | global_page_state(NR_PAGETABLE), |
Bartlomiej Zolnierkiewicz | d1ce749 | 2012-10-08 16:32:02 -0700 | [diff] [blame] | 4275 | global_page_state(NR_BOUNCE), |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4276 | global_page_state(NR_FREE_PAGES), |
| 4277 | free_pcp, |
Bartlomiej Zolnierkiewicz | d1ce749 | 2012-10-08 16:32:02 -0700 | [diff] [blame] | 4278 | global_page_state(NR_FREE_CMA_PAGES)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4279 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 4280 | for_each_online_pgdat(pgdat) { |
| 4281 | printk("Node %d" |
| 4282 | " active_anon:%lukB" |
| 4283 | " inactive_anon:%lukB" |
| 4284 | " active_file:%lukB" |
| 4285 | " inactive_file:%lukB" |
| 4286 | " unevictable:%lukB" |
| 4287 | " isolated(anon):%lukB" |
| 4288 | " isolated(file):%lukB" |
Mel Gorman | 50658e2 | 2016-07-28 15:46:14 -0700 | [diff] [blame] | 4289 | " mapped:%lukB" |
Mel Gorman | 11fb998 | 2016-07-28 15:46:20 -0700 | [diff] [blame] | 4290 | " dirty:%lukB" |
| 4291 | " writeback:%lukB" |
| 4292 | " shmem:%lukB" |
| 4293 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 4294 | " shmem_thp: %lukB" |
| 4295 | " shmem_pmdmapped: %lukB" |
| 4296 | " anon_thp: %lukB" |
| 4297 | #endif |
| 4298 | " writeback_tmp:%lukB" |
| 4299 | " unstable:%lukB" |
Minchan Kim | 33e077b | 2016-07-28 15:47:14 -0700 | [diff] [blame] | 4300 | " pages_scanned:%lu" |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 4301 | " all_unreclaimable? %s" |
| 4302 | "\n", |
| 4303 | pgdat->node_id, |
| 4304 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), |
| 4305 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), |
| 4306 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), |
| 4307 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), |
| 4308 | K(node_page_state(pgdat, NR_UNEVICTABLE)), |
| 4309 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), |
| 4310 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), |
Mel Gorman | 50658e2 | 2016-07-28 15:46:14 -0700 | [diff] [blame] | 4311 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
Mel Gorman | 11fb998 | 2016-07-28 15:46:20 -0700 | [diff] [blame] | 4312 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
| 4313 | K(node_page_state(pgdat, NR_WRITEBACK)), |
| 4314 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 4315 | K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR), |
| 4316 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) |
| 4317 | * HPAGE_PMD_NR), |
| 4318 | K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR), |
| 4319 | #endif |
| 4320 | K(node_page_state(pgdat, NR_SHMEM)), |
| 4321 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
| 4322 | K(node_page_state(pgdat, NR_UNSTABLE_NFS)), |
Minchan Kim | 33e077b | 2016-07-28 15:47:14 -0700 | [diff] [blame] | 4323 | node_page_state(pgdat, NR_PAGES_SCANNED), |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 4324 | !pgdat_reclaimable(pgdat) ? "yes" : "no"); |
| 4325 | } |
| 4326 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 4327 | for_each_populated_zone(zone) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4328 | int i; |
| 4329 | |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 4330 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4331 | continue; |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4332 | |
| 4333 | free_pcp = 0; |
| 4334 | for_each_online_cpu(cpu) |
| 4335 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; |
| 4336 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4337 | show_node(zone); |
| 4338 | printk("%s" |
| 4339 | " free:%lukB" |
| 4340 | " min:%lukB" |
| 4341 | " low:%lukB" |
| 4342 | " high:%lukB" |
Minchan Kim | 71c799f | 2016-07-28 15:47:26 -0700 | [diff] [blame] | 4343 | " active_anon:%lukB" |
| 4344 | " inactive_anon:%lukB" |
| 4345 | " active_file:%lukB" |
| 4346 | " inactive_file:%lukB" |
| 4347 | " unevictable:%lukB" |
Mel Gorman | 5a1c84b | 2016-07-28 15:47:31 -0700 | [diff] [blame] | 4348 | " writepending:%lukB" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4349 | " present:%lukB" |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 4350 | " managed:%lukB" |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4351 | " mlocked:%lukB" |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4352 | " slab_reclaimable:%lukB" |
| 4353 | " slab_unreclaimable:%lukB" |
KOSAKI Motohiro | c6a7f57 | 2009-09-21 17:01:32 -0700 | [diff] [blame] | 4354 | " kernel_stack:%lukB" |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4355 | " pagetables:%lukB" |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4356 | " bounce:%lukB" |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4357 | " free_pcp:%lukB" |
| 4358 | " local_pcp:%ukB" |
Bartlomiej Zolnierkiewicz | d1ce749 | 2012-10-08 16:32:02 -0700 | [diff] [blame] | 4359 | " free_cma:%lukB" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4360 | "\n", |
| 4361 | zone->name, |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 4362 | K(zone_page_state(zone, NR_FREE_PAGES)), |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 4363 | K(min_wmark_pages(zone)), |
| 4364 | K(low_wmark_pages(zone)), |
| 4365 | K(high_wmark_pages(zone)), |
Minchan Kim | 71c799f | 2016-07-28 15:47:26 -0700 | [diff] [blame] | 4366 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
| 4367 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), |
| 4368 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), |
| 4369 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), |
| 4370 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), |
Mel Gorman | 5a1c84b | 2016-07-28 15:47:31 -0700 | [diff] [blame] | 4371 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4372 | K(zone->present_pages), |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 4373 | K(zone->managed_pages), |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4374 | K(zone_page_state(zone, NR_MLOCK)), |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4375 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
| 4376 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), |
Andy Lutomirski | d30dd8b | 2016-07-28 15:48:14 -0700 | [diff] [blame] | 4377 | zone_page_state(zone, NR_KERNEL_STACK_KB), |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4378 | K(zone_page_state(zone, NR_PAGETABLE)), |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 4379 | K(zone_page_state(zone, NR_BOUNCE)), |
Konstantin Khlebnikov | d1bfcdb | 2015-04-14 15:45:30 -0700 | [diff] [blame] | 4380 | K(free_pcp), |
| 4381 | K(this_cpu_read(zone->pageset->pcp.count)), |
Minchan Kim | 33e077b | 2016-07-28 15:47:14 -0700 | [diff] [blame] | 4382 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4383 | printk("lowmem_reserve[]:"); |
| 4384 | for (i = 0; i < MAX_NR_ZONES; i++) |
Mel Gorman | 3484b2d | 2014-08-06 16:07:14 -0700 | [diff] [blame] | 4385 | printk(" %ld", zone->lowmem_reserve[i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4386 | printk("\n"); |
| 4387 | } |
| 4388 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 4389 | for_each_populated_zone(zone) { |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 4390 | unsigned int order; |
| 4391 | unsigned long nr[MAX_ORDER], flags, total = 0; |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4392 | unsigned char types[MAX_ORDER]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4393 | |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 4394 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 4395 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4396 | show_node(zone); |
| 4397 | printk("%s: ", zone->name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4398 | |
| 4399 | spin_lock_irqsave(&zone->lock, flags); |
| 4400 | for (order = 0; order < MAX_ORDER; order++) { |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4401 | struct free_area *area = &zone->free_area[order]; |
| 4402 | int type; |
| 4403 | |
| 4404 | nr[order] = area->nr_free; |
Kirill Korotaev | 8f9de51 | 2006-06-23 02:03:50 -0700 | [diff] [blame] | 4405 | total += nr[order] << order; |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4406 | |
| 4407 | types[order] = 0; |
| 4408 | for (type = 0; type < MIGRATE_TYPES; type++) { |
| 4409 | if (!list_empty(&area->free_list[type])) |
| 4410 | types[order] |= 1 << type; |
| 4411 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4412 | } |
| 4413 | spin_unlock_irqrestore(&zone->lock, flags); |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4414 | for (order = 0; order < MAX_ORDER; order++) { |
Kirill Korotaev | 8f9de51 | 2006-06-23 02:03:50 -0700 | [diff] [blame] | 4415 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
Rabin Vincent | 377e4f1 | 2012-12-11 16:00:24 -0800 | [diff] [blame] | 4416 | if (nr[order]) |
| 4417 | show_migration_types(types[order]); |
| 4418 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4419 | printk("= %lukB\n", K(total)); |
| 4420 | } |
| 4421 | |
David Rientjes | 949f7ec | 2013-04-29 15:07:48 -0700 | [diff] [blame] | 4422 | hugetlb_show_meminfo(); |
| 4423 | |
Mel Gorman | 11fb998 | 2016-07-28 15:46:20 -0700 | [diff] [blame] | 4424 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
Larry Woodman | e6f3602 | 2008-02-04 22:29:30 -0800 | [diff] [blame] | 4425 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4426 | show_swap_cache_info(); |
| 4427 | } |
| 4428 | |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 4429 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
| 4430 | { |
| 4431 | zoneref->zone = zone; |
| 4432 | zoneref->zone_idx = zone_idx(zone); |
| 4433 | } |
| 4434 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4435 | /* |
| 4436 | * Builds allocation fallback zone lists. |
Christoph Lameter | 1a93205 | 2006-01-06 00:11:16 -0800 | [diff] [blame] | 4437 | * |
| 4438 | * Add all populated zones of a node to the zonelist. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4439 | */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4440 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4441 | int nr_zones) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4442 | { |
Christoph Lameter | 1a93205 | 2006-01-06 00:11:16 -0800 | [diff] [blame] | 4443 | struct zone *zone; |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4444 | enum zone_type zone_type = MAX_NR_ZONES; |
Christoph Lameter | 02a68a5 | 2006-01-06 00:11:18 -0800 | [diff] [blame] | 4445 | |
| 4446 | do { |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 4447 | zone_type--; |
Christoph Lameter | 070f803 | 2006-01-06 00:11:19 -0800 | [diff] [blame] | 4448 | zone = pgdat->node_zones + zone_type; |
Mel Gorman | 6aa303d | 2016-09-01 16:14:55 -0700 | [diff] [blame] | 4449 | if (managed_zone(zone)) { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4450 | zoneref_set_zone(zone, |
| 4451 | &zonelist->_zonerefs[nr_zones++]); |
Christoph Lameter | 070f803 | 2006-01-06 00:11:19 -0800 | [diff] [blame] | 4452 | check_highest_zone(zone_type); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4453 | } |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 4454 | } while (zone_type); |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4455 | |
Christoph Lameter | 070f803 | 2006-01-06 00:11:19 -0800 | [diff] [blame] | 4456 | return nr_zones; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4457 | } |
| 4458 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4459 | |
| 4460 | /* |
| 4461 | * zonelist_order: |
| 4462 | * 0 = automatic detection of better ordering. |
| 4463 | * 1 = order by ([node] distance, -zonetype) |
| 4464 | * 2 = order by (-zonetype, [node] distance) |
| 4465 | * |
| 4466 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create |
| 4467 | * the same zonelist. So only NUMA can configure this param. |
| 4468 | */ |
| 4469 | #define ZONELIST_ORDER_DEFAULT 0 |
| 4470 | #define ZONELIST_ORDER_NODE 1 |
| 4471 | #define ZONELIST_ORDER_ZONE 2 |
| 4472 | |
| 4473 | /* zonelist order in the kernel. |
| 4474 | * set_zonelist_order() will set this to NODE or ZONE. |
| 4475 | */ |
| 4476 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; |
| 4477 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; |
| 4478 | |
| 4479 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4480 | #ifdef CONFIG_NUMA |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4481 | /* The value user specified ....changed by config */ |
| 4482 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; |
| 4483 | /* string for sysctl */ |
| 4484 | #define NUMA_ZONELIST_ORDER_LEN 16 |
| 4485 | char numa_zonelist_order[16] = "default"; |
| 4486 | |
| 4487 | /* |
| 4488 | * interface for configure zonelist ordering. |
| 4489 | * command line option "numa_zonelist_order" |
| 4490 | * = "[dD]efault - default, automatic configuration. |
| 4491 | * = "[nN]ode - order by node locality, then by zone within node |
| 4492 | * = "[zZ]one - order by zone, then by locality within zone |
| 4493 | */ |
| 4494 | |
| 4495 | static int __parse_numa_zonelist_order(char *s) |
| 4496 | { |
| 4497 | if (*s == 'd' || *s == 'D') { |
| 4498 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; |
| 4499 | } else if (*s == 'n' || *s == 'N') { |
| 4500 | user_zonelist_order = ZONELIST_ORDER_NODE; |
| 4501 | } else if (*s == 'z' || *s == 'Z') { |
| 4502 | user_zonelist_order = ZONELIST_ORDER_ZONE; |
| 4503 | } else { |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 4504 | pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4505 | return -EINVAL; |
| 4506 | } |
| 4507 | return 0; |
| 4508 | } |
| 4509 | |
| 4510 | static __init int setup_numa_zonelist_order(char *s) |
| 4511 | { |
Volodymyr G. Lukiianyk | ecb256f | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 4512 | int ret; |
| 4513 | |
| 4514 | if (!s) |
| 4515 | return 0; |
| 4516 | |
| 4517 | ret = __parse_numa_zonelist_order(s); |
| 4518 | if (ret == 0) |
| 4519 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); |
| 4520 | |
| 4521 | return ret; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4522 | } |
| 4523 | early_param("numa_zonelist_order", setup_numa_zonelist_order); |
| 4524 | |
| 4525 | /* |
| 4526 | * sysctl handler for numa_zonelist_order |
| 4527 | */ |
Joe Perches | cccad5b | 2014-06-06 14:38:09 -0700 | [diff] [blame] | 4528 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 4529 | void __user *buffer, size_t *length, |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4530 | loff_t *ppos) |
| 4531 | { |
| 4532 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; |
| 4533 | int ret; |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 4534 | static DEFINE_MUTEX(zl_order_mutex); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4535 | |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 4536 | mutex_lock(&zl_order_mutex); |
Chen Gang | dacbde0 | 2013-07-03 15:02:35 -0700 | [diff] [blame] | 4537 | if (write) { |
| 4538 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { |
| 4539 | ret = -EINVAL; |
| 4540 | goto out; |
| 4541 | } |
| 4542 | strcpy(saved_string, (char *)table->data); |
| 4543 | } |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 4544 | ret = proc_dostring(table, write, buffer, length, ppos); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4545 | if (ret) |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 4546 | goto out; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4547 | if (write) { |
| 4548 | int oldval = user_zonelist_order; |
Chen Gang | dacbde0 | 2013-07-03 15:02:35 -0700 | [diff] [blame] | 4549 | |
| 4550 | ret = __parse_numa_zonelist_order((char *)table->data); |
| 4551 | if (ret) { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4552 | /* |
| 4553 | * bogus value. restore saved string |
| 4554 | */ |
Chen Gang | dacbde0 | 2013-07-03 15:02:35 -0700 | [diff] [blame] | 4555 | strncpy((char *)table->data, saved_string, |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4556 | NUMA_ZONELIST_ORDER_LEN); |
| 4557 | user_zonelist_order = oldval; |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 4558 | } else if (oldval != user_zonelist_order) { |
| 4559 | mutex_lock(&zonelists_mutex); |
Jiang Liu | 9adb62a | 2012-07-31 16:43:28 -0700 | [diff] [blame] | 4560 | build_all_zonelists(NULL, NULL); |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 4561 | mutex_unlock(&zonelists_mutex); |
| 4562 | } |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4563 | } |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 4564 | out: |
| 4565 | mutex_unlock(&zl_order_mutex); |
| 4566 | return ret; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4567 | } |
| 4568 | |
| 4569 | |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 4570 | #define MAX_NODE_LOAD (nr_online_nodes) |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4571 | static int node_load[MAX_NUMNODES]; |
| 4572 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4573 | /** |
Pavel Pisa | 4dc3b16 | 2005-05-01 08:59:25 -0700 | [diff] [blame] | 4574 | * 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] | 4575 | * @node: node whose fallback list we're appending |
| 4576 | * @used_node_mask: nodemask_t of already used nodes |
| 4577 | * |
| 4578 | * We use a number of factors to determine which is the next node that should |
| 4579 | * appear on a given node's fallback list. The node should not have appeared |
| 4580 | * already in @node's fallback list, and it should be the next closest node |
| 4581 | * according to the distance array (which contains arbitrary distance values |
| 4582 | * from each node to each node in the system), and should also prefer nodes |
| 4583 | * with no CPUs, since presumably they'll have very little allocation pressure |
| 4584 | * on them otherwise. |
| 4585 | * It returns -1 if no node is found. |
| 4586 | */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4587 | 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] | 4588 | { |
Linus Torvalds | 4cf808eb | 2006-02-17 20:38:21 +0100 | [diff] [blame] | 4589 | int n, val; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4590 | int min_val = INT_MAX; |
David Rientjes | 00ef2d2 | 2013-02-22 16:35:36 -0800 | [diff] [blame] | 4591 | int best_node = NUMA_NO_NODE; |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 4592 | const struct cpumask *tmp = cpumask_of_node(0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4593 | |
Linus Torvalds | 4cf808eb | 2006-02-17 20:38:21 +0100 | [diff] [blame] | 4594 | /* Use the local node if we haven't already */ |
| 4595 | if (!node_isset(node, *used_node_mask)) { |
| 4596 | node_set(node, *used_node_mask); |
| 4597 | return node; |
| 4598 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4599 | |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 4600 | for_each_node_state(n, N_MEMORY) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4601 | |
| 4602 | /* Don't want a node to appear more than once */ |
| 4603 | if (node_isset(n, *used_node_mask)) |
| 4604 | continue; |
| 4605 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4606 | /* Use the distance array to find the distance */ |
| 4607 | val = node_distance(node, n); |
| 4608 | |
Linus Torvalds | 4cf808eb | 2006-02-17 20:38:21 +0100 | [diff] [blame] | 4609 | /* Penalize nodes under us ("prefer the next node") */ |
| 4610 | val += (n < node); |
| 4611 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4612 | /* Give preference to headless and unused nodes */ |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 4613 | tmp = cpumask_of_node(n); |
| 4614 | if (!cpumask_empty(tmp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4615 | val += PENALTY_FOR_NODE_WITH_CPUS; |
| 4616 | |
| 4617 | /* Slight preference for less loaded node */ |
| 4618 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); |
| 4619 | val += node_load[n]; |
| 4620 | |
| 4621 | if (val < min_val) { |
| 4622 | min_val = val; |
| 4623 | best_node = n; |
| 4624 | } |
| 4625 | } |
| 4626 | |
| 4627 | if (best_node >= 0) |
| 4628 | node_set(best_node, *used_node_mask); |
| 4629 | |
| 4630 | return best_node; |
| 4631 | } |
| 4632 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4633 | |
| 4634 | /* |
| 4635 | * Build zonelists ordered by node and zones within node. |
| 4636 | * This results in maximum locality--normal zone overflows into local |
| 4637 | * DMA zone, if any--but risks exhausting DMA zone. |
| 4638 | */ |
| 4639 | 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] | 4640 | { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4641 | int j; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4642 | struct zonelist *zonelist; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4643 | |
Aneesh Kumar K.V | c9634cf | 2016-10-07 16:59:12 -0700 | [diff] [blame] | 4644 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4645 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4646 | ; |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4647 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4648 | zonelist->_zonerefs[j].zone = NULL; |
| 4649 | zonelist->_zonerefs[j].zone_idx = 0; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4650 | } |
| 4651 | |
| 4652 | /* |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 4653 | * Build gfp_thisnode zonelists |
| 4654 | */ |
| 4655 | static void build_thisnode_zonelists(pg_data_t *pgdat) |
| 4656 | { |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 4657 | int j; |
| 4658 | struct zonelist *zonelist; |
| 4659 | |
Aneesh Kumar K.V | c9634cf | 2016-10-07 16:59:12 -0700 | [diff] [blame] | 4660 | zonelist = &pgdat->node_zonelists[ZONELIST_NOFALLBACK]; |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4661 | j = build_zonelists_node(pgdat, zonelist, 0); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4662 | zonelist->_zonerefs[j].zone = NULL; |
| 4663 | zonelist->_zonerefs[j].zone_idx = 0; |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 4664 | } |
| 4665 | |
| 4666 | /* |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4667 | * Build zonelists ordered by zone and nodes within zones. |
| 4668 | * This results in conserving DMA zone[s] until all Normal memory is |
| 4669 | * exhausted, but results in overflowing to remote node while memory |
| 4670 | * may still exist in local DMA zone. |
| 4671 | */ |
| 4672 | static int node_order[MAX_NUMNODES]; |
| 4673 | |
| 4674 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) |
| 4675 | { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4676 | int pos, j, node; |
| 4677 | int zone_type; /* needs to be signed */ |
| 4678 | struct zone *z; |
| 4679 | struct zonelist *zonelist; |
| 4680 | |
Aneesh Kumar K.V | c9634cf | 2016-10-07 16:59:12 -0700 | [diff] [blame] | 4681 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4682 | pos = 0; |
| 4683 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { |
| 4684 | for (j = 0; j < nr_nodes; j++) { |
| 4685 | node = node_order[j]; |
| 4686 | z = &NODE_DATA(node)->node_zones[zone_type]; |
Mel Gorman | 6aa303d | 2016-09-01 16:14:55 -0700 | [diff] [blame] | 4687 | if (managed_zone(z)) { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4688 | zoneref_set_zone(z, |
