blob: ef19f22b2b7de1728fb4ed8d6451d29bb993a928 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
Mel Gorman0e1cc952015-06-30 14:57:27 -070021#include <linux/rwsem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/pagemap.h>
KOSAKI Motohiro10ed2732008-03-04 14:28:32 -080023#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/bootmem.h>
Yinghai Luedbe7d22010-08-25 13:39:16 -070025#include <linux/memblock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070027#include <linux/kernel.h>
Vegard Nossumb1eeab62008-11-25 16:55:53 +010028#include <linux/kmemcheck.h>
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -080029#include <linux/kasan.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/module.h>
31#include <linux/suspend.h>
32#include <linux/pagevec.h>
33#include <linux/blkdev.h>
34#include <linux/slab.h>
Dave Hansena238ab52011-05-24 17:12:16 -070035#include <linux/ratelimit.h>
David Rientjes5a3135c22007-10-16 23:25:53 -070036#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/notifier.h>
38#include <linux/topology.h>
39#include <linux/sysctl.h>
40#include <linux/cpu.h>
41#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070042#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <linux/nodemask.h>
44#include <linux/vmalloc.h>
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -070045#include <linux/vmstat.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080046#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070047#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070048#include <linux/sort.h>
49#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070050#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080051#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070052#include <linux/page-isolation.h>
Joonsoo Kimeefa8642014-12-12 16:55:46 -080053#include <linux/page_ext.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070054#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010055#include <linux/kmemleak.h>
Mel Gorman56de7262010-05-24 14:32:30 -070056#include <linux/compaction.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070057#include <trace/events/kmem.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070058#include <linux/prefetch.h>
Lisa Du6e543d52013-09-11 14:22:36 -070059#include <linux/mm_inline.h>
Michal Nazarewicz041d3a82011-12-29 13:09:50 +010060#include <linux/migrate.h>
Joonsoo Kime30825f2014-12-12 16:55:49 -080061#include <linux/page_ext.h>
David Rientjes949f7ec2013-04-29 15:07:48 -070062#include <linux/hugetlb.h>
Clark Williams8bd75c72013-02-07 09:47:07 -060063#include <linux/sched/rt.h>
Joonsoo Kim48c96a32014-12-12 16:56:01 -080064#include <linux/page_owner.h>
Mel Gorman0e1cc952015-06-30 14:57:27 -070065#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Jiang Liu7ee3d4e2013-07-03 15:03:41 -070067#include <asm/sections.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070069#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include "internal.h"
71
Cody P Schaferc8e251f2013-07-03 15:01:29 -070072/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
73static DEFINE_MUTEX(pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -070074#define MIN_PERCPU_PAGELIST_FRACTION (8)
Cody P Schaferc8e251f2013-07-03 15:01:29 -070075
Lee Schermerhorn72812012010-05-26 14:44:56 -070076#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
77DEFINE_PER_CPU(int, numa_node);
78EXPORT_PER_CPU_SYMBOL(numa_node);
79#endif
80
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070081#ifdef CONFIG_HAVE_MEMORYLESS_NODES
82/*
83 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
84 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
85 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
86 * defined in <linux/topology.h>.
87 */
88DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
89EXPORT_PER_CPU_SYMBOL(_numa_mem_);
Joonsoo Kimad2c8142014-10-09 15:26:13 -070090int _node_numa_mem_[MAX_NUMNODES];
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070091#endif
92
Linus Torvalds1da177e2005-04-16 15:20:36 -070093/*
Christoph Lameter13808912007-10-16 01:25:27 -070094 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 */
Christoph Lameter13808912007-10-16 01:25:27 -070096nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
97 [N_POSSIBLE] = NODE_MASK_ALL,
98 [N_ONLINE] = { { [0] = 1UL } },
99#ifndef CONFIG_NUMA
100 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
101#ifdef CONFIG_HIGHMEM
102 [N_HIGH_MEMORY] = { { [0] = 1UL } },
103#endif
Lai Jiangshan20b2f522012-12-12 13:52:00 -0800104#ifdef CONFIG_MOVABLE_NODE
105 [N_MEMORY] = { { [0] = 1UL } },
106#endif
Christoph Lameter13808912007-10-16 01:25:27 -0700107 [N_CPU] = { { [0] = 1UL } },
108#endif /* NUMA */
109};
110EXPORT_SYMBOL(node_states);
111
Jiang Liuc3d5f5f2013-07-03 15:03:14 -0700112/* Protect totalram_pages and zone->managed_pages */
113static DEFINE_SPINLOCK(managed_page_count_lock);
114
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700115unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -0700116unsigned long totalreserve_pages __read_mostly;
Pintu Kumare48322a2014-12-18 16:17:15 -0800117unsigned long totalcma_pages __read_mostly;
Johannes Weinerab8fabd2012-01-10 15:07:42 -0800118/*
119 * When calculating the number of globally allowed dirty pages, there
120 * is a certain number of per-zone reserves that should not be
121 * considered dirtyable memory. This is the sum of those reserves
122 * over all existing zones that contribute dirtyable memory.
123 */
124unsigned long dirty_balance_reserve __read_mostly;
125
Hugh Dickins1b76b022012-05-11 01:00:07 -0700126int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000127gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800129#ifdef CONFIG_PM_SLEEP
130/*
131 * The following functions are used by the suspend/hibernate code to temporarily
132 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
133 * while devices are suspended. To avoid races with the suspend/hibernate code,
134 * they should always be called with pm_mutex held (gfp_allowed_mask also should
135 * only be modified with pm_mutex held, unless the suspend/hibernate code is
136 * guaranteed not to run in parallel with that modification).
137 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100138
139static gfp_t saved_gfp_mask;
140
141void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800142{
143 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100144 if (saved_gfp_mask) {
145 gfp_allowed_mask = saved_gfp_mask;
146 saved_gfp_mask = 0;
147 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800148}
149
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100150void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800151{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800152 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100153 WARN_ON(saved_gfp_mask);
154 saved_gfp_mask = gfp_allowed_mask;
155 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800156}
Mel Gormanf90ac392012-01-10 15:07:15 -0800157
158bool pm_suspended_storage(void)
159{
160 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
161 return false;
162 return true;
163}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800164#endif /* CONFIG_PM_SLEEP */
165
Mel Gormand9c23402007-10-16 01:26:01 -0700166#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
167int pageblock_order __read_mostly;
168#endif
169
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800170static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800171
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172/*
173 * results with 256, 32 in the lowmem_reserve sysctl:
174 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
175 * 1G machine -> (16M dma, 784M normal, 224M high)
176 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
177 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
Yaowei Bai84109e12015-02-12 15:00:22 -0800178 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100179 *
180 * TBD: should special case ZONE_DMA32 machines here - in those we normally
181 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700183int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800184#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700185 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800186#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700187#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700188 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700189#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700190#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700191 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700192#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700193 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700194};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
196EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
Helge Deller15ad7cd2006-12-06 20:40:36 -0800198static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800199#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700200 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800201#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700202#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700203 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700204#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700205 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700206#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700207 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700208#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700209 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700210};
211
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212int min_free_kbytes = 1024;
Han Pingtian42aa83c2014-01-23 15:53:28 -0800213int user_min_free_kbytes = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214
Jan Beulich2c85f512009-09-21 17:03:07 -0700215static unsigned long __meminitdata nr_kernel_pages;
216static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700217static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
Tejun Heo0ee332c2011-12-08 10:22:09 -0800219#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
220static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
221static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
222static unsigned long __initdata required_kernelcore;
223static unsigned long __initdata required_movablecore;
224static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gormanc7132162006-09-27 01:49:43 -0700225
Tejun Heo0ee332c2011-12-08 10:22:09 -0800226/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
227int movable_zone;
228EXPORT_SYMBOL(movable_zone);
229#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700230
Miklos Szeredi418508c2007-05-23 13:57:55 -0700231#if MAX_NUMNODES > 1
232int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700233int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700234EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700235EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700236#endif
237
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700238int page_group_by_mobility_disabled __read_mostly;
239
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700240#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
241static inline void reset_deferred_meminit(pg_data_t *pgdat)
242{
243 pgdat->first_deferred_pfn = ULONG_MAX;
244}
245
246/* Returns true if the struct page for the pfn is uninitialised */
Mel Gorman0e1cc952015-06-30 14:57:27 -0700247static inline bool __meminit early_page_uninitialised(unsigned long pfn)
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700248{
Mel Gormanae026b22015-07-17 16:23:48 -0700249 if (pfn >= NODE_DATA(early_pfn_to_nid(pfn))->first_deferred_pfn)
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700250 return true;
251
252 return false;
253}
254
Mel Gorman7e18adb2015-06-30 14:57:05 -0700255static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
256{
257 if (pfn >= NODE_DATA(nid)->first_deferred_pfn)
258 return true;
259
260 return false;
261}
262
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700263/*
264 * Returns false when the remaining initialisation should be deferred until
265 * later in the boot cycle when it can be parallelised.
266 */
267static inline bool update_defer_init(pg_data_t *pgdat,
268 unsigned long pfn, unsigned long zone_end,
269 unsigned long *nr_initialised)
270{
271 /* Always populate low zones for address-contrained allocations */
272 if (zone_end < pgdat_end_pfn(pgdat))
273 return true;
274
275 /* Initialise at least 2G of the highest zone */
276 (*nr_initialised)++;
277 if (*nr_initialised > (2UL << (30 - PAGE_SHIFT)) &&
278 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
279 pgdat->first_deferred_pfn = pfn;
280 return false;
281 }
282
283 return true;
284}
285#else
286static inline void reset_deferred_meminit(pg_data_t *pgdat)
287{
288}
289
290static inline bool early_page_uninitialised(unsigned long pfn)
291{
292 return false;
293}
294
Mel Gorman7e18adb2015-06-30 14:57:05 -0700295static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
296{
297 return false;
298}
299
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700300static inline bool update_defer_init(pg_data_t *pgdat,
301 unsigned long pfn, unsigned long zone_end,
302 unsigned long *nr_initialised)
303{
304 return true;
305}
306#endif
307
308
Minchan Kimee6f5092012-07-31 16:43:50 -0700309void set_pageblock_migratetype(struct page *page, int migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700310{
KOSAKI Motohiro5d0f3f72013-11-12 15:08:18 -0800311 if (unlikely(page_group_by_mobility_disabled &&
312 migratetype < MIGRATE_PCPTYPES))
Mel Gorman49255c62009-06-16 15:31:58 -0700313 migratetype = MIGRATE_UNMOVABLE;
314
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700315 set_pageblock_flags_group(page, (unsigned long)migratetype,
316 PB_migrate, PB_migrate_end);
317}
318
Nick Piggin13e74442006-01-06 00:10:58 -0800319#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700320static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700322 int ret = 0;
323 unsigned seq;
324 unsigned long pfn = page_to_pfn(page);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800325 unsigned long sp, start_pfn;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700326
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700327 do {
328 seq = zone_span_seqbegin(zone);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800329 start_pfn = zone->zone_start_pfn;
330 sp = zone->spanned_pages;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800331 if (!zone_spans_pfn(zone, pfn))
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700332 ret = 1;
333 } while (zone_span_seqretry(zone, seq));
334
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800335 if (ret)
Dave Hansen613813e2014-06-04 16:07:27 -0700336 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
337 pfn, zone_to_nid(zone), zone->name,
338 start_pfn, start_pfn + sp);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800339
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700340 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700341}
342
343static int page_is_consistent(struct zone *zone, struct page *page)
344{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700345 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700346 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700348 return 0;
349
350 return 1;
351}
352/*
353 * Temporary debugging check for pages not lying within a given zone.
354 */
355static int bad_range(struct zone *zone, struct page *page)
356{
357 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700359 if (!page_is_consistent(zone, page))
360 return 1;
361
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 return 0;
363}
Nick Piggin13e74442006-01-06 00:10:58 -0800364#else
365static inline int bad_range(struct zone *zone, struct page *page)
366{
367 return 0;
368}
369#endif
370
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700371static void bad_page(struct page *page, const char *reason,
372 unsigned long bad_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800374 static unsigned long resume;
375 static unsigned long nr_shown;
376 static unsigned long nr_unshown;
377
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200378 /* Don't complain about poisoned pages */
379 if (PageHWPoison(page)) {
Mel Gorman22b751c2013-02-22 16:34:59 -0800380 page_mapcount_reset(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200381 return;
382 }
383
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800384 /*
385 * Allow a burst of 60 reports, then keep quiet for that minute;
386 * or allow a steady drip of one report per second.
387 */
388 if (nr_shown == 60) {
389 if (time_before(jiffies, resume)) {
390 nr_unshown++;
391 goto out;
392 }
393 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800394 printk(KERN_ALERT
395 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800396 nr_unshown);
397 nr_unshown = 0;
398 }
399 nr_shown = 0;
400 }
401 if (nr_shown++ == 0)
402 resume = jiffies + 60 * HZ;
403
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800404 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800405 current->comm, page_to_pfn(page));
Dave Hansenf0b791a2014-01-23 15:52:49 -0800406 dump_page_badflags(page, reason, bad_flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700407
Dave Jones4f318882011-10-31 17:07:24 -0700408 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800410out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800411 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800412 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030413 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414}
415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416/*
417 * Higher-order pages are called "compound pages". They are structured thusly:
418 *
419 * The first PAGE_SIZE page is called the "head page".
420 *
421 * The remaining PAGE_SIZE pages are called "tail pages".
422 *
Wang Sheng-Hui6416b9f2011-11-17 10:53:50 +0100423 * All pages have PG_compound set. All tail pages have their ->first_page
424 * pointing at the head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800426 * The first tail page's ->lru.next holds the address of the compound page's
427 * put_page() function. Its ->lru.prev holds the order of allocation.
428 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800430
431static void free_compound_page(struct page *page)
432{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700433 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800434}
435
Andi Kleen01ad1c02008-07-23 21:27:46 -0700436void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437{
438 int i;
439 int nr_pages = 1 << order;
440
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800441 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700442 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700443 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800444 for (i = 1; i < nr_pages; i++) {
445 struct page *p = page + i;
Youquan Song58a84aa2011-12-08 14:34:18 -0800446 set_page_count(p, 0);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700447 p->first_page = page;
David Rientjes668f9abb2014-03-03 15:38:18 -0800448 /* Make sure p->first_page is always valid for PageTail() */
449 smp_wmb();
450 __SetPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 }
452}
453
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800454#ifdef CONFIG_DEBUG_PAGEALLOC
455unsigned int _debug_guardpage_minorder;
Joonsoo Kim031bc572014-12-12 16:55:52 -0800456bool _debug_pagealloc_enabled __read_mostly;
Joonsoo Kime30825f2014-12-12 16:55:49 -0800457bool _debug_guardpage_enabled __read_mostly;
458
Joonsoo Kim031bc572014-12-12 16:55:52 -0800459static int __init early_debug_pagealloc(char *buf)
460{
461 if (!buf)
462 return -EINVAL;
463
464 if (strcmp(buf, "on") == 0)
465 _debug_pagealloc_enabled = true;
466
467 return 0;
468}
469early_param("debug_pagealloc", early_debug_pagealloc);
470
Joonsoo Kime30825f2014-12-12 16:55:49 -0800471static bool need_debug_guardpage(void)
472{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800473 /* If we don't use debug_pagealloc, we don't need guard page */
474 if (!debug_pagealloc_enabled())
475 return false;
476
Joonsoo Kime30825f2014-12-12 16:55:49 -0800477 return true;
478}
479
480static void init_debug_guardpage(void)
481{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800482 if (!debug_pagealloc_enabled())
483 return;
484
Joonsoo Kime30825f2014-12-12 16:55:49 -0800485 _debug_guardpage_enabled = true;
486}
487
488struct page_ext_operations debug_guardpage_ops = {
489 .need = need_debug_guardpage,
490 .init = init_debug_guardpage,
491};
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800492
493static int __init debug_guardpage_minorder_setup(char *buf)
494{
495 unsigned long res;
496
497 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
498 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
499 return 0;
500 }
501 _debug_guardpage_minorder = res;
502 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
503 return 0;
504}
505__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
506
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800507static inline void set_page_guard(struct zone *zone, struct page *page,
508 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800509{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800510 struct page_ext *page_ext;
511
512 if (!debug_guardpage_enabled())
513 return;
514
515 page_ext = lookup_page_ext(page);
516 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
517
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800518 INIT_LIST_HEAD(&page->lru);
519 set_page_private(page, order);
520 /* Guard pages are not available for any usage */
521 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800522}
523
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800524static inline void clear_page_guard(struct zone *zone, struct page *page,
525 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800526{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800527 struct page_ext *page_ext;
528
529 if (!debug_guardpage_enabled())
530 return;
531
532 page_ext = lookup_page_ext(page);
533 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
534
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800535 set_page_private(page, 0);
536 if (!is_migrate_isolate(migratetype))
537 __mod_zone_freepage_state(zone, (1 << order), migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800538}
539#else
Joonsoo Kime30825f2014-12-12 16:55:49 -0800540struct page_ext_operations debug_guardpage_ops = { NULL, };
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800541static inline void set_page_guard(struct zone *zone, struct page *page,
542 unsigned int order, int migratetype) {}
543static inline void clear_page_guard(struct zone *zone, struct page *page,
544 unsigned int order, int migratetype) {}
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800545#endif
546
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700547static inline void set_page_order(struct page *page, unsigned int order)
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700548{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700549 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000550 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551}
552
553static inline void rmv_page_order(struct page *page)
554{
Nick Piggin676165a2006-04-10 11:21:48 +1000555 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700556 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557}
558
559/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 * This function checks whether a page is free && is the buddy
561 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800562 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000563 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700564 * (c) a page and its buddy have the same order &&
565 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 *
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700567 * For recording whether a page is in the buddy system, we set ->_mapcount
568 * PAGE_BUDDY_MAPCOUNT_VALUE.