| 4689 | &zonelist->_zonerefs[pos++]); |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4690 | check_highest_zone(zone_type); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4691 | } |
| 4692 | } |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4693 | } |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4694 | zonelist->_zonerefs[pos].zone = NULL; |
| 4695 | zonelist->_zonerefs[pos].zone_idx = 0; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4696 | } |
| 4697 | |
Mel Gorman | 3193913 | 2014-10-09 15:28:30 -0700 | [diff] [blame] | 4698 | #if defined(CONFIG_64BIT) |
| 4699 | /* |
| 4700 | * Devices that require DMA32/DMA are relatively rare and do not justify a |
| 4701 | * penalty to every machine in case the specialised case applies. Default |
| 4702 | * to Node-ordering on 64-bit NUMA machines |
| 4703 | */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4704 | static int default_zonelist_order(void) |
| 4705 | { |
Mel Gorman | 3193913 | 2014-10-09 15:28:30 -0700 | [diff] [blame] | 4706 | return ZONELIST_ORDER_NODE; |
| 4707 | } |
| 4708 | #else |
| 4709 | /* |
| 4710 | * On 32-bit, the Normal zone needs to be preserved for allocations accessible |
| 4711 | * by the kernel. If processes running on node 0 deplete the low memory zone |
| 4712 | * then reclaim will occur more frequency increasing stalls and potentially |
| 4713 | * be easier to OOM if a large percentage of the zone is under writeback or |
| 4714 | * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set. |
| 4715 | * Hence, default to zone ordering on 32-bit. |
| 4716 | */ |
| 4717 | static int default_zonelist_order(void) |
| 4718 | { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4719 | return ZONELIST_ORDER_ZONE; |
| 4720 | } |
Mel Gorman | 3193913 | 2014-10-09 15:28:30 -0700 | [diff] [blame] | 4721 | #endif /* CONFIG_64BIT */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4722 | |
| 4723 | static void set_zonelist_order(void) |
| 4724 | { |
| 4725 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) |
| 4726 | current_zonelist_order = default_zonelist_order(); |
| 4727 | else |
| 4728 | current_zonelist_order = user_zonelist_order; |
| 4729 | } |
| 4730 | |
| 4731 | static void build_zonelists(pg_data_t *pgdat) |
| 4732 | { |
Yaowei Bai | c00eb15 | 2016-01-14 15:19:00 -0800 | [diff] [blame] | 4733 | int i, node, load; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4734 | nodemask_t used_mask; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4735 | int local_node, prev_node; |
| 4736 | struct zonelist *zonelist; |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 4737 | unsigned int order = current_zonelist_order; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4738 | |
| 4739 | /* initialize zonelists */ |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 4740 | for (i = 0; i < MAX_ZONELISTS; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4741 | zonelist = pgdat->node_zonelists + i; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4742 | zonelist->_zonerefs[0].zone = NULL; |
| 4743 | zonelist->_zonerefs[0].zone_idx = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4744 | } |
| 4745 | |
| 4746 | /* NUMA-aware ordering of nodes */ |
| 4747 | local_node = pgdat->node_id; |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 4748 | load = nr_online_nodes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4749 | prev_node = local_node; |
| 4750 | nodes_clear(used_mask); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4751 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4752 | memset(node_order, 0, sizeof(node_order)); |
Yaowei Bai | c00eb15 | 2016-01-14 15:19:00 -0800 | [diff] [blame] | 4753 | i = 0; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4754 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4755 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
| 4756 | /* |
| 4757 | * We don't want to pressure a particular node. |
| 4758 | * So adding penalty to the first node in same |
| 4759 | * distance group to make it round-robin. |
| 4760 | */ |
David Rientjes | 957f822 | 2012-10-08 16:33:24 -0700 | [diff] [blame] | 4761 | if (node_distance(local_node, node) != |
| 4762 | node_distance(local_node, prev_node)) |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4763 | node_load[node] = load; |
| 4764 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4765 | prev_node = node; |
| 4766 | load--; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4767 | if (order == ZONELIST_ORDER_NODE) |
| 4768 | build_zonelists_in_node_order(pgdat, node); |
| 4769 | else |
Yaowei Bai | c00eb15 | 2016-01-14 15:19:00 -0800 | [diff] [blame] | 4770 | node_order[i++] = node; /* remember order */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4771 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4772 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4773 | if (order == ZONELIST_ORDER_ZONE) { |
| 4774 | /* calculate node order -- i.e., DMA last! */ |
Yaowei Bai | c00eb15 | 2016-01-14 15:19:00 -0800 | [diff] [blame] | 4775 | build_zonelists_in_zone_order(pgdat, i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4776 | } |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 4777 | |
| 4778 | build_thisnode_zonelists(pgdat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4779 | } |
| 4780 | |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 4781 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
| 4782 | /* |
| 4783 | * Return node id of node used for "local" allocations. |
| 4784 | * I.e., first node id of first zone in arg node's generic zonelist. |
| 4785 | * Used for initializing percpu 'numa_mem', which is used primarily |
| 4786 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. |
| 4787 | */ |
| 4788 | int local_memory_node(int node) |
| 4789 | { |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 4790 | struct zoneref *z; |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 4791 | |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 4792 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 4793 | gfp_zone(GFP_KERNEL), |
Mel Gorman | c33d6c0 | 2016-05-19 17:14:10 -0700 | [diff] [blame] | 4794 | NULL); |
| 4795 | return z->zone->node; |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 4796 | } |
| 4797 | #endif |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4798 | |
Joonsoo Kim | 6423aa8 | 2016-08-10 16:27:49 -0700 | [diff] [blame] | 4799 | static void setup_min_unmapped_ratio(void); |
| 4800 | static void setup_min_slab_ratio(void); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4801 | #else /* CONFIG_NUMA */ |
| 4802 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4803 | static void set_zonelist_order(void) |
| 4804 | { |
| 4805 | current_zonelist_order = ZONELIST_ORDER_ZONE; |
| 4806 | } |
| 4807 | |
| 4808 | static void build_zonelists(pg_data_t *pgdat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4809 | { |
Christoph Lameter | 19655d3 | 2006-09-25 23:31:19 -0700 | [diff] [blame] | 4810 | int node, local_node; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4811 | enum zone_type j; |
| 4812 | struct zonelist *zonelist; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4813 | |
| 4814 | local_node = pgdat->node_id; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4815 | |
Aneesh Kumar K.V | c9634cf | 2016-10-07 16:59:12 -0700 | [diff] [blame] | 4816 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4817 | j = build_zonelists_node(pgdat, zonelist, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4818 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4819 | /* |
| 4820 | * Now we build the zonelist so that it contains the zones |
| 4821 | * of all the other nodes. |
| 4822 | * We don't want to pressure a particular node, so when |
| 4823 | * building the zones for node N, we make sure that the |
| 4824 | * zones coming right after the local ones are those from |
| 4825 | * node N+1 (modulo N) |
| 4826 | */ |
| 4827 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { |
| 4828 | if (!node_online(node)) |
| 4829 | continue; |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4830 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4831 | } |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4832 | for (node = 0; node < local_node; node++) { |
| 4833 | if (!node_online(node)) |
| 4834 | continue; |
Zhang Yanfei | bc732f1 | 2013-07-08 16:00:06 -0700 | [diff] [blame] | 4835 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 4836 | } |
| 4837 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 4838 | zonelist->_zonerefs[j].zone = NULL; |
| 4839 | zonelist->_zonerefs[j].zone_idx = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4840 | } |
| 4841 | |
| 4842 | #endif /* CONFIG_NUMA */ |
| 4843 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 4844 | /* |
| 4845 | * Boot pageset table. One per cpu which is going to be used for all |
| 4846 | * zones and all nodes. The parameters will be set in such a way |
| 4847 | * that an item put on a list will immediately be handed over to |
| 4848 | * the buddy list. This is safe since pageset manipulation is done |
| 4849 | * with interrupts disabled. |
| 4850 | * |
| 4851 | * The boot_pagesets must be kept even after bootup is complete for |
| 4852 | * unused processors and/or zones. They do play a role for bootstrapping |
| 4853 | * hotplugged processors. |
| 4854 | * |
| 4855 | * zoneinfo_show() and maybe other functions do |
| 4856 | * not check if the processor is online before following the pageset pointer. |
| 4857 | * Other parts of the kernel may not check if the zone is available. |
| 4858 | */ |
| 4859 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); |
| 4860 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); |
Haicheng Li | 1f52250 | 2010-05-24 14:32:51 -0700 | [diff] [blame] | 4861 | static void setup_zone_pageset(struct zone *zone); |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 4862 | |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 4863 | /* |
| 4864 | * Global mutex to protect against size modification of zonelists |
| 4865 | * as well as to serialize pageset setup for the new populated zone. |
| 4866 | */ |
| 4867 | DEFINE_MUTEX(zonelists_mutex); |
| 4868 | |
Rusty Russell | 9b1a4d3 | 2008-07-28 12:16:30 -0500 | [diff] [blame] | 4869 | /* return values int ....just for stop_machine() */ |
Jiang Liu | 4ed7e02 | 2012-07-31 16:43:35 -0700 | [diff] [blame] | 4870 | static int __build_all_zonelists(void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4871 | { |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 4872 | int nid; |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 4873 | int cpu; |
Jiang Liu | 9adb62a | 2012-07-31 16:43:28 -0700 | [diff] [blame] | 4874 | pg_data_t *self = data; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 4875 | |
Bo Liu | 7f9cfb3 | 2009-08-18 14:11:19 -0700 | [diff] [blame] | 4876 | #ifdef CONFIG_NUMA |
| 4877 | memset(node_load, 0, sizeof(node_load)); |
| 4878 | #endif |
Jiang Liu | 9adb62a | 2012-07-31 16:43:28 -0700 | [diff] [blame] | 4879 | |
| 4880 | if (self && !node_online(self->node_id)) { |
| 4881 | build_zonelists(self); |
Jiang Liu | 9adb62a | 2012-07-31 16:43:28 -0700 | [diff] [blame] | 4882 | } |
| 4883 | |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 4884 | for_each_online_node(nid) { |
Christoph Lameter | 7ea1530 | 2007-10-16 01:25:29 -0700 | [diff] [blame] | 4885 | pg_data_t *pgdat = NODE_DATA(nid); |
| 4886 | |
| 4887 | build_zonelists(pgdat); |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 4888 | } |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 4889 | |
| 4890 | /* |
| 4891 | * Initialize the boot_pagesets that are going to be used |
| 4892 | * for bootstrapping processors. The real pagesets for |
| 4893 | * each zone will be allocated later when the per cpu |
| 4894 | * allocator is available. |
| 4895 | * |
| 4896 | * boot_pagesets are used also for bootstrapping offline |
| 4897 | * cpus if the system is already booted because the pagesets |
| 4898 | * are needed to initialize allocators on a specific cpu too. |
| 4899 | * F.e. the percpu allocator needs the page allocator which |
| 4900 | * needs the percpu allocator in order to allocate its pagesets |
| 4901 | * (a chicken-egg dilemma). |
| 4902 | */ |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 4903 | for_each_possible_cpu(cpu) { |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 4904 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
| 4905 | |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 4906 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
| 4907 | /* |
| 4908 | * We now know the "local memory node" for each node-- |
| 4909 | * i.e., the node of the first zone in the generic zonelist. |
| 4910 | * Set up numa_mem percpu variable for on-line cpus. During |
| 4911 | * boot, only the boot cpu should be on-line; we'll init the |
| 4912 | * secondary cpus' numa_mem as they come on-line. During |
| 4913 | * node/memory hotplug, we'll fixup all on-line cpus. |
| 4914 | */ |
| 4915 | if (cpu_online(cpu)) |
| 4916 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
| 4917 | #endif |
| 4918 | } |
| 4919 | |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 4920 | return 0; |
| 4921 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4922 | |
Rasmus Villemoes | 061f67b | 2015-02-12 15:00:06 -0800 | [diff] [blame] | 4923 | static noinline void __init |
| 4924 | build_all_zonelists_init(void) |
| 4925 | { |
| 4926 | __build_all_zonelists(NULL); |
| 4927 | mminit_verify_zonelist(); |
| 4928 | cpuset_init_current_mems_allowed(); |
| 4929 | } |
| 4930 | |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 4931 | /* |
| 4932 | * Called with zonelists_mutex held always |
| 4933 | * unless system_state == SYSTEM_BOOTING. |
Rasmus Villemoes | 061f67b | 2015-02-12 15:00:06 -0800 | [diff] [blame] | 4934 | * |
| 4935 | * __ref due to (1) call of __meminit annotated setup_zone_pageset |
| 4936 | * [we're only called with non-NULL zone through __meminit paths] and |
| 4937 | * (2) call of __init annotated helper build_all_zonelists_init |
| 4938 | * [protected by SYSTEM_BOOTING]. |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 4939 | */ |
Jiang Liu | 9adb62a | 2012-07-31 16:43:28 -0700 | [diff] [blame] | 4940 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 4941 | { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4942 | set_zonelist_order(); |
| 4943 | |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 4944 | if (system_state == SYSTEM_BOOTING) { |
Rasmus Villemoes | 061f67b | 2015-02-12 15:00:06 -0800 | [diff] [blame] | 4945 | build_all_zonelists_init(); |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 4946 | } else { |
KAMEZAWA Hiroyuki | e9959f0 | 2010-11-24 12:57:09 -0800 | [diff] [blame] | 4947 | #ifdef CONFIG_MEMORY_HOTPLUG |
Jiang Liu | 9adb62a | 2012-07-31 16:43:28 -0700 | [diff] [blame] | 4948 | if (zone) |
| 4949 | setup_zone_pageset(zone); |
KAMEZAWA Hiroyuki | e9959f0 | 2010-11-24 12:57:09 -0800 | [diff] [blame] | 4950 | #endif |
Cody P Schafer | dd1895e | 2013-07-03 15:01:36 -0700 | [diff] [blame] | 4951 | /* we have to stop all cpus to guarantee there is no user |
| 4952 | of zonelist */ |
Jiang Liu | 9adb62a | 2012-07-31 16:43:28 -0700 | [diff] [blame] | 4953 | stop_machine(__build_all_zonelists, pgdat, NULL); |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 4954 | /* cpuset refresh routine should be here */ |
| 4955 | } |
Andrew Morton | bd1e22b | 2006-06-23 02:03:47 -0700 | [diff] [blame] | 4956 | vm_total_pages = nr_free_pagecache_pages(); |
Mel Gorman | 9ef9acb | 2007-10-16 01:25:54 -0700 | [diff] [blame] | 4957 | /* |
| 4958 | * Disable grouping by mobility if the number of pages in the |
| 4959 | * system is too low to allow the mechanism to work. It would be |
| 4960 | * more accurate, but expensive to check per-zone. This check is |
| 4961 | * made on memory-hotadd so a system can start with mobility |
| 4962 | * disabled and enable it later |
| 4963 | */ |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 4964 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
Mel Gorman | 9ef9acb | 2007-10-16 01:25:54 -0700 | [diff] [blame] | 4965 | page_group_by_mobility_disabled = 1; |
| 4966 | else |
| 4967 | page_group_by_mobility_disabled = 0; |
| 4968 | |
Joe Perches | 756a025 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 4969 | pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n", |
| 4970 | nr_online_nodes, |
| 4971 | zonelist_order_name[current_zonelist_order], |
| 4972 | page_group_by_mobility_disabled ? "off" : "on", |
| 4973 | vm_total_pages); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4974 | #ifdef CONFIG_NUMA |
Anton Blanchard | f88dfff | 2014-12-10 15:42:53 -0800 | [diff] [blame] | 4975 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 4976 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4977 | } |
| 4978 | |
| 4979 | /* |
| 4980 | * Helper functions to size the waitqueue hash table. |
| 4981 | * Essentially these want to choose hash table sizes sufficiently |
| 4982 | * large so that collisions trying to wait on pages are rare. |
| 4983 | * But in fact, the number of active page waitqueues on typical |
| 4984 | * systems is ridiculously low, less than 200. So this is even |
| 4985 | * conservative, even though it seems large. |
| 4986 | * |
| 4987 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to |
| 4988 | * waitqueues, i.e. the size of the waitq table given the number of pages. |
| 4989 | */ |
| 4990 | #define PAGES_PER_WAITQUEUE 256 |
| 4991 | |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 4992 | #ifndef CONFIG_MEMORY_HOTPLUG |
Yasunori Goto | 02b694d | 2006-06-23 02:03:08 -0700 | [diff] [blame] | 4993 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4994 | { |
| 4995 | unsigned long size = 1; |
| 4996 | |
| 4997 | pages /= PAGES_PER_WAITQUEUE; |
| 4998 | |
| 4999 | while (size < pages) |
| 5000 | size <<= 1; |
| 5001 | |
| 5002 | /* |
| 5003 | * Once we have dozens or even hundreds of threads sleeping |
| 5004 | * on IO we've got bigger problems than wait queue collision. |
| 5005 | * Limit the size of the wait table to a reasonable size. |
| 5006 | */ |
| 5007 | size = min(size, 4096UL); |
| 5008 | |
| 5009 | return max(size, 4UL); |
| 5010 | } |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5011 | #else |
| 5012 | /* |
| 5013 | * A zone's size might be changed by hot-add, so it is not possible to determine |
| 5014 | * a suitable size for its wait_table. So we use the maximum size now. |
| 5015 | * |
| 5016 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: |
| 5017 | * |
| 5018 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. |
| 5019 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. |
| 5020 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. |
| 5021 | * |
| 5022 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages |
| 5023 | * or more by the traditional way. (See above). It equals: |
| 5024 | * |
| 5025 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. |
| 5026 | * ia64(16K page size) : = ( 8G + 4M)byte. |
| 5027 | * powerpc (64K page size) : = (32G +16M)byte. |
| 5028 | */ |
| 5029 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
| 5030 | { |
| 5031 | return 4096UL; |
| 5032 | } |
| 5033 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5034 | |
| 5035 | /* |
| 5036 | * This is an integer logarithm so that shifts can be used later |
| 5037 | * to extract the more random high bits from the multiplicative |
| 5038 | * hash function before the remainder is taken. |
| 5039 | */ |
| 5040 | static inline unsigned long wait_table_bits(unsigned long size) |
| 5041 | { |
| 5042 | return ffz(~size); |
| 5043 | } |
| 5044 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 5045 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5046 | * Initially all pages are reserved - free ones are freed |
| 5047 | * up by free_all_bootmem() once the early boot process is |
| 5048 | * done. Non-atomic initialization, single-pass. |
| 5049 | */ |
Matt Tolentino | c09b424 | 2006-01-17 07:03:44 +0100 | [diff] [blame] | 5050 | 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] | 5051 | unsigned long start_pfn, enum memmap_context context) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5052 | { |
Dan Williams | 4b94ffd | 2016-01-15 16:56:22 -0800 | [diff] [blame] | 5053 | struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn)); |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 5054 | unsigned long end_pfn = start_pfn + size; |
Dan Williams | 4b94ffd | 2016-01-15 16:56:22 -0800 | [diff] [blame] | 5055 | pg_data_t *pgdat = NODE_DATA(nid); |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 5056 | unsigned long pfn; |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 5057 | unsigned long nr_initialised = 0; |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5058 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
| 5059 | struct memblock_region *r = NULL, *tmp; |
| 5060 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5061 | |
Hugh Dickins | 22b31ee | 2009-01-06 14:40:09 -0800 | [diff] [blame] | 5062 | if (highest_memmap_pfn < end_pfn - 1) |
| 5063 | highest_memmap_pfn = end_pfn - 1; |
| 5064 | |
Dan Williams | 4b94ffd | 2016-01-15 16:56:22 -0800 | [diff] [blame] | 5065 | /* |
| 5066 | * Honor reservation requested by the driver for this ZONE_DEVICE |
| 5067 | * memory |
| 5068 | */ |
| 5069 | if (altmap && start_pfn == altmap->base_pfn) |
| 5070 | start_pfn += altmap->reserve; |
| 5071 | |
Greg Ungerer | cbe8dd4 | 2006-01-12 01:05:24 -0800 | [diff] [blame] | 5072 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 5073 | /* |
Andrew Morton | b72d0ff | 2016-03-15 14:55:25 -0700 | [diff] [blame] | 5074 | * There can be holes in boot-time mem_map[]s handed to this |
| 5075 | * function. They do not exist on hotplugged memory. |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 5076 | */ |
Andrew Morton | b72d0ff | 2016-03-15 14:55:25 -0700 | [diff] [blame] | 5077 | if (context != MEMMAP_EARLY) |
| 5078 | goto not_early; |
| 5079 | |
| 5080 | if (!early_pfn_valid(pfn)) |
| 5081 | continue; |
| 5082 | if (!early_pfn_in_nid(pfn, nid)) |
| 5083 | continue; |
| 5084 | if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised)) |
| 5085 | break; |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5086 | |
| 5087 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
Andrew Morton | b72d0ff | 2016-03-15 14:55:25 -0700 | [diff] [blame] | 5088 | /* |
Andrew Morton | b72d0ff | 2016-03-15 14:55:25 -0700 | [diff] [blame] | 5089 | * Check given memblock attribute by firmware which can affect |
| 5090 | * kernel memory layout. If zone==ZONE_MOVABLE but memory is |
| 5091 | * mirrored, it's an overlapped memmap init. skip it. |
| 5092 | */ |
| 5093 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { |
| 5094 | if (!r || pfn >= memblock_region_memory_end_pfn(r)) { |
| 5095 | for_each_memblock(memory, tmp) |
| 5096 | if (pfn < memblock_region_memory_end_pfn(tmp)) |
| 5097 | break; |
| 5098 | r = tmp; |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5099 | } |
Andrew Morton | b72d0ff | 2016-03-15 14:55:25 -0700 | [diff] [blame] | 5100 | if (pfn >= memblock_region_memory_base_pfn(r) && |
| 5101 | memblock_is_mirror(r)) { |
| 5102 | /* already initialized as NORMAL */ |
| 5103 | pfn = memblock_region_memory_end_pfn(r); |
| 5104 | continue; |
| 5105 | } |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 5106 | } |
Andrew Morton | b72d0ff | 2016-03-15 14:55:25 -0700 | [diff] [blame] | 5107 | #endif |
Mel Gorman | ac5d253 | 2015-06-30 14:57:20 -0700 | [diff] [blame] | 5108 | |
Andrew Morton | b72d0ff | 2016-03-15 14:55:25 -0700 | [diff] [blame] | 5109 | not_early: |
Mel Gorman | ac5d253 | 2015-06-30 14:57:20 -0700 | [diff] [blame] | 5110 | /* |
| 5111 | * Mark the block movable so that blocks are reserved for |
| 5112 | * movable at startup. This will force kernel allocations |
| 5113 | * to reserve their blocks rather than leaking throughout |
| 5114 | * the address space during boot when many long-lived |
Mel Gorman | 974a786 | 2015-11-06 16:28:34 -0800 | [diff] [blame] | 5115 | * kernel allocations are made. |
Mel Gorman | ac5d253 | 2015-06-30 14:57:20 -0700 | [diff] [blame] | 5116 | * |
| 5117 | * bitmap is created for zone's valid pfn range. but memmap |
| 5118 | * can be created for invalid pages (for alignment) |
| 5119 | * check here not to call set_pageblock_migratetype() against |
| 5120 | * pfn out of zone. |
| 5121 | */ |
| 5122 | if (!(pfn & (pageblock_nr_pages - 1))) { |
| 5123 | struct page *page = pfn_to_page(pfn); |
| 5124 | |
| 5125 | __init_single_page(page, pfn, zone, nid); |
| 5126 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
| 5127 | } else { |
| 5128 | __init_single_pfn(pfn, zone, nid); |
| 5129 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5130 | } |
| 5131 | } |
| 5132 | |
Andi Kleen | 1e548de | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 5133 | static void __meminit zone_init_free_lists(struct zone *zone) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5134 | { |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 5135 | unsigned int order, t; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 5136 | for_each_migratetype_order(order, t) { |
| 5137 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5138 | zone->free_area[order].nr_free = 0; |
| 5139 | } |
| 5140 | } |
| 5141 | |
| 5142 | #ifndef __HAVE_ARCH_MEMMAP_INIT |
| 5143 | #define memmap_init(size, nid, zone, start_pfn) \ |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 5144 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5145 | #endif |
| 5146 | |
David Rientjes | 7cd2b0a | 2014-06-23 13:22:04 -0700 | [diff] [blame] | 5147 | static int zone_batchsize(struct zone *zone) |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5148 | { |
David Howells | 3a6be87 | 2009-05-06 16:03:03 -0700 | [diff] [blame] | 5149 | #ifdef CONFIG_MMU |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5150 | int batch; |
| 5151 | |
| 5152 | /* |
| 5153 | * The per-cpu-pages pools are set to around 1000th of the |
Seth, Rohit | ba56e91 | 2005-10-29 18:15:47 -0700 | [diff] [blame] | 5154 | * 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] | 5155 | * |
| 5156 | * OK, so we don't know how big the cache is. So guess. |
| 5157 | */ |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 5158 | batch = zone->managed_pages / 1024; |
Seth, Rohit | ba56e91 | 2005-10-29 18:15:47 -0700 | [diff] [blame] | 5159 | if (batch * PAGE_SIZE > 512 * 1024) |
| 5160 | batch = (512 * 1024) / PAGE_SIZE; |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5161 | batch /= 4; /* We effectively *= 4 below */ |
| 5162 | if (batch < 1) |
| 5163 | batch = 1; |
| 5164 | |
| 5165 | /* |
Nick Piggin | 0ceaacc | 2005-12-04 13:55:25 +1100 | [diff] [blame] | 5166 | * Clamp the batch to a 2^n - 1 value. Having a power |
| 5167 | * of 2 value was found to be more likely to have |
| 5168 | * suboptimal cache aliasing properties in some cases. |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5169 | * |
Nick Piggin | 0ceaacc | 2005-12-04 13:55:25 +1100 | [diff] [blame] | 5170 | * For example if 2 tasks are alternately allocating |
| 5171 | * batches of pages, one task can end up with a lot |
| 5172 | * of pages of one half of the possible page colors |
| 5173 | * and the other with pages of the other colors. |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5174 | */ |
David Howells | 9155203 | 2009-05-06 16:03:02 -0700 | [diff] [blame] | 5175 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
Seth, Rohit | ba56e91 | 2005-10-29 18:15:47 -0700 | [diff] [blame] | 5176 | |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5177 | return batch; |
David Howells | 3a6be87 | 2009-05-06 16:03:03 -0700 | [diff] [blame] | 5178 | |
| 5179 | #else |
| 5180 | /* The deferral and batching of frees should be suppressed under NOMMU |
| 5181 | * conditions. |
| 5182 | * |
| 5183 | * The problem is that NOMMU needs to be able to allocate large chunks |
| 5184 | * of contiguous memory as there's no hardware page translation to |
| 5185 | * assemble apparent contiguous memory from discontiguous pages. |
| 5186 | * |
| 5187 | * Queueing large contiguous runs of pages for batching, however, |
| 5188 | * causes the pages to actually be freed in smaller chunks. As there |
| 5189 | * can be a significant delay between the individual batches being |
| 5190 | * recycled, this leads to the once large chunks of space being |
| 5191 | * fragmented and becoming unavailable for high-order allocations. |
| 5192 | */ |
| 5193 | return 0; |
| 5194 | #endif |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5195 | } |
| 5196 | |
Cody P Schafer | 8d7a8fa | 2013-07-03 15:01:31 -0700 | [diff] [blame] | 5197 | /* |
| 5198 | * pcp->high and pcp->batch values are related and dependent on one another: |
| 5199 | * ->batch must never be higher then ->high. |
| 5200 | * The following function updates them in a safe manner without read side |
| 5201 | * locking. |
| 5202 | * |
| 5203 | * Any new users of pcp->batch and pcp->high should ensure they can cope with |
| 5204 | * those fields changing asynchronously (acording the the above rule). |
| 5205 | * |
| 5206 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function |
| 5207 | * outside of boot time (or some other assurance that no concurrent updaters |
| 5208 | * exist). |
| 5209 | */ |
| 5210 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, |
| 5211 | unsigned long batch) |
| 5212 | { |
| 5213 | /* start with a fail safe value for batch */ |
| 5214 | pcp->batch = 1; |
| 5215 | smp_wmb(); |
| 5216 | |
| 5217 | /* Update high, then batch, in order */ |
| 5218 | pcp->high = high; |
| 5219 | smp_wmb(); |
| 5220 | |
| 5221 | pcp->batch = batch; |
| 5222 | } |
| 5223 | |
Cody P Schafer | 3664033 | 2013-07-03 15:01:40 -0700 | [diff] [blame] | 5224 | /* a companion to pageset_set_high() */ |
Cody P Schafer | 4008bab | 2013-07-03 15:01:28 -0700 | [diff] [blame] | 5225 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
| 5226 | { |
Cody P Schafer | 8d7a8fa | 2013-07-03 15:01:31 -0700 | [diff] [blame] | 5227 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
Cody P Schafer | 4008bab | 2013-07-03 15:01:28 -0700 | [diff] [blame] | 5228 | } |
| 5229 | |
Cody P Schafer | 88c90db | 2013-07-03 15:01:35 -0700 | [diff] [blame] | 5230 | static void pageset_init(struct per_cpu_pageset *p) |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 5231 | { |
| 5232 | struct per_cpu_pages *pcp; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 5233 | int migratetype; |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 5234 | |
Magnus Damm | 1c6fe94 | 2005-10-26 01:58:59 -0700 | [diff] [blame] | 5235 | memset(p, 0, sizeof(*p)); |
| 5236 | |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 5237 | pcp = &p->pcp; |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 5238 | pcp->count = 0; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 5239 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
| 5240 | INIT_LIST_HEAD(&pcp->lists[migratetype]); |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 5241 | } |
| 5242 | |
Cody P Schafer | 88c90db | 2013-07-03 15:01:35 -0700 | [diff] [blame] | 5243 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
| 5244 | { |
| 5245 | pageset_init(p); |
| 5246 | pageset_set_batch(p, batch); |
| 5247 | } |
| 5248 | |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5249 | /* |
Cody P Schafer | 3664033 | 2013-07-03 15:01:40 -0700 | [diff] [blame] | 5250 | * 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] | 5251 | * to the value high for the pageset p. |
| 5252 | */ |
Cody P Schafer | 3664033 | 2013-07-03 15:01:40 -0700 | [diff] [blame] | 5253 | static void pageset_set_high(struct per_cpu_pageset *p, |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5254 | unsigned long high) |
| 5255 | { |
Cody P Schafer | 8d7a8fa | 2013-07-03 15:01:31 -0700 | [diff] [blame] | 5256 | unsigned long batch = max(1UL, high / 4); |
| 5257 | if ((high / 4) > (PAGE_SHIFT * 8)) |
| 5258 | batch = PAGE_SHIFT * 8; |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5259 | |
Cody P Schafer | 8d7a8fa | 2013-07-03 15:01:31 -0700 | [diff] [blame] | 5260 | pageset_update(&p->pcp, high, batch); |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5261 | } |
| 5262 | |
David Rientjes | 7cd2b0a | 2014-06-23 13:22:04 -0700 | [diff] [blame] | 5263 | static void pageset_set_high_and_batch(struct zone *zone, |
| 5264 | struct per_cpu_pageset *pcp) |
Cody P Schafer | 56cef2b | 2013-07-03 15:01:38 -0700 | [diff] [blame] | 5265 | { |
Cody P Schafer | 56cef2b | 2013-07-03 15:01:38 -0700 | [diff] [blame] | 5266 | if (percpu_pagelist_fraction) |
Cody P Schafer | 3664033 | 2013-07-03 15:01:40 -0700 | [diff] [blame] | 5267 | pageset_set_high(pcp, |
Cody P Schafer | 56cef2b | 2013-07-03 15:01:38 -0700 | [diff] [blame] | 5268 | (zone->managed_pages / |
| 5269 | percpu_pagelist_fraction)); |
| 5270 | else |
| 5271 | pageset_set_batch(pcp, zone_batchsize(zone)); |
| 5272 | } |
| 5273 | |
Cody P Schafer | 169f6c1 | 2013-07-03 15:01:41 -0700 | [diff] [blame] | 5274 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
| 5275 | { |
| 5276 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); |
| 5277 | |
| 5278 | pageset_init(pcp); |
| 5279 | pageset_set_high_and_batch(zone, pcp); |
| 5280 | } |
| 5281 | |
Jiang Liu | 4ed7e02 | 2012-07-31 16:43:35 -0700 | [diff] [blame] | 5282 | static void __meminit setup_zone_pageset(struct zone *zone) |
Wu Fengguang | 319774e | 2010-05-24 14:32:49 -0700 | [diff] [blame] | 5283 | { |
| 5284 | int cpu; |
Wu Fengguang | 319774e | 2010-05-24 14:32:49 -0700 | [diff] [blame] | 5285 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
Cody P Schafer | 56cef2b | 2013-07-03 15:01:38 -0700 | [diff] [blame] | 5286 | for_each_possible_cpu(cpu) |
| 5287 | zone_pageset_init(zone, cpu); |
Wu Fengguang | 319774e | 2010-05-24 14:32:49 -0700 | [diff] [blame] | 5288 | } |
| 5289 | |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5290 | /* |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 5291 | * Allocate per cpu pagesets and initialize them. |
| 5292 | * Before this call only boot pagesets were available. |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 5293 | */ |
Al Viro | 78d9955 | 2005-12-15 09:18:25 +0000 | [diff] [blame] | 5294 | void __init setup_per_cpu_pageset(void) |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5295 | { |
Mel Gorman | b4911ea | 2016-08-04 15:31:49 -0700 | [diff] [blame] | 5296 | struct pglist_data *pgdat; |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 5297 | struct zone *zone; |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5298 | |
Wu Fengguang | 319774e | 2010-05-24 14:32:49 -0700 | [diff] [blame] | 5299 | for_each_populated_zone(zone) |
| 5300 | setup_zone_pageset(zone); |
Mel Gorman | b4911ea | 2016-08-04 15:31:49 -0700 | [diff] [blame] | 5301 | |
| 5302 | for_each_online_pgdat(pgdat) |
| 5303 | pgdat->per_cpu_nodestats = |
| 5304 | alloc_percpu(struct per_cpu_nodestat); |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 5305 | } |
| 5306 | |
Fabian Frederick | bd721ea | 2016-08-02 14:03:33 -0700 | [diff] [blame] | 5307 | static noinline __ref |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5308 | 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] | 5309 | { |
| 5310 | int i; |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5311 | size_t alloc_size; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5312 | |
| 5313 | /* |
| 5314 | * The per-page waitqueue mechanism uses hashed waitqueues |
| 5315 | * per zone. |
| 5316 | */ |
Yasunori Goto | 02b694d | 2006-06-23 02:03:08 -0700 | [diff] [blame] | 5317 | zone->wait_table_hash_nr_entries = |
| 5318 | wait_table_hash_nr_entries(zone_size_pages); |
| 5319 | zone->wait_table_bits = |
| 5320 | wait_table_bits(zone->wait_table_hash_nr_entries); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5321 | alloc_size = zone->wait_table_hash_nr_entries |
| 5322 | * sizeof(wait_queue_head_t); |
| 5323 | |
Heiko Carstens | cd94b9d | 2008-05-23 13:04:52 -0700 | [diff] [blame] | 5324 | if (!slab_is_available()) { |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5325 | zone->wait_table = (wait_queue_head_t *) |
Santosh Shilimkar | 6782832 | 2014-01-21 15:50:25 -0800 | [diff] [blame] | 5326 | memblock_virt_alloc_node_nopanic( |
| 5327 | alloc_size, zone->zone_pgdat->node_id); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5328 | } else { |
| 5329 | /* |
| 5330 | * This case means that a zone whose size was 0 gets new memory |
| 5331 | * via memory hot-add. |
| 5332 | * But it may be the case that a new node was hot-added. In |
| 5333 | * this case vmalloc() will not be able to use this new node's |
| 5334 | * memory - this wait_table must be initialized to use this new |
| 5335 | * node itself as well. |
| 5336 | * To use this new node's memory, further consideration will be |
| 5337 | * necessary. |
| 5338 | */ |
Jesper Juhl | 8691f3a | 2007-10-16 01:24:49 -0700 | [diff] [blame] | 5339 | zone->wait_table = vmalloc(alloc_size); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5340 | } |
| 5341 | if (!zone->wait_table) |
| 5342 | return -ENOMEM; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5343 | |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 5344 | for (i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5345 | init_waitqueue_head(zone->wait_table + i); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5346 | |
| 5347 | return 0; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5348 | } |
| 5349 | |
Matt Tolentino | c09b424 | 2006-01-17 07:03:44 +0100 | [diff] [blame] | 5350 | static __meminit void zone_pcp_init(struct zone *zone) |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5351 | { |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 5352 | /* |
| 5353 | * per cpu subsystem is not up at this point. The following code |
| 5354 | * relies on the ability of the linker to provide the |
| 5355 | * offset of a (static) per cpu variable into the per cpu area. |
| 5356 | */ |
| 5357 | zone->pageset = &boot_pageset; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5358 | |
Xishi Qiu | b38a872 | 2013-11-12 15:07:20 -0800 | [diff] [blame] | 5359 | if (populated_zone(zone)) |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 5360 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
| 5361 | zone->name, zone->present_pages, |
| 5362 | zone_batchsize(zone)); |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5363 | } |
| 5364 | |
Jiang Liu | 4ed7e02 | 2012-07-31 16:43:35 -0700 | [diff] [blame] | 5365 | int __meminit init_currently_empty_zone(struct zone *zone, |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5366 | unsigned long zone_start_pfn, |
Yaowei Bai | b171e40 | 2015-11-05 18:47:06 -0800 | [diff] [blame] | 5367 | unsigned long size) |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5368 | { |
| 5369 | struct pglist_data *pgdat = zone->zone_pgdat; |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5370 | int ret; |
| 5371 | ret = zone_wait_table_init(zone, size); |
| 5372 | if (ret) |
| 5373 | return ret; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5374 | pgdat->nr_zones = zone_idx(zone) + 1; |
| 5375 | |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5376 | zone->zone_start_pfn = zone_start_pfn; |
| 5377 | |
Mel Gorman | 708614e | 2008-07-23 21:26:51 -0700 | [diff] [blame] | 5378 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
| 5379 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", |
| 5380 | pgdat->node_id, |
| 5381 | (unsigned long)zone_idx(zone), |
| 5382 | zone_start_pfn, (zone_start_pfn + size)); |
| 5383 | |
Andi Kleen | 1e548de | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 5384 | zone_init_free_lists(zone); |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5385 | |
| 5386 | return 0; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5387 | } |
| 5388 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 5389 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5390 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
Mel Gorman | 8a942fd | 2015-06-30 14:56:55 -0700 | [diff] [blame] | 5391 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5392 | /* |
| 5393 | * 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] | 5394 | */ |
Mel Gorman | 8a942fd | 2015-06-30 14:56:55 -0700 | [diff] [blame] | 5395 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
| 5396 | struct mminit_pfnnid_cache *state) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5397 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5398 | unsigned long start_pfn, end_pfn; |
Yinghai Lu | e76b63f | 2013-09-11 14:22:17 -0700 | [diff] [blame] | 5399 | int nid; |
Russ Anderson | 7c243c7 | 2013-04-29 15:07:59 -0700 | [diff] [blame] | 5400 | |
Mel Gorman | 8a942fd | 2015-06-30 14:56:55 -0700 | [diff] [blame] | 5401 | if (state->last_start <= pfn && pfn < state->last_end) |
| 5402 | return state->last_nid; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5403 | |
Yinghai Lu | e76b63f | 2013-09-11 14:22:17 -0700 | [diff] [blame] | 5404 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
| 5405 | if (nid != -1) { |
Mel Gorman | 8a942fd | 2015-06-30 14:56:55 -0700 | [diff] [blame] | 5406 | state->last_start = start_pfn; |
| 5407 | state->last_end = end_pfn; |
| 5408 | state->last_nid = nid; |
Yinghai Lu | e76b63f | 2013-09-11 14:22:17 -0700 | [diff] [blame] | 5409 | } |
| 5410 | |
| 5411 | return nid; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5412 | } |
| 5413 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ |
| 5414 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5415 | /** |
Santosh Shilimkar | 6782832 | 2014-01-21 15:50:25 -0800 | [diff] [blame] | 5416 | * 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] | 5417 | * @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] | 5418 | * @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] | 5419 | * |
Zhang Zhen | 7d01817 | 2014-06-04 16:10:53 -0700 | [diff] [blame] | 5420 | * If an architecture guarantees that all ranges registered contain no holes |
| 5421 | * and may be freed, this this function may be used instead of calling |
| 5422 | * memblock_free_early_nid() manually. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5423 | */ |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5424 | 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] | 5425 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5426 | unsigned long start_pfn, end_pfn; |
| 5427 | int i, this_nid; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5428 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5429 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
| 5430 | start_pfn = min(start_pfn, max_low_pfn); |
| 5431 | end_pfn = min(end_pfn, max_low_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5432 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5433 | if (start_pfn < end_pfn) |
Santosh Shilimkar | 6782832 | 2014-01-21 15:50:25 -0800 | [diff] [blame] | 5434 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
| 5435 | (end_pfn - start_pfn) << PAGE_SHIFT, |
| 5436 | this_nid); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5437 | } |
| 5438 | } |
| 5439 | |
| 5440 | /** |
| 5441 | * 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] | 5442 | * @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] | 5443 | * |
Zhang Zhen | 7d01817 | 2014-06-04 16:10:53 -0700 | [diff] [blame] | 5444 | * If an architecture guarantees that all ranges registered contain no holes and may |
| 5445 | * 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] | 5446 | */ |
| 5447 | void __init sparse_memory_present_with_active_regions(int nid) |
| 5448 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5449 | unsigned long start_pfn, end_pfn; |
| 5450 | int i, this_nid; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5451 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5452 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
| 5453 | memory_present(this_nid, start_pfn, end_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5454 | } |
| 5455 | |
| 5456 | /** |
| 5457 | * 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] | 5458 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
| 5459 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. |
| 5460 | * @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] | 5461 | * |
| 5462 | * 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] | 5463 | * provided by memblock_set_node(). If called for a node |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5464 | * 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] | 5465 | * PFNs will be 0. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5466 | */ |
Yasunori Goto | a3142c8 | 2007-05-08 00:23:07 -0700 | [diff] [blame] | 5467 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5468 | unsigned long *start_pfn, unsigned long *end_pfn) |
| 5469 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5470 | unsigned long this_start_pfn, this_end_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5471 | int i; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5472 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5473 | *start_pfn = -1UL; |
| 5474 | *end_pfn = 0; |
| 5475 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 5476 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
| 5477 | *start_pfn = min(*start_pfn, this_start_pfn); |
| 5478 | *end_pfn = max(*end_pfn, this_end_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5479 | } |
| 5480 | |
Christoph Lameter | 633c066 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 5481 | if (*start_pfn == -1UL) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5482 | *start_pfn = 0; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5483 | } |
| 5484 | |
| 5485 | /* |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 5486 | * This finds a zone that can be used for ZONE_MOVABLE pages. The |