569 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
570 * serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000571 *
572 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700574static inline int page_is_buddy(struct page *page, struct page *buddy,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700575 unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700577 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800578 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800579
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800580 if (page_is_guard(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700581 if (page_zone_id(page) != page_zone_id(buddy))
582 return 0;
583
Weijie Yang4c5018c2015-02-10 14:11:39 -0800584 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
585
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800586 return 1;
587 }
588
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700589 if (PageBuddy(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700590 /*
591 * zone check is done late to avoid uselessly
592 * calculating zone/node ids for pages that could
593 * never merge.
594 */
595 if (page_zone_id(page) != page_zone_id(buddy))
596 return 0;
597
Weijie Yang4c5018c2015-02-10 14:11:39 -0800598 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
599
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700600 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000601 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700602 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603}
604
605/*
606 * Freeing function for a buddy system allocator.
607 *
608 * The concept of a buddy system is to maintain direct-mapped table
609 * (containing bit values) for memory blocks of various "orders".
610 * The bottom level table contains the map for the smallest allocatable
611 * units of memory (here, pages), and each level above it describes
612 * pairs of units from the levels below, hence, "buddies".
613 * At a high level, all that happens here is marking the table entry
614 * at the bottom level available, and propagating the changes upward
615 * as necessary, plus some accounting needed to play nicely with other
616 * parts of the VM system.
617 * At each level, we keep a list of pages, which are heads of continuous
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700618 * free pages of length of (1 << order) and marked with _mapcount
619 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
620 * field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100622 * other. That is, if we allocate a small block, and both were
623 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100625 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100627 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 */
629
Nick Piggin48db57f2006-01-08 01:00:42 -0800630static inline void __free_one_page(struct page *page,
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700631 unsigned long pfn,
Mel Gormaned0ae212009-06-16 15:32:07 -0700632 struct zone *zone, unsigned int order,
633 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
635 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700636 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800637 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700638 struct page *buddy;
Joonsoo Kim3c605092014-11-13 15:19:21 -0800639 int max_order = MAX_ORDER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640
Cody P Schaferd29bb972013-02-22 16:35:25 -0800641 VM_BUG_ON(!zone_is_initialized(zone));
Kirill A. Shutemov6e9f0d52015-02-11 15:25:50 -0800642 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
Mel Gormaned0ae212009-06-16 15:32:07 -0700644 VM_BUG_ON(migratetype == -1);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800645 if (is_migrate_isolate(migratetype)) {
646 /*
647 * We restrict max order of merging to prevent merge
648 * between freepages on isolate pageblock and normal
649 * pageblock. Without this, pageblock isolation
650 * could cause incorrect freepage accounting.
651 */
652 max_order = min(MAX_ORDER, pageblock_order + 1);
653 } else {
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800654 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800655 }
Mel Gormaned0ae212009-06-16 15:32:07 -0700656
Joonsoo Kim3c605092014-11-13 15:19:21 -0800657 page_idx = pfn & ((1 << max_order) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
Sasha Levin309381fea2014-01-23 15:52:54 -0800659 VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
660 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
Joonsoo Kim3c605092014-11-13 15:19:21 -0800662 while (order < max_order - 1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800663 buddy_idx = __find_buddy_index(page_idx, order);
664 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700665 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700666 break;
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800667 /*
668 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
669 * merge with it and move up one order.
670 */
671 if (page_is_guard(buddy)) {
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800672 clear_page_guard(zone, buddy, order, migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800673 } else {
674 list_del(&buddy->lru);
675 zone->free_area[order].nr_free--;
676 rmv_page_order(buddy);
677 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800678 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 page = page + (combined_idx - page_idx);
680 page_idx = combined_idx;
681 order++;
682 }
683 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700684
685 /*
686 * If this is not the largest possible page, check if the buddy
687 * of the next-highest order is free. If it is, it's possible
688 * that pages are being freed that will coalesce soon. In case,
689 * that is happening, add the free page to the tail of the list
690 * so it's less likely to be used soon and more likely to be merged
691 * as a higher order page
692 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700693 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700694 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800695 combined_idx = buddy_idx & page_idx;
696 higher_page = page + (combined_idx - page_idx);
697 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700698 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700699 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
700 list_add_tail(&page->lru,
701 &zone->free_area[order].free_list[migratetype]);
702 goto out;
703 }
704 }
705
706 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
707out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 zone->free_area[order].nr_free++;
709}
710
Nick Piggin224abf92006-01-06 00:11:11 -0800711static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700713 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800714 unsigned long bad_flags = 0;
715
716 if (unlikely(page_mapcount(page)))
717 bad_reason = "nonzero mapcount";
718 if (unlikely(page->mapping != NULL))
719 bad_reason = "non-NULL mapping";
720 if (unlikely(atomic_read(&page->_count) != 0))
721 bad_reason = "nonzero _count";
722 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
723 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
724 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
725 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -0800726#ifdef CONFIG_MEMCG
727 if (unlikely(page->mem_cgroup))
728 bad_reason = "page still charged to cgroup";
729#endif
Dave Hansenf0b791a2014-01-23 15:52:49 -0800730 if (unlikely(bad_reason)) {
731 bad_page(page, bad_reason, bad_flags);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800732 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800733 }
Peter Zijlstra90572892013-10-07 11:29:20 +0100734 page_cpupid_reset_last(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800735 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
736 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
737 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738}
739
740/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700741 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700743 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 *
745 * If the zone was previously in an "all pages pinned" state then look to
746 * see if this freeing clears that state.
747 *
748 * And clear the zone's pages_scanned counter, to hold off the "all pages are
749 * pinned" detection logic.
750 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700751static void free_pcppages_bulk(struct zone *zone, int count,
752 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700754 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700755 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700756 int to_free = count;
Mel Gorman0d5d8232014-08-06 16:07:16 -0700757 unsigned long nr_scanned;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700758
Nick Pigginc54ad302006-01-06 00:10:56 -0800759 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700760 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
761 if (nr_scanned)
762 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700763
Mel Gorman72853e22010-09-09 16:38:16 -0700764 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800765 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700766 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800767
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700768 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700769 * Remove pages from lists in a round-robin fashion. A
770 * batch_free count is maintained that is incremented when an
771 * empty list is encountered. This is so more pages are freed
772 * off fuller lists instead of spinning excessively around empty
773 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700774 */
775 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700776 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700777 if (++migratetype == MIGRATE_PCPTYPES)
778 migratetype = 0;
779 list = &pcp->lists[migratetype];
780 } while (list_empty(list));
781
Namhyung Kim1d168712011-03-22 16:32:45 -0700782 /* This is the only non-empty list. Free them all. */
783 if (batch_free == MIGRATE_PCPTYPES)
784 batch_free = to_free;
785
Mel Gormana6f9edd2009-09-21 17:03:20 -0700786 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700787 int mt; /* migratetype of the to-be-freed page */
788
Mel Gormana6f9edd2009-09-21 17:03:20 -0700789 page = list_entry(list->prev, struct page, lru);
790 /* must delete as __free_one_page list manipulates */
791 list_del(&page->lru);
Minchan Kimb12c4ad2012-10-08 16:32:08 -0700792 mt = get_freepage_migratetype(page);
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800793 if (unlikely(has_isolate_pageblock(zone)))
Joonsoo Kim51bb1a42014-11-13 15:19:14 -0800794 mt = get_pageblock_migratetype(page);
Joonsoo Kim51bb1a42014-11-13 15:19:14 -0800795
Hugh Dickinsa7016232010-01-29 17:46:34 +0000796 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700797 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700798 trace_mm_page_pcpu_drain(page, 0, mt);
Mel Gorman72853e22010-09-09 16:38:16 -0700799 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800801 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802}
803
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700804static void free_one_page(struct zone *zone,
805 struct page *page, unsigned long pfn,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700806 unsigned int order,
Mel Gormaned0ae212009-06-16 15:32:07 -0700807 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800808{
Mel Gorman0d5d8232014-08-06 16:07:16 -0700809 unsigned long nr_scanned;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700810 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700811 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
812 if (nr_scanned)
813 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700814
Joonsoo Kimad53f922014-11-13 15:19:11 -0800815 if (unlikely(has_isolate_pageblock(zone) ||
816 is_migrate_isolate(migratetype))) {
817 migratetype = get_pfnblock_migratetype(page, pfn);
Joonsoo Kimad53f922014-11-13 15:19:11 -0800818 }
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700819 __free_one_page(page, pfn, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700820 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800821}
822
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800823static int free_tail_pages_check(struct page *head_page, struct page *page)
824{
825 if (!IS_ENABLED(CONFIG_DEBUG_VM))
826 return 0;
827 if (unlikely(!PageTail(page))) {
828 bad_page(page, "PageTail not set", 0);
829 return 1;
830 }
831 if (unlikely(page->first_page != head_page)) {
832 bad_page(page, "first_page not consistent", 0);
833 return 1;
834 }
835 return 0;
836}
837
Robin Holt1e8ce832015-06-30 14:56:45 -0700838static void __meminit __init_single_page(struct page *page, unsigned long pfn,
839 unsigned long zone, int nid)
840{
Robin Holt1e8ce832015-06-30 14:56:45 -0700841 set_page_links(page, zone, nid, pfn);
Robin Holt1e8ce832015-06-30 14:56:45 -0700842 init_page_count(page);
843 page_mapcount_reset(page);
844 page_cpupid_reset_last(page);
Robin Holt1e8ce832015-06-30 14:56:45 -0700845
Robin Holt1e8ce832015-06-30 14:56:45 -0700846 INIT_LIST_HEAD(&page->lru);
847#ifdef WANT_PAGE_VIRTUAL
848 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
849 if (!is_highmem_idx(zone))
850 set_page_address(page, __va(pfn << PAGE_SHIFT));
851#endif
852}
853
854static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
855 int nid)
856{
857 return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
858}
859
Mel Gorman7e18adb2015-06-30 14:57:05 -0700860#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
861static void init_reserved_page(unsigned long pfn)
862{
863 pg_data_t *pgdat;
864 int nid, zid;
865
866 if (!early_page_uninitialised(pfn))
867 return;
868
869 nid = early_pfn_to_nid(pfn);
870 pgdat = NODE_DATA(nid);
871
872 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
873 struct zone *zone = &pgdat->node_zones[zid];
874
875 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
876 break;
877 }
878 __init_single_pfn(pfn, zid, nid);
879}
880#else
881static inline void init_reserved_page(unsigned long pfn)
882{
883}
884#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
885
Nathan Zimmer92923ca2015-06-30 14:56:48 -0700886/*
887 * Initialised pages do not have PageReserved set. This function is
888 * called for each range allocated by the bootmem allocator and
889 * marks the pages PageReserved. The remaining valid pages are later
890 * sent to the buddy page allocator.
891 */
Mel Gorman7e18adb2015-06-30 14:57:05 -0700892void __meminit reserve_bootmem_region(unsigned long start, unsigned long end)
Nathan Zimmer92923ca2015-06-30 14:56:48 -0700893{
894 unsigned long start_pfn = PFN_DOWN(start);
895 unsigned long end_pfn = PFN_UP(end);
896
Mel Gorman7e18adb2015-06-30 14:57:05 -0700897 for (; start_pfn < end_pfn; start_pfn++) {
898 if (pfn_valid(start_pfn)) {
899 struct page *page = pfn_to_page(start_pfn);
900
901 init_reserved_page(start_pfn);
902 SetPageReserved(page);
903 }
904 }
Nathan Zimmer92923ca2015-06-30 14:56:48 -0700905}
906
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700907static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908{
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800909 bool compound = PageCompound(page);
910 int i, bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
Yu Zhaoab1f3062014-12-10 15:43:17 -0800912 VM_BUG_ON_PAGE(PageTail(page), page);
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800913 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
Yu Zhaoab1f3062014-12-10 15:43:17 -0800914
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -0800915 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100916 kmemcheck_free_shadow(page, order);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -0800917 kasan_free_pages(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100918
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800919 if (PageAnon(page))
920 page->mapping = NULL;
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800921 bad += free_pages_check(page);
922 for (i = 1; i < (1 << order); i++) {
923 if (compound)
924 bad += free_tail_pages_check(page, page + i);
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800925 bad += free_pages_check(page + i);
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800926 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800927 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700928 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800929
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800930 reset_page_owner(page, order);
931
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700932 if (!PageHighMem(page)) {
Pintu Kumarb8af2942013-09-11 14:20:34 -0700933 debug_check_no_locks_freed(page_address(page),
934 PAGE_SIZE << order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700935 debug_check_no_obj_freed(page_address(page),
936 PAGE_SIZE << order);
937 }
Nick Piggindafb1362006-10-11 01:21:30 -0700938 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800939 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700940
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700941 return true;
942}
943
944static void __free_pages_ok(struct page *page, unsigned int order)
945{
946 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -0700947 int migratetype;
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700948 unsigned long pfn = page_to_pfn(page);
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700949
950 if (!free_pages_prepare(page, order))
951 return;
952
Mel Gormancfc47a22014-06-04 16:10:19 -0700953 migratetype = get_pfnblock_migratetype(page, pfn);
Nick Pigginc54ad302006-01-06 00:10:56 -0800954 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700955 __count_vm_events(PGFREE, 1 << order);
Minchan Kim95e34412012-10-08 16:32:11 -0700956 set_freepage_migratetype(page, migratetype);
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700957 free_one_page(page_zone(page), page, pfn, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -0800958 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959}
960
Mel Gorman0e1cc952015-06-30 14:57:27 -0700961static void __init __free_pages_boot_core(struct page *page,
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700962 unsigned long pfn, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800963{
Johannes Weinerc3993072012-01-10 15:08:10 -0800964 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700965 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -0800966 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -0800967
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700968 prefetchw(p);
969 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
970 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -0800971 __ClearPageReserved(p);
972 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -0800973 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700974 __ClearPageReserved(p);
975 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -0800976
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700977 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -0800978 set_page_refcounted(page);
979 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800980}
981
Mel Gorman75a592a2015-06-30 14:56:59 -0700982#if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
983 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
984/* Only safe to use early in boot when initialisation is single-threaded */
985static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
986
987int __meminit early_pfn_to_nid(unsigned long pfn)
988{
989 int nid;
990
991 /* The system will behave unpredictably otherwise */
992 BUG_ON(system_state != SYSTEM_BOOTING);
993
994 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
995 if (nid >= 0)
996 return nid;
997 /* just returns 0 */
998 return 0;
999}
1000#endif
1001
1002#ifdef CONFIG_NODES_SPAN_OTHER_NODES
1003static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1004 struct mminit_pfnnid_cache *state)
1005{
1006 int nid;
1007
1008 nid = __early_pfn_to_nid(pfn, state);
1009 if (nid >= 0 && nid != node)
1010 return false;
1011 return true;
1012}
1013
1014/* Only safe to use early in boot when initialisation is single-threaded */
1015static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1016{
1017 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1018}
1019
1020#else
1021
1022static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1023{
1024 return true;
1025}
1026static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1027 struct mminit_pfnnid_cache *state)
1028{
1029 return true;
1030}
1031#endif
1032
1033
Mel Gorman0e1cc952015-06-30 14:57:27 -07001034void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001035 unsigned int order)
1036{
1037 if (early_page_uninitialised(pfn))
1038 return;
1039 return __free_pages_boot_core(page, pfn, order);
1040}
1041
Mel Gorman7e18adb2015-06-30 14:57:05 -07001042#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
Mel Gorman0e1cc952015-06-30 14:57:27 -07001043static void __init deferred_free_range(struct page *page,
Mel Gormana4de83d2015-06-30 14:57:16 -07001044 unsigned long pfn, int nr_pages)
1045{
1046 int i;
1047
1048 if (!page)
1049 return;
1050
1051 /* Free a large naturally-aligned chunk if possible */