| 5487 | * assumption is made that zones within a node are ordered in monotonic |
| 5488 | * increasing memory addresses so that the "highest" populated zone is used |
| 5489 | */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 5490 | static void __init find_usable_zone_for_movable(void) |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 5491 | { |
| 5492 | int zone_index; |
| 5493 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { |
| 5494 | if (zone_index == ZONE_MOVABLE) |
| 5495 | continue; |
| 5496 | |
| 5497 | if (arch_zone_highest_possible_pfn[zone_index] > |
| 5498 | arch_zone_lowest_possible_pfn[zone_index]) |
| 5499 | break; |
| 5500 | } |
| 5501 | |
| 5502 | VM_BUG_ON(zone_index == -1); |
| 5503 | movable_zone = zone_index; |
| 5504 | } |
| 5505 | |
| 5506 | /* |
| 5507 | * 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] | 5508 | * because it is sized independent of architecture. Unlike the other zones, |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 5509 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
| 5510 | * in each node depending on the size of each node and how evenly kernelcore |
| 5511 | * is distributed. This helper function adjusts the zone ranges |
| 5512 | * provided by the architecture for a given node by using the end of the |
| 5513 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that |
| 5514 | * zones within a node are in order of monotonic increases memory addresses |
| 5515 | */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 5516 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 5517 | unsigned long zone_type, |
| 5518 | unsigned long node_start_pfn, |
| 5519 | unsigned long node_end_pfn, |
| 5520 | unsigned long *zone_start_pfn, |
| 5521 | unsigned long *zone_end_pfn) |
| 5522 | { |
| 5523 | /* Only adjust if ZONE_MOVABLE is on this node */ |
| 5524 | if (zone_movable_pfn[nid]) { |
| 5525 | /* Size ZONE_MOVABLE */ |
| 5526 | if (zone_type == ZONE_MOVABLE) { |
| 5527 | *zone_start_pfn = zone_movable_pfn[nid]; |
| 5528 | *zone_end_pfn = min(node_end_pfn, |
| 5529 | arch_zone_highest_possible_pfn[movable_zone]); |
| 5530 | |
Xishi Qiu | e506b99 | 2016-10-07 16:58:06 -0700 | [diff] [blame] | 5531 | /* Adjust for ZONE_MOVABLE starting within this range */ |
| 5532 | } else if (!mirrored_kernelcore && |
| 5533 | *zone_start_pfn < zone_movable_pfn[nid] && |
| 5534 | *zone_end_pfn > zone_movable_pfn[nid]) { |
| 5535 | *zone_end_pfn = zone_movable_pfn[nid]; |
| 5536 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 5537 | /* Check if this whole range is within ZONE_MOVABLE */ |
| 5538 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) |
| 5539 | *zone_start_pfn = *zone_end_pfn; |
| 5540 | } |
| 5541 | } |
| 5542 | |
| 5543 | /* |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5544 | * Return the number of pages a zone spans in a node, including holes |
| 5545 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() |
| 5546 | */ |
Paul Mundt | 6ea6e68 | 2007-07-15 23:38:20 -0700 | [diff] [blame] | 5547 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5548 | unsigned long zone_type, |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5549 | unsigned long node_start_pfn, |
| 5550 | unsigned long node_end_pfn, |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5551 | unsigned long *zone_start_pfn, |
| 5552 | unsigned long *zone_end_pfn, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5553 | unsigned long *ignored) |
| 5554 | { |
Xishi Qiu | b5685e9 | 2015-09-08 15:04:16 -0700 | [diff] [blame] | 5555 | /* 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] | 5556 | if (!node_start_pfn && !node_end_pfn) |
| 5557 | return 0; |
| 5558 | |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5559 | /* Get the start and end of the zone */ |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5560 | *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
| 5561 | *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 5562 | adjust_zone_range_for_zone_movable(nid, zone_type, |
| 5563 | node_start_pfn, node_end_pfn, |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5564 | zone_start_pfn, zone_end_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5565 | |
| 5566 | /* Check that this node has pages within the zone's required range */ |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5567 | 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] | 5568 | return 0; |
| 5569 | |
| 5570 | /* Move the zone boundaries inside the node if necessary */ |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5571 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
| 5572 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5573 | |
| 5574 | /* Return the spanned pages */ |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5575 | return *zone_end_pfn - *zone_start_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5576 | } |
| 5577 | |
| 5578 | /* |
| 5579 | * 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] | 5580 | * then all holes in the requested range will be accounted for. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5581 | */ |
Yinghai Lu | 3299625 | 2009-12-15 17:59:02 -0800 | [diff] [blame] | 5582 | unsigned long __meminit __absent_pages_in_range(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5583 | unsigned long range_start_pfn, |
| 5584 | unsigned long range_end_pfn) |
| 5585 | { |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 5586 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
| 5587 | unsigned long start_pfn, end_pfn; |
| 5588 | int i; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5589 | |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 5590 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
| 5591 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); |
| 5592 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); |
| 5593 | nr_absent -= end_pfn - start_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5594 | } |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 5595 | return nr_absent; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5596 | } |
| 5597 | |
| 5598 | /** |
| 5599 | * absent_pages_in_range - Return number of page frames in holes within a range |
| 5600 | * @start_pfn: The start PFN to start searching for holes |
| 5601 | * @end_pfn: The end PFN to stop searching for holes |
| 5602 | * |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 5603 | * 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] | 5604 | */ |
| 5605 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, |
| 5606 | unsigned long end_pfn) |
| 5607 | { |
| 5608 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); |
| 5609 | } |
| 5610 | |
| 5611 | /* 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] | 5612 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5613 | unsigned long zone_type, |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5614 | unsigned long node_start_pfn, |
| 5615 | unsigned long node_end_pfn, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5616 | unsigned long *ignored) |
| 5617 | { |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 5618 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
| 5619 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; |
Mel Gorman | 9c7cd68 | 2006-09-27 01:49:58 -0700 | [diff] [blame] | 5620 | unsigned long zone_start_pfn, zone_end_pfn; |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5621 | unsigned long nr_absent; |
Mel Gorman | 9c7cd68 | 2006-09-27 01:49:58 -0700 | [diff] [blame] | 5622 | |
Xishi Qiu | b5685e9 | 2015-09-08 15:04:16 -0700 | [diff] [blame] | 5623 | /* 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] | 5624 | if (!node_start_pfn && !node_end_pfn) |
| 5625 | return 0; |
| 5626 | |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 5627 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
| 5628 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); |
Mel Gorman | 9c7cd68 | 2006-09-27 01:49:58 -0700 | [diff] [blame] | 5629 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 5630 | adjust_zone_range_for_zone_movable(nid, zone_type, |
| 5631 | node_start_pfn, node_end_pfn, |
| 5632 | &zone_start_pfn, &zone_end_pfn); |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5633 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
| 5634 | |
| 5635 | /* |
| 5636 | * ZONE_MOVABLE handling. |
| 5637 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages |
| 5638 | * and vice versa. |
| 5639 | */ |
Xishi Qiu | e506b99 | 2016-10-07 16:58:06 -0700 | [diff] [blame] | 5640 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
| 5641 | unsigned long start_pfn, end_pfn; |
| 5642 | struct memblock_region *r; |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5643 | |
Xishi Qiu | e506b99 | 2016-10-07 16:58:06 -0700 | [diff] [blame] | 5644 | for_each_memblock(memory, r) { |
| 5645 | start_pfn = clamp(memblock_region_memory_base_pfn(r), |
| 5646 | zone_start_pfn, zone_end_pfn); |
| 5647 | end_pfn = clamp(memblock_region_memory_end_pfn(r), |
| 5648 | zone_start_pfn, zone_end_pfn); |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5649 | |
Xishi Qiu | e506b99 | 2016-10-07 16:58:06 -0700 | [diff] [blame] | 5650 | if (zone_type == ZONE_MOVABLE && |
| 5651 | memblock_is_mirror(r)) |
| 5652 | nr_absent += end_pfn - start_pfn; |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5653 | |
Xishi Qiu | e506b99 | 2016-10-07 16:58:06 -0700 | [diff] [blame] | 5654 | if (zone_type == ZONE_NORMAL && |
| 5655 | !memblock_is_mirror(r)) |
| 5656 | nr_absent += end_pfn - start_pfn; |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 5657 | } |
| 5658 | } |
| 5659 | |
| 5660 | return nr_absent; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5661 | } |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 5662 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 5663 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Paul Mundt | 6ea6e68 | 2007-07-15 23:38:20 -0700 | [diff] [blame] | 5664 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5665 | unsigned long zone_type, |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5666 | unsigned long node_start_pfn, |
| 5667 | unsigned long node_end_pfn, |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5668 | unsigned long *zone_start_pfn, |
| 5669 | unsigned long *zone_end_pfn, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5670 | unsigned long *zones_size) |
| 5671 | { |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5672 | unsigned int zone; |
| 5673 | |
| 5674 | *zone_start_pfn = node_start_pfn; |
| 5675 | for (zone = 0; zone < zone_type; zone++) |
| 5676 | *zone_start_pfn += zones_size[zone]; |
| 5677 | |
| 5678 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; |
| 5679 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5680 | return zones_size[zone_type]; |
| 5681 | } |
| 5682 | |
Paul Mundt | 6ea6e68 | 2007-07-15 23:38:20 -0700 | [diff] [blame] | 5683 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5684 | unsigned long zone_type, |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5685 | unsigned long node_start_pfn, |
| 5686 | unsigned long node_end_pfn, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5687 | unsigned long *zholes_size) |
| 5688 | { |
| 5689 | if (!zholes_size) |
| 5690 | return 0; |
| 5691 | |
| 5692 | return zholes_size[zone_type]; |
| 5693 | } |
Yinghai Lu | 20e6926 | 2013-03-01 14:51:27 -0800 | [diff] [blame] | 5694 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 5695 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5696 | |
Yasunori Goto | a3142c8 | 2007-05-08 00:23:07 -0700 | [diff] [blame] | 5697 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5698 | unsigned long node_start_pfn, |
| 5699 | unsigned long node_end_pfn, |
| 5700 | unsigned long *zones_size, |
| 5701 | unsigned long *zholes_size) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5702 | { |
Gu Zheng | febd594 | 2015-06-24 16:57:02 -0700 | [diff] [blame] | 5703 | unsigned long realtotalpages = 0, totalpages = 0; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5704 | enum zone_type i; |
| 5705 | |
Gu Zheng | febd594 | 2015-06-24 16:57:02 -0700 | [diff] [blame] | 5706 | for (i = 0; i < MAX_NR_ZONES; i++) { |
| 5707 | struct zone *zone = pgdat->node_zones + i; |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5708 | unsigned long zone_start_pfn, zone_end_pfn; |
Gu Zheng | febd594 | 2015-06-24 16:57:02 -0700 | [diff] [blame] | 5709 | unsigned long size, real_size; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5710 | |
Gu Zheng | febd594 | 2015-06-24 16:57:02 -0700 | [diff] [blame] | 5711 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
| 5712 | node_start_pfn, |
| 5713 | node_end_pfn, |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5714 | &zone_start_pfn, |
| 5715 | &zone_end_pfn, |
Gu Zheng | febd594 | 2015-06-24 16:57:02 -0700 | [diff] [blame] | 5716 | zones_size); |
| 5717 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5718 | node_start_pfn, node_end_pfn, |
| 5719 | zholes_size); |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5720 | if (size) |
| 5721 | zone->zone_start_pfn = zone_start_pfn; |
| 5722 | else |
| 5723 | zone->zone_start_pfn = 0; |
Gu Zheng | febd594 | 2015-06-24 16:57:02 -0700 | [diff] [blame] | 5724 | zone->spanned_pages = size; |
| 5725 | zone->present_pages = real_size; |
| 5726 | |
| 5727 | totalpages += size; |
| 5728 | realtotalpages += real_size; |
| 5729 | } |
| 5730 | |
| 5731 | pgdat->node_spanned_pages = totalpages; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5732 | pgdat->node_present_pages = realtotalpages; |
| 5733 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, |
| 5734 | realtotalpages); |
| 5735 | } |
| 5736 | |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5737 | #ifndef CONFIG_SPARSEMEM |
| 5738 | /* |
| 5739 | * Calculate the size of the zone->blockflags rounded to an unsigned long |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5740 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
| 5741 | * 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] | 5742 | * round what is now in bits to nearest long in bits, then return it in |
| 5743 | * bytes. |
| 5744 | */ |
Linus Torvalds | 7c45512 | 2013-02-18 09:58:02 -0800 | [diff] [blame] | 5745 | 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] | 5746 | { |
| 5747 | unsigned long usemapsize; |
| 5748 | |
Linus Torvalds | 7c45512 | 2013-02-18 09:58:02 -0800 | [diff] [blame] | 5749 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5750 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
| 5751 | usemapsize = usemapsize >> pageblock_order; |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5752 | usemapsize *= NR_PAGEBLOCK_BITS; |
| 5753 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); |
| 5754 | |
| 5755 | return usemapsize / 8; |
| 5756 | } |
| 5757 | |
| 5758 | static void __init setup_usemap(struct pglist_data *pgdat, |
Linus Torvalds | 7c45512 | 2013-02-18 09:58:02 -0800 | [diff] [blame] | 5759 | struct zone *zone, |
| 5760 | unsigned long zone_start_pfn, |
| 5761 | unsigned long zonesize) |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5762 | { |
Linus Torvalds | 7c45512 | 2013-02-18 09:58:02 -0800 | [diff] [blame] | 5763 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5764 | zone->pageblock_flags = NULL; |
Julia Lawall | 58a01a4 | 2009-01-06 14:39:28 -0800 | [diff] [blame] | 5765 | if (usemapsize) |
Santosh Shilimkar | 6782832 | 2014-01-21 15:50:25 -0800 | [diff] [blame] | 5766 | zone->pageblock_flags = |
| 5767 | memblock_virt_alloc_node_nopanic(usemapsize, |
| 5768 | pgdat->node_id); |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5769 | } |
| 5770 | #else |
Linus Torvalds | 7c45512 | 2013-02-18 09:58:02 -0800 | [diff] [blame] | 5771 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
| 5772 | unsigned long zone_start_pfn, unsigned long zonesize) {} |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5773 | #endif /* CONFIG_SPARSEMEM */ |
| 5774 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5775 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 5776 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5777 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
Chen Gang | 15ca220 | 2013-09-11 14:20:27 -0700 | [diff] [blame] | 5778 | void __paginginit set_pageblock_order(void) |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5779 | { |
Andrew Morton | 955c1cd | 2012-05-29 15:06:31 -0700 | [diff] [blame] | 5780 | unsigned int order; |
| 5781 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5782 | /* Check that pageblock_nr_pages has not already been setup */ |
| 5783 | if (pageblock_order) |
| 5784 | return; |
| 5785 | |
Andrew Morton | 955c1cd | 2012-05-29 15:06:31 -0700 | [diff] [blame] | 5786 | if (HPAGE_SHIFT > PAGE_SHIFT) |
| 5787 | order = HUGETLB_PAGE_ORDER; |
| 5788 | else |
| 5789 | order = MAX_ORDER - 1; |
| 5790 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5791 | /* |
| 5792 | * Assume the largest contiguous order of interest is a huge page. |
Andrew Morton | 955c1cd | 2012-05-29 15:06:31 -0700 | [diff] [blame] | 5793 | * This value may be variable depending on boot parameters on IA64 and |
| 5794 | * powerpc. |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5795 | */ |
| 5796 | pageblock_order = order; |
| 5797 | } |
| 5798 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ |
| 5799 | |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 5800 | /* |
| 5801 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() |
Andrew Morton | 955c1cd | 2012-05-29 15:06:31 -0700 | [diff] [blame] | 5802 | * is unused as pageblock_order is set at compile-time. See |
| 5803 | * include/linux/pageblock-flags.h for the values of pageblock_order based on |
| 5804 | * the kernel config |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 5805 | */ |
Chen Gang | 15ca220 | 2013-09-11 14:20:27 -0700 | [diff] [blame] | 5806 | void __paginginit set_pageblock_order(void) |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 5807 | { |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 5808 | } |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5809 | |
| 5810 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ |
| 5811 | |
Jiang Liu | 01cefae | 2012-12-12 13:52:19 -0800 | [diff] [blame] | 5812 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
| 5813 | unsigned long present_pages) |
| 5814 | { |
| 5815 | unsigned long pages = spanned_pages; |
| 5816 | |
| 5817 | /* |
| 5818 | * Provide a more accurate estimation if there are holes within |
| 5819 | * the zone and SPARSEMEM is in use. If there are holes within the |
| 5820 | * zone, each populated memory region may cost us one or two extra |
| 5821 | * memmap pages due to alignment because memmap pages for each |
| 5822 | * populated regions may not naturally algined on page boundary. |
| 5823 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
| 5824 | */ |
| 5825 | if (spanned_pages > present_pages + (present_pages >> 4) && |
| 5826 | IS_ENABLED(CONFIG_SPARSEMEM)) |
| 5827 | pages = present_pages; |
| 5828 | |
| 5829 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; |
| 5830 | } |
| 5831 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5832 | /* |
| 5833 | * Set up the zone data structures: |
| 5834 | * - mark all pages reserved |
| 5835 | * - mark all memory queues empty |
| 5836 | * - clear the memory bitmaps |
Minchan Kim | 6527af5 | 2012-07-31 16:46:16 -0700 | [diff] [blame] | 5837 | * |
| 5838 | * NOTE: pgdat should get zeroed by caller. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5839 | */ |
Wei Yang | 7f3eb55 | 2015-09-08 14:59:50 -0700 | [diff] [blame] | 5840 | static void __paginginit free_area_init_core(struct pglist_data *pgdat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5841 | { |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 5842 | enum zone_type j; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5843 | int nid = pgdat->node_id; |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5844 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5845 | |
Dave Hansen | 208d54e | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 5846 | pgdat_resize_init(pgdat); |
Andrea Arcangeli | 8177a42 | 2012-03-23 20:56:34 +0100 | [diff] [blame] | 5847 | #ifdef CONFIG_NUMA_BALANCING |
| 5848 | spin_lock_init(&pgdat->numabalancing_migrate_lock); |
| 5849 | pgdat->numabalancing_migrate_nr_pages = 0; |
| 5850 | pgdat->numabalancing_migrate_next_window = jiffies; |
| 5851 | #endif |
Kirill A. Shutemov | a3d0a918 | 2016-02-02 16:57:08 -0800 | [diff] [blame] | 5852 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 5853 | spin_lock_init(&pgdat->split_queue_lock); |
| 5854 | INIT_LIST_HEAD(&pgdat->split_queue); |
| 5855 | pgdat->split_queue_len = 0; |
| 5856 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5857 | init_waitqueue_head(&pgdat->kswapd_wait); |
Mel Gorman | 5515061 | 2012-07-31 16:44:35 -0700 | [diff] [blame] | 5858 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
Vlastimil Babka | 698b1b3 | 2016-03-17 14:18:08 -0700 | [diff] [blame] | 5859 | #ifdef CONFIG_COMPACTION |
| 5860 | init_waitqueue_head(&pgdat->kcompactd_wait); |
| 5861 | #endif |
Joonsoo Kim | eefa864b | 2014-12-12 16:55:46 -0800 | [diff] [blame] | 5862 | pgdat_page_ext_init(pgdat); |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame] | 5863 | spin_lock_init(&pgdat->lru_lock); |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 5864 | lruvec_init(node_lruvec(pgdat)); |
Michal Nazarewicz | 5f63b72 | 2012-01-11 15:16:11 +0100 | [diff] [blame] | 5865 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5866 | for (j = 0; j < MAX_NR_ZONES; j++) { |
| 5867 | struct zone *zone = pgdat->node_zones + j; |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 5868 | unsigned long size, realsize, freesize, memmap_pages; |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5869 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5870 | |
Gu Zheng | febd594 | 2015-06-24 16:57:02 -0700 | [diff] [blame] | 5871 | size = zone->spanned_pages; |
| 5872 | realsize = freesize = zone->present_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5873 | |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 5874 | /* |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 5875 | * Adjust freesize so that it accounts for how much memory |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 5876 | * is used by this zone for memmap. This affects the watermark |
| 5877 | * and per-cpu initialisations |
| 5878 | */ |
Jiang Liu | 01cefae | 2012-12-12 13:52:19 -0800 | [diff] [blame] | 5879 | memmap_pages = calc_memmap_size(size, realsize); |
Zhong Hongbo | ba914f4 | 2014-12-12 16:56:21 -0800 | [diff] [blame] | 5880 | if (!is_highmem_idx(j)) { |
| 5881 | if (freesize >= memmap_pages) { |
| 5882 | freesize -= memmap_pages; |
| 5883 | if (memmap_pages) |
| 5884 | printk(KERN_DEBUG |
| 5885 | " %s zone: %lu pages used for memmap\n", |
| 5886 | zone_names[j], memmap_pages); |
| 5887 | } else |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 5888 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
Zhong Hongbo | ba914f4 | 2014-12-12 16:56:21 -0800 | [diff] [blame] | 5889 | zone_names[j], memmap_pages, freesize); |
| 5890 | } |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 5891 | |
Christoph Lameter | 6267276 | 2007-02-10 01:43:07 -0800 | [diff] [blame] | 5892 | /* Account for reserved pages */ |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 5893 | if (j == 0 && freesize > dma_reserve) { |
| 5894 | freesize -= dma_reserve; |
Yinghai Lu | d903ef9 | 2008-10-18 20:27:06 -0700 | [diff] [blame] | 5895 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
Christoph Lameter | 6267276 | 2007-02-10 01:43:07 -0800 | [diff] [blame] | 5896 | zone_names[0], dma_reserve); |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 5897 | } |
| 5898 | |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 5899 | if (!is_highmem_idx(j)) |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 5900 | nr_kernel_pages += freesize; |
Jiang Liu | 01cefae | 2012-12-12 13:52:19 -0800 | [diff] [blame] | 5901 | /* Charge for highmem memmap if there are enough kernel pages */ |
| 5902 | else if (nr_kernel_pages > memmap_pages * 2) |
| 5903 | nr_kernel_pages -= memmap_pages; |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 5904 | nr_all_pages += freesize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5905 | |
Jiang Liu | 9feedc9 | 2012-12-12 13:52:12 -0800 | [diff] [blame] | 5906 | /* |
| 5907 | * Set an approximate value for lowmem here, it will be adjusted |
| 5908 | * when the bootmem allocator frees pages into the buddy system. |
| 5909 | * And all highmem pages will be managed by the buddy system. |
| 5910 | */ |
| 5911 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 5912 | #ifdef CONFIG_NUMA |
Christoph Lameter | d5f541e | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 5913 | zone->node = nid; |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 5914 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5915 | zone->name = zone_names[j]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5916 | zone->zone_pgdat = pgdat; |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame] | 5917 | spin_lock_init(&zone->lock); |
| 5918 | zone_seqlock_init(zone); |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 5919 | zone_pcp_init(zone); |
Johannes Weiner | 81c0a2b | 2013-09-11 14:20:47 -0700 | [diff] [blame] | 5920 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5921 | if (!size) |
| 5922 | continue; |
| 5923 | |
Andrew Morton | 955c1cd | 2012-05-29 15:06:31 -0700 | [diff] [blame] | 5924 | set_pageblock_order(); |
Linus Torvalds | 7c45512 | 2013-02-18 09:58:02 -0800 | [diff] [blame] | 5925 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
Yaowei Bai | b171e40 | 2015-11-05 18:47:06 -0800 | [diff] [blame] | 5926 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 5927 | BUG_ON(ret); |
Heiko Carstens | 76cdd58 | 2008-05-14 16:05:52 -0700 | [diff] [blame] | 5928 | memmap_init(size, nid, j, zone_start_pfn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5929 | } |
| 5930 | } |
| 5931 | |
Fabian Frederick | bd721ea | 2016-08-02 14:03:33 -0700 | [diff] [blame] | 5932 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5933 | { |
Tony Luck | b0aeba7 | 2015-11-10 10:09:47 -0800 | [diff] [blame] | 5934 | unsigned long __maybe_unused start = 0; |
Laura Abbott | a1c34a3 | 2015-11-05 18:48:46 -0800 | [diff] [blame] | 5935 | unsigned long __maybe_unused offset = 0; |
| 5936 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5937 | /* Skip empty nodes */ |
| 5938 | if (!pgdat->node_spanned_pages) |
| 5939 | return; |
| 5940 | |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 5941 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
Tony Luck | b0aeba7 | 2015-11-10 10:09:47 -0800 | [diff] [blame] | 5942 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
| 5943 | offset = pgdat->node_start_pfn - start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5944 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
| 5945 | if (!pgdat->node_mem_map) { |
Tony Luck | b0aeba7 | 2015-11-10 10:09:47 -0800 | [diff] [blame] | 5946 | unsigned long size, end; |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 5947 | struct page *map; |
| 5948 | |
Bob Picco | e984bb4 | 2006-05-20 15:00:31 -0700 | [diff] [blame] | 5949 | /* |
| 5950 | * The zone's endpoints aren't required to be MAX_ORDER |
| 5951 | * aligned but the node_mem_map endpoints must be in order |
| 5952 | * for the buddy allocator to function correctly. |
| 5953 | */ |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 5954 | end = pgdat_end_pfn(pgdat); |
Bob Picco | e984bb4 | 2006-05-20 15:00:31 -0700 | [diff] [blame] | 5955 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
| 5956 | size = (end - start) * sizeof(struct page); |
Dave Hansen | 6f167ec | 2005-06-23 00:07:39 -0700 | [diff] [blame] | 5957 | map = alloc_remap(pgdat->node_id, size); |
| 5958 | if (!map) |
Santosh Shilimkar | 6782832 | 2014-01-21 15:50:25 -0800 | [diff] [blame] | 5959 | map = memblock_virt_alloc_node_nopanic(size, |
| 5960 | pgdat->node_id); |
Laura Abbott | a1c34a3 | 2015-11-05 18:48:46 -0800 | [diff] [blame] | 5961 | pgdat->node_mem_map = map + offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5962 | } |
Roman Zippel | 12d810c | 2007-05-31 00:40:54 -0700 | [diff] [blame] | 5963 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5964 | /* |
| 5965 | * With no DISCONTIG, the global mem_map is just set as node 0's |
| 5966 | */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5967 | if (pgdat == NODE_DATA(0)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5968 | mem_map = NODE_DATA(0)->node_mem_map; |
Laura Abbott | a1c34a3 | 2015-11-05 18:48:46 -0800 | [diff] [blame] | 5969 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5970 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
Laura Abbott | a1c34a3 | 2015-11-05 18:48:46 -0800 | [diff] [blame] | 5971 | mem_map -= offset; |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 5972 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 5973 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5974 | #endif |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 5975 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5976 | } |
| 5977 | |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 5978 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
| 5979 | unsigned long node_start_pfn, unsigned long *zholes_size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5980 | { |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 5981 | pg_data_t *pgdat = NODE_DATA(nid); |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5982 | unsigned long start_pfn = 0; |
| 5983 | unsigned long end_pfn = 0; |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 5984 | |
Minchan Kim | 88fdf75 | 2012-07-31 16:46:14 -0700 | [diff] [blame] | 5985 | /* pg_data_t should be reset to zero when it's allocated */ |
Mel Gorman | 38087d9 | 2016-07-28 15:45:49 -0700 | [diff] [blame] | 5986 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx); |
Minchan Kim | 88fdf75 | 2012-07-31 16:46:14 -0700 | [diff] [blame] | 5987 | |
Mel Gorman | 3a80a7f | 2015-06-30 14:57:02 -0700 | [diff] [blame] | 5988 | reset_deferred_meminit(pgdat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5989 | pgdat->node_id = nid; |
| 5990 | pgdat->node_start_pfn = node_start_pfn; |
Mel Gorman | 75ef718 | 2016-07-28 15:45:24 -0700 | [diff] [blame] | 5991 | pgdat->per_cpu_nodestats = NULL; |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5992 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
| 5993 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
Juergen Gross | 8d29e18 | 2015-02-11 15:26:01 -0800 | [diff] [blame] | 5994 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
Zhen Lei | 4ada0c5 | 2015-09-08 15:04:19 -0700 | [diff] [blame] | 5995 | (u64)start_pfn << PAGE_SHIFT, |
| 5996 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); |
Taku Izumi | d91749c | 2016-03-15 14:55:18 -0700 | [diff] [blame] | 5997 | #else |
| 5998 | start_pfn = node_start_pfn; |
Zhang Yanfei | 7960aed | 2013-07-08 15:59:52 -0700 | [diff] [blame] | 5999 | #endif |
| 6000 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, |
| 6001 | zones_size, zholes_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6002 | |
| 6003 | alloc_node_mem_map(pgdat); |
Yinghai Lu | e8c27ac | 2008-06-01 13:15:22 -0700 | [diff] [blame] | 6004 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
| 6005 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", |
| 6006 | nid, (unsigned long)pgdat, |
| 6007 | (unsigned long)pgdat->node_mem_map); |
| 6008 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6009 | |
Wei Yang | 7f3eb55 | 2015-09-08 14:59:50 -0700 | [diff] [blame] | 6010 | free_area_init_core(pgdat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6011 | } |
| 6012 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 6013 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 6014 | |
| 6015 | #if MAX_NUMNODES > 1 |
| 6016 | /* |
| 6017 | * Figure out the number of possible node ids. |
| 6018 | */ |
Cody P Schafer | f9872ca | 2013-04-29 15:08:01 -0700 | [diff] [blame] | 6019 | void __init setup_nr_node_ids(void) |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 6020 | { |
Wei Yang | 904a955 | 2015-09-08 14:59:48 -0700 | [diff] [blame] | 6021 | unsigned int highest; |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 6022 | |
Wei Yang | 904a955 | 2015-09-08 14:59:48 -0700 | [diff] [blame] | 6023 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 6024 | nr_node_ids = highest + 1; |
| 6025 | } |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 6026 | #endif |
| 6027 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6028 | /** |
Tejun Heo | 1e01979 | 2011-07-12 09:45:34 +0200 | [diff] [blame] | 6029 | * node_map_pfn_alignment - determine the maximum internode alignment |
| 6030 | * |
| 6031 | * This function should be called after node map is populated and sorted. |
| 6032 | * It calculates the maximum power of two alignment which can distinguish |
| 6033 | * all the nodes. |
| 6034 | * |
| 6035 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value |
| 6036 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the |
| 6037 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is |
| 6038 | * shifted, 1GiB is enough and this function will indicate so. |
| 6039 | * |
| 6040 | * This is used to test whether pfn -> nid mapping of the chosen memory |
| 6041 | * model has fine enough granularity to avoid incorrect mapping for the |
| 6042 | * populated node map. |
| 6043 | * |
| 6044 | * Returns the determined alignment in pfn's. 0 if there is no alignment |
| 6045 | * requirement (single node). |
| 6046 | */ |
| 6047 | unsigned long __init node_map_pfn_alignment(void) |
| 6048 | { |
| 6049 | unsigned long accl_mask = 0, last_end = 0; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6050 | unsigned long start, end, mask; |
Tejun Heo | 1e01979 | 2011-07-12 09:45:34 +0200 | [diff] [blame] | 6051 | int last_nid = -1; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6052 | int i, nid; |
Tejun Heo | 1e01979 | 2011-07-12 09:45:34 +0200 | [diff] [blame] | 6053 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6054 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
Tejun Heo | 1e01979 | 2011-07-12 09:45:34 +0200 | [diff] [blame] | 6055 | if (!start || last_nid < 0 || last_nid == nid) { |
| 6056 | last_nid = nid; |
| 6057 | last_end = end; |
| 6058 | continue; |
| 6059 | } |
| 6060 | |
| 6061 | /* |
| 6062 | * Start with a mask granular enough to pin-point to the |
| 6063 | * start pfn and tick off bits one-by-one until it becomes |
| 6064 | * too coarse to separate the current node from the last. |
| 6065 | */ |
| 6066 | mask = ~((1 << __ffs(start)) - 1); |
| 6067 | while (mask && last_end <= (start & (mask << 1))) |
| 6068 | mask <<= 1; |
| 6069 | |
| 6070 | /* accumulate all internode masks */ |
| 6071 | accl_mask |= mask; |
| 6072 | } |
| 6073 | |
| 6074 | /* convert mask to number of pages */ |
| 6075 | return ~accl_mask + 1; |
| 6076 | } |
| 6077 | |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 6078 | /* Find the lowest pfn for a node */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 6079 | static unsigned long __init find_min_pfn_for_node(int nid) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6080 | { |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 6081 | unsigned long min_pfn = ULONG_MAX; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6082 | unsigned long start_pfn; |
| 6083 | int i; |
Mel Gorman | 1abbfb4 | 2006-11-23 12:01:41 +0000 | [diff] [blame] | 6084 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6085 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
| 6086 | min_pfn = min(min_pfn, start_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6087 | |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 6088 | if (min_pfn == ULONG_MAX) { |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 6089 | pr_warn("Could not find start_pfn for node %d\n", nid); |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 6090 | return 0; |
| 6091 | } |
| 6092 | |
| 6093 | return min_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6094 | } |
| 6095 | |
| 6096 | /** |
| 6097 | * find_min_pfn_with_active_regions - Find the minimum PFN registered |
| 6098 | * |
| 6099 | * It returns the minimum PFN based on information provided via |
Zhang Zhen | 7d01817 | 2014-06-04 16:10:53 -0700 | [diff] [blame] | 6100 | * memblock_set_node(). |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6101 | */ |
| 6102 | unsigned long __init find_min_pfn_with_active_regions(void) |
| 6103 | { |
| 6104 | return find_min_pfn_for_node(MAX_NUMNODES); |
| 6105 | } |
| 6106 | |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6107 | /* |
| 6108 | * early_calculate_totalpages() |
| 6109 | * Sum pages in active regions for movable zone. |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6110 | * Populate N_MEMORY for calculating usable_nodes. |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6111 | */ |
Adrian Bunk | 484f51f | 2007-10-16 01:26:03 -0700 | [diff] [blame] | 6112 | static unsigned long __init early_calculate_totalpages(void) |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6113 | { |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6114 | unsigned long totalpages = 0; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6115 | unsigned long start_pfn, end_pfn; |
| 6116 | int i, nid; |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6117 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6118 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
| 6119 | unsigned long pages = end_pfn - start_pfn; |
| 6120 | |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6121 | totalpages += pages; |
| 6122 | if (pages) |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6123 | node_set_state(nid, N_MEMORY); |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6124 | } |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 6125 | return totalpages; |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6126 | } |
| 6127 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6128 | /* |
| 6129 | * Find the PFN the Movable zone begins in each node. Kernel memory |
| 6130 | * is spread evenly between nodes as long as the nodes have enough |
| 6131 | * memory. When they don't, some nodes will have more kernelcore than |
| 6132 | * others |
| 6133 | */ |
Kautuk Consul | b224ef8 | 2012-03-21 16:34:15 -0700 | [diff] [blame] | 6134 | static void __init find_zone_movable_pfns_for_nodes(void) |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6135 | { |
| 6136 | int i, nid; |
| 6137 | unsigned long usable_startpfn; |
| 6138 | unsigned long kernelcore_node, kernelcore_remaining; |
Yinghai Lu | 66918dc | 2009-06-30 11:41:37 -0700 | [diff] [blame] | 6139 | /* save the state before borrow the nodemask */ |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6140 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6141 | unsigned long totalpages = early_calculate_totalpages(); |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6142 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
Emil Medve | 136199f | 2014-04-07 15:37:52 -0700 | [diff] [blame] | 6143 | struct memblock_region *r; |
Tang Chen | b2f3eeb | 2014-01-21 15:49:38 -0800 | [diff] [blame] | 6144 | |
| 6145 | /* Need to find movable_zone earlier when movable_node is specified. */ |
| 6146 | find_usable_zone_for_movable(); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6147 | |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6148 | /* |
Tang Chen | b2f3eeb | 2014-01-21 15:49:38 -0800 | [diff] [blame] | 6149 | * If movable_node is specified, ignore kernelcore and movablecore |
| 6150 | * options. |
| 6151 | */ |
| 6152 | if (movable_node_is_enabled()) { |
Emil Medve | 136199f | 2014-04-07 15:37:52 -0700 | [diff] [blame] | 6153 | for_each_memblock(memory, r) { |
| 6154 | if (!memblock_is_hotpluggable(r)) |
Tang Chen | b2f3eeb | 2014-01-21 15:49:38 -0800 | [diff] [blame] | 6155 | continue; |
| 6156 | |
Emil Medve | 136199f | 2014-04-07 15:37:52 -0700 | [diff] [blame] | 6157 | nid = r->nid; |
Tang Chen | b2f3eeb | 2014-01-21 15:49:38 -0800 | [diff] [blame] | 6158 | |
Emil Medve | 136199f | 2014-04-07 15:37:52 -0700 | [diff] [blame] | 6159 | usable_startpfn = PFN_DOWN(r->base); |
Tang Chen | b2f3eeb | 2014-01-21 15:49:38 -0800 | [diff] [blame] | 6160 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
| 6161 | min(usable_startpfn, zone_movable_pfn[nid]) : |
| 6162 | usable_startpfn; |
| 6163 | } |
| 6164 | |
| 6165 | goto out2; |
| 6166 | } |
| 6167 | |
| 6168 | /* |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 6169 | * If kernelcore=mirror is specified, ignore movablecore option |
| 6170 | */ |
| 6171 | if (mirrored_kernelcore) { |
| 6172 | bool mem_below_4gb_not_mirrored = false; |
| 6173 | |
| 6174 | for_each_memblock(memory, r) { |
| 6175 | if (memblock_is_mirror(r)) |
| 6176 | continue; |
| 6177 | |
| 6178 | nid = r->nid; |
| 6179 | |
| 6180 | usable_startpfn = memblock_region_memory_base_pfn(r); |
| 6181 | |
| 6182 | if (usable_startpfn < 0x100000) { |
| 6183 | mem_below_4gb_not_mirrored = true; |
| 6184 | continue; |
| 6185 | } |
| 6186 | |
| 6187 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
| 6188 | min(usable_startpfn, zone_movable_pfn[nid]) : |
| 6189 | usable_startpfn; |
| 6190 | } |
| 6191 | |
| 6192 | if (mem_below_4gb_not_mirrored) |
| 6193 | pr_warn("This configuration results in unmirrored kernel memory."); |
| 6194 | |
| 6195 | goto out2; |
| 6196 | } |
| 6197 | |
| 6198 | /* |
Tang Chen | b2f3eeb | 2014-01-21 15:49:38 -0800 | [diff] [blame] | 6199 | * If movablecore=nn[KMG] was specified, calculate what size of |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6200 | * kernelcore that corresponds so that memory usable for |
| 6201 | * any allocation type is evenly spread. If both kernelcore |
| 6202 | * and movablecore are specified, then the value of kernelcore |
| 6203 | * will be used for required_kernelcore if it's greater than |
| 6204 | * what movablecore would have allowed. |
| 6205 | */ |
| 6206 | if (required_movablecore) { |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6207 | unsigned long corepages; |
| 6208 | |
| 6209 | /* |
| 6210 | * Round-up so that ZONE_MOVABLE is at least as large as what |
| 6211 | * was requested by the user |
| 6212 | */ |
| 6213 | required_movablecore = |
| 6214 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); |
Xishi Qiu | 9fd745d | 2015-11-05 18:48:11 -0800 | [diff] [blame] | 6215 | required_movablecore = min(totalpages, required_movablecore); |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6216 | corepages = totalpages - required_movablecore; |
| 6217 | |
| 6218 | required_kernelcore = max(required_kernelcore, corepages); |
| 6219 | } |
| 6220 | |
Xishi Qiu | bde304b | 2015-11-05 18:48:56 -0800 | [diff] [blame] | 6221 | /* |
| 6222 | * If kernelcore was not specified or kernelcore size is larger |
| 6223 | * than totalpages, there is no ZONE_MOVABLE. |
| 6224 | */ |
| 6225 | if (!required_kernelcore || required_kernelcore >= totalpages) |
Yinghai Lu | 66918dc | 2009-06-30 11:41:37 -0700 | [diff] [blame] | 6226 | goto out; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6227 | |
| 6228 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6229 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
| 6230 | |
| 6231 | restart: |
| 6232 | /* Spread kernelcore memory as evenly as possible throughout nodes */ |
| 6233 | kernelcore_node = required_kernelcore / usable_nodes; |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6234 | for_each_node_state(nid, N_MEMORY) { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6235 | unsigned long start_pfn, end_pfn; |
| 6236 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6237 | /* |
| 6238 | * Recalculate kernelcore_node if the division per node |
| 6239 | * now exceeds what is necessary to satisfy the requested |
| 6240 | * amount of memory for the kernel |
| 6241 | */ |
| 6242 | if (required_kernelcore < kernelcore_node) |
| 6243 | kernelcore_node = required_kernelcore / usable_nodes; |
| 6244 | |
| 6245 | /* |
| 6246 | * As the map is walked, we track how much memory is usable |
| 6247 | * by the kernel using kernelcore_remaining. When it is |
| 6248 | * 0, the rest of the node is usable by ZONE_MOVABLE |
| 6249 | */ |
| 6250 | kernelcore_remaining = kernelcore_node; |
| 6251 | |
| 6252 | /* Go through each range of PFNs within this node */ |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6253 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6254 | unsigned long size_pages; |
| 6255 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6256 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6257 | if (start_pfn >= end_pfn) |
| 6258 | continue; |
| 6259 | |
| 6260 | /* Account for what is only usable for kernelcore */ |
| 6261 | if (start_pfn < usable_startpfn) { |
| 6262 | unsigned long kernel_pages; |
| 6263 | kernel_pages = min(end_pfn, usable_startpfn) |
| 6264 | - start_pfn; |
| 6265 | |
| 6266 | kernelcore_remaining -= min(kernel_pages, |
| 6267 | kernelcore_remaining); |
| 6268 | required_kernelcore -= min(kernel_pages, |
| 6269 | required_kernelcore); |
| 6270 | |
| 6271 | /* Continue if range is now fully accounted */ |
| 6272 | if (end_pfn <= usable_startpfn) { |
| 6273 | |
| 6274 | /* |
| 6275 | * Push zone_movable_pfn to the end so |
| 6276 | * that if we have to rebalance |
| 6277 | * kernelcore across nodes, we will |
| 6278 | * not double account here |
| 6279 | */ |
| 6280 | zone_movable_pfn[nid] = end_pfn; |
| 6281 | continue; |
| 6282 | } |
| 6283 | start_pfn = usable_startpfn; |
| 6284 | } |
| 6285 | |
| 6286 | /* |
| 6287 | * The usable PFN range for ZONE_MOVABLE is from |
| 6288 | * start_pfn->end_pfn. Calculate size_pages as the |