1052 if (nr_pages == MAX_ORDER_NR_PAGES &&
1053 (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
Mel Gormanac5d2532015-06-30 14:57:20 -07001054 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormana4de83d2015-06-30 14:57:16 -07001055 __free_pages_boot_core(page, pfn, MAX_ORDER-1);
1056 return;
1057 }
1058
1059 for (i = 0; i < nr_pages; i++, page++, pfn++)
1060 __free_pages_boot_core(page, pfn, 0);
1061}
1062
Mel Gorman0e1cc952015-06-30 14:57:27 -07001063static __initdata DECLARE_RWSEM(pgdat_init_rwsem);
1064
Mel Gorman7e18adb2015-06-30 14:57:05 -07001065/* Initialise remaining memory on a node */
Mel Gorman0e1cc952015-06-30 14:57:27 -07001066static int __init deferred_init_memmap(void *data)
Mel Gorman7e18adb2015-06-30 14:57:05 -07001067{
Mel Gorman0e1cc952015-06-30 14:57:27 -07001068 pg_data_t *pgdat = data;
1069 int nid = pgdat->node_id;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001070 struct mminit_pfnnid_cache nid_init_state = { };
1071 unsigned long start = jiffies;
1072 unsigned long nr_pages = 0;
1073 unsigned long walk_start, walk_end;
1074 int i, zid;
1075 struct zone *zone;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001076 unsigned long first_init_pfn = pgdat->first_deferred_pfn;
Mel Gorman0e1cc952015-06-30 14:57:27 -07001077 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001078
Mel Gorman0e1cc952015-06-30 14:57:27 -07001079 if (first_init_pfn == ULONG_MAX) {
1080 up_read(&pgdat_init_rwsem);
1081 return 0;
1082 }
1083
1084 /* Bind memory initialisation thread to a local node if possible */
1085 if (!cpumask_empty(cpumask))
1086 set_cpus_allowed_ptr(current, cpumask);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001087
1088 /* Sanity check boundaries */
1089 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1090 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1091 pgdat->first_deferred_pfn = ULONG_MAX;
1092
1093 /* Only the highest zone is deferred so find it */
1094 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1095 zone = pgdat->node_zones + zid;
1096 if (first_init_pfn < zone_end_pfn(zone))
1097 break;
1098 }
1099
1100 for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1101 unsigned long pfn, end_pfn;
Mel Gorman54608c32015-06-30 14:57:09 -07001102 struct page *page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001103 struct page *free_base_page = NULL;
1104 unsigned long free_base_pfn = 0;
1105 int nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001106
1107 end_pfn = min(walk_end, zone_end_pfn(zone));
1108 pfn = first_init_pfn;
1109 if (pfn < walk_start)
1110 pfn = walk_start;
1111 if (pfn < zone->zone_start_pfn)
1112 pfn = zone->zone_start_pfn;
1113
1114 for (; pfn < end_pfn; pfn++) {
Mel Gorman54608c32015-06-30 14:57:09 -07001115 if (!pfn_valid_within(pfn))
Mel Gormana4de83d2015-06-30 14:57:16 -07001116 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001117
Mel Gorman54608c32015-06-30 14:57:09 -07001118 /*
1119 * Ensure pfn_valid is checked every
1120 * MAX_ORDER_NR_PAGES for memory holes
1121 */
1122 if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1123 if (!pfn_valid(pfn)) {
1124 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001125 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001126 }
1127 }
1128
1129 if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1130 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001131 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001132 }
1133
1134 /* Minimise pfn page lookups and scheduler checks */
1135 if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1136 page++;
1137 } else {
Mel Gormana4de83d2015-06-30 14:57:16 -07001138 nr_pages += nr_to_free;
1139 deferred_free_range(free_base_page,
1140 free_base_pfn, nr_to_free);
1141 free_base_page = NULL;
1142 free_base_pfn = nr_to_free = 0;
1143
Mel Gorman54608c32015-06-30 14:57:09 -07001144 page = pfn_to_page(pfn);
1145 cond_resched();
1146 }
Mel Gorman7e18adb2015-06-30 14:57:05 -07001147
1148 if (page->flags) {
1149 VM_BUG_ON(page_zone(page) != zone);
Mel Gormana4de83d2015-06-30 14:57:16 -07001150 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001151 }
1152
1153 __init_single_page(page, pfn, zid, nid);
Mel Gormana4de83d2015-06-30 14:57:16 -07001154 if (!free_base_page) {
1155 free_base_page = page;
1156 free_base_pfn = pfn;
1157 nr_to_free = 0;
1158 }
1159 nr_to_free++;
1160
1161 /* Where possible, batch up pages for a single free */
1162 continue;
1163free_range:
1164 /* Free the current block of pages to allocator */
1165 nr_pages += nr_to_free;
1166 deferred_free_range(free_base_page, free_base_pfn,
1167 nr_to_free);
1168 free_base_page = NULL;
1169 free_base_pfn = nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001170 }
Mel Gormana4de83d2015-06-30 14:57:16 -07001171
Mel Gorman7e18adb2015-06-30 14:57:05 -07001172 first_init_pfn = max(end_pfn, first_init_pfn);
1173 }
1174
1175 /* Sanity check that the next zone really is unpopulated */
1176 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1177
Mel Gorman0e1cc952015-06-30 14:57:27 -07001178 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
Mel Gorman7e18adb2015-06-30 14:57:05 -07001179 jiffies_to_msecs(jiffies - start));
Mel Gorman0e1cc952015-06-30 14:57:27 -07001180 up_read(&pgdat_init_rwsem);
1181 return 0;
1182}
1183
1184void __init page_alloc_init_late(void)
1185{
1186 int nid;
1187
1188 for_each_node_state(nid, N_MEMORY) {
1189 down_read(&pgdat_init_rwsem);
1190 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1191 }
1192
1193 /* Block until all are initialised */
1194 down_write(&pgdat_init_rwsem);
1195 up_write(&pgdat_init_rwsem);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001196}
1197#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1198
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001199#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +08001200/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001201void __init init_cma_reserved_pageblock(struct page *page)
1202{
1203 unsigned i = pageblock_nr_pages;
1204 struct page *p = page;
1205
1206 do {
1207 __ClearPageReserved(p);
1208 set_page_count(p, 0);
1209 } while (++p, --i);
1210
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001211 set_pageblock_migratetype(page, MIGRATE_CMA);
Michal Nazarewiczdc783272014-07-02 15:22:35 -07001212
1213 if (pageblock_order >= MAX_ORDER) {
1214 i = pageblock_nr_pages;
1215 p = page;
1216 do {
1217 set_page_refcounted(p);
1218 __free_pages(p, MAX_ORDER - 1);
1219 p += MAX_ORDER_NR_PAGES;
1220 } while (i -= MAX_ORDER_NR_PAGES);
1221 } else {
1222 set_page_refcounted(page);
1223 __free_pages(page, pageblock_order);
1224 }
1225
Jiang Liu3dcc0572013-07-03 15:03:21 -07001226 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001227}
1228#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
1230/*
1231 * The order of subdivision here is critical for the IO subsystem.
1232 * Please do not alter this order without good reasons and regression
1233 * testing. Specifically, as large blocks of memory are subdivided,
1234 * the order in which smaller blocks are delivered depends on the order
1235 * they're subdivided in this function. This is the primary factor
1236 * influencing the order in which pages are delivered to the IO
1237 * subsystem according to empirical testing, and this is also justified
1238 * by considering the behavior of a buddy system containing a single
1239 * large block of memory acted on by a series of small allocations.
1240 * This behavior is a critical factor in sglist merging's success.
1241 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +01001242 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 */
Nick Piggin085cc7d2006-01-06 00:11:01 -08001244static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001245 int low, int high, struct free_area *area,
1246 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247{
1248 unsigned long size = 1 << high;
1249
1250 while (high > low) {
1251 area--;
1252 high--;
1253 size >>= 1;
Sasha Levin309381fea2014-01-23 15:52:54 -08001254 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08001255
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001256 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
Joonsoo Kime30825f2014-12-12 16:55:49 -08001257 debug_guardpage_enabled() &&
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001258 high < debug_guardpage_minorder()) {
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08001259 /*
1260 * Mark as guard pages (or page), that will allow to
1261 * merge back to allocator when buddy will be freed.
1262 * Corresponding page table entries will not be touched,
1263 * pages will stay not present in virtual address space
1264 */
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001265 set_page_guard(zone, &page[size], high, migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08001266 continue;
1267 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001268 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 area->nr_free++;
1270 set_page_order(&page[size], high);
1271 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272}
1273
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274/*
1275 * This page is about to be returned from the page allocator
1276 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001277static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -07001279 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -08001280 unsigned long bad_flags = 0;
1281
1282 if (unlikely(page_mapcount(page)))
1283 bad_reason = "nonzero mapcount";
1284 if (unlikely(page->mapping != NULL))
1285 bad_reason = "non-NULL mapping";
1286 if (unlikely(atomic_read(&page->_count) != 0))
1287 bad_reason = "nonzero _count";
1288 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1289 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1290 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1291 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -08001292#ifdef CONFIG_MEMCG
1293 if (unlikely(page->mem_cgroup))
1294 bad_reason = "page still charged to cgroup";
1295#endif
Dave Hansenf0b791a2014-01-23 15:52:49 -08001296 if (unlikely(bad_reason)) {
1297 bad_page(page, bad_reason, bad_flags);
Hugh Dickins689bceb2005-11-21 21:32:20 -08001298 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -08001299 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001300 return 0;
1301}
1302
Vlastimil Babka75379192015-02-11 15:25:38 -08001303static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1304 int alloc_flags)
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001305{
1306 int i;
1307
1308 for (i = 0; i < (1 << order); i++) {
1309 struct page *p = page + i;
1310 if (unlikely(check_new_page(p)))
1311 return 1;
1312 }
Hugh Dickins689bceb2005-11-21 21:32:20 -08001313
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001314 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -08001315 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -08001316
1317 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 kernel_map_pages(page, 1 << order, 1);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -08001319 kasan_alloc_pages(page, order);
Nick Piggin17cf4402006-03-22 00:08:41 -08001320
1321 if (gfp_flags & __GFP_ZERO)
Anisse Astierf4d28972015-06-24 16:56:36 -07001322 for (i = 0; i < (1 << order); i++)
1323 clear_highpage(page + i);
Nick Piggin17cf4402006-03-22 00:08:41 -08001324
1325 if (order && (gfp_flags & __GFP_COMP))
1326 prep_compound_page(page, order);
1327
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001328 set_page_owner(page, order, gfp_flags);
1329
Vlastimil Babka75379192015-02-11 15:25:38 -08001330 /*
1331 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was necessary to
1332 * allocate the page. The expectation is that the caller is taking
1333 * steps that will free more memory. The caller should avoid the page
1334 * being used for !PFMEMALLOC purposes.
1335 */
1336 page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
1337
Hugh Dickins689bceb2005-11-21 21:32:20 -08001338 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339}
1340
Mel Gorman56fd56b2007-10-16 01:25:58 -07001341/*
1342 * Go through the free lists for the given migratetype and remove
1343 * the smallest available page from the freelists
1344 */
Mel Gorman728ec982009-06-16 15:32:04 -07001345static inline
1346struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -07001347 int migratetype)
1348{
1349 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -07001350 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001351 struct page *page;
1352
1353 /* Find a page of the appropriate size in the preferred list */
1354 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1355 area = &(zone->free_area[current_order]);
1356 if (list_empty(&area->free_list[migratetype]))
1357 continue;
1358
1359 page = list_entry(area->free_list[migratetype].next,
1360 struct page, lru);
1361 list_del(&page->lru);
1362 rmv_page_order(page);
1363 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001364 expand(zone, page, order, current_order, area, migratetype);
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001365 set_freepage_migratetype(page, migratetype);
Mel Gorman56fd56b2007-10-16 01:25:58 -07001366 return page;
1367 }
1368
1369 return NULL;
1370}
1371
1372
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001373/*
1374 * This array describes the order lists are fallen back to when
1375 * the free lists for the desirable migrate type are depleted
1376 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001377static int fallbacks[MIGRATE_TYPES][4] = {
1378 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
1379 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001380 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
Joonsoo Kimdc676472015-04-14 15:45:15 -07001381#ifdef CONFIG_CMA
1382 [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001383#endif
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001384 [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001385#ifdef CONFIG_MEMORY_ISOLATION
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001386 [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001387#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001388};
1389
Joonsoo Kimdc676472015-04-14 15:45:15 -07001390#ifdef CONFIG_CMA
1391static struct page *__rmqueue_cma_fallback(struct zone *zone,
1392 unsigned int order)
1393{
1394 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1395}
1396#else
1397static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1398 unsigned int order) { return NULL; }
1399#endif
1400
Mel Gormanc361be52007-10-16 01:25:51 -07001401/*
1402 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -07001403 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -07001404 * boundary. If alignment is required, use move_freepages_block()
1405 */
Minchan Kim435b4052012-10-08 16:32:16 -07001406int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -07001407 struct page *start_page, struct page *end_page,
1408 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001409{
1410 struct page *page;
1411 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -07001412 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -07001413
1414#ifndef CONFIG_HOLES_IN_ZONE
1415 /*
1416 * page_zone is not safe to call in this context when
1417 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1418 * anyway as we check zone boundaries in move_freepages_block().
1419 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -07001420 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -07001421 */
Mel Gorman97ee4ba2014-10-09 15:28:28 -07001422 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
Mel Gormanc361be52007-10-16 01:25:51 -07001423#endif
1424
1425 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -07001426 /* Make sure we are not inadvertently changing nodes */
Sasha Levin309381fea2014-01-23 15:52:54 -08001427 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
Adam Litke344c7902008-09-02 14:35:38 -07001428
Mel Gormanc361be52007-10-16 01:25:51 -07001429 if (!pfn_valid_within(page_to_pfn(page))) {
1430 page++;
1431 continue;
1432 }
1433
1434 if (!PageBuddy(page)) {
1435 page++;
1436 continue;
1437 }
1438
1439 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001440 list_move(&page->lru,
1441 &zone->free_area[order].free_list[migratetype]);
Minchan Kim95e34412012-10-08 16:32:11 -07001442 set_freepage_migratetype(page, migratetype);
Mel Gormanc361be52007-10-16 01:25:51 -07001443 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001444 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001445 }
1446
Mel Gormand1003132007-10-16 01:26:00 -07001447 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001448}
1449
Minchan Kimee6f5092012-07-31 16:43:50 -07001450int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001451 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001452{
1453 unsigned long start_pfn, end_pfn;
1454 struct page *start_page, *end_page;
1455
1456 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001457 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001458 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001459 end_page = start_page + pageblock_nr_pages - 1;
1460 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001461
1462 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001463 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001464 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001465 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001466 return 0;
1467
1468 return move_freepages(zone, start_page, end_page, migratetype);
1469}
1470
Mel Gorman2f66a682009-09-21 17:02:31 -07001471static void change_pageblock_range(struct page *pageblock_page,
1472 int start_order, int migratetype)
1473{
1474 int nr_pageblocks = 1 << (start_order - pageblock_order);
1475
1476 while (nr_pageblocks--) {
1477 set_pageblock_migratetype(pageblock_page, migratetype);
1478 pageblock_page += pageblock_nr_pages;
1479 }
1480}
1481
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001482/*
Vlastimil Babka9c0415e2015-02-11 15:28:21 -08001483 * When we are falling back to another migratetype during allocation, try to
1484 * steal extra free pages from the same pageblocks to satisfy further
1485 * allocations, instead of polluting multiple pageblocks.
1486 *
1487 * If we are stealing a relatively large buddy page, it is likely there will
1488 * be more free pages in the pageblock, so try to steal them all. For
1489 * reclaimable and unmovable allocations, we steal regardless of page size,
1490 * as fragmentation caused by those allocations polluting movable pageblocks
1491 * is worse than movable allocations stealing from unmovable and reclaimable
1492 * pageblocks.
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001493 */
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001494static bool can_steal_fallback(unsigned int order, int start_mt)
1495{
1496 /*
1497 * Leaving this order check is intended, although there is
1498 * relaxed order check in next check. The reason is that
1499 * we can actually steal whole pageblock if this condition met,
1500 * but, below check doesn't guarantee it and that is just heuristic
1501 * so could be changed anytime.
1502 */
1503 if (order >= pageblock_order)
1504 return true;
1505
1506 if (order >= pageblock_order / 2 ||
1507 start_mt == MIGRATE_RECLAIMABLE ||
1508 start_mt == MIGRATE_UNMOVABLE ||
1509 page_group_by_mobility_disabled)
1510 return true;
1511
1512 return false;
1513}
1514
1515/*
1516 * This function implements actual steal behaviour. If order is large enough,
1517 * we can steal whole pageblock. If not, we first move freepages in this
1518 * pageblock and check whether half of pages are moved or not. If half of
1519 * pages are moved, we can change migratetype of pageblock and permanently
1520 * use it's pages as requested migratetype in the future.
1521 */
1522static void steal_suitable_fallback(struct zone *zone, struct page *page,
1523 int start_type)
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001524{
1525 int current_order = page_order(page);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001526 int pages;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001527
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001528 /* Take ownership for orders >= pageblock_order */
1529 if (current_order >= pageblock_order) {
1530 change_pageblock_range(page, current_order, start_type);
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001531 return;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001532 }
1533
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001534 pages = move_freepages_block(zone, page, start_type);
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001535
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001536 /* Claim the whole block if over half of it is free */
1537 if (pages >= (1 << (pageblock_order-1)) ||
1538 page_group_by_mobility_disabled)
1539 set_pageblock_migratetype(page, start_type);
1540}
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001541
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001542/*
1543 * Check whether there is a suitable fallback freepage with requested order.
1544 * If only_stealable is true, this function returns fallback_mt only if
1545 * we can steal other freepages all together. This would help to reduce
1546 * fragmentation due to mixed migratetype pages in one pageblock.