| 6289 | * number of pages used as kernelcore |
| 6290 | */ |
| 6291 | size_pages = end_pfn - start_pfn; |
| 6292 | if (size_pages > kernelcore_remaining) |
| 6293 | size_pages = kernelcore_remaining; |
| 6294 | zone_movable_pfn[nid] = start_pfn + size_pages; |
| 6295 | |
| 6296 | /* |
| 6297 | * Some kernelcore has been met, update counts and |
| 6298 | * break if the kernelcore for this node has been |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 6299 | * satisfied |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6300 | */ |
| 6301 | required_kernelcore -= min(required_kernelcore, |
| 6302 | size_pages); |
| 6303 | kernelcore_remaining -= size_pages; |
| 6304 | if (!kernelcore_remaining) |
| 6305 | break; |
| 6306 | } |
| 6307 | } |
| 6308 | |
| 6309 | /* |
| 6310 | * If there is still required_kernelcore, we do another pass with one |
| 6311 | * less node in the count. This will push zone_movable_pfn[nid] further |
| 6312 | * along on the nodes that still have memory until kernelcore is |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 6313 | * satisfied |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6314 | */ |
| 6315 | usable_nodes--; |
| 6316 | if (usable_nodes && required_kernelcore > usable_nodes) |
| 6317 | goto restart; |
| 6318 | |
Tang Chen | b2f3eeb | 2014-01-21 15:49:38 -0800 | [diff] [blame] | 6319 | out2: |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6320 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
| 6321 | for (nid = 0; nid < MAX_NUMNODES; nid++) |
| 6322 | zone_movable_pfn[nid] = |
| 6323 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); |
Yinghai Lu | 66918dc | 2009-06-30 11:41:37 -0700 | [diff] [blame] | 6324 | |
Yinghai Lu | 20e6926 | 2013-03-01 14:51:27 -0800 | [diff] [blame] | 6325 | out: |
Yinghai Lu | 66918dc | 2009-06-30 11:41:37 -0700 | [diff] [blame] | 6326 | /* restore the node_state */ |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6327 | node_states[N_MEMORY] = saved_node_state; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6328 | } |
| 6329 | |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6330 | /* Any regular or high memory on that node ? */ |
| 6331 | static void check_for_memory(pg_data_t *pgdat, int nid) |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6332 | { |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6333 | enum zone_type zone_type; |
| 6334 | |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6335 | if (N_MEMORY == N_NORMAL_MEMORY) |
| 6336 | return; |
| 6337 | |
| 6338 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6339 | struct zone *zone = &pgdat->node_zones[zone_type]; |
Xishi Qiu | b38a872 | 2013-11-12 15:07:20 -0800 | [diff] [blame] | 6340 | if (populated_zone(zone)) { |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6341 | node_set_state(nid, N_HIGH_MEMORY); |
| 6342 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && |
| 6343 | zone_type <= ZONE_NORMAL) |
| 6344 | node_set_state(nid, N_NORMAL_MEMORY); |
Bob Liu | d0048b0 | 2012-01-12 17:19:07 -0800 | [diff] [blame] | 6345 | break; |
| 6346 | } |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6347 | } |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6348 | } |
| 6349 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6350 | /** |
| 6351 | * free_area_init_nodes - Initialise all pg_data_t and zone data |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 6352 | * @max_zone_pfn: an array of max PFNs for each zone |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6353 | * |
| 6354 | * 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] | 6355 | * 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] | 6356 | * zone in each node and their holes is calculated. If the maximum PFN |
| 6357 | * between two adjacent zones match, it is assumed that the zone is empty. |
| 6358 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed |
| 6359 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone |
| 6360 | * starts where the previous one ended. For example, ZONE_DMA32 starts |
| 6361 | * at arch_max_dma_pfn. |
| 6362 | */ |
| 6363 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) |
| 6364 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6365 | unsigned long start_pfn, end_pfn; |
| 6366 | int i, nid; |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 6367 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6368 | /* Record where the zone boundaries are */ |
| 6369 | memset(arch_zone_lowest_possible_pfn, 0, |
| 6370 | sizeof(arch_zone_lowest_possible_pfn)); |
| 6371 | memset(arch_zone_highest_possible_pfn, 0, |
| 6372 | sizeof(arch_zone_highest_possible_pfn)); |
Oliver O'Halloran | 90cae1f | 2016-07-26 15:22:17 -0700 | [diff] [blame] | 6373 | |
| 6374 | start_pfn = find_min_pfn_with_active_regions(); |
| 6375 | |
| 6376 | for (i = 0; i < MAX_NR_ZONES; i++) { |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6377 | if (i == ZONE_MOVABLE) |
| 6378 | continue; |
Oliver O'Halloran | 90cae1f | 2016-07-26 15:22:17 -0700 | [diff] [blame] | 6379 | |
| 6380 | end_pfn = max(max_zone_pfn[i], start_pfn); |
| 6381 | arch_zone_lowest_possible_pfn[i] = start_pfn; |
| 6382 | arch_zone_highest_possible_pfn[i] = end_pfn; |
| 6383 | |
| 6384 | start_pfn = end_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6385 | } |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6386 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
| 6387 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; |
| 6388 | |
| 6389 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ |
| 6390 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); |
Kautuk Consul | b224ef8 | 2012-03-21 16:34:15 -0700 | [diff] [blame] | 6391 | find_zone_movable_pfns_for_nodes(); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6392 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6393 | /* Print out the zone ranges */ |
Anton Blanchard | f88dfff | 2014-12-10 15:42:53 -0800 | [diff] [blame] | 6394 | pr_info("Zone ranges:\n"); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6395 | for (i = 0; i < MAX_NR_ZONES; i++) { |
| 6396 | if (i == ZONE_MOVABLE) |
| 6397 | continue; |
Anton Blanchard | f88dfff | 2014-12-10 15:42:53 -0800 | [diff] [blame] | 6398 | pr_info(" %-8s ", zone_names[i]); |
David Rientjes | 72f0ba0 | 2010-03-05 13:42:14 -0800 | [diff] [blame] | 6399 | if (arch_zone_lowest_possible_pfn[i] == |
| 6400 | arch_zone_highest_possible_pfn[i]) |
Anton Blanchard | f88dfff | 2014-12-10 15:42:53 -0800 | [diff] [blame] | 6401 | pr_cont("empty\n"); |
David Rientjes | 72f0ba0 | 2010-03-05 13:42:14 -0800 | [diff] [blame] | 6402 | else |
Juergen Gross | 8d29e18 | 2015-02-11 15:26:01 -0800 | [diff] [blame] | 6403 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
| 6404 | (u64)arch_zone_lowest_possible_pfn[i] |
| 6405 | << PAGE_SHIFT, |
| 6406 | ((u64)arch_zone_highest_possible_pfn[i] |
Bjorn Helgaas | a62e2f4 | 2012-05-29 15:06:30 -0700 | [diff] [blame] | 6407 | << PAGE_SHIFT) - 1); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6408 | } |
| 6409 | |
| 6410 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ |
Anton Blanchard | f88dfff | 2014-12-10 15:42:53 -0800 | [diff] [blame] | 6411 | pr_info("Movable zone start for each node\n"); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6412 | for (i = 0; i < MAX_NUMNODES; i++) { |
| 6413 | if (zone_movable_pfn[i]) |
Juergen Gross | 8d29e18 | 2015-02-11 15:26:01 -0800 | [diff] [blame] | 6414 | pr_info(" Node %d: %#018Lx\n", i, |
| 6415 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6416 | } |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6417 | |
Wanpeng Li | f2d52fe | 2012-10-08 16:32:24 -0700 | [diff] [blame] | 6418 | /* Print out the early node map */ |
Anton Blanchard | f88dfff | 2014-12-10 15:42:53 -0800 | [diff] [blame] | 6419 | pr_info("Early memory node ranges\n"); |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 6420 | 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] | 6421 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
| 6422 | (u64)start_pfn << PAGE_SHIFT, |
| 6423 | ((u64)end_pfn << PAGE_SHIFT) - 1); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6424 | |
| 6425 | /* Initialise every node */ |
Mel Gorman | 708614e | 2008-07-23 21:26:51 -0700 | [diff] [blame] | 6426 | mminit_verify_pageflags_layout(); |
Christoph Lameter | 8ef8286 | 2007-02-20 13:57:52 -0800 | [diff] [blame] | 6427 | setup_nr_node_ids(); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6428 | for_each_online_node(nid) { |
| 6429 | pg_data_t *pgdat = NODE_DATA(nid); |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 6430 | free_area_init_node(nid, NULL, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6431 | find_min_pfn_for_node(nid), NULL); |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 6432 | |
| 6433 | /* Any memory on that node */ |
| 6434 | if (pgdat->node_present_pages) |
Lai Jiangshan | 4b0ef1fe | 2012-12-12 13:51:46 -0800 | [diff] [blame] | 6435 | node_set_state(nid, N_MEMORY); |
| 6436 | check_for_memory(pgdat, nid); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6437 | } |
| 6438 | } |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6439 | |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6440 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6441 | { |
| 6442 | unsigned long long coremem; |
| 6443 | if (!p) |
| 6444 | return -EINVAL; |
| 6445 | |
| 6446 | coremem = memparse(p, &p); |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6447 | *core = coremem >> PAGE_SHIFT; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6448 | |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6449 | /* Paranoid check that UL is enough for the coremem value */ |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 6450 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
| 6451 | |
| 6452 | return 0; |
| 6453 | } |
Mel Gorman | ed7ed36 | 2007-07-17 04:03:14 -0700 | [diff] [blame] | 6454 | |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6455 | /* |
| 6456 | * kernelcore=size sets the amount of memory for use for allocations that |
| 6457 | * cannot be reclaimed or migrated. |
| 6458 | */ |
| 6459 | static int __init cmdline_parse_kernelcore(char *p) |
| 6460 | { |
Taku Izumi | 342332e | 2016-03-15 14:55:22 -0700 | [diff] [blame] | 6461 | /* parse kernelcore=mirror */ |
| 6462 | if (parse_option_str(p, "mirror")) { |
| 6463 | mirrored_kernelcore = true; |
| 6464 | return 0; |
| 6465 | } |
| 6466 | |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6467 | return cmdline_parse_core(p, &required_kernelcore); |
| 6468 | } |
| 6469 | |
| 6470 | /* |
| 6471 | * movablecore=size sets the amount of memory for use for allocations that |
| 6472 | * can be reclaimed or migrated. |
| 6473 | */ |
| 6474 | static int __init cmdline_parse_movablecore(char *p) |
| 6475 | { |
| 6476 | return cmdline_parse_core(p, &required_movablecore); |
| 6477 | } |
| 6478 | |
Mel Gorman | ed7ed36 | 2007-07-17 04:03:14 -0700 | [diff] [blame] | 6479 | early_param("kernelcore", cmdline_parse_kernelcore); |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 6480 | early_param("movablecore", cmdline_parse_movablecore); |
Mel Gorman | ed7ed36 | 2007-07-17 04:03:14 -0700 | [diff] [blame] | 6481 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 6482 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 6483 | |
Jiang Liu | c3d5f5f | 2013-07-03 15:03:14 -0700 | [diff] [blame] | 6484 | void adjust_managed_page_count(struct page *page, long count) |
| 6485 | { |
| 6486 | spin_lock(&managed_page_count_lock); |
| 6487 | page_zone(page)->managed_pages += count; |
| 6488 | totalram_pages += count; |
Jiang Liu | 3dcc057 | 2013-07-03 15:03:21 -0700 | [diff] [blame] | 6489 | #ifdef CONFIG_HIGHMEM |
| 6490 | if (PageHighMem(page)) |
| 6491 | totalhigh_pages += count; |
| 6492 | #endif |
Jiang Liu | c3d5f5f | 2013-07-03 15:03:14 -0700 | [diff] [blame] | 6493 | spin_unlock(&managed_page_count_lock); |
| 6494 | } |
Jiang Liu | 3dcc057 | 2013-07-03 15:03:21 -0700 | [diff] [blame] | 6495 | EXPORT_SYMBOL(adjust_managed_page_count); |
Jiang Liu | c3d5f5f | 2013-07-03 15:03:14 -0700 | [diff] [blame] | 6496 | |
Jiang Liu | 1119969 | 2013-07-03 15:02:48 -0700 | [diff] [blame] | 6497 | 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] | 6498 | { |
Jiang Liu | 1119969 | 2013-07-03 15:02:48 -0700 | [diff] [blame] | 6499 | void *pos; |
| 6500 | unsigned long pages = 0; |
Jiang Liu | 69afade | 2013-04-29 15:06:21 -0700 | [diff] [blame] | 6501 | |
Jiang Liu | 1119969 | 2013-07-03 15:02:48 -0700 | [diff] [blame] | 6502 | start = (void *)PAGE_ALIGN((unsigned long)start); |
| 6503 | end = (void *)((unsigned long)end & PAGE_MASK); |
| 6504 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { |
Jiang Liu | dbe67df | 2013-07-03 15:02:51 -0700 | [diff] [blame] | 6505 | if ((unsigned int)poison <= 0xFF) |
Jiang Liu | 1119969 | 2013-07-03 15:02:48 -0700 | [diff] [blame] | 6506 | memset(pos, poison, PAGE_SIZE); |
| 6507 | free_reserved_page(virt_to_page(pos)); |
Jiang Liu | 69afade | 2013-04-29 15:06:21 -0700 | [diff] [blame] | 6508 | } |
| 6509 | |
| 6510 | if (pages && s) |
Jiang Liu | 1119969 | 2013-07-03 15:02:48 -0700 | [diff] [blame] | 6511 | pr_info("Freeing %s memory: %ldK (%p - %p)\n", |
Jiang Liu | 69afade | 2013-04-29 15:06:21 -0700 | [diff] [blame] | 6512 | s, pages << (PAGE_SHIFT - 10), start, end); |
| 6513 | |
| 6514 | return pages; |
| 6515 | } |
Jiang Liu | 1119969 | 2013-07-03 15:02:48 -0700 | [diff] [blame] | 6516 | EXPORT_SYMBOL(free_reserved_area); |
Jiang Liu | 69afade | 2013-04-29 15:06:21 -0700 | [diff] [blame] | 6517 | |
Jiang Liu | cfa11e0 | 2013-04-29 15:07:00 -0700 | [diff] [blame] | 6518 | #ifdef CONFIG_HIGHMEM |
| 6519 | void free_highmem_page(struct page *page) |
| 6520 | { |
| 6521 | __free_reserved_page(page); |
| 6522 | totalram_pages++; |
Jiang Liu | 7b4b2a0 | 2013-07-03 15:03:11 -0700 | [diff] [blame] | 6523 | page_zone(page)->managed_pages++; |
Jiang Liu | cfa11e0 | 2013-04-29 15:07:00 -0700 | [diff] [blame] | 6524 | totalhigh_pages++; |
| 6525 | } |
| 6526 | #endif |
| 6527 | |
Jiang Liu | 7ee3d4e | 2013-07-03 15:03:41 -0700 | [diff] [blame] | 6528 | |
| 6529 | void __init mem_init_print_info(const char *str) |
| 6530 | { |
| 6531 | unsigned long physpages, codesize, datasize, rosize, bss_size; |
| 6532 | unsigned long init_code_size, init_data_size; |
| 6533 | |
| 6534 | physpages = get_num_physpages(); |
| 6535 | codesize = _etext - _stext; |
| 6536 | datasize = _edata - _sdata; |
| 6537 | rosize = __end_rodata - __start_rodata; |
| 6538 | bss_size = __bss_stop - __bss_start; |
| 6539 | init_data_size = __init_end - __init_begin; |
| 6540 | init_code_size = _einittext - _sinittext; |
| 6541 | |
| 6542 | /* |
| 6543 | * Detect special cases and adjust section sizes accordingly: |
| 6544 | * 1) .init.* may be embedded into .data sections |
| 6545 | * 2) .init.text.* may be out of [__init_begin, __init_end], |
| 6546 | * please refer to arch/tile/kernel/vmlinux.lds.S. |
| 6547 | * 3) .rodata.* may be embedded into .text or .data sections. |
| 6548 | */ |
| 6549 | #define adj_init_size(start, end, size, pos, adj) \ |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 6550 | do { \ |
| 6551 | if (start <= pos && pos < end && size > adj) \ |
| 6552 | size -= adj; \ |
| 6553 | } while (0) |
Jiang Liu | 7ee3d4e | 2013-07-03 15:03:41 -0700 | [diff] [blame] | 6554 | |
| 6555 | adj_init_size(__init_begin, __init_end, init_data_size, |
| 6556 | _sinittext, init_code_size); |
| 6557 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); |
| 6558 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); |
| 6559 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); |
| 6560 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); |
| 6561 | |
| 6562 | #undef adj_init_size |
| 6563 | |
Joe Perches | 756a025 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 6564 | 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] | 6565 | #ifdef CONFIG_HIGHMEM |
Joe Perches | 756a025 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 6566 | ", %luK highmem" |
Jiang Liu | 7ee3d4e | 2013-07-03 15:03:41 -0700 | [diff] [blame] | 6567 | #endif |
Joe Perches | 756a025 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 6568 | "%s%s)\n", |
| 6569 | nr_free_pages() << (PAGE_SHIFT - 10), |
| 6570 | physpages << (PAGE_SHIFT - 10), |
| 6571 | codesize >> 10, datasize >> 10, rosize >> 10, |
| 6572 | (init_data_size + init_code_size) >> 10, bss_size >> 10, |
| 6573 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10), |
| 6574 | totalcma_pages << (PAGE_SHIFT - 10), |
Jiang Liu | 7ee3d4e | 2013-07-03 15:03:41 -0700 | [diff] [blame] | 6575 | #ifdef CONFIG_HIGHMEM |
Joe Perches | 756a025 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 6576 | totalhigh_pages << (PAGE_SHIFT - 10), |
Jiang Liu | 7ee3d4e | 2013-07-03 15:03:41 -0700 | [diff] [blame] | 6577 | #endif |
Joe Perches | 756a025 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 6578 | str ? ", " : "", str ? str : ""); |
Jiang Liu | 7ee3d4e | 2013-07-03 15:03:41 -0700 | [diff] [blame] | 6579 | } |
| 6580 | |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 6581 | /** |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 6582 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
| 6583 | * @new_dma_reserve: The number of pages to mark reserved |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 6584 | * |
Yaowei Bai | 013110a | 2015-09-08 15:04:10 -0700 | [diff] [blame] | 6585 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 6586 | * In the DMA zone, a significant percentage may be consumed by kernel image |
| 6587 | * and other unfreeable allocations which can skew the watermarks badly. This |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 6588 | * function may optionally be used to account for unfreeable pages in the |
| 6589 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and |
| 6590 | * smaller per-cpu batchsize. |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 6591 | */ |
| 6592 | void __init set_dma_reserve(unsigned long new_dma_reserve) |
| 6593 | { |
| 6594 | dma_reserve = new_dma_reserve; |
| 6595 | } |
| 6596 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6597 | void __init free_area_init(unsigned long *zones_size) |
| 6598 | { |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 6599 | free_area_init_node(0, zones_size, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6600 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
| 6601 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6602 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6603 | static int page_alloc_cpu_notify(struct notifier_block *self, |
| 6604 | unsigned long action, void *hcpu) |
| 6605 | { |
| 6606 | int cpu = (unsigned long)hcpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6607 | |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 6608 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
Konstantin Khlebnikov | f0cb3c7 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 6609 | lru_add_drain_cpu(cpu); |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 6610 | drain_pages(cpu); |
| 6611 | |
| 6612 | /* |
| 6613 | * Spill the event counters of the dead processor |
| 6614 | * into the current processors event counters. |
| 6615 | * This artificially elevates the count of the current |
| 6616 | * processor. |
| 6617 | */ |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 6618 | vm_events_fold_cpu(cpu); |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 6619 | |
| 6620 | /* |
| 6621 | * Zero the differential counters of the dead processor |
| 6622 | * so that the vm statistics are consistent. |
| 6623 | * |
| 6624 | * This is only okay since the processor is dead and cannot |
| 6625 | * race with what we are doing. |
| 6626 | */ |
Christoph Lameter | 2bb921e | 2013-09-11 14:21:30 -0700 | [diff] [blame] | 6627 | cpu_vm_stats_fold(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6628 | } |
| 6629 | return NOTIFY_OK; |
| 6630 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6631 | |
| 6632 | void __init page_alloc_init(void) |
| 6633 | { |
| 6634 | hotcpu_notifier(page_alloc_cpu_notify, 0); |
| 6635 | } |
| 6636 | |
| 6637 | /* |
Yaowei Bai | 34b1006 | 2015-09-08 15:04:13 -0700 | [diff] [blame] | 6638 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6639 | * or min_free_kbytes changes. |
| 6640 | */ |
| 6641 | static void calculate_totalreserve_pages(void) |
| 6642 | { |
| 6643 | struct pglist_data *pgdat; |
| 6644 | unsigned long reserve_pages = 0; |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 6645 | enum zone_type i, j; |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6646 | |
| 6647 | for_each_online_pgdat(pgdat) { |
Mel Gorman | 281e372 | 2016-07-28 15:46:11 -0700 | [diff] [blame] | 6648 | |
| 6649 | pgdat->totalreserve_pages = 0; |
| 6650 | |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6651 | for (i = 0; i < MAX_NR_ZONES; i++) { |
| 6652 | struct zone *zone = pgdat->node_zones + i; |
Mel Gorman | 3484b2d | 2014-08-06 16:07:14 -0700 | [diff] [blame] | 6653 | long max = 0; |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6654 | |
| 6655 | /* Find valid and maximum lowmem_reserve in the zone */ |
| 6656 | for (j = i; j < MAX_NR_ZONES; j++) { |
| 6657 | if (zone->lowmem_reserve[j] > max) |
| 6658 | max = zone->lowmem_reserve[j]; |
| 6659 | } |
| 6660 | |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 6661 | /* we treat the high watermark as reserved pages. */ |
| 6662 | max += high_wmark_pages(zone); |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6663 | |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6664 | if (max > zone->managed_pages) |
| 6665 | max = zone->managed_pages; |
Johannes Weiner | a8d0143 | 2016-01-14 15:20:15 -0800 | [diff] [blame] | 6666 | |
Mel Gorman | 281e372 | 2016-07-28 15:46:11 -0700 | [diff] [blame] | 6667 | pgdat->totalreserve_pages += max; |
Johannes Weiner | a8d0143 | 2016-01-14 15:20:15 -0800 | [diff] [blame] | 6668 | |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6669 | reserve_pages += max; |
| 6670 | } |
| 6671 | } |
| 6672 | totalreserve_pages = reserve_pages; |
| 6673 | } |
| 6674 | |
| 6675 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6676 | * setup_per_zone_lowmem_reserve - called whenever |
Yaowei Bai | 34b1006 | 2015-09-08 15:04:13 -0700 | [diff] [blame] | 6677 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6678 | * has a correct pages reserved value, so an adequate number of |
| 6679 | * pages are left in the zone after a successful __alloc_pages(). |
| 6680 | */ |
| 6681 | static void setup_per_zone_lowmem_reserve(void) |
| 6682 | { |
| 6683 | struct pglist_data *pgdat; |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 6684 | enum zone_type j, idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6685 | |
KAMEZAWA Hiroyuki | ec936fc | 2006-03-27 01:15:59 -0800 | [diff] [blame] | 6686 | for_each_online_pgdat(pgdat) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6687 | for (j = 0; j < MAX_NR_ZONES; j++) { |
| 6688 | struct zone *zone = pgdat->node_zones + j; |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6689 | unsigned long managed_pages = zone->managed_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6690 | |
| 6691 | zone->lowmem_reserve[j] = 0; |
| 6692 | |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 6693 | idx = j; |
| 6694 | while (idx) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6695 | struct zone *lower_zone; |
| 6696 | |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 6697 | idx--; |
| 6698 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6699 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
| 6700 | sysctl_lowmem_reserve_ratio[idx] = 1; |
| 6701 | |
| 6702 | lower_zone = pgdat->node_zones + idx; |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6703 | lower_zone->lowmem_reserve[j] = managed_pages / |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6704 | sysctl_lowmem_reserve_ratio[idx]; |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6705 | managed_pages += lower_zone->managed_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6706 | } |
| 6707 | } |
| 6708 | } |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6709 | |
| 6710 | /* update totalreserve_pages */ |
| 6711 | calculate_totalreserve_pages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6712 | } |
| 6713 | |
Mel Gorman | cfd3da1 | 2011-04-25 21:36:42 +0000 | [diff] [blame] | 6714 | static void __setup_per_zone_wmarks(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6715 | { |
| 6716 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); |
| 6717 | unsigned long lowmem_pages = 0; |
| 6718 | struct zone *zone; |
| 6719 | unsigned long flags; |
| 6720 | |
| 6721 | /* Calculate total number of !ZONE_HIGHMEM pages */ |
| 6722 | for_each_zone(zone) { |
| 6723 | if (!is_highmem(zone)) |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6724 | lowmem_pages += zone->managed_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6725 | } |
| 6726 | |
| 6727 | for_each_zone(zone) { |
Andrew Morton | ac924c6 | 2006-05-15 09:43:59 -0700 | [diff] [blame] | 6728 | u64 tmp; |
| 6729 | |
Gerald Schaefer | 1125b4e | 2008-10-18 20:27:11 -0700 | [diff] [blame] | 6730 | spin_lock_irqsave(&zone->lock, flags); |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6731 | tmp = (u64)pages_min * zone->managed_pages; |
Andrew Morton | ac924c6 | 2006-05-15 09:43:59 -0700 | [diff] [blame] | 6732 | do_div(tmp, lowmem_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6733 | if (is_highmem(zone)) { |
| 6734 | /* |
Nick Piggin | 669ed17 | 2005-11-13 16:06:45 -0800 | [diff] [blame] | 6735 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
| 6736 | * need highmem pages, so cap pages_min to a small |
| 6737 | * value here. |
| 6738 | * |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 6739 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
Yaowei Bai | 42ff270 | 2015-04-14 15:47:14 -0700 | [diff] [blame] | 6740 | * deltas control asynch page reclaim, and so should |
Nick Piggin | 669ed17 | 2005-11-13 16:06:45 -0800 | [diff] [blame] | 6741 | * not be capped for highmem. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6742 | */ |
Andrew Morton | 90ae8d6 | 2013-02-22 16:32:22 -0800 | [diff] [blame] | 6743 | unsigned long min_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6744 | |
Jiang Liu | b40da04 | 2013-02-22 16:33:52 -0800 | [diff] [blame] | 6745 | min_pages = zone->managed_pages / 1024; |
Andrew Morton | 90ae8d6 | 2013-02-22 16:32:22 -0800 | [diff] [blame] | 6746 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 6747 | zone->watermark[WMARK_MIN] = min_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6748 | } else { |
Nick Piggin | 669ed17 | 2005-11-13 16:06:45 -0800 | [diff] [blame] | 6749 | /* |
| 6750 | * If it's a lowmem zone, reserve a number of pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6751 | * proportionate to the zone's size. |
| 6752 | */ |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 6753 | zone->watermark[WMARK_MIN] = tmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6754 | } |
| 6755 | |
Johannes Weiner | 795ae7a | 2016-03-17 14:19:14 -0700 | [diff] [blame] | 6756 | /* |
| 6757 | * Set the kswapd watermarks distance according to the |
| 6758 | * scale factor in proportion to available memory, but |
| 6759 | * ensure a minimum size on small systems. |
| 6760 | */ |
| 6761 | tmp = max_t(u64, tmp >> 2, |
| 6762 | mult_frac(zone->managed_pages, |
| 6763 | watermark_scale_factor, 10000)); |
| 6764 | |
| 6765 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; |
| 6766 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; |
Marek Szyprowski | 49f223a | 2012-01-25 12:49:24 +0100 | [diff] [blame] | 6767 | |
Gerald Schaefer | 1125b4e | 2008-10-18 20:27:11 -0700 | [diff] [blame] | 6768 | spin_unlock_irqrestore(&zone->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6769 | } |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 6770 | |
| 6771 | /* update totalreserve_pages */ |
| 6772 | calculate_totalreserve_pages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6773 | } |
| 6774 | |
Mel Gorman | cfd3da1 | 2011-04-25 21:36:42 +0000 | [diff] [blame] | 6775 | /** |
| 6776 | * setup_per_zone_wmarks - called when min_free_kbytes changes |
| 6777 | * or when memory is hot-{added|removed} |
| 6778 | * |
| 6779 | * Ensures that the watermark[min,low,high] values for each zone are set |
| 6780 | * correctly with respect to min_free_kbytes. |
| 6781 | */ |
| 6782 | void setup_per_zone_wmarks(void) |
| 6783 | { |
| 6784 | mutex_lock(&zonelists_mutex); |
| 6785 | __setup_per_zone_wmarks(); |
| 6786 | mutex_unlock(&zonelists_mutex); |
| 6787 | } |
| 6788 | |
Randy Dunlap | 55a4462 | 2009-09-21 17:01:20 -0700 | [diff] [blame] | 6789 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6790 | * Initialise min_free_kbytes. |
| 6791 | * |
| 6792 | * For small machines we want it small (128k min). For large machines |
| 6793 | * we want it large (64MB max). But it is not linear, because network |
| 6794 | * bandwidth does not increase linearly with machine size. We use |
| 6795 | * |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 6796 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6797 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
| 6798 | * |
| 6799 | * which yields |
| 6800 | * |
| 6801 | * 16MB: 512k |
| 6802 | * 32MB: 724k |
| 6803 | * 64MB: 1024k |
| 6804 | * 128MB: 1448k |
| 6805 | * 256MB: 2048k |
| 6806 | * 512MB: 2896k |
| 6807 | * 1024MB: 4096k |
| 6808 | * 2048MB: 5792k |
| 6809 | * 4096MB: 8192k |
| 6810 | * 8192MB: 11584k |
| 6811 | * 16384MB: 16384k |
| 6812 | */ |
KOSAKI Motohiro | 1b79acc | 2011-05-24 17:11:32 -0700 | [diff] [blame] | 6813 | int __meminit init_per_zone_wmark_min(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6814 | { |
| 6815 | unsigned long lowmem_kbytes; |
Michal Hocko | 5f12733 | 2013-07-08 16:00:40 -0700 | [diff] [blame] | 6816 | int new_min_free_kbytes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6817 | |
| 6818 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); |
Michal Hocko | 5f12733 | 2013-07-08 16:00:40 -0700 | [diff] [blame] | 6819 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6820 | |
Michal Hocko | 5f12733 | 2013-07-08 16:00:40 -0700 | [diff] [blame] | 6821 | if (new_min_free_kbytes > user_min_free_kbytes) { |
| 6822 | min_free_kbytes = new_min_free_kbytes; |
| 6823 | if (min_free_kbytes < 128) |
| 6824 | min_free_kbytes = 128; |
| 6825 | if (min_free_kbytes > 65536) |
| 6826 | min_free_kbytes = 65536; |
| 6827 | } else { |
| 6828 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", |
| 6829 | new_min_free_kbytes, user_min_free_kbytes); |
| 6830 | } |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 6831 | setup_per_zone_wmarks(); |
KOSAKI Motohiro | a6cccdc | 2011-05-24 17:11:33 -0700 | [diff] [blame] | 6832 | refresh_zone_stat_thresholds(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6833 | setup_per_zone_lowmem_reserve(); |
Joonsoo Kim | 6423aa8 | 2016-08-10 16:27:49 -0700 | [diff] [blame] | 6834 | |
| 6835 | #ifdef CONFIG_NUMA |
| 6836 | setup_min_unmapped_ratio(); |
| 6837 | setup_min_slab_ratio(); |
| 6838 | #endif |
| 6839 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6840 | return 0; |
| 6841 | } |
Jason Baron | bc22af7 | 2016-05-05 16:22:12 -0700 | [diff] [blame] | 6842 | core_initcall(init_per_zone_wmark_min) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6843 | |
| 6844 | /* |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 6845 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6846 | * that we can call two helper functions whenever min_free_kbytes |
| 6847 | * changes. |
| 6848 | */ |
Joe Perches | cccad5b | 2014-06-06 14:38:09 -0700 | [diff] [blame] | 6849 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 6850 | void __user *buffer, size_t *length, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6851 | { |
Han Pingtian | da8c757 | 2014-01-23 15:53:17 -0800 | [diff] [blame] | 6852 | int rc; |
| 6853 | |
| 6854 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
| 6855 | if (rc) |
| 6856 | return rc; |
| 6857 | |
Michal Hocko | 5f12733 | 2013-07-08 16:00:40 -0700 | [diff] [blame] | 6858 | if (write) { |
| 6859 | user_min_free_kbytes = min_free_kbytes; |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 6860 | setup_per_zone_wmarks(); |
Michal Hocko | 5f12733 | 2013-07-08 16:00:40 -0700 | [diff] [blame] | 6861 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6862 | return 0; |
| 6863 | } |
| 6864 | |
Johannes Weiner | 795ae7a | 2016-03-17 14:19:14 -0700 | [diff] [blame] | 6865 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
| 6866 | void __user *buffer, size_t *length, loff_t *ppos) |
| 6867 | { |
| 6868 | int rc; |
| 6869 | |
| 6870 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
| 6871 | if (rc) |
| 6872 | return rc; |
| 6873 | |
| 6874 | if (write) |
| 6875 | setup_per_zone_wmarks(); |
| 6876 | |
| 6877 | return 0; |
| 6878 | } |
| 6879 | |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 6880 | #ifdef CONFIG_NUMA |
Joonsoo Kim | 6423aa8 | 2016-08-10 16:27:49 -0700 | [diff] [blame] | 6881 | static void setup_min_unmapped_ratio(void) |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 6882 | { |
Joonsoo Kim | 6423aa8 | 2016-08-10 16:27:49 -0700 | [diff] [blame] | 6883 | pg_data_t *pgdat; |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 6884 | struct zone *zone; |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 6885 | |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 6886 | for_each_online_pgdat(pgdat) |
Joonsoo Kim | 81cbcbc | 2016-08-10 16:27:46 -0700 | [diff] [blame] | 6887 | pgdat->min_unmapped_pages = 0; |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 6888 | |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 6889 | for_each_zone(zone) |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 6890 | zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages * |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 6891 | sysctl_min_unmapped_ratio) / 100; |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 6892 | } |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 6893 | |
Joonsoo Kim | 6423aa8 | 2016-08-10 16:27:49 -0700 | [diff] [blame] | 6894 | |
| 6895 | 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] | 6896 | void __user *buffer, size_t *length, loff_t *ppos) |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 6897 | { |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 6898 | int rc; |
| 6899 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 6900 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 6901 | if (rc) |
| 6902 | return rc; |
| 6903 | |
Joonsoo Kim | 6423aa8 | 2016-08-10 16:27:49 -0700 | [diff] [blame] | 6904 | setup_min_unmapped_ratio(); |
| 6905 | |
| 6906 | return 0; |
| 6907 | } |
| 6908 | |
| 6909 | static void setup_min_slab_ratio(void) |
| 6910 | { |
| 6911 | pg_data_t *pgdat; |
| 6912 | struct zone *zone; |
| 6913 | |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 6914 | for_each_online_pgdat(pgdat) |
| 6915 | pgdat->min_slab_pages = 0; |
| 6916 | |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 6917 | for_each_zone(zone) |
Mel Gorman | a5f5f91 | 2016-07-28 15:46:32 -0700 | [diff] [blame] | 6918 | zone->zone_pgdat->min_slab_pages += (zone->managed_pages * |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 6919 | sysctl_min_slab_ratio) / 100; |
Joonsoo Kim | 6423aa8 | 2016-08-10 16:27:49 -0700 | [diff] [blame] | 6920 | } |
| 6921 | |
| 6922 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, |
| 6923 | void __user *buffer, size_t *length, loff_t *ppos) |
| 6924 | { |
| 6925 | int rc; |
| 6926 | |
| 6927 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
| 6928 | if (rc) |
| 6929 | return rc; |
| 6930 | |
| 6931 | setup_min_slab_ratio(); |
| 6932 | |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 6933 | return 0; |
| 6934 | } |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 6935 | #endif |
| 6936 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6937 | /* |
| 6938 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around |
| 6939 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() |
| 6940 | * whenever sysctl_lowmem_reserve_ratio changes. |
| 6941 | * |
| 6942 | * The reserve ratio obviously has absolutely no relation with the |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 6943 | * minimum watermarks. The lowmem reserve ratio can only make sense |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6944 | * if in function of the boot time zone sizes. |
| 6945 | */ |
Joe Perches | cccad5b | 2014-06-06 14:38:09 -0700 | [diff] [blame] | 6946 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 6947 | void __user *buffer, size_t *length, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6948 | { |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 6949 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6950 | setup_per_zone_lowmem_reserve(); |
| 6951 | return 0; |
| 6952 | } |
| 6953 | |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 6954 | /* |
| 6955 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 6956 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
| 6957 | * pagelist can have before it gets flushed back to buddy allocator. |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 6958 | */ |
Joe Perches | cccad5b | 2014-06-06 14:38:09 -0700 | [diff] [blame] | 6959 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 6960 | void __user *buffer, size_t *length, loff_t *ppos) |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 6961 | { |
| 6962 | struct zone *zone; |
David Rientjes | 7cd2b0a | 2014-06-23 13:22:04 -0700 | [diff] [blame] | 6963 | int old_percpu_pagelist_fraction; |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 6964 | int ret; |
| 6965 | |
Cody P Schafer | c8e251f | 2013-07-03 15:01:29 -0700 | [diff] [blame] | 6966 | mutex_lock(&pcp_batch_high_lock); |
David Rientjes | 7cd2b0a | 2014-06-23 13:22:04 -0700 | [diff] [blame] | 6967 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; |
| 6968 | |
| 6969 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
| 6970 | if (!write || ret < 0) |
| 6971 | goto out; |
| 6972 | |
| 6973 | /* Sanity checking to avoid pcp imbalance */ |
| 6974 | if (percpu_pagelist_fraction && |
| 6975 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { |
| 6976 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; |
| 6977 | ret = -EINVAL; |
| 6978 | goto out; |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 6979 | } |
David Rientjes | 7cd2b0a | 2014-06-23 13:22:04 -0700 | [diff] [blame] | 6980 | |
| 6981 | /* No change? */ |
| 6982 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) |
| 6983 | goto out; |
| 6984 | |
| 6985 | for_each_populated_zone(zone) { |
| 6986 | unsigned int cpu; |
| 6987 | |
| 6988 | for_each_possible_cpu(cpu) |
| 6989 | pageset_set_high_and_batch(zone, |
| 6990 | per_cpu_ptr(zone->pageset, cpu)); |
| 6991 | } |
| 6992 | out: |
Cody P Schafer | c8e251f | 2013-07-03 15:01:29 -0700 | [diff] [blame] | 6993 | mutex_unlock(&pcp_batch_high_lock); |
David Rientjes | 7cd2b0a | 2014-06-23 13:22:04 -0700 | [diff] [blame] | 6994 | return ret; |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 6995 | } |
| 6996 | |
Rasmus Villemoes | a9919c7 | 2015-06-24 16:56:28 -0700 | [diff] [blame] | 6997 | #ifdef CONFIG_NUMA |
David S. Miller | f034b5d | 2006-08-24 03:08:07 -0700 | [diff] [blame] | 6998 | int hashdist = HASHDIST_DEFAULT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6999 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7000 | static int __init set_hashdist(char *str) |
| 7001 | { |
| 7002 | if (!str) |
| 7003 | return 0; |
| 7004 | hashdist = simple_strtoul(str, &str, 0); |
| 7005 | return 1; |
| 7006 | } |
| 7007 | __setup("hashdist=", set_hashdist); |
| 7008 | #endif |
| 7009 | |
Srikar Dronamraju | f6f34b4 | 2016-10-07 16:59:15 -0700 | [diff] [blame] | 7010 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
| 7011 | /* |
| 7012 | * Returns the number of pages that arch has reserved but |
| 7013 | * is not known to alloc_large_system_hash(). |
| 7014 | */ |
| 7015 | static unsigned long __init arch_reserved_kernel_pages(void) |
| 7016 | { |
| 7017 | return 0; |
| 7018 | } |
| 7019 | #endif |
| 7020 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7021 | /* |
| 7022 | * allocate a large system hash table from bootmem |
| 7023 | * - it is assumed that the hash table must contain an exact power-of-2 |
| 7024 | * quantity of entries |
| 7025 | * - limit is the number of hash buckets, not the total allocation size |
| 7026 | */ |
| 7027 | void *__init alloc_large_system_hash(const char *tablename, |
| 7028 | unsigned long bucketsize, |
| 7029 | unsigned long numentries, |
| 7030 | int scale, |
| 7031 | int flags, |
| 7032 | unsigned int *_hash_shift, |
| 7033 | unsigned int *_hash_mask, |
Tim Bird | 31fe62b | 2012-05-23 13:33:35 +0000 | [diff] [blame] | 7034 | unsigned long low_limit, |
| 7035 | unsigned long high_limit) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7036 | { |
Tim Bird | 31fe62b | 2012-05-23 13:33:35 +0000 | [diff] [blame] | 7037 | unsigned long long max = high_limit; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7038 | unsigned long log2qty, size; |
| 7039 | void *table = NULL; |
| 7040 | |
| 7041 | /* allow the kernel cmdline to have a say */ |
| 7042 | if (!numentries) { |
| 7043 | /* round applicable memory size up to nearest megabyte */ |
Andrew Morton | 0490366 | 2006-12-06 20:37:33 -0800 | [diff] [blame] | 7044 | numentries = nr_kernel_pages; |
Srikar Dronamraju | f6f34b4 | 2016-10-07 16:59:15 -0700 | [diff] [blame] | 7045 | numentries -= arch_reserved_kernel_pages(); |
Jerry Zhou | a7e8331 | 2013-09-11 14:20:26 -0700 | [diff] [blame] | 7046 | |
| 7047 | /* It isn't necessary when PAGE_SIZE >= 1MB */ |
| 7048 | if (PAGE_SHIFT < 20) |
| 7049 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7050 | |
| 7051 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
| 7052 | if (scale > PAGE_SHIFT) |
| 7053 | numentries >>= (scale - PAGE_SHIFT); |
| 7054 | else |
| 7055 | numentries <<= (PAGE_SHIFT - scale); |
Paul Mundt | 9ab37b8 | 2007-01-05 16:36:30 -0800 | [diff] [blame] | 7056 | |
| 7057 | /* Make sure we've got at least a 0-order allocation.. */ |
Jan Beulich | 2c85f51 | 2009-09-21 17:03:07 -0700 | [diff] [blame] | 7058 | if (unlikely(flags & HASH_SMALL)) { |
| 7059 | /* Makes no sense without HASH_EARLY */ |
| 7060 | WARN_ON(!(flags & HASH_EARLY)); |
| 7061 | if (!(numentries >> *_hash_shift)) { |
| 7062 | numentries = 1UL << *_hash_shift; |
| 7063 | BUG_ON(!numentries); |
| 7064 | } |
| 7065 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) |
Paul Mundt | 9ab37b8 | 2007-01-05 16:36:30 -0800 | [diff] [blame] | 7066 | numentries = PAGE_SIZE / bucketsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7067 | } |
John Hawkes | 6e692ed | 2006-03-25 03:08:02 -0800 | [diff] [blame] | 7068 | numentries = roundup_pow_of_two(numentries); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7069 | |
| 7070 | /* limit allocation size to 1/16 total memory by default */ |
| 7071 | if (max == 0) { |
| 7072 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; |
| 7073 | do_div(max, bucketsize); |
| 7074 | } |
Dimitri Sivanich | 074b851 | 2012-02-08 12:39:07 -0800 | [diff] [blame] | 7075 | max = min(max, 0x80000000ULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7076 | |
Tim Bird | 31fe62b | 2012-05-23 13:33:35 +0000 | [diff] [blame] | 7077 | if (numentries < low_limit) |
| 7078 | numentries = low_limit; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7079 | if (numentries > max) |
| 7080 | numentries = max; |
| 7081 | |
David Howells | f0d1b0b | 2006-12-08 02:37:49 -0800 | [diff] [blame] | 7082 | log2qty = ilog2(numentries); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7083 | |
| 7084 | do { |
| 7085 | size = bucketsize << log2qty; |
| 7086 | if (flags & HASH_EARLY) |
Santosh Shilimkar | 6782832 | 2014-01-21 15:50:25 -0800 | [diff] [blame] | 7087 | table = memblock_virt_alloc_nopanic(size, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7088 | else if (hashdist) |
| 7089 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); |
| 7090 | else { |
Eric Dumazet | 1037b83 | 2007-07-15 23:38:05 -0700 | [diff] [blame] | 7091 | /* |
| 7092 | * If bucketsize is not a power-of-two, we may free |
Mel Gorman | a1dd268 | 2009-06-16 15:32:19 -0700 | [diff] [blame] | 7093 | * some pages at the end of hash table which |
| 7094 | * alloc_pages_exact() automatically does |
Eric Dumazet | 1037b83 | 2007-07-15 23:38:05 -0700 | [diff] [blame] | 7095 | */ |
Catalin Marinas | 264ef8a | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 7096 | if (get_order(size) < MAX_ORDER) { |
Mel Gorman | a1dd268 | 2009-06-16 15:32:19 -0700 | [diff] [blame] | 7097 | table = alloc_pages_exact(size, GFP_ATOMIC); |
Catalin Marinas | 264ef8a | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 7098 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
| 7099 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7100 | } |
| 7101 | } while (!table && size > PAGE_SIZE && --log2qty); |
| 7102 | |
| 7103 | if (!table) |
| 7104 | panic("Failed to allocate %s hash table\n", tablename); |
| 7105 | |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 7106 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n", |
| 7107 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7108 | |
| 7109 | if (_hash_shift) |
| 7110 | *_hash_shift = log2qty; |
| 7111 | if (_hash_mask) |
| 7112 | *_hash_mask = (1 << log2qty) - 1; |
| 7113 | |
| 7114 | return table; |
| 7115 | } |
KAMEZAWA Hiroyuki | a117e66 | 2006-03-27 01:15:25 -0800 | [diff] [blame] | 7116 | |
KAMEZAWA Hiroyuki | a5d76b54 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 7117 | /* |
Minchan Kim | 8093451 | 2012-07-31 16:43:01 -0700 | [diff] [blame] | 7118 | * This function checks whether pageblock includes unmovable pages or not. |
| 7119 | * If @count is not zero, it is okay to include less @count unmovable pages |