1547 */
1548int find_suitable_fallback(struct free_area *area, unsigned int order,
1549 int migratetype, bool only_stealable, bool *can_steal)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001550{
1551 int i;
1552 int fallback_mt;
1553
1554 if (area->nr_free == 0)
1555 return -1;
1556
1557 *can_steal = false;
1558 for (i = 0;; i++) {
1559 fallback_mt = fallbacks[migratetype][i];
1560 if (fallback_mt == MIGRATE_RESERVE)
1561 break;
1562
1563 if (list_empty(&area->free_list[fallback_mt]))
1564 continue;
1565
1566 if (can_steal_fallback(order, migratetype))
1567 *can_steal = true;
1568
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001569 if (!only_stealable)
1570 return fallback_mt;
1571
1572 if (*can_steal)
1573 return fallback_mt;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001574 }
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001575
1576 return -1;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001577}
1578
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001579/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001580static inline struct page *
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001581__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001582{
Pintu Kumarb8af2942013-09-11 14:20:34 -07001583 struct free_area *area;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001584 unsigned int current_order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001585 struct page *page;
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001586 int fallback_mt;
1587 bool can_steal;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001588
1589 /* Find the largest possible block of pages in the other list */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001590 for (current_order = MAX_ORDER-1;
1591 current_order >= order && current_order <= MAX_ORDER-1;
1592 --current_order) {
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001593 area = &(zone->free_area[current_order]);
1594 fallback_mt = find_suitable_fallback(area, current_order,
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001595 start_migratetype, false, &can_steal);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001596 if (fallback_mt == -1)
1597 continue;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001598
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001599 page = list_entry(area->free_list[fallback_mt].next,
1600 struct page, lru);
1601 if (can_steal)
1602 steal_suitable_fallback(zone, page, start_migratetype);
Mel Gormane0104872007-10-16 01:25:53 -07001603
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001604 /* Remove the page from the freelists */
1605 area->nr_free--;
1606 list_del(&page->lru);
1607 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001608
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001609 expand(zone, page, order, current_order, area,
1610 start_migratetype);
1611 /*
1612 * The freepage_migratetype may differ from pageblock's
1613 * migratetype depending on the decisions in
1614 * try_to_steal_freepages(). This is OK as long as it
1615 * does not differ for MIGRATE_CMA pageblocks. For CMA
1616 * we need to make sure unallocated pages flushed from
1617 * pcp lists are returned to the correct freelist.
1618 */
1619 set_freepage_migratetype(page, start_migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001620
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001621 trace_mm_page_alloc_extfrag(page, order, current_order,
1622 start_migratetype, fallback_mt);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001623
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001624 return page;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001625 }
1626
Mel Gorman728ec982009-06-16 15:32:04 -07001627 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001628}
1629
Mel Gorman56fd56b2007-10-16 01:25:58 -07001630/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 * Do the hard work of removing an element from the buddy allocator.
1632 * Call me with the zone->lock already held.
1633 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001634static struct page *__rmqueue(struct zone *zone, unsigned int order,
1635 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 struct page *page;
1638
Mel Gorman728ec982009-06-16 15:32:04 -07001639retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001640 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641
Mel Gorman728ec982009-06-16 15:32:04 -07001642 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Joonsoo Kimdc676472015-04-14 15:45:15 -07001643 if (migratetype == MIGRATE_MOVABLE)
1644 page = __rmqueue_cma_fallback(zone, order);
1645
1646 if (!page)
1647 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001648
Mel Gorman728ec982009-06-16 15:32:04 -07001649 /*
1650 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1651 * is used because __rmqueue_smallest is an inline function
1652 * and we want just one call site
1653 */
1654 if (!page) {
1655 migratetype = MIGRATE_RESERVE;
1656 goto retry_reserve;
1657 }
1658 }
1659
Mel Gorman0d3d0622009-09-21 17:02:44 -07001660 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001661 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662}
1663
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001664/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 * Obtain a specified number of elements from the buddy allocator, all under
1666 * a single hold of the lock, for efficiency. Add them to the supplied list.
1667 * Returns the number of new pages which were placed at *list.
1668 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001669static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001670 unsigned long count, struct list_head *list,
Mel Gormanb745bc82014-06-04 16:10:22 -07001671 int migratetype, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672{
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001673 int i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001674
Nick Pigginc54ad302006-01-06 00:10:56 -08001675 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001677 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001678 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001680
1681 /*
1682 * Split buddy pages returned by expand() are received here
1683 * in physical page order. The page is added to the callers and
1684 * list and the list head then moves forward. From the callers
1685 * perspective, the linked list is ordered by page number in
1686 * some conditions. This is useful for IO devices that can
1687 * merge IO requests if the physical pages are ordered
1688 * properly.
1689 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001690 if (likely(!cold))
Mel Gormane084b2d2009-07-29 15:02:04 -07001691 list_add(&page->lru, list);
1692 else
1693 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001694 list = &page->lru;
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001695 if (is_migrate_cma(get_freepage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001696 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1697 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001699 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001700 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001701 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702}
1703
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001704#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001705/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001706 * Called from the vmstat counter updater to drain pagesets of this
1707 * currently executing processor on remote nodes after they have
1708 * expired.
1709 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001710 * Note that this function must be called with the thread pinned to
1711 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001712 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001713void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001714{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001715 unsigned long flags;
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001716 int to_drain, batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001717
Christoph Lameter4037d452007-05-09 02:35:14 -07001718 local_irq_save(flags);
Jason Low4db0c3c2015-04-15 16:14:08 -07001719 batch = READ_ONCE(pcp->batch);
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001720 to_drain = min(pcp->count, batch);
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07001721 if (to_drain > 0) {
1722 free_pcppages_bulk(zone, to_drain, pcp);
1723 pcp->count -= to_drain;
1724 }
Christoph Lameter4037d452007-05-09 02:35:14 -07001725 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001726}
1727#endif
1728
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001729/*
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001730 * Drain pcplists of the indicated processor and zone.
1731 *
1732 * The processor must either be the current processor and the
1733 * thread pinned to the current processor or a processor that
1734 * is not online.
1735 */
1736static void drain_pages_zone(unsigned int cpu, struct zone *zone)
1737{
1738 unsigned long flags;
1739 struct per_cpu_pageset *pset;
1740 struct per_cpu_pages *pcp;
1741
1742 local_irq_save(flags);
1743 pset = per_cpu_ptr(zone->pageset, cpu);
1744
1745 pcp = &pset->pcp;
1746 if (pcp->count) {
1747 free_pcppages_bulk(zone, pcp->count, pcp);
1748 pcp->count = 0;
1749 }
1750 local_irq_restore(flags);
1751}
1752
1753/*
1754 * Drain pcplists of all zones on the indicated processor.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001755 *
1756 * The processor must either be the current processor and the
1757 * thread pinned to the current processor or a processor that
1758 * is not online.
1759 */
1760static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761{
1762 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001764 for_each_populated_zone(zone) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001765 drain_pages_zone(cpu, zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 }
1767}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001769/*
1770 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001771 *
1772 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1773 * the single zone's pages.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001774 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001775void drain_local_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001776{
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001777 int cpu = smp_processor_id();
1778
1779 if (zone)
1780 drain_pages_zone(cpu, zone);
1781 else
1782 drain_pages(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001783}
1784
1785/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001786 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1787 *
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001788 * When zone parameter is non-NULL, spill just the single zone's pages.
1789 *
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001790 * Note that this code is protected against sending an IPI to an offline
1791 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1792 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1793 * nothing keeps CPUs from showing up after we populated the cpumask and
1794 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001795 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001796void drain_all_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001797{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001798 int cpu;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001799
1800 /*
1801 * Allocate in the BSS so we wont require allocation in
1802 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1803 */
1804 static cpumask_t cpus_with_pcps;
1805
1806 /*
1807 * We don't care about racing with CPU hotplug event
1808 * as offline notification will cause the notified
1809 * cpu to drain that CPU pcps and on_each_cpu_mask
1810 * disables preemption as part of its processing
1811 */
1812 for_each_online_cpu(cpu) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001813 struct per_cpu_pageset *pcp;
1814 struct zone *z;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001815 bool has_pcps = false;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001816
1817 if (zone) {
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001818 pcp = per_cpu_ptr(zone->pageset, cpu);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001819 if (pcp->pcp.count)
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001820 has_pcps = true;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001821 } else {
1822 for_each_populated_zone(z) {
1823 pcp = per_cpu_ptr(z->pageset, cpu);
1824 if (pcp->pcp.count) {
1825 has_pcps = true;
1826 break;
1827 }
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001828 }
1829 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001830
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001831 if (has_pcps)
1832 cpumask_set_cpu(cpu, &cpus_with_pcps);
1833 else
1834 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1835 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001836 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
1837 zone, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001838}
1839
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001840#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841
1842void mark_free_pages(struct zone *zone)
1843{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001844 unsigned long pfn, max_zone_pfn;
1845 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001846 unsigned int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 struct list_head *curr;
1848
Xishi Qiu8080fc02013-09-11 14:21:45 -07001849 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 return;
1851
1852 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001853
Cody P Schafer108bcc92013-02-22 16:35:23 -08001854 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001855 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1856 if (pfn_valid(pfn)) {
1857 struct page *page = pfn_to_page(pfn);
1858
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001859 if (!swsusp_page_is_forbidden(page))
1860 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001861 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001863 for_each_migratetype_order(order, t) {
1864 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001865 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001867 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1868 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001869 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001870 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001871 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 spin_unlock_irqrestore(&zone->lock, flags);
1873}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001874#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
1876/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 * Free a 0-order page
Mel Gormanb745bc82014-06-04 16:10:22 -07001878 * cold == true ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001880void free_hot_cold_page(struct page *page, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881{
1882 struct zone *zone = page_zone(page);
1883 struct per_cpu_pages *pcp;
1884 unsigned long flags;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001885 unsigned long pfn = page_to_pfn(page);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001886 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001888 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001889 return;
1890
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001891 migratetype = get_pfnblock_migratetype(page, pfn);
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001892 set_freepage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001894 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001895
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001896 /*
1897 * We only track unmovable, reclaimable and movable on pcp lists.
1898 * Free ISOLATE pages back to the allocator because they are being
1899 * offlined but treat RESERVE as movable pages so we can get those
1900 * areas back if necessary. Otherwise, we may have to free
1901 * excessively into the page allocator
1902 */
1903 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08001904 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001905 free_one_page(zone, page, pfn, 0, migratetype);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001906 goto out;
1907 }
1908 migratetype = MIGRATE_MOVABLE;
1909 }
1910
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001911 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Mel Gormanb745bc82014-06-04 16:10:22 -07001912 if (!cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001913 list_add(&page->lru, &pcp->lists[migratetype]);
Mel Gormanb745bc82014-06-04 16:10:22 -07001914 else
1915 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001917 if (pcp->count >= pcp->high) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001918 unsigned long batch = READ_ONCE(pcp->batch);
Cody P Schafer998d39c2013-07-03 15:01:32 -07001919 free_pcppages_bulk(zone, batch, pcp);
1920 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08001921 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001922
1923out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925}
1926
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001927/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001928 * Free a list of 0-order pages
1929 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001930void free_hot_cold_page_list(struct list_head *list, bool cold)
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001931{
1932 struct page *page, *next;
1933
1934 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001935 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001936 free_hot_cold_page(page, cold);
1937 }
1938}
1939
1940/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001941 * split_page takes a non-compound higher-order page, and splits it into
1942 * n (1<<order) sub-pages: page[0..n]
1943 * Each sub-page must be freed individually.
1944 *
1945 * Note: this is probably too low level an operation for use in drivers.
1946 * Please consult with lkml before using this in your driver.
1947 */
1948void split_page(struct page *page, unsigned int order)
1949{
1950 int i;
Joonsoo Kime2cfc912015-07-17 16:24:18 -07001951 gfp_t gfp_mask;
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001952
Sasha Levin309381fea2014-01-23 15:52:54 -08001953 VM_BUG_ON_PAGE(PageCompound(page), page);
1954 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001955
1956#ifdef CONFIG_KMEMCHECK
1957 /*
1958 * Split shadow pages too, because free(page[0]) would
1959 * otherwise free the whole shadow.
1960 */
1961 if (kmemcheck_page_is_tracked(page))
1962 split_page(virt_to_page(page[0].shadow), order);
1963#endif
1964
Joonsoo Kime2cfc912015-07-17 16:24:18 -07001965 gfp_mask = get_page_owner_gfp(page);
1966 set_page_owner(page, 0, gfp_mask);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001967 for (i = 1; i < (1 << order); i++) {
Nick Piggin7835e982006-03-22 00:08:40 -08001968 set_page_refcounted(page + i);
Joonsoo Kime2cfc912015-07-17 16:24:18 -07001969 set_page_owner(page + i, 0, gfp_mask);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001970 }
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001971}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07001972EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001973
Joonsoo Kim3c605092014-11-13 15:19:21 -08001974int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07001975{
Mel Gorman748446b2010-05-24 14:32:27 -07001976 unsigned long watermark;
1977 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001978 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07001979
1980 BUG_ON(!PageBuddy(page));
1981
1982 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001983 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07001984
Minchan Kim194159f2013-02-22 16:33:58 -08001985 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001986 /* Obey watermarks as if the page was being allocated */
1987 watermark = low_wmark_pages(zone) + (1 << order);
1988 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1989 return 0;
1990
Mel Gorman8fb74b92013-01-11 14:32:16 -08001991 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08001992 }
Mel Gorman748446b2010-05-24 14:32:27 -07001993
1994 /* Remove page from free list */
1995 list_del(&page->lru);
1996 zone->free_area[order].nr_free--;
1997 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001998
Joonsoo Kime2cfc912015-07-17 16:24:18 -07001999 set_page_owner(page, order, __GFP_MOVABLE);
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002000
Mel Gorman8fb74b92013-01-11 14:32:16 -08002001 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07002002 if (order >= pageblock_order - 1) {
2003 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002004 for (; page < endpage; page += pageblock_nr_pages) {
2005 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08002006 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002007 set_pageblock_migratetype(page,
2008 MIGRATE_MOVABLE);
2009 }
Mel Gorman748446b2010-05-24 14:32:27 -07002010 }
2011
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002012
Mel Gorman8fb74b92013-01-11 14:32:16 -08002013 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002014}
2015
2016/*
2017 * Similar to split_page except the page is already free. As this is only
2018 * being used for migration, the migratetype of the block also changes.
2019 * As this is called with interrupts disabled, the caller is responsible
2020 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2021 * are enabled.
2022 *
2023 * Note: this is probably too low level an operation for use in drivers.
2024 * Please consult with lkml before using this in your driver.
2025 */
2026int split_free_page(struct page *page)
2027{
2028 unsigned int order;
2029 int nr_pages;
2030
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002031 order = page_order(page);
2032
Mel Gorman8fb74b92013-01-11 14:32:16 -08002033 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002034 if (!nr_pages)
2035 return 0;
2036
2037 /* Split into individual pages */
2038 set_page_refcounted(page);
2039 split_page(page, order);
2040 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07002041}
2042
2043/*
Vlastimil Babka75379192015-02-11 15:25:38 -08002044 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07002046static inline
2047struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002048 struct zone *zone, unsigned int order,
2049 gfp_t gfp_flags, int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050{
2051 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08002052 struct page *page;
Mel Gormanb745bc82014-06-04 16:10:22 -07002053 bool cold = ((gfp_flags & __GFP_COLD) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
Nick Piggin48db57f2006-01-08 01:00:42 -08002055 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002057 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002060 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2061 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002062 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07002063 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002064 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07002065 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002066 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07002067 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07002068 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07002069
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002070 if (cold)
2071 page = list_entry(list->prev, struct page, lru);
2072 else
2073 page = list_entry(list->next, struct page, lru);
2074
Mel Gormanb92a6ed2007-10-16 01:25:50 -07002075 list_del(&page->lru);
2076 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002077 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07002078 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
2079 /*
2080 * __GFP_NOFAIL is not to be used in new code.
2081 *
2082 * All __GFP_NOFAIL callers should be fixed so that they
2083 * properly detect and handle allocation failures.
2084 *
2085 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07002086 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07002087 * __GFP_NOFAIL.
2088 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07002089 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07002090 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002092 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08002093 spin_unlock(&zone->lock);
2094 if (!page)
2095 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07002096 __mod_zone_freepage_state(zone, -(1 << order),
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07002097 get_freepage_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 }
2099
Johannes Weiner3a025762014-04-07 15:37:48 -07002100 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Johannes Weinerabe5f972014-10-02 16:21:10 -07002101 if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
Johannes Weiner57054652014-10-09 15:28:17 -07002102 !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
2103 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Johannes Weiner27329362014-03-03 15:38:41 -08002104
Christoph Lameterf8891e52006-06-30 01:55:45 -07002105 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07002106 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002107 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
Sasha Levin309381fea2014-01-23 15:52:54 -08002109 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08002111
2112failed:
2113 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002114 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115}
2116
Akinobu Mita933e3122006-12-08 02:39:45 -08002117#ifdef CONFIG_FAIL_PAGE_ALLOC
2118
Akinobu Mitab2588c42011-07-26 16:09:03 -07002119static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08002120 struct fault_attr attr;
2121
2122 u32 ignore_gfp_highmem;
2123 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07002124 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08002125} fail_page_alloc = {
2126 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08002127 .ignore_gfp_wait = 1,
2128 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07002129 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08002130};
2131
2132static int __init setup_fail_page_alloc(char *str)
2133{
2134 return setup_fault_attr(&fail_page_alloc.attr, str);
2135}
2136__setup("fail_page_alloc=", setup_fail_page_alloc);
2137
Gavin Shandeaf3862012-07-31 16:41:51 -07002138static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002139{
Akinobu Mita54114992007-07-15 23:40:23 -07002140 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07002141 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002142 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07002143 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002144 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002145 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002146 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
Gavin Shandeaf3862012-07-31 16:41:51 -07002147 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002148
2149 return should_fail(&fail_page_alloc.attr, 1 << order);
2150}
2151
2152#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2153
2154static int __init fail_page_alloc_debugfs(void)
2155{
Al Virof4ae40a2011-07-24 04:33:43 -04002156 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08002157 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08002158
Akinobu Mitadd48c082011-08-03 16:21:01 -07002159 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2160 &fail_page_alloc.attr);
2161 if (IS_ERR(dir))
2162 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002163
Akinobu Mitab2588c42011-07-26 16:09:03 -07002164 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
2165 &fail_page_alloc.ignore_gfp_wait))
2166 goto fail;
2167 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2168 &fail_page_alloc.ignore_gfp_highmem))
2169 goto fail;
2170 if (!debugfs_create_u32("min-order", mode, dir,
2171 &fail_page_alloc.min_order))
2172 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08002173
Akinobu Mitab2588c42011-07-26 16:09:03 -07002174 return 0;
2175fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07002176 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002177
Akinobu Mitab2588c42011-07-26 16:09:03 -07002178 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08002179}
2180
2181late_initcall(fail_page_alloc_debugfs);
2182
2183#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2184
2185#else /* CONFIG_FAIL_PAGE_ALLOC */
2186
Gavin Shandeaf3862012-07-31 16:41:51 -07002187static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002188{
Gavin Shandeaf3862012-07-31 16:41:51 -07002189 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002190}
2191
2192#endif /* CONFIG_FAIL_PAGE_ALLOC */
2193
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08002195 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 * of the allocation.