| 7120 | * |
Pintu Kumar | b8af294 | 2013-09-11 14:20:34 -0700 | [diff] [blame] | 7121 | * PageLRU check without isolation or lru_lock could race so that |
Minchan Kim | 8093451 | 2012-07-31 16:43:01 -0700 | [diff] [blame] | 7122 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't |
| 7123 | * expect this function should be exact. |
KAMEZAWA Hiroyuki | a5d76b54 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 7124 | */ |
Wen Congyang | b023f46 | 2012-12-11 16:00:45 -0800 | [diff] [blame] | 7125 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
| 7126 | bool skip_hwpoisoned_pages) |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7127 | { |
| 7128 | unsigned long pfn, iter, found; |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7129 | int mt; |
| 7130 | |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7131 | /* |
| 7132 | * For avoiding noise data, lru_add_drain_all() should be called |
Minchan Kim | 8093451 | 2012-07-31 16:43:01 -0700 | [diff] [blame] | 7133 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7134 | */ |
| 7135 | if (zone_idx(zone) == ZONE_MOVABLE) |
Minchan Kim | 8093451 | 2012-07-31 16:43:01 -0700 | [diff] [blame] | 7136 | return false; |
Michal Nazarewicz | 47118af | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7137 | mt = get_pageblock_migratetype(page); |
| 7138 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) |
Minchan Kim | 8093451 | 2012-07-31 16:43:01 -0700 | [diff] [blame] | 7139 | return false; |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7140 | |
| 7141 | pfn = page_to_pfn(page); |
| 7142 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { |
| 7143 | unsigned long check = pfn + iter; |
| 7144 | |
Namhyung Kim | 29723fc | 2011-02-25 14:44:25 -0800 | [diff] [blame] | 7145 | if (!pfn_valid_within(check)) |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7146 | continue; |
Namhyung Kim | 29723fc | 2011-02-25 14:44:25 -0800 | [diff] [blame] | 7147 | |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7148 | page = pfn_to_page(check); |
Naoya Horiguchi | c8721bb | 2013-09-11 14:22:09 -0700 | [diff] [blame] | 7149 | |
| 7150 | /* |
| 7151 | * Hugepages are not in LRU lists, but they're movable. |
| 7152 | * We need not scan over tail pages bacause we don't |
| 7153 | * handle each tail page individually in migration. |
| 7154 | */ |
| 7155 | if (PageHuge(page)) { |
| 7156 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; |
| 7157 | continue; |
| 7158 | } |
| 7159 | |
Minchan Kim | 97d255c | 2012-07-31 16:42:59 -0700 | [diff] [blame] | 7160 | /* |
| 7161 | * We can't use page_count without pin a page |
| 7162 | * because another CPU can free compound page. |
| 7163 | * This check already skips compound tails of THP |
Joonsoo Kim | 0139aa7 | 2016-05-19 17:10:49 -0700 | [diff] [blame] | 7164 | * because their page->_refcount is zero at all time. |
Minchan Kim | 97d255c | 2012-07-31 16:42:59 -0700 | [diff] [blame] | 7165 | */ |
Joonsoo Kim | fe896d1 | 2016-03-17 14:19:26 -0700 | [diff] [blame] | 7166 | if (!page_ref_count(page)) { |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7167 | if (PageBuddy(page)) |
| 7168 | iter += (1 << page_order(page)) - 1; |
| 7169 | continue; |
| 7170 | } |
Minchan Kim | 97d255c | 2012-07-31 16:42:59 -0700 | [diff] [blame] | 7171 | |
Wen Congyang | b023f46 | 2012-12-11 16:00:45 -0800 | [diff] [blame] | 7172 | /* |
| 7173 | * The HWPoisoned page may be not in buddy system, and |
| 7174 | * page_count() is not 0. |
| 7175 | */ |
| 7176 | if (skip_hwpoisoned_pages && PageHWPoison(page)) |
| 7177 | continue; |
| 7178 | |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7179 | if (!PageLRU(page)) |
| 7180 | found++; |
| 7181 | /* |
Johannes Weiner | 6b4f779 | 2014-12-12 16:56:13 -0800 | [diff] [blame] | 7182 | * If there are RECLAIMABLE pages, we need to check |
| 7183 | * it. But now, memory offline itself doesn't call |
| 7184 | * shrink_node_slabs() and it still to be fixed. |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7185 | */ |
| 7186 | /* |
| 7187 | * If the page is not RAM, page_count()should be 0. |
| 7188 | * we don't need more check. This is an _used_ not-movable page. |
| 7189 | * |
| 7190 | * The problematic thing here is PG_reserved pages. PG_reserved |
| 7191 | * is set to both of a memory hole page and a _used_ kernel |
| 7192 | * page at boot. |
| 7193 | */ |
| 7194 | if (found > count) |
Minchan Kim | 8093451 | 2012-07-31 16:43:01 -0700 | [diff] [blame] | 7195 | return true; |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7196 | } |
Minchan Kim | 8093451 | 2012-07-31 16:43:01 -0700 | [diff] [blame] | 7197 | return false; |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 7198 | } |
| 7199 | |
| 7200 | bool is_pageblock_removable_nolock(struct page *page) |
| 7201 | { |
Michal Hocko | 656a070 | 2012-01-20 14:33:58 -0800 | [diff] [blame] | 7202 | struct zone *zone; |
| 7203 | unsigned long pfn; |
Michal Hocko | 687875f | 2012-01-20 14:33:55 -0800 | [diff] [blame] | 7204 | |
| 7205 | /* |
| 7206 | * We have to be careful here because we are iterating over memory |
| 7207 | * sections which are not zone aware so we might end up outside of |
| 7208 | * the zone but still within the section. |
Michal Hocko | 656a070 | 2012-01-20 14:33:58 -0800 | [diff] [blame] | 7209 | * We have to take care about the node as well. If the node is offline |
| 7210 | * its NODE_DATA will be NULL - see page_zone. |
Michal Hocko | 687875f | 2012-01-20 14:33:55 -0800 | [diff] [blame] | 7211 | */ |
Michal Hocko | 656a070 | 2012-01-20 14:33:58 -0800 | [diff] [blame] | 7212 | if (!node_online(page_to_nid(page))) |
| 7213 | return false; |
| 7214 | |
| 7215 | zone = page_zone(page); |
| 7216 | pfn = page_to_pfn(page); |
Cody P Schafer | 108bcc9 | 2013-02-22 16:35:23 -0800 | [diff] [blame] | 7217 | if (!zone_spans_pfn(zone, pfn)) |
Michal Hocko | 687875f | 2012-01-20 14:33:55 -0800 | [diff] [blame] | 7218 | return false; |
| 7219 | |
Wen Congyang | b023f46 | 2012-12-11 16:00:45 -0800 | [diff] [blame] | 7220 | return !has_unmovable_pages(zone, page, 0, true); |
KAMEZAWA Hiroyuki | a5d76b54 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 7221 | } |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 7222 | |
Vlastimil Babka | 080fe20 | 2016-02-05 15:36:41 -0800 | [diff] [blame] | 7223 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7224 | |
| 7225 | static unsigned long pfn_max_align_down(unsigned long pfn) |
| 7226 | { |
| 7227 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, |
| 7228 | pageblock_nr_pages) - 1); |
| 7229 | } |
| 7230 | |
| 7231 | static unsigned long pfn_max_align_up(unsigned long pfn) |
| 7232 | { |
| 7233 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, |
| 7234 | pageblock_nr_pages)); |
| 7235 | } |
| 7236 | |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7237 | /* [start, end) must belong to a single zone. */ |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 7238 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
| 7239 | unsigned long start, unsigned long end) |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7240 | { |
| 7241 | /* This function is based on compact_zone() from compaction.c. */ |
Minchan Kim | beb51ea | 2012-10-08 16:33:51 -0700 | [diff] [blame] | 7242 | unsigned long nr_reclaimed; |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7243 | unsigned long pfn = start; |
| 7244 | unsigned int tries = 0; |
| 7245 | int ret = 0; |
| 7246 | |
Marek Szyprowski | be49a6e | 2012-12-12 13:51:19 -0800 | [diff] [blame] | 7247 | migrate_prep(); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7248 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 7249 | while (pfn < end || !list_empty(&cc->migratepages)) { |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7250 | if (fatal_signal_pending(current)) { |
| 7251 | ret = -EINTR; |
| 7252 | break; |
| 7253 | } |
| 7254 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 7255 | if (list_empty(&cc->migratepages)) { |
| 7256 | cc->nr_migratepages = 0; |
Vlastimil Babka | edc2ca6 | 2014-10-09 15:27:09 -0700 | [diff] [blame] | 7257 | pfn = isolate_migratepages_range(cc, pfn, end); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7258 | if (!pfn) { |
| 7259 | ret = -EINTR; |
| 7260 | break; |
| 7261 | } |
| 7262 | tries = 0; |
| 7263 | } else if (++tries == 5) { |
| 7264 | ret = ret < 0 ? ret : -EBUSY; |
| 7265 | break; |
| 7266 | } |
| 7267 | |
Minchan Kim | beb51ea | 2012-10-08 16:33:51 -0700 | [diff] [blame] | 7268 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
| 7269 | &cc->migratepages); |
| 7270 | cc->nr_migratepages -= nr_reclaimed; |
Minchan Kim | 02c6de8 | 2012-10-08 16:31:55 -0700 | [diff] [blame] | 7271 | |
Hugh Dickins | 9c620e2 | 2013-02-22 16:35:14 -0800 | [diff] [blame] | 7272 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 7273 | NULL, 0, cc->mode, MR_CMA); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7274 | } |
Srinivas Pandruvada | 2a6f512 | 2013-02-22 16:32:09 -0800 | [diff] [blame] | 7275 | if (ret < 0) { |
| 7276 | putback_movable_pages(&cc->migratepages); |
| 7277 | return ret; |
| 7278 | } |
| 7279 | return 0; |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7280 | } |
| 7281 | |
| 7282 | /** |
| 7283 | * alloc_contig_range() -- tries to allocate given range of pages |
| 7284 | * @start: start PFN to allocate |
| 7285 | * @end: one-past-the-last PFN to allocate |
Michal Nazarewicz | 0815f3d | 2012-04-03 15:06:15 +0200 | [diff] [blame] | 7286 | * @migratetype: migratetype of the underlaying pageblocks (either |
| 7287 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks |
| 7288 | * in range must have the same migratetype and it must |
| 7289 | * be either of the two. |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7290 | * |
| 7291 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES |
| 7292 | * aligned, however it's the caller's responsibility to guarantee that |
| 7293 | * we are the only thread that changes migrate type of pageblocks the |
| 7294 | * pages fall in. |
| 7295 | * |
| 7296 | * The PFN range must belong to a single zone. |
| 7297 | * |
| 7298 | * Returns zero on success or negative error code. On success all |
| 7299 | * pages which PFN is in [start, end) are allocated for the caller and |
| 7300 | * need to be freed with free_contig_range(). |
| 7301 | */ |
Michal Nazarewicz | 0815f3d | 2012-04-03 15:06:15 +0200 | [diff] [blame] | 7302 | int alloc_contig_range(unsigned long start, unsigned long end, |
| 7303 | unsigned migratetype) |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7304 | { |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7305 | unsigned long outer_start, outer_end; |
Kirill A. Shutemov | d00181b | 2015-11-06 16:29:57 -0800 | [diff] [blame] | 7306 | unsigned int order; |
| 7307 | int ret = 0; |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7308 | |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 7309 | struct compact_control cc = { |
| 7310 | .nr_migratepages = 0, |
| 7311 | .order = -1, |
| 7312 | .zone = page_zone(pfn_to_page(start)), |
David Rientjes | e0b9dae | 2014-06-04 16:08:28 -0700 | [diff] [blame] | 7313 | .mode = MIGRATE_SYNC, |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 7314 | .ignore_skip_hint = true, |
| 7315 | }; |
| 7316 | INIT_LIST_HEAD(&cc.migratepages); |
| 7317 | |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7318 | /* |
| 7319 | * What we do here is we mark all pageblocks in range as |
| 7320 | * MIGRATE_ISOLATE. Because pageblock and max order pages may |
| 7321 | * have different sizes, and due to the way page allocator |
| 7322 | * work, we align the range to biggest of the two pages so |
| 7323 | * that page allocator won't try to merge buddies from |
| 7324 | * different pageblocks and change MIGRATE_ISOLATE to some |
| 7325 | * other migration type. |
| 7326 | * |
| 7327 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we |
| 7328 | * migrate the pages from an unaligned range (ie. pages that |
| 7329 | * we are interested in). This will put all the pages in |
| 7330 | * range back to page allocator as MIGRATE_ISOLATE. |
| 7331 | * |
| 7332 | * When this is done, we take the pages in range from page |
| 7333 | * allocator removing them from the buddy system. This way |
| 7334 | * page allocator will never consider using them. |
| 7335 | * |
| 7336 | * This lets us mark the pageblocks back as |
| 7337 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the |
| 7338 | * aligned range but not in the unaligned, original range are |
| 7339 | * put back to page allocator so that buddy can use them. |
| 7340 | */ |
| 7341 | |
| 7342 | ret = start_isolate_page_range(pfn_max_align_down(start), |
Wen Congyang | b023f46 | 2012-12-11 16:00:45 -0800 | [diff] [blame] | 7343 | pfn_max_align_up(end), migratetype, |
| 7344 | false); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7345 | if (ret) |
Bob Liu | 86a595f | 2012-10-25 13:37:56 -0700 | [diff] [blame] | 7346 | return ret; |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7347 | |
Joonsoo Kim | 8ef5849 | 2016-01-14 15:18:45 -0800 | [diff] [blame] | 7348 | /* |
| 7349 | * In case of -EBUSY, we'd like to know which page causes problem. |
| 7350 | * So, just fall through. We will check it in test_pages_isolated(). |
| 7351 | */ |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 7352 | ret = __alloc_contig_migrate_range(&cc, start, end); |
Joonsoo Kim | 8ef5849 | 2016-01-14 15:18:45 -0800 | [diff] [blame] | 7353 | if (ret && ret != -EBUSY) |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7354 | goto done; |
| 7355 | |
| 7356 | /* |
| 7357 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES |
| 7358 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's |
| 7359 | * more, all pages in [start, end) are free in page allocator. |
| 7360 | * What we are going to do is to allocate all pages from |
| 7361 | * [start, end) (that is remove them from page allocator). |
| 7362 | * |
| 7363 | * The only problem is that pages at the beginning and at the |
| 7364 | * end of interesting range may be not aligned with pages that |
| 7365 | * page allocator holds, ie. they can be part of higher order |
| 7366 | * pages. Because of this, we reserve the bigger range and |
| 7367 | * once this is done free the pages we are not interested in. |
| 7368 | * |
| 7369 | * We don't have to hold zone->lock here because the pages are |
| 7370 | * isolated thus they won't get removed from buddy. |
| 7371 | */ |
| 7372 | |
| 7373 | lru_add_drain_all(); |
Vlastimil Babka | 510f550 | 2014-12-10 15:43:07 -0800 | [diff] [blame] | 7374 | drain_all_pages(cc.zone); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7375 | |
| 7376 | order = 0; |
| 7377 | outer_start = start; |
| 7378 | while (!PageBuddy(pfn_to_page(outer_start))) { |
| 7379 | if (++order >= MAX_ORDER) { |
Joonsoo Kim | 8ef5849 | 2016-01-14 15:18:45 -0800 | [diff] [blame] | 7380 | outer_start = start; |
| 7381 | break; |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7382 | } |
| 7383 | outer_start &= ~0UL << order; |
| 7384 | } |
| 7385 | |
Joonsoo Kim | 8ef5849 | 2016-01-14 15:18:45 -0800 | [diff] [blame] | 7386 | if (outer_start != start) { |
| 7387 | order = page_order(pfn_to_page(outer_start)); |
| 7388 | |
| 7389 | /* |
| 7390 | * outer_start page could be small order buddy page and |
| 7391 | * it doesn't include start page. Adjust outer_start |
| 7392 | * in this case to report failed page properly |
| 7393 | * on tracepoint in test_pages_isolated() |
| 7394 | */ |
| 7395 | if (outer_start + (1UL << order) <= start) |
| 7396 | outer_start = start; |
| 7397 | } |
| 7398 | |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7399 | /* Make sure the range is really isolated. */ |
Wen Congyang | b023f46 | 2012-12-11 16:00:45 -0800 | [diff] [blame] | 7400 | if (test_pages_isolated(outer_start, end, false)) { |
Michal Nazarewicz | dae803e | 2014-11-13 15:19:27 -0800 | [diff] [blame] | 7401 | pr_info("%s: [%lx, %lx) PFNs busy\n", |
| 7402 | __func__, outer_start, end); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7403 | ret = -EBUSY; |
| 7404 | goto done; |
| 7405 | } |
| 7406 | |
Marek Szyprowski | 49f223a | 2012-01-25 12:49:24 +0100 | [diff] [blame] | 7407 | /* Grab isolated pages from freelists. */ |
Mel Gorman | bb13ffe | 2012-10-08 16:32:41 -0700 | [diff] [blame] | 7408 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7409 | if (!outer_end) { |
| 7410 | ret = -EBUSY; |
| 7411 | goto done; |
| 7412 | } |
| 7413 | |
| 7414 | /* Free head and tail (if any) */ |
| 7415 | if (start != outer_start) |
| 7416 | free_contig_range(outer_start, start - outer_start); |
| 7417 | if (end != outer_end) |
| 7418 | free_contig_range(end, outer_end - end); |
| 7419 | |
| 7420 | done: |
| 7421 | undo_isolate_page_range(pfn_max_align_down(start), |
Michal Nazarewicz | 0815f3d | 2012-04-03 15:06:15 +0200 | [diff] [blame] | 7422 | pfn_max_align_up(end), migratetype); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7423 | return ret; |
| 7424 | } |
| 7425 | |
| 7426 | void free_contig_range(unsigned long pfn, unsigned nr_pages) |
| 7427 | { |
Marek Szyprowski | bcc2b02 | 2012-12-20 15:05:18 -0800 | [diff] [blame] | 7428 | unsigned int count = 0; |
| 7429 | |
| 7430 | for (; nr_pages--; pfn++) { |
| 7431 | struct page *page = pfn_to_page(pfn); |
| 7432 | |
| 7433 | count += page_count(page) != 1; |
| 7434 | __free_page(page); |
| 7435 | } |
| 7436 | WARN(count != 0, "%d pages are still in use!\n", count); |
Michal Nazarewicz | 041d3a8 | 2011-12-29 13:09:50 +0100 | [diff] [blame] | 7437 | } |
| 7438 | #endif |
| 7439 | |
Jiang Liu | 4ed7e02 | 2012-07-31 16:43:35 -0700 | [diff] [blame] | 7440 | #ifdef CONFIG_MEMORY_HOTPLUG |
Cody P Schafer | 0a647f3 | 2013-07-03 15:01:33 -0700 | [diff] [blame] | 7441 | /* |
| 7442 | * The zone indicated has a new number of managed_pages; batch sizes and percpu |
| 7443 | * page high values need to be recalulated. |
| 7444 | */ |
Jiang Liu | 4ed7e02 | 2012-07-31 16:43:35 -0700 | [diff] [blame] | 7445 | void __meminit zone_pcp_update(struct zone *zone) |
| 7446 | { |
Cody P Schafer | 0a647f3 | 2013-07-03 15:01:33 -0700 | [diff] [blame] | 7447 | unsigned cpu; |
Cody P Schafer | c8e251f | 2013-07-03 15:01:29 -0700 | [diff] [blame] | 7448 | mutex_lock(&pcp_batch_high_lock); |
Cody P Schafer | 0a647f3 | 2013-07-03 15:01:33 -0700 | [diff] [blame] | 7449 | for_each_possible_cpu(cpu) |
Cody P Schafer | 169f6c1 | 2013-07-03 15:01:41 -0700 | [diff] [blame] | 7450 | pageset_set_high_and_batch(zone, |
| 7451 | per_cpu_ptr(zone->pageset, cpu)); |
Cody P Schafer | c8e251f | 2013-07-03 15:01:29 -0700 | [diff] [blame] | 7452 | mutex_unlock(&pcp_batch_high_lock); |
Jiang Liu | 4ed7e02 | 2012-07-31 16:43:35 -0700 | [diff] [blame] | 7453 | } |
| 7454 | #endif |
| 7455 | |
Jiang Liu | 340175b | 2012-07-31 16:43:32 -0700 | [diff] [blame] | 7456 | void zone_pcp_reset(struct zone *zone) |
| 7457 | { |
| 7458 | unsigned long flags; |
Minchan Kim | 5a88381 | 2012-10-08 16:33:39 -0700 | [diff] [blame] | 7459 | int cpu; |
| 7460 | struct per_cpu_pageset *pset; |
Jiang Liu | 340175b | 2012-07-31 16:43:32 -0700 | [diff] [blame] | 7461 | |
| 7462 | /* avoid races with drain_pages() */ |
| 7463 | local_irq_save(flags); |
| 7464 | if (zone->pageset != &boot_pageset) { |
Minchan Kim | 5a88381 | 2012-10-08 16:33:39 -0700 | [diff] [blame] | 7465 | for_each_online_cpu(cpu) { |
| 7466 | pset = per_cpu_ptr(zone->pageset, cpu); |
| 7467 | drain_zonestat(zone, pset); |
| 7468 | } |
Jiang Liu | 340175b | 2012-07-31 16:43:32 -0700 | [diff] [blame] | 7469 | free_percpu(zone->pageset); |
| 7470 | zone->pageset = &boot_pageset; |
| 7471 | } |
| 7472 | local_irq_restore(flags); |
| 7473 | } |
| 7474 | |
Wen Congyang | 6dcd73d | 2012-12-11 16:01:01 -0800 | [diff] [blame] | 7475 | #ifdef CONFIG_MEMORY_HOTREMOVE |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 7476 | /* |
Joonsoo Kim | b9eb631 | 2016-05-19 17:12:06 -0700 | [diff] [blame] | 7477 | * All pages in the range must be in a single zone and isolated |
| 7478 | * before calling this. |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 7479 | */ |
| 7480 | void |
| 7481 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
| 7482 | { |
| 7483 | struct page *page; |
| 7484 | struct zone *zone; |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 7485 | unsigned int order, i; |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 7486 | unsigned long pfn; |
| 7487 | unsigned long flags; |
| 7488 | /* find the first valid pfn */ |
| 7489 | for (pfn = start_pfn; pfn < end_pfn; pfn++) |
| 7490 | if (pfn_valid(pfn)) |
| 7491 | break; |
| 7492 | if (pfn == end_pfn) |
| 7493 | return; |
| 7494 | zone = page_zone(pfn_to_page(pfn)); |
| 7495 | spin_lock_irqsave(&zone->lock, flags); |
| 7496 | pfn = start_pfn; |
| 7497 | while (pfn < end_pfn) { |
| 7498 | if (!pfn_valid(pfn)) { |
| 7499 | pfn++; |
| 7500 | continue; |
| 7501 | } |
| 7502 | page = pfn_to_page(pfn); |
Wen Congyang | b023f46 | 2012-12-11 16:00:45 -0800 | [diff] [blame] | 7503 | /* |
| 7504 | * The HWPoisoned page may be not in buddy system, and |
| 7505 | * page_count() is not 0. |
| 7506 | */ |
| 7507 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { |
| 7508 | pfn++; |
| 7509 | SetPageReserved(page); |
| 7510 | continue; |
| 7511 | } |
| 7512 | |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 7513 | BUG_ON(page_count(page)); |
| 7514 | BUG_ON(!PageBuddy(page)); |
| 7515 | order = page_order(page); |
| 7516 | #ifdef CONFIG_DEBUG_VM |
Joe Perches | 1170532 | 2016-03-17 14:19:50 -0700 | [diff] [blame] | 7517 | pr_info("remove from free list %lx %d %lx\n", |
| 7518 | pfn, 1 << order, end_pfn); |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 7519 | #endif |
| 7520 | list_del(&page->lru); |
| 7521 | rmv_page_order(page); |
| 7522 | zone->free_area[order].nr_free--; |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 7523 | for (i = 0; i < (1 << order); i++) |
| 7524 | SetPageReserved((page+i)); |
| 7525 | pfn += (1 << order); |
| 7526 | } |
| 7527 | spin_unlock_irqrestore(&zone->lock, flags); |
| 7528 | } |
| 7529 | #endif |
Wu Fengguang | 8d22ba1 | 2009-12-16 12:19:58 +0100 | [diff] [blame] | 7530 | |
Wu Fengguang | 8d22ba1 | 2009-12-16 12:19:58 +0100 | [diff] [blame] | 7531 | bool is_free_buddy_page(struct page *page) |
| 7532 | { |
| 7533 | struct zone *zone = page_zone(page); |
| 7534 | unsigned long pfn = page_to_pfn(page); |
| 7535 | unsigned long flags; |
Mel Gorman | 7aeb09f | 2014-06-04 16:10:21 -0700 | [diff] [blame] | 7536 | unsigned int order; |
Wu Fengguang | 8d22ba1 | 2009-12-16 12:19:58 +0100 | [diff] [blame] | 7537 | |
| 7538 | spin_lock_irqsave(&zone->lock, flags); |
| 7539 | for (order = 0; order < MAX_ORDER; order++) { |
| 7540 | struct page *page_head = page - (pfn & ((1 << order) - 1)); |
| 7541 | |
| 7542 | if (PageBuddy(page_head) && page_order(page_head) >= order) |
| 7543 | break; |
| 7544 | } |
| 7545 | spin_unlock_irqrestore(&zone->lock, flags); |
| 7546 | |
| 7547 | return order < MAX_ORDER; |
| 7548 | } |