2197 */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002198static bool __zone_watermark_ok(struct zone *z, unsigned int order,
2199 unsigned long mark, int classzone_idx, int alloc_flags,
2200 long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201{
Wei Yuan26086de2014-12-10 15:44:44 -08002202 /* free_pages may go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08002203 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 int o;
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002205 long free_cma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206
Michal Hockodf0a6da2012-01-10 15:08:02 -08002207 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002208 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002210 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 min -= min / 4;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002212#ifdef CONFIG_CMA
2213 /* If allocation can't use CMA areas don't use free CMA pages */
2214 if (!(alloc_flags & ALLOC_CMA))
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002215 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002216#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002217
Mel Gorman3484b2d2014-08-06 16:07:14 -07002218 if (free_pages - free_cma <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08002219 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 for (o = 0; o < order; o++) {
2221 /* At the next order, this order's pages become unavailable */
2222 free_pages -= z->free_area[o].nr_free << o;
2223
2224 /* Require fewer higher order pages to be free */
2225 min >>= 1;
2226
2227 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002228 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08002230 return true;
2231}
2232
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002233bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002234 int classzone_idx, int alloc_flags)
2235{
2236 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2237 zone_page_state(z, NR_FREE_PAGES));
2238}
2239
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002240bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
2241 unsigned long mark, int classzone_idx, int alloc_flags)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002242{
2243 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2244
2245 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2246 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2247
2248 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2249 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250}
2251
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002252#ifdef CONFIG_NUMA
2253/*
2254 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
2255 * skip over zones that are not allowed by the cpuset, or that have
2256 * been recently (in last second) found to be nearly full. See further
2257 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02002258 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002259 *
Zhi Yong Wua1aeb652013-11-12 15:08:29 -08002260 * If the zonelist cache is present in the passed zonelist, then
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002261 * returns a pointer to the allowed node mask (either the current
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08002262 * tasks mems_allowed, or node_states[N_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002263 *
2264 * If the zonelist cache is not available for this zonelist, does
2265 * nothing and returns NULL.
2266 *
2267 * If the fullzones BITMAP in the zonelist cache is stale (more than
2268 * a second since last zap'd) then we zap it out (clear its bits.)
2269 *
2270 * We hold off even calling zlc_setup, until after we've checked the
2271 * first zone in the zonelist, on the theory that most allocations will
2272 * be satisfied from that first zone, so best to examine that zone as
2273 * quickly as we can.
2274 */
2275static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
2276{
2277 struct zonelist_cache *zlc; /* cached zonelist speedup info */
2278 nodemask_t *allowednodes; /* zonelist_cache approximation */
2279
2280 zlc = zonelist->zlcache_ptr;
2281 if (!zlc)
2282 return NULL;
2283
S.Caglar Onurf05111f2008-04-28 02:12:38 -07002284 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002285 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2286 zlc->last_full_zap = jiffies;
2287 }
2288
2289 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
2290 &cpuset_current_mems_allowed :
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08002291 &node_states[N_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002292 return allowednodes;
2293}
2294
2295/*
2296 * Given 'z' scanning a zonelist, run a couple of quick checks to see
2297 * if it is worth looking at further for free memory:
2298 * 1) Check that the zone isn't thought to be full (doesn't have its
2299 * bit set in the zonelist_cache fullzones BITMAP).
2300 * 2) Check that the zones node (obtained from the zonelist_cache
2301 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
2302 * Return true (non-zero) if zone is worth looking at further, or
2303 * else return false (zero) if it is not.
2304 *
2305 * This check -ignores- the distinction between various watermarks,
2306 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
2307 * found to be full for any variation of these watermarks, it will
2308 * be considered full for up to one second by all requests, unless
2309 * we are so low on memory on all allowed nodes that we are forced
2310 * into the second scan of the zonelist.
2311 *
2312 * In the second scan we ignore this zonelist cache and exactly
2313 * apply the watermarks to all zones, even it is slower to do so.
2314 * We are low on memory in the second scan, and should leave no stone
2315 * unturned looking for a free page.
2316 */
Mel Gormandd1a2392008-04-28 02:12:17 -07002317static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002318 nodemask_t *allowednodes)
2319{
2320 struct zonelist_cache *zlc; /* cached zonelist speedup info */
2321 int i; /* index of *z in zonelist zones */
2322 int n; /* node that zone *z is on */
2323
2324 zlc = zonelist->zlcache_ptr;
2325 if (!zlc)
2326 return 1;
2327
Mel Gormandd1a2392008-04-28 02:12:17 -07002328 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002329 n = zlc->z_to_n[i];
2330
2331 /* This zone is worth trying if it is allowed but not full */
2332 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
2333}
2334
2335/*
2336 * Given 'z' scanning a zonelist, set the corresponding bit in
2337 * zlc->fullzones, so that subsequent attempts to allocate a page
2338 * from that zone don't waste time re-examining it.
2339 */
Mel Gormandd1a2392008-04-28 02:12:17 -07002340static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002341{
2342 struct zonelist_cache *zlc; /* cached zonelist speedup info */
2343 int i; /* index of *z in zonelist zones */
2344
2345 zlc = zonelist->zlcache_ptr;
2346 if (!zlc)
2347 return;
2348
Mel Gormandd1a2392008-04-28 02:12:17 -07002349 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002350
2351 set_bit(i, zlc->fullzones);
2352}
2353
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002354/*
2355 * clear all zones full, called after direct reclaim makes progress so that
2356 * a zone that was recently full is not skipped over for up to a second
2357 */
2358static void zlc_clear_zones_full(struct zonelist *zonelist)
2359{
2360 struct zonelist_cache *zlc; /* cached zonelist speedup info */
2361
2362 zlc = zonelist->zlcache_ptr;
2363 if (!zlc)
2364 return;
2365
2366 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2367}
2368
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002369static bool zone_local(struct zone *local_zone, struct zone *zone)
2370{
Johannes Weinerfff40682013-12-20 14:54:12 +00002371 return local_zone->node == zone->node;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002372}
2373
David Rientjes957f8222012-10-08 16:33:24 -07002374static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2375{
Mel Gorman5f7a75a2014-06-04 16:07:15 -07002376 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2377 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07002378}
2379
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002380#else /* CONFIG_NUMA */
2381
2382static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
2383{
2384 return NULL;
2385}
2386
Mel Gormandd1a2392008-04-28 02:12:17 -07002387static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002388 nodemask_t *allowednodes)
2389{
2390 return 1;
2391}
2392
Mel Gormandd1a2392008-04-28 02:12:17 -07002393static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002394{
2395}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002396
2397static void zlc_clear_zones_full(struct zonelist *zonelist)
2398{
2399}
David Rientjes957f8222012-10-08 16:33:24 -07002400
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002401static bool zone_local(struct zone *local_zone, struct zone *zone)
2402{
2403 return true;
2404}
2405
David Rientjes957f8222012-10-08 16:33:24 -07002406static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2407{
2408 return true;
2409}
2410
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002411#endif /* CONFIG_NUMA */
2412
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002413static void reset_alloc_batches(struct zone *preferred_zone)
2414{
2415 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2416
2417 do {
2418 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2419 high_wmark_pages(zone) - low_wmark_pages(zone) -
2420 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner57054652014-10-09 15:28:17 -07002421 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002422 } while (zone++ != preferred_zone);
2423}
2424
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002425/*
Paul Jackson0798e512006-12-06 20:31:38 -08002426 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002427 * a page.
2428 */
2429static struct page *
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002430get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2431 const struct alloc_context *ac)
Martin Hicks753ee722005-06-21 17:14:41 -07002432{
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002433 struct zonelist *zonelist = ac->zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07002434 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002435 struct page *page = NULL;
Mel Gorman5117f452009-06-16 15:31:59 -07002436 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002437 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
2438 int zlc_active = 0; /* set if using zonelist_cache */
2439 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gormana6e21b142014-06-04 16:10:12 -07002440 bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) &&
2441 (gfp_mask & __GFP_WRITE);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002442 int nr_fair_skipped = 0;
2443 bool zonelist_rescan;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002444
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002445zonelist_scan:
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002446 zonelist_rescan = false;
2447
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002448 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002449 * Scan zonelist, looking for a zone with enough free.
Vladimir Davydov344736f2014-10-20 15:50:30 +04002450 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002451 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002452 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2453 ac->nodemask) {
Johannes Weinere085dbc2013-09-11 14:20:46 -07002454 unsigned long mark;
2455
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002456 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002457 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2458 continue;
Mel Gorman664eedd2014-06-04 16:10:08 -07002459 if (cpusets_enabled() &&
2460 (alloc_flags & ALLOC_CPUSET) &&
Vladimir Davydov344736f2014-10-20 15:50:30 +04002461 !cpuset_zone_allowed(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07002462 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08002463 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002464 * Distribute pages in proportion to the individual
2465 * zone size to ensure fair page aging. The zone a
2466 * page was allocated in should have no effect on the
2467 * time the page has in memory before being reclaimed.
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002468 */
Johannes Weiner3a025762014-04-07 15:37:48 -07002469 if (alloc_flags & ALLOC_FAIR) {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002470 if (!zone_local(ac->preferred_zone, zone))
Mel Gormanf7b5d642014-08-06 16:07:20 -07002471 break;
Johannes Weiner57054652014-10-09 15:28:17 -07002472 if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002473 nr_fair_skipped++;
Johannes Weiner3a025762014-04-07 15:37:48 -07002474 continue;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002475 }
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002476 }
2477 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08002478 * When allocating a page cache page for writing, we
2479 * want to get it from a zone that is within its dirty
2480 * limit, such that no single zone holds more than its
2481 * proportional share of globally allowed dirty pages.
2482 * The dirty limits take into account the zone's
2483 * lowmem reserves and high watermark so that kswapd
2484 * should be able to balance it without having to
2485 * write pages from its LRU list.
2486 *
2487 * This may look like it could increase pressure on
2488 * lower zones by failing allocations in higher zones
2489 * before they are full. But the pages that do spill
2490 * over are limited as the lower zones are protected
2491 * by this very same mechanism. It should not become
2492 * a practical burden to them.
2493 *
2494 * XXX: For now, allow allocations to potentially
2495 * exceed the per-zone dirty limit in the slowpath
2496 * (ALLOC_WMARK_LOW unset) before going into reclaim,
2497 * which is important when on a NUMA setup the allowed
2498 * zones are together not big enough to reach the
2499 * global limit. The proper fix for these situations
2500 * will require awareness of zones in the
2501 * dirty-throttling and the flusher threads.
2502 */
Mel Gormana6e21b142014-06-04 16:10:12 -07002503 if (consider_zone_dirty && !zone_dirty_ok(zone))
Mel Gorman800a1e72014-06-04 16:10:06 -07002504 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002505
Johannes Weinere085dbc2013-09-11 14:20:46 -07002506 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2507 if (!zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002508 ac->classzone_idx, alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07002509 int ret;
2510
Mel Gorman5dab2912014-06-04 16:10:14 -07002511 /* Checked here to keep the fast path fast */
2512 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2513 if (alloc_flags & ALLOC_NO_WATERMARKS)
2514 goto try_this_zone;
2515
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002516 if (IS_ENABLED(CONFIG_NUMA) &&
2517 !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormancd38b112011-07-25 17:12:29 -07002518 /*
2519 * we do zlc_setup if there are multiple nodes
2520 * and before considering the first zone allowed
2521 * by the cpuset.
2522 */
2523 allowednodes = zlc_setup(zonelist, alloc_flags);
2524 zlc_active = 1;
2525 did_zlc_setup = 1;
2526 }
2527
David Rientjes957f8222012-10-08 16:33:24 -07002528 if (zone_reclaim_mode == 0 ||
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002529 !zone_allows_reclaim(ac->preferred_zone, zone))
Mel Gormanfa5e0842009-06-16 15:33:22 -07002530 goto this_zone_full;
2531
Mel Gormancd38b112011-07-25 17:12:29 -07002532 /*
2533 * As we may have just activated ZLC, check if the first
2534 * eligible zone has failed zone_reclaim recently.
2535 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002536 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Mel Gormancd38b112011-07-25 17:12:29 -07002537 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2538 continue;
2539
Mel Gormanfa5e0842009-06-16 15:33:22 -07002540 ret = zone_reclaim(zone, gfp_mask, order);
2541 switch (ret) {
2542 case ZONE_RECLAIM_NOSCAN:
2543 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07002544 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002545 case ZONE_RECLAIM_FULL:
2546 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002547 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002548 default:
2549 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002550 if (zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002551 ac->classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002552 goto try_this_zone;
2553
2554 /*
2555 * Failed to reclaim enough to meet watermark.
2556 * Only mark the zone full if checking the min
2557 * watermark or if we failed to reclaim just
2558 * 1<<order pages or else the page allocator
2559 * fastpath will prematurely mark zones full
2560 * when the watermark is between the low and
2561 * min watermarks.
2562 */
2563 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2564 ret == ZONE_RECLAIM_SOME)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002565 goto this_zone_full;
Mel Gormanfed27192013-04-29 15:07:57 -07002566
2567 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002568 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002569 }
2570
Mel Gormanfa5e0842009-06-16 15:33:22 -07002571try_this_zone:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002572 page = buffered_rmqueue(ac->preferred_zone, zone, order,
2573 gfp_mask, ac->migratetype);
Vlastimil Babka75379192015-02-11 15:25:38 -08002574 if (page) {
2575 if (prep_new_page(page, order, gfp_mask, alloc_flags))
2576 goto try_this_zone;
2577 return page;
2578 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002579this_zone_full:
Mel Gorman65bb3712014-06-04 16:10:05 -07002580 if (IS_ENABLED(CONFIG_NUMA) && zlc_active)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002581 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002582 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002583
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002584 /*
2585 * The first pass makes sure allocations are spread fairly within the
2586 * local node. However, the local node might have free pages left
2587 * after the fairness batches are exhausted, and remote zones haven't
2588 * even been considered yet. Try once more without fairness, and
2589 * include remote zones now, before entering the slowpath and waking
2590 * kswapd: prefer spilling to a remote zone over swapping locally.
2591 */
2592 if (alloc_flags & ALLOC_FAIR) {
2593 alloc_flags &= ~ALLOC_FAIR;
2594 if (nr_fair_skipped) {
2595 zonelist_rescan = true;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002596 reset_alloc_batches(ac->preferred_zone);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002597 }
2598 if (nr_online_nodes > 1)
2599 zonelist_rescan = true;
2600 }
2601
2602 if (unlikely(IS_ENABLED(CONFIG_NUMA) && zlc_active)) {
2603 /* Disable zlc cache for second zonelist scan */
2604 zlc_active = 0;
2605 zonelist_rescan = true;
2606 }
2607
2608 if (zonelist_rescan)
2609 goto zonelist_scan;
2610
2611 return NULL;
Martin Hicks753ee722005-06-21 17:14:41 -07002612}
2613
David Rientjes29423e772011-03-22 16:30:47 -07002614/*
2615 * Large machines with many possible nodes should not always dump per-node
2616 * meminfo in irq context.
2617 */
2618static inline bool should_suppress_show_mem(void)
2619{
2620 bool ret = false;
2621
2622#if NODES_SHIFT > 8
2623 ret = in_interrupt();
2624#endif
2625 return ret;
2626}
2627
Dave Hansena238ab52011-05-24 17:12:16 -07002628static DEFINE_RATELIMIT_STATE(nopage_rs,
2629 DEFAULT_RATELIMIT_INTERVAL,
2630 DEFAULT_RATELIMIT_BURST);
2631
2632void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2633{
Dave Hansena238ab52011-05-24 17:12:16 -07002634 unsigned int filter = SHOW_MEM_FILTER_NODES;
2635
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08002636 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2637 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002638 return;
2639
2640 /*
2641 * This documents exceptions given to allocations in certain
2642 * contexts that are allowed to allocate outside current's set
2643 * of allowed nodes.
2644 */
2645 if (!(gfp_mask & __GFP_NOMEMALLOC))
2646 if (test_thread_flag(TIF_MEMDIE) ||
2647 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2648 filter &= ~SHOW_MEM_FILTER_NODES;
2649 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2650 filter &= ~SHOW_MEM_FILTER_NODES;
2651
2652 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002653 struct va_format vaf;
2654 va_list args;
2655
Dave Hansena238ab52011-05-24 17:12:16 -07002656 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002657
2658 vaf.fmt = fmt;
2659 vaf.va = &args;
2660
2661 pr_warn("%pV", &vaf);
2662
Dave Hansena238ab52011-05-24 17:12:16 -07002663 va_end(args);
2664 }
2665
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002666 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2667 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002668
2669 dump_stack();
2670 if (!should_suppress_show_mem())
2671 show_mem(filter);
2672}
2673
Mel Gorman11e33f62009-06-16 15:31:57 -07002674static inline struct page *
2675__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002676 const struct alloc_context *ac, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07002677{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679
Johannes Weiner9879de72015-01-26 12:58:32 -08002680 *did_some_progress = 0;
2681
Johannes Weiner9879de72015-01-26 12:58:32 -08002682 /*
Johannes Weinerdc564012015-06-24 16:57:19 -07002683 * Acquire the oom lock. If that fails, somebody else is
2684 * making progress for us.
Johannes Weiner9879de72015-01-26 12:58:32 -08002685 */
Johannes Weinerdc564012015-06-24 16:57:19 -07002686 if (!mutex_trylock(&oom_lock)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002687 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002688 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 return NULL;
2690 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002691
Mel Gorman11e33f62009-06-16 15:31:57 -07002692 /*
2693 * Go through the zonelist yet one more time, keep very high watermark
2694 * here, this is only to catch a parallel oom killing, we must fail if
2695 * we're still under heavy pressure.
2696 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002697 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2698 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002699 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002700 goto out;
2701
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002702 if (!(gfp_mask & __GFP_NOFAIL)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002703 /* Coredumps can quickly deplete all memory reserves */
2704 if (current->flags & PF_DUMPCORE)
2705 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002706 /* The OOM killer will not help higher order allocs */
2707 if (order > PAGE_ALLOC_COSTLY_ORDER)
2708 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002709 /* The OOM killer does not needlessly kill tasks for lowmem */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002710 if (ac->high_zoneidx < ZONE_NORMAL)
David Rientjes03668b32010-08-09 17:18:54 -07002711 goto out;
Johannes Weiner90839052015-06-24 16:57:21 -07002712 /* The OOM killer does not compensate for IO-less reclaim */
Johannes Weinercc873172015-02-27 15:52:09 -08002713 if (!(gfp_mask & __GFP_FS)) {
2714 /*
2715 * XXX: Page reclaim didn't yield anything,
2716 * and the OOM killer can't be invoked, but
Johannes Weiner90839052015-06-24 16:57:21 -07002717 * keep looping as per tradition.
Johannes Weinercc873172015-02-27 15:52:09 -08002718 */
2719 *did_some_progress = 1;
Johannes Weiner9879de72015-01-26 12:58:32 -08002720 goto out;
Johannes Weinercc873172015-02-27 15:52:09 -08002721 }
Johannes Weiner90839052015-06-24 16:57:21 -07002722 if (pm_suspended_storage())
2723 goto out;
David Rientjes4167e9b2015-04-14 15:46:55 -07002724 /* The OOM killer may not free memory on a specific node */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002725 if (gfp_mask & __GFP_THISNODE)
2726 goto out;
2727 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002728 /* Exhausted what can be done so it's blamo time */
Michal Hockoe009d5d2015-03-12 16:25:52 -07002729 if (out_of_memory(ac->zonelist, gfp_mask, order, ac->nodemask, false)
2730 || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL))
Michal Hockoc32b3cb2015-02-11 15:26:24 -08002731 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002732out:
Johannes Weinerdc564012015-06-24 16:57:19 -07002733 mutex_unlock(&oom_lock);
Mel Gorman11e33f62009-06-16 15:31:57 -07002734 return page;
2735}
2736
Mel Gorman56de7262010-05-24 14:32:30 -07002737#ifdef CONFIG_COMPACTION
2738/* Try memory compaction for high-order allocations before reclaim */
2739static struct page *
2740__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002741 int alloc_flags, const struct alloc_context *ac,
2742 enum migrate_mode mode, int *contended_compaction,
2743 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002744{
Vlastimil Babka53853e22014-10-09 15:27:02 -07002745 unsigned long compact_result;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002746 struct page *page;
Vlastimil Babka53853e22014-10-09 15:27:02 -07002747
Mel Gorman66199712012-01-12 17:19:41 -08002748 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002749 return NULL;
2750
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002751 current->flags |= PF_MEMALLOC;
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08002752 compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2753 mode, contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002754 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002755
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002756 switch (compact_result) {
2757 case COMPACT_DEFERRED:
Vlastimil Babka53853e22014-10-09 15:27:02 -07002758 *deferred_compaction = true;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002759 /* fall-through */
2760 case COMPACT_SKIPPED:
2761 return NULL;
2762 default:
2763 break;
Mel Gorman56de7262010-05-24 14:32:30 -07002764 }
2765
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002766 /*
2767 * At least in one zone compaction wasn't deferred or skipped, so let's
2768 * count a compaction stall
2769 */
2770 count_vm_event(COMPACTSTALL);
2771
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002772 page = get_page_from_freelist(gfp_mask, order,
2773 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002774
2775 if (page) {
2776 struct zone *zone = page_zone(page);
2777
2778 zone->compact_blockskip_flush = false;
2779 compaction_defer_reset(zone, order, true);
2780 count_vm_event(COMPACTSUCCESS);
2781 return page;
2782 }
2783
2784 /*
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002785 * It's bad if compaction run occurs and fails. The most likely reason
2786 * is that pages exist, but not enough to satisfy watermarks.
2787 */
2788 count_vm_event(COMPACTFAIL);
2789
2790 cond_resched();
2791
Mel Gorman56de7262010-05-24 14:32:30 -07002792 return NULL;
2793}
2794#else
2795static inline struct page *
2796__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002797 int alloc_flags, const struct alloc_context *ac,
2798 enum migrate_mode mode, int *contended_compaction,
2799 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002800{
2801 return NULL;
2802}
2803#endif /* CONFIG_COMPACTION */
2804
Marek Szyprowskibba90712012-01-25 12:09:52 +01002805/* Perform direct synchronous page reclaim */
2806static int
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002807__perform_reclaim(gfp_t gfp_mask, unsigned int order,
2808 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002809{
Mel Gorman11e33f62009-06-16 15:31:57 -07002810 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002811 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002812
2813 cond_resched();
2814
2815 /* We now go into synchronous reclaim */
2816 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002817 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002818 lockdep_set_current_reclaim_state(gfp_mask);
2819 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002820 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002821
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002822 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
2823 ac->nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002824
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002825 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002826 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002827 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002828
2829 cond_resched();
2830
Marek Szyprowskibba90712012-01-25 12:09:52 +01002831 return progress;
2832}
2833
2834/* The really slow allocator path where we enter direct reclaim */
2835static inline struct page *
2836__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002837 int alloc_flags, const struct alloc_context *ac,
2838 unsigned long *did_some_progress)
Marek Szyprowskibba90712012-01-25 12:09:52 +01002839{
2840 struct page *page = NULL;
2841 bool drained = false;
2842
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002843 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002844 if (unlikely(!(*did_some_progress)))
2845 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002846
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002847 /* After successful reclaim, reconsider all zones for allocation */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002848 if (IS_ENABLED(CONFIG_NUMA))
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002849 zlc_clear_zones_full(ac->zonelist);
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002850
Mel Gorman9ee493c2010-09-09 16:38:18 -07002851retry:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002852 page = get_page_from_freelist(gfp_mask, order,
2853 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002854
2855 /*
2856 * If an allocation failed after direct reclaim, it could be because
2857 * pages are pinned on the per-cpu lists. Drain them and try again
2858 */
2859 if (!page && !drained) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002860 drain_all_pages(NULL);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002861 drained = true;
2862 goto retry;
2863 }
2864
Mel Gorman11e33f62009-06-16 15:31:57 -07002865 return page;
2866}
2867
Mel Gorman11e33f62009-06-16 15:31:57 -07002868/*
2869 * This is called in the allocator slow-path if the allocation request is of
2870 * sufficient urgency to ignore watermarks and take other desperate measures
2871 */
2872static inline struct page *
2873__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002874 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002875{
2876 struct page *page;
2877
2878 do {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002879 page = get_page_from_freelist(gfp_mask, order,
2880 ALLOC_NO_WATERMARKS, ac);
Mel Gorman11e33f62009-06-16 15:31:57 -07002881
2882 if (!page && gfp_mask & __GFP_NOFAIL)
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002883 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC,
2884 HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002885 } while (!page && (gfp_mask & __GFP_NOFAIL));
2886
2887 return page;
2888}
2889
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002890static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002891{
2892 struct zoneref *z;
2893 struct zone *zone;
2894
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002895 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
2896 ac->high_zoneidx, ac->nodemask)
2897 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
Mel Gorman11e33f62009-06-16 15:31:57 -07002898}
2899
Peter Zijlstra341ce062009-06-16 15:32:02 -07002900static inline int
2901gfp_to_alloc_flags(gfp_t gfp_mask)
2902{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002903 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
David Rientjesb104a352014-07-30 16:08:24 -07002904 const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD));
Peter Zijlstra341ce062009-06-16 15:32:02 -07002905
Mel Gormana56f57f2009-06-16 15:32:02 -07002906 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002907 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002908
Peter Zijlstra341ce062009-06-16 15:32:02 -07002909 /*
2910 * The caller may dip into page reserves a bit more if the caller
2911 * cannot run direct reclaim, or if the caller has realtime scheduling
2912 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
David Rientjesb104a352014-07-30 16:08:24 -07002913 * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
Peter Zijlstra341ce062009-06-16 15:32:02 -07002914 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002915 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002916
David Rientjesb104a352014-07-30 16:08:24 -07002917 if (atomic) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002918 /*
David Rientjesb104a352014-07-30 16:08:24 -07002919 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2920 * if it can't schedule.
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002921 */
David Rientjesb104a352014-07-30 16:08:24 -07002922 if (!(gfp_mask & __GFP_NOMEMALLOC))
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002923 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002924 /*
David Rientjesb104a352014-07-30 16:08:24 -07002925 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
Vladimir Davydov344736f2014-10-20 15:50:30 +04002926 * comment for __cpuset_node_allowed().
Peter Zijlstra341ce062009-06-16 15:32:02 -07002927 */
2928 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002929 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002930 alloc_flags |= ALLOC_HARDER;
2931
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002932 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2933 if (gfp_mask & __GFP_MEMALLOC)
2934 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07002935 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2936 alloc_flags |= ALLOC_NO_WATERMARKS;
2937 else if (!in_interrupt() &&
2938 ((current->flags & PF_MEMALLOC) ||
2939 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07002940 alloc_flags |= ALLOC_NO_WATERMARKS;
2941 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002942#ifdef CONFIG_CMA
David Rientjes43e7a342014-10-09 15:27:25 -07002943 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002944 alloc_flags |= ALLOC_CMA;
2945#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002946 return alloc_flags;
2947}
2948
Mel Gorman072bb0a2012-07-31 16:43:58 -07002949bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2950{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002951 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002952}
2953
Mel Gorman11e33f62009-06-16 15:31:57 -07002954static inline struct page *
2955__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002956 struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002957{
2958 const gfp_t wait = gfp_mask & __GFP_WAIT;
2959 struct page *page = NULL;
2960 int alloc_flags;
2961 unsigned long pages_reclaimed = 0;
2962 unsigned long did_some_progress;
David Rientjese0b9dae2014-06-04 16:08:28 -07002963 enum migrate_mode migration_mode = MIGRATE_ASYNC;
Mel Gorman66199712012-01-12 17:19:41 -08002964 bool deferred_compaction = false;
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07002965 int contended_compaction = COMPACT_CONTENDED_NONE;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002966
Christoph Lameter952f3b52006-12-06 20:33:26 -08002967 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002968 * In the slowpath, we sanity check order to avoid ever trying to
2969 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2970 * be using allocators in order of preference for an area that is
2971 * too large.
2972 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002973 if (order >= MAX_ORDER) {
2974 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002975 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002976 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002977
Christoph Lameter952f3b52006-12-06 20:33:26 -08002978 /*
David Rientjes4167e9b2015-04-14 15:46:55 -07002979 * If this allocation cannot block and it is for a specific node, then
2980 * fail early. There's no need to wakeup kswapd or retry for a
2981 * speculative node-specific allocation.
Christoph Lameter952f3b52006-12-06 20:33:26 -08002982 */
David Rientjes4167e9b2015-04-14 15:46:55 -07002983 if (IS_ENABLED(CONFIG_NUMA) && (gfp_mask & __GFP_THISNODE) && !wait)
Christoph Lameter952f3b52006-12-06 20:33:26 -08002984 goto nopage;
2985
Johannes Weiner9879de72015-01-26 12:58:32 -08002986retry:
Johannes Weiner3a025762014-04-07 15:37:48 -07002987 if (!(gfp_mask & __GFP_NO_KSWAPD))
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002988 wake_all_kswapds(order, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002989
Paul Jackson9bf22292005-09-06 15:18:12 -07002990 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002991 * OK, we're below the kswapd watermark and have kicked background
2992 * reclaim. Now things get more complex, so set up alloc_flags according
2993 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002994 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002995 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996
David Rientjesf33261d2011-01-25 15:07:20 -08002997 /*
2998 * Find the true preferred zone if the allocation is unconstrained by
2999 * cpusets.
3000 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003001 if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
Mel Gormand8846372014-06-04 16:10:33 -07003002 struct zoneref *preferred_zoneref;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003003 preferred_zoneref = first_zones_zonelist(ac->zonelist,
3004 ac->high_zoneidx, NULL, &ac->preferred_zone);
3005 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gormand8846372014-06-04 16:10:33 -07003006 }
David Rientjesf33261d2011-01-25 15:07:20 -08003007
Peter Zijlstra341ce062009-06-16 15:32:02 -07003008 /* This is the last chance, in general, before the goto nopage. */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003009 page = get_page_from_freelist(gfp_mask, order,
3010 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003011 if (page)
3012 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013
Mel Gorman11e33f62009-06-16 15:31:57 -07003014 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07003015 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07003016 /*
3017 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3018 * the allocation is high priority and these type of
3019 * allocations are system rather than user orientated
3020 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003021 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
Mel Gorman183f6372012-07-31 16:44:12 -07003022
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003023 page = __alloc_pages_high_priority(gfp_mask, order, ac);
3024
Mel Gormancfd19c52012-07-31 16:44:10 -07003025 if (page) {
Peter Zijlstra341ce062009-06-16 15:32:02 -07003026 goto got_pg;
Mel Gormancfd19c52012-07-31 16:44:10 -07003027 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 }
3029
3030 /* Atomic allocations - we can't balance anything */
David Rientjesaed0a0e2014-01-21 15:51:12 -08003031 if (!wait) {
3032 /*
3033 * All existing users of the deprecated __GFP_NOFAIL are
3034 * blockable, so warn of any new users that actually allow this
3035 * type of allocation to fail.
3036 */
3037 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 goto nopage;
David Rientjesaed0a0e2014-01-21 15:51:12 -08003039 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040
Peter Zijlstra341ce062009-06-16 15:32:02 -07003041 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003042 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07003043 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044
David Rientjes6583bb62009-07-29 15:02:06 -07003045 /* Avoid allocations with no watermarks from looping endlessly */
3046 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3047 goto nopage;
3048
Mel Gorman77f1fe62011-01-13 15:45:57 -08003049 /*
3050 * Try direct compaction. The first pass is asynchronous. Subsequent
3051 * attempts after direct reclaim are synchronous
3052 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003053 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3054 migration_mode,
3055 &contended_compaction,
Vlastimil Babka53853e22014-10-09 15:27:02 -07003056 &deferred_compaction);
Mel Gorman56de7262010-05-24 14:32:30 -07003057 if (page)
3058 goto got_pg;
David Rientjes75f30862014-06-04 16:08:30 -07003059
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003060 /* Checks for THP-specific high-order allocations */
3061 if ((gfp_mask & GFP_TRANSHUGE) == GFP_TRANSHUGE) {
3062 /*
3063 * If compaction is deferred for high-order allocations, it is
3064 * because sync compaction recently failed. If this is the case
3065 * and the caller requested a THP allocation, we do not want
3066 * to heavily disrupt the system, so we fail the allocation
3067 * instead of entering direct reclaim.
3068 */
3069 if (deferred_compaction)
3070 goto nopage;
3071
3072 /*
3073 * In all zones where compaction was attempted (and not
3074 * deferred or skipped), lock contention has been detected.
3075 * For THP allocation we do not want to disrupt the others
3076 * so we fallback to base pages instead.
3077 */
3078 if (contended_compaction == COMPACT_CONTENDED_LOCK)
3079 goto nopage;
3080
3081 /*
3082 * If compaction was aborted due to need_resched(), we do not
3083 * want to further increase allocation latency, unless it is
3084 * khugepaged trying to collapse.
3085 */
3086 if (contended_compaction == COMPACT_CONTENDED_SCHED
3087 && !(current->flags & PF_KTHREAD))
3088 goto nopage;
3089 }
Mel Gorman66199712012-01-12 17:19:41 -08003090
David Rientjes8fe78042014-08-06 16:07:54 -07003091 /*
3092 * It can become very expensive to allocate transparent hugepages at
3093 * fault, so use asynchronous memory compaction for THP unless it is
3094 * khugepaged trying to collapse.
3095 */
3096 if ((gfp_mask & GFP_TRANSHUGE) != GFP_TRANSHUGE ||
3097 (current->flags & PF_KTHREAD))
3098 migration_mode = MIGRATE_SYNC_LIGHT;
3099
Mel Gorman11e33f62009-06-16 15:31:57 -07003100 /* Try direct reclaim and then allocating */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003101 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3102 &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07003103 if (page)
3104 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105
Johannes Weiner90839052015-06-24 16:57:21 -07003106 /* Do not loop if specifically requested */
3107 if (gfp_mask & __GFP_NORETRY)
3108 goto noretry;
3109
3110 /* Keep reclaiming pages as long as there is reasonable progress */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07003111 pages_reclaimed += did_some_progress;
Johannes Weiner90839052015-06-24 16:57:21 -07003112 if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) ||
3113 ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) {
Mel Gorman11e33f62009-06-16 15:31:57 -07003114 /* Wait for some write requests to complete then retry */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003115 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
Johannes Weiner9879de72015-01-26 12:58:32 -08003116 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117 }
3118
Johannes Weiner90839052015-06-24 16:57:21 -07003119 /* Reclaim has failed us, start killing things */
3120 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3121 if (page)
3122 goto got_pg;
3123
3124 /* Retry as long as the OOM killer is making progress */
3125 if (did_some_progress)
3126 goto retry;
3127
3128noretry:
3129 /*
3130 * High-order allocations do not necessarily loop after
3131 * direct reclaim and reclaim/compaction depends on compaction
3132 * being called after reclaim so call directly if necessary
3133 */
3134 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags,
3135 ac, migration_mode,
3136 &contended_compaction,
3137 &deferred_compaction);
3138 if (page)
3139 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07003141 warn_alloc_failed(gfp_mask, order, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142got_pg:
Mel Gorman072bb0a2012-07-31 16:43:58 -07003143 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144}
Mel Gorman11e33f62009-06-16 15:31:57 -07003145
3146/*
3147 * This is the 'heart' of the zoned buddy allocator.
3148 */
3149struct page *
3150__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3151 struct zonelist *zonelist, nodemask_t *nodemask)
3152{
Mel Gormand8846372014-06-04 16:10:33 -07003153 struct zoneref *preferred_zoneref;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003154 struct page *page = NULL;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003155 unsigned int cpuset_mems_cookie;
Johannes Weiner3a025762014-04-07 15:37:48 -07003156 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
Andrew Morton91fbdc02015-02-11 15:25:04 -08003157 gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003158 struct alloc_context ac = {
3159 .high_zoneidx = gfp_zone(gfp_mask),
3160 .nodemask = nodemask,
3161 .migratetype = gfpflags_to_migratetype(gfp_mask),
3162 };
Mel Gorman11e33f62009-06-16 15:31:57 -07003163
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003164 gfp_mask &= gfp_allowed_mask;
3165
Mel Gorman11e33f62009-06-16 15:31:57 -07003166 lockdep_trace_alloc(gfp_mask);
3167
3168 might_sleep_if(gfp_mask & __GFP_WAIT);
3169
3170 if (should_fail_alloc_page(gfp_mask, order))
3171 return NULL;
3172
3173 /*
3174 * Check the zones suitable for the gfp_mask contain at least one
3175 * valid zone. It's possible to have an empty zonelist as a result
David Rientjes4167e9b2015-04-14 15:46:55 -07003176 * of __GFP_THISNODE and a memoryless node
Mel Gorman11e33f62009-06-16 15:31:57 -07003177 */
3178 if (unlikely(!zonelist->_zonerefs->zone))
3179 return NULL;
3180
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003181 if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
Vlastimil Babka21bb9bd2014-10-09 15:26:51 -07003182 alloc_flags |= ALLOC_CMA;
3183
Mel Gormancc9a6c82012-03-21 16:34:11 -07003184retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07003185 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003186
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003187 /* We set it here, as __alloc_pages_slowpath might have changed it */
3188 ac.zonelist = zonelist;
Mel Gorman5117f452009-06-16 15:31:59 -07003189 /* The preferred zone is used for statistics later */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003190 preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
3191 ac.nodemask ? : &cpuset_current_mems_allowed,
3192 &ac.preferred_zone);
3193 if (!ac.preferred_zone)
Mel Gormancc9a6c82012-03-21 16:34:11 -07003194 goto out;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003195 ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gorman5117f452009-06-16 15:31:59 -07003196
3197 /* First allocation attempt */
Andrew Morton91fbdc02015-02-11 15:25:04 -08003198 alloc_mask = gfp_mask|__GFP_HARDWALL;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003199 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08003200 if (unlikely(!page)) {
3201 /*
3202 * Runtime PM, block IO and its error handling path
3203 * can deadlock because I/O on the device might not
3204 * complete.
3205 */
Andrew Morton91fbdc02015-02-11 15:25:04 -08003206 alloc_mask = memalloc_noio_flags(gfp_mask);
3207
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003208 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08003209 }
Mel Gorman11e33f62009-06-16 15:31:57 -07003210
Xishi Qiu23f086f2015-02-11 15:25:07 -08003211 if (kmemcheck_enabled && page)
3212 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3213
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003214 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07003215
3216out:
3217 /*
3218 * When updating a task's mems_allowed, it is possible to race with
3219 * parallel threads in such a way that an allocation can fail while
3220 * the mask is being updated. If a page allocation is about to fail,
3221 * check if the cpuset changed during allocation and if so, retry.
3222 */
Mel Gormand26914d2014-04-03 14:47:24 -07003223 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07003224 goto retry_cpuset;
3225
Mel Gorman11e33f62009-06-16 15:31:57 -07003226 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227}
Mel Gormand2391712009-06-16 15:31:52 -07003228EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229
3230/*
3231 * Common helper functions.
3232 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08003233unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234{
Akinobu Mita945a1112009-09-21 17:01:47 -07003235 struct page *page;
3236
3237 /*
3238 * __get_free_pages() returns a 32-bit address, which cannot represent
3239 * a highmem page
3240 */
3241 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3242
Linus Torvalds1da177e2005-04-16 15:20:36 -07003243 page = alloc_pages(gfp_mask, order);
3244 if (!page)
3245 return 0;
3246 return (unsigned long) page_address(page);
3247}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248EXPORT_SYMBOL(__get_free_pages);
3249
Harvey Harrison920c7a52008-02-04 22:29:26 -08003250unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251{
Akinobu Mita945a1112009-09-21 17:01:47 -07003252 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254EXPORT_SYMBOL(get_zeroed_page);
3255
Harvey Harrison920c7a52008-02-04 22:29:26 -08003256void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257{
Nick Pigginb5810032005-10-29 18:16:12 -07003258 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259 if (order == 0)
Mel Gormanb745bc82014-06-04 16:10:22 -07003260 free_hot_cold_page(page, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261 else
3262 __free_pages_ok(page, order);
3263 }
3264}
3265
3266EXPORT_SYMBOL(__free_pages);
3267
Harvey Harrison920c7a52008-02-04 22:29:26 -08003268void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269{
3270 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07003271 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003272 __free_pages(virt_to_page((void *)addr), order);
3273 }
3274}
3275
3276EXPORT_SYMBOL(free_pages);
3277
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003278/*
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003279 * Page Fragment:
3280 * An arbitrary-length arbitrary-offset area of memory which resides
3281 * within a 0 or higher order page. Multiple fragments within that page
3282 * are individually refcounted, in the page's reference counter.
3283 *
3284 * The page_frag functions below provide a simple allocation framework for
3285 * page fragments. This is used by the network stack and network device
3286 * drivers to provide a backing region of memory for use as either an
3287 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3288 */
3289static struct page *__page_frag_refill(struct page_frag_cache *nc,
3290 gfp_t gfp_mask)
3291{
3292 struct page *page = NULL;
3293 gfp_t gfp = gfp_mask;
3294
3295#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3296 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3297 __GFP_NOMEMALLOC;
3298 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3299 PAGE_FRAG_CACHE_MAX_ORDER);
3300 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3301#endif
3302 if (unlikely(!page))
3303 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3304
3305 nc->va = page ? page_address(page) : NULL;
3306
3307 return page;
3308}
3309
3310void *__alloc_page_frag(struct page_frag_cache *nc,
3311 unsigned int fragsz, gfp_t gfp_mask)
3312{
3313 unsigned int size = PAGE_SIZE;
3314 struct page *page;
3315 int offset;
3316
3317 if (unlikely(!nc->va)) {
3318refill:
3319 page = __page_frag_refill(nc, gfp_mask);
3320 if (!page)
3321 return NULL;
3322
3323#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3324 /* if size can vary use size else just use PAGE_SIZE */
3325 size = nc->size;
3326#endif
3327 /* Even if we own the page, we do not use atomic_set().
3328 * This would break get_page_unless_zero() users.
3329 */
3330 atomic_add(size - 1, &page->_count);
3331
3332 /* reset page count bias and offset to start of new frag */
3333 nc->pfmemalloc = page->pfmemalloc;
3334 nc->pagecnt_bias = size;
3335 nc->offset = size;
3336 }
3337
3338 offset = nc->offset - fragsz;
3339 if (unlikely(offset < 0)) {
3340 page = virt_to_page(nc->va);
3341
3342 if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count))
3343 goto refill;
3344
3345#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3346 /* if size can vary use size else just use PAGE_SIZE */
3347 size = nc->size;
3348#endif
3349 /* OK, page count is 0, we can safely set it */
3350 atomic_set(&page->_count, size);
3351
3352 /* reset page count bias and offset to start of new frag */
3353 nc->pagecnt_bias = size;
3354 offset = size - fragsz;
3355 }
3356
3357 nc->pagecnt_bias--;
3358 nc->offset = offset;
3359
3360 return nc->va + offset;
3361}
3362EXPORT_SYMBOL(__alloc_page_frag);
3363
3364/*
3365 * Frees a page fragment allocated out of either a compound or order 0 page.
3366 */
3367void __free_page_frag(void *addr)
3368{
3369 struct page *page = virt_to_head_page(addr);
3370
3371 if (unlikely(put_page_testzero(page)))
3372 __free_pages_ok(page, compound_order(page));
3373}
3374EXPORT_SYMBOL(__free_page_frag);
3375
3376/*
Vladimir Davydov52383432014-06-04 16:06:39 -07003377 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3378 * of the current memory cgroup.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003379 *
Vladimir Davydov52383432014-06-04 16:06:39 -07003380 * It should be used when the caller would like to use kmalloc, but since the
3381 * allocation is large, it has to fall back to the page allocator.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003382 */
Vladimir Davydov52383432014-06-04 16:06:39 -07003383struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
3384{
3385 struct page *page;
3386 struct mem_cgroup *memcg = NULL;
3387
3388 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
3389 return NULL;
3390 page = alloc_pages(gfp_mask, order);
3391 memcg_kmem_commit_charge(page, memcg, order);
3392 return page;
3393}
3394
3395struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
3396{
3397 struct page *page;
3398 struct mem_cgroup *memcg = NULL;
3399
3400 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
3401 return NULL;
3402 page = alloc_pages_node(nid, gfp_mask, order);
3403 memcg_kmem_commit_charge(page, memcg, order);
3404 return page;
3405}
3406
3407/*
3408 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3409 * alloc_kmem_pages.
3410 */
3411void __free_kmem_pages(struct page *page, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003412{
3413 memcg_kmem_uncharge_pages(page, order);
3414 __free_pages(page, order);
3415}
3416
Vladimir Davydov52383432014-06-04 16:06:39 -07003417void free_kmem_pages(unsigned long addr, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003418{
3419 if (addr != 0) {
3420 VM_BUG_ON(!virt_addr_valid((void *)addr));
Vladimir Davydov52383432014-06-04 16:06:39 -07003421 __free_kmem_pages(virt_to_page((void *)addr), order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003422 }
3423}
3424
Andi Kleenee85c2e2011-05-11 15:13:34 -07003425static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
3426{
3427 if (addr) {
3428 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3429 unsigned long used = addr + PAGE_ALIGN(size);
3430
3431 split_page(virt_to_page((void *)addr), order);
3432 while (used < alloc_end) {
3433 free_page(used);
3434 used += PAGE_SIZE;
3435 }
3436 }
3437 return (void *)addr;
3438}
3439
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003440/**
3441 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3442 * @size: the number of bytes to allocate
3443 * @gfp_mask: GFP flags for the allocation
3444 *
3445 * This function is similar to alloc_pages(), except that it allocates the
3446 * minimum number of pages to satisfy the request. alloc_pages() can only
3447 * allocate memory in power-of-two pages.
3448 *
3449 * This function is also limited by MAX_ORDER.
3450 *
3451 * Memory allocated by this function must be released by free_pages_exact().
3452 */
3453void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3454{
3455 unsigned int order = get_order(size);
3456 unsigned long addr;
3457
3458 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07003459 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003460}
3461EXPORT_SYMBOL(alloc_pages_exact);
3462
3463/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07003464 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3465 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07003466 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07003467 * @size: the number of bytes to allocate
3468 * @gfp_mask: GFP flags for the allocation
3469 *
3470 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3471 * back.
3472 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
3473 * but is not exact.
3474 */
Fabian Fredericke1931812014-08-06 16:04:59 -07003475void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
Andi Kleenee85c2e2011-05-11 15:13:34 -07003476{
3477 unsigned order = get_order(size);
3478 struct page *p = alloc_pages_node(nid, gfp_mask, order);
3479 if (!p)
3480 return NULL;
3481 return make_alloc_exact((unsigned long)page_address(p), order, size);
3482}
Andi Kleenee85c2e2011-05-11 15:13:34 -07003483
3484/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003485 * free_pages_exact - release memory allocated via alloc_pages_exact()
3486 * @virt: the value returned by alloc_pages_exact.
3487 * @size: size of allocation, same value as passed to alloc_pages_exact().
3488 *
3489 * Release the memory allocated by a previous call to alloc_pages_exact.
3490 */
3491void free_pages_exact(void *virt, size_t size)
3492{
3493 unsigned long addr = (unsigned long)virt;
3494 unsigned long end = addr + PAGE_ALIGN(size);
3495
3496 while (addr < end) {
3497 free_page(addr);
3498 addr += PAGE_SIZE;
3499 }
3500}
3501EXPORT_SYMBOL(free_pages_exact);
3502
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003503/**
3504 * nr_free_zone_pages - count number of pages beyond high watermark
3505 * @offset: The zone index of the highest zone
3506 *
3507 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3508 * high watermark within all zones at or below a given zone index. For each
3509 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07003510 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003511 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003512static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003513{
Mel Gormandd1a2392008-04-28 02:12:17 -07003514 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003515 struct zone *zone;
3516
Martin J. Blighe310fd42005-07-29 22:59:18 -07003517 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003518 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519
Mel Gorman0e884602008-04-28 02:12:14 -07003520 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003521
Mel Gorman54a6eb52008-04-28 02:12:16 -07003522 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08003523 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003524 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07003525 if (size > high)
3526 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003527 }
3528
3529 return sum;
3530}
3531
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003532/**
3533 * nr_free_buffer_pages - count number of pages beyond high watermark
3534 *
3535 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3536 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003537 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003538unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003539{
Al Viroaf4ca452005-10-21 02:55:38 -04003540 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003541}
Meelap Shahc2f1a552007-07-17 04:04:39 -07003542EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003543
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003544/**
3545 * nr_free_pagecache_pages - count number of pages beyond high watermark
3546 *
3547 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3548 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003549 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003550unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003551{
Mel Gorman2a1e2742007-07-17 04:03:12 -07003552 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003553}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07003554
3555static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003556{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08003557 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08003558 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003559}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561void si_meminfo(struct sysinfo *val)
3562{
3563 val->totalram = totalram_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003564 val->sharedram = global_page_state(NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003565 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003566 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567 val->totalhigh = totalhigh_pages;
3568 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003569 val->mem_unit = PAGE_SIZE;
3570}
3571
3572EXPORT_SYMBOL(si_meminfo);
3573
3574#ifdef CONFIG_NUMA
3575void si_meminfo_node(struct sysinfo *val, int nid)
3576{
Jiang Liucdd91a72013-07-03 15:03:27 -07003577 int zone_type; /* needs to be signed */
3578 unsigned long managed_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579 pg_data_t *pgdat = NODE_DATA(nid);
3580
Jiang Liucdd91a72013-07-03 15:03:27 -07003581 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3582 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3583 val->totalram = managed_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003584 val->sharedram = node_page_state(nid, NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003585 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003586#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08003587 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08003588 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3589 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003590#else
3591 val->totalhigh = 0;
3592 val->freehigh = 0;
3593#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594 val->mem_unit = PAGE_SIZE;
3595}
3596#endif
3597
David Rientjesddd588b2011-03-22 16:30:46 -07003598/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07003599 * Determine whether the node should be displayed or not, depending on whether
3600 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07003601 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003602bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07003603{
3604 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003605 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07003606
3607 if (!(flags & SHOW_MEM_FILTER_NODES))
3608 goto out;
3609
Mel Gormancc9a6c82012-03-21 16:34:11 -07003610 do {
Mel Gormand26914d2014-04-03 14:47:24 -07003611 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003612 ret = !node_isset(nid, cpuset_current_mems_allowed);
Mel Gormand26914d2014-04-03 14:47:24 -07003613 } while (read_mems_allowed_retry(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07003614out:
3615 return ret;
3616}
3617
Linus Torvalds1da177e2005-04-16 15:20:36 -07003618#define K(x) ((x) << (PAGE_SHIFT-10))
3619
Rabin Vincent377e4f12012-12-11 16:00:24 -08003620static void show_migration_types(unsigned char type)
3621{
3622 static const char types[MIGRATE_TYPES] = {
3623 [MIGRATE_UNMOVABLE] = 'U',
3624 [MIGRATE_RECLAIMABLE] = 'E',
3625 [MIGRATE_MOVABLE] = 'M',
3626 [MIGRATE_RESERVE] = 'R',
3627#ifdef CONFIG_CMA
3628 [MIGRATE_CMA] = 'C',
3629#endif
Minchan Kim194159f2013-02-22 16:33:58 -08003630#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08003631 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08003632#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08003633 };
3634 char tmp[MIGRATE_TYPES + 1];
3635 char *p = tmp;
3636 int i;
3637
3638 for (i = 0; i < MIGRATE_TYPES; i++) {
3639 if (type & (1 << i))
3640 *p++ = types[i];
3641 }
3642
3643 *p = '\0';
3644 printk("(%s) ", tmp);
3645}
3646
Linus Torvalds1da177e2005-04-16 15:20:36 -07003647/*
3648 * Show free area list (used inside shift_scroll-lock stuff)
3649 * We also calculate the percentage fragmentation. We do this by counting the
3650 * memory on each free list with the exception of the first item on the list.
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003651 *
3652 * Bits in @filter:
3653 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3654 * cpuset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003656void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657{
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003658 unsigned long free_pcp = 0;
Jes Sorensenc7241912006-09-27 01:50:05 -07003659 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660 struct zone *zone;
3661
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003662 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07003663 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003664 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003665
Konstantin Khlebnikov761b0672015-04-14 15:45:32 -07003666 for_each_online_cpu(cpu)
3667 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003668 }
3669
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003670 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3671 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003672 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3673 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003674 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003675 " free:%lu free_pcp:%lu free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003676 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003677 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003678 global_page_state(NR_ISOLATED_ANON),
3679 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003680 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003681 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003682 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003683 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003684 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003685 global_page_state(NR_UNSTABLE_NFS),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003686 global_page_state(NR_SLAB_RECLAIMABLE),
3687 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003688 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003689 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003690 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003691 global_page_state(NR_BOUNCE),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003692 global_page_state(NR_FREE_PAGES),
3693 free_pcp,
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003694 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003695
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003696 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003697 int i;
3698
David Rientjes7bf02ea2011-05-24 17:11:16 -07003699 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003700 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003701
3702 free_pcp = 0;
3703 for_each_online_cpu(cpu)
3704 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3705
Linus Torvalds1da177e2005-04-16 15:20:36 -07003706 show_node(zone);
3707 printk("%s"
3708 " free:%lukB"
3709 " min:%lukB"
3710 " low:%lukB"
3711 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003712 " active_anon:%lukB"
3713 " inactive_anon:%lukB"
3714 " active_file:%lukB"
3715 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003716 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003717 " isolated(anon):%lukB"
3718 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003719 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003720 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003721 " mlocked:%lukB"
3722 " dirty:%lukB"
3723 " writeback:%lukB"
3724 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003725 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003726 " slab_reclaimable:%lukB"
3727 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003728 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003729 " pagetables:%lukB"
3730 " unstable:%lukB"
3731 " bounce:%lukB"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003732 " free_pcp:%lukB"
3733 " local_pcp:%ukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003734 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003735 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003736 " pages_scanned:%lu"
3737 " all_unreclaimable? %s"
3738 "\n",
3739 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003740 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003741 K(min_wmark_pages(zone)),
3742 K(low_wmark_pages(zone)),
3743 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003744 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3745 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3746 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3747 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003748 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003749 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3750 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003752 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003753 K(zone_page_state(zone, NR_MLOCK)),
3754 K(zone_page_state(zone, NR_FILE_DIRTY)),
3755 K(zone_page_state(zone, NR_WRITEBACK)),
3756 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003757 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003758 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3759 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003760 zone_page_state(zone, NR_KERNEL_STACK) *
3761 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003762 K(zone_page_state(zone, NR_PAGETABLE)),
3763 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3764 K(zone_page_state(zone, NR_BOUNCE)),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003765 K(free_pcp),
3766 K(this_cpu_read(zone->pageset->pcp.count)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003767 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003768 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Mel Gorman0d5d8232014-08-06 16:07:16 -07003769 K(zone_page_state(zone, NR_PAGES_SCANNED)),
Lisa Du6e543d52013-09-11 14:22:36 -07003770 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771 );
3772 printk("lowmem_reserve[]:");
3773 for (i = 0; i < MAX_NR_ZONES; i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07003774 printk(" %ld", zone->lowmem_reserve[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775 printk("\n");
3776 }
3777
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003778 for_each_populated_zone(zone) {
Pintu Kumarb8af2942013-09-11 14:20:34 -07003779 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003780 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781
David Rientjes7bf02ea2011-05-24 17:11:16 -07003782 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003783 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784 show_node(zone);
3785 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786
3787 spin_lock_irqsave(&zone->lock, flags);
3788 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08003789 struct free_area *area = &zone->free_area[order];
3790 int type;
3791
3792 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003793 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003794
3795 types[order] = 0;
3796 for (type = 0; type < MIGRATE_TYPES; type++) {
3797 if (!list_empty(&area->free_list[type]))
3798 types[order] |= 1 << type;
3799 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800 }
3801 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003802 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003803 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003804 if (nr[order])
3805 show_migration_types(types[order]);
3806 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807 printk("= %lukB\n", K(total));
3808 }
3809
David Rientjes949f7ec2013-04-29 15:07:48 -07003810 hugetlb_show_meminfo();
3811
Larry Woodmane6f36022008-02-04 22:29:30 -08003812 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3813
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814 show_swap_cache_info();
3815}
3816
Mel Gorman19770b32008-04-28 02:12:18 -07003817static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3818{
3819 zoneref->zone = zone;
3820 zoneref->zone_idx = zone_idx(zone);
3821}
3822
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823/*
3824 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003825 *
3826 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003828static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003829 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830{
Christoph Lameter1a932052006-01-06 00:11:16 -08003831 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003832 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003833
3834 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003835 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003836 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003837 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003838 zoneref_set_zone(zone,
3839 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003840 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003841 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003842 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003843
Christoph Lameter070f8032006-01-06 00:11:19 -08003844 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003845}
3846
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003847
3848/*
3849 * zonelist_order:
3850 * 0 = automatic detection of better ordering.
3851 * 1 = order by ([node] distance, -zonetype)
3852 * 2 = order by (-zonetype, [node] distance)
3853 *
3854 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3855 * the same zonelist. So only NUMA can configure this param.
3856 */
3857#define ZONELIST_ORDER_DEFAULT 0
3858#define ZONELIST_ORDER_NODE 1
3859#define ZONELIST_ORDER_ZONE 2
3860
3861/* zonelist order in the kernel.
3862 * set_zonelist_order() will set this to NODE or ZONE.
3863 */
3864static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3865static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3866
3867
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003869/* The value user specified ....changed by config */
3870static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3871/* string for sysctl */
3872#define NUMA_ZONELIST_ORDER_LEN 16
3873char numa_zonelist_order[16] = "default";
3874
3875/*
3876 * interface for configure zonelist ordering.
3877 * command line option "numa_zonelist_order"
3878 * = "[dD]efault - default, automatic configuration.
3879 * = "[nN]ode - order by node locality, then by zone within node
3880 * = "[zZ]one - order by zone, then by locality within zone
3881 */
3882
3883static int __parse_numa_zonelist_order(char *s)
3884{
3885 if (*s == 'd' || *s == 'D') {
3886 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3887 } else if (*s == 'n' || *s == 'N') {
3888 user_zonelist_order = ZONELIST_ORDER_NODE;
3889 } else if (*s == 'z' || *s == 'Z') {
3890 user_zonelist_order = ZONELIST_ORDER_ZONE;
3891 } else {
3892 printk(KERN_WARNING
3893 "Ignoring invalid numa_zonelist_order value: "
3894 "%s\n", s);
3895 return -EINVAL;
3896 }
3897 return 0;
3898}
3899
3900static __init int setup_numa_zonelist_order(char *s)
3901{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003902 int ret;
3903
3904 if (!s)
3905 return 0;
3906
3907 ret = __parse_numa_zonelist_order(s);
3908 if (ret == 0)
3909 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3910
3911 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003912}
3913early_param("numa_zonelist_order", setup_numa_zonelist_order);
3914
3915/*
3916 * sysctl handler for numa_zonelist_order
3917 */
Joe Perchescccad5b2014-06-06 14:38:09 -07003918int numa_zonelist_order_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003919 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003920 loff_t *ppos)
3921{
3922 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3923 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003924 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003925
Andi Kleen443c6f12009-12-23 21:00:47 +01003926 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07003927 if (write) {
3928 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3929 ret = -EINVAL;
3930 goto out;
3931 }
3932 strcpy(saved_string, (char *)table->data);
3933 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003934 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003935 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003936 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003937 if (write) {
3938 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07003939
3940 ret = __parse_numa_zonelist_order((char *)table->data);
3941 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003942 /*
3943 * bogus value. restore saved string
3944 */
Chen Gangdacbde02013-07-03 15:02:35 -07003945 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003946 NUMA_ZONELIST_ORDER_LEN);
3947 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003948 } else if (oldval != user_zonelist_order) {
3949 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07003950 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003951 mutex_unlock(&zonelists_mutex);
3952 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003953 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003954out:
3955 mutex_unlock(&zl_order_mutex);
3956 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003957}
3958
3959
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003960#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003961static int node_load[MAX_NUMNODES];
3962
Linus Torvalds1da177e2005-04-16 15:20:36 -07003963/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07003964 * find_next_best_node - find the next node that should appear in a given node's fallback list
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965 * @node: node whose fallback list we're appending
3966 * @used_node_mask: nodemask_t of already used nodes
3967 *
3968 * We use a number of factors to determine which is the next node that should
3969 * appear on a given node's fallback list. The node should not have appeared
3970 * already in @node's fallback list, and it should be the next closest node
3971 * according to the distance array (which contains arbitrary distance values
3972 * from each node to each node in the system), and should also prefer nodes
3973 * with no CPUs, since presumably they'll have very little allocation pressure
3974 * on them otherwise.
3975 * It returns -1 if no node is found.
3976 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003977static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003978{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003979 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003980 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08003981 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10303982 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003984 /* Use the local node if we haven't already */
3985 if (!node_isset(node, *used_node_mask)) {
3986 node_set(node, *used_node_mask);
3987 return node;
3988 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08003990 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991
3992 /* Don't want a node to appear more than once */
3993 if (node_isset(n, *used_node_mask))
3994 continue;
3995
Linus Torvalds1da177e2005-04-16 15:20:36 -07003996 /* Use the distance array to find the distance */
3997 val = node_distance(node, n);
3998
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003999 /* Penalize nodes under us ("prefer the next node") */
4000 val += (n < node);
4001
Linus Torvalds1da177e2005-04-16 15:20:36 -07004002 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10304003 tmp = cpumask_of_node(n);
4004 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005 val += PENALTY_FOR_NODE_WITH_CPUS;
4006
4007 /* Slight preference for less loaded node */
4008 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4009 val += node_load[n];
4010
4011 if (val < min_val) {
4012 min_val = val;
4013 best_node = n;
4014 }
4015 }
4016
4017 if (best_node >= 0)
4018 node_set(best_node, *used_node_mask);
4019
4020 return best_node;
4021}
4022
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004023
4024/*
4025 * Build zonelists ordered by node and zones within node.
4026 * This results in maximum locality--normal zone overflows into local
4027 * DMA zone, if any--but risks exhausting DMA zone.
4028 */
4029static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004030{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004031 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004032 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004033
Mel Gorman54a6eb52008-04-28 02:12:16 -07004034 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07004035 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07004036 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004037 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07004038 zonelist->_zonerefs[j].zone = NULL;
4039 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004040}
4041
4042/*
Christoph Lameter523b9452007-10-16 01:25:37 -07004043 * Build gfp_thisnode zonelists
4044 */
4045static void build_thisnode_zonelists(pg_data_t *pgdat)
4046{
Christoph Lameter523b9452007-10-16 01:25:37 -07004047 int j;
4048 struct zonelist *zonelist;
4049
Mel Gorman54a6eb52008-04-28 02:12:16 -07004050 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004051 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07004052 zonelist->_zonerefs[j].zone = NULL;
4053 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07004054}
4055
4056/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004057 * Build zonelists ordered by zone and nodes within zones.
4058 * This results in conserving DMA zone[s] until all Normal memory is
4059 * exhausted, but results in overflowing to remote node while memory
4060 * may still exist in local DMA zone.
4061 */
4062static int node_order[MAX_NUMNODES];
4063
4064static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4065{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004066 int pos, j, node;
4067 int zone_type; /* needs to be signed */
4068 struct zone *z;
4069 struct zonelist *zonelist;
4070
Mel Gorman54a6eb52008-04-28 02:12:16 -07004071 zonelist = &pgdat->node_zonelists[0];
4072 pos = 0;
4073 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4074 for (j = 0; j < nr_nodes; j++) {
4075 node = node_order[j];
4076 z = &NODE_DATA(node)->node_zones[zone_type];
4077 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07004078 zoneref_set_zone(z,
4079 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004080 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004081 }
4082 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004083 }
Mel Gormandd1a2392008-04-28 02:12:17 -07004084 zonelist->_zonerefs[pos].zone = NULL;
4085 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004086}
4087
Mel Gorman31939132014-10-09 15:28:30 -07004088#if defined(CONFIG_64BIT)
4089/*
4090 * Devices that require DMA32/DMA are relatively rare and do not justify a
4091 * penalty to every machine in case the specialised case applies. Default
4092 * to Node-ordering on 64-bit NUMA machines
4093 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004094static int default_zonelist_order(void)
4095{
Mel Gorman31939132014-10-09 15:28:30 -07004096 return ZONELIST_ORDER_NODE;
4097}
4098#else
4099/*
4100 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4101 * by the kernel. If processes running on node 0 deplete the low memory zone
4102 * then reclaim will occur more frequency increasing stalls and potentially
4103 * be easier to OOM if a large percentage of the zone is under writeback or
4104 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4105 * Hence, default to zone ordering on 32-bit.
4106 */
4107static int default_zonelist_order(void)
4108{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004109 return ZONELIST_ORDER_ZONE;
4110}
Mel Gorman31939132014-10-09 15:28:30 -07004111#endif /* CONFIG_64BIT */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004112
4113static void set_zonelist_order(void)
4114{
4115 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4116 current_zonelist_order = default_zonelist_order();
4117 else
4118 current_zonelist_order = user_zonelist_order;
4119}
4120
4121static void build_zonelists(pg_data_t *pgdat)
4122{
4123 int j, node, load;
4124 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004125 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004126 int local_node, prev_node;
4127 struct zonelist *zonelist;
4128 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004129
4130 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07004131 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004132 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07004133 zonelist->_zonerefs[0].zone = NULL;
4134 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004135 }
4136
4137 /* NUMA-aware ordering of nodes */
4138 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004139 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004140 prev_node = local_node;
4141 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004142
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004143 memset(node_order, 0, sizeof(node_order));
4144 j = 0;
4145
Linus Torvalds1da177e2005-04-16 15:20:36 -07004146 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4147 /*
4148 * We don't want to pressure a particular node.
4149 * So adding penalty to the first node in same
4150 * distance group to make it round-robin.
4151 */
David Rientjes957f8222012-10-08 16:33:24 -07004152 if (node_distance(local_node, node) !=
4153 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004154 node_load[node] = load;
4155
Linus Torvalds1da177e2005-04-16 15:20:36 -07004156 prev_node = node;
4157 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004158 if (order == ZONELIST_ORDER_NODE)
4159 build_zonelists_in_node_order(pgdat, node);
4160 else
4161 node_order[j++] = node; /* remember order */
4162 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004163
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004164 if (order == ZONELIST_ORDER_ZONE) {
4165 /* calculate node order -- i.e., DMA last! */
4166 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004167 }
Christoph Lameter523b9452007-10-16 01:25:37 -07004168
4169 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004170}
4171
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004172/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004173static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004174{
Mel Gorman54a6eb52008-04-28 02:12:16 -07004175 struct zonelist *zonelist;
4176 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07004177 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004178
Mel Gorman54a6eb52008-04-28 02:12:16 -07004179 zonelist = &pgdat->node_zonelists[0];
4180 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
4181 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07004182 for (z = zonelist->_zonerefs; z->zone; z++)
4183 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004184}
4185
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004186#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4187/*
4188 * Return node id of node used for "local" allocations.
4189 * I.e., first node id of first zone in arg node's generic zonelist.
4190 * Used for initializing percpu 'numa_mem', which is used primarily
4191 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4192 */
4193int local_memory_node(int node)
4194{
4195 struct zone *zone;
4196
4197 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
4198 gfp_zone(GFP_KERNEL),
4199 NULL,
4200 &zone);
4201 return zone->node;
4202}
4203#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004204
Linus Torvalds1da177e2005-04-16 15:20:36 -07004205#else /* CONFIG_NUMA */
4206
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004207static void set_zonelist_order(void)
4208{
4209 current_zonelist_order = ZONELIST_ORDER_ZONE;
4210}
4211
4212static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004213{
Christoph Lameter19655d32006-09-25 23:31:19 -07004214 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07004215 enum zone_type j;
4216 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004217
4218 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004219
Mel Gorman54a6eb52008-04-28 02:12:16 -07004220 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004221 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004222
Mel Gorman54a6eb52008-04-28 02:12:16 -07004223 /*
4224 * Now we build the zonelist so that it contains the zones
4225 * of all the other nodes.
4226 * We don't want to pressure a particular node, so when
4227 * building the zones for node N, we make sure that the
4228 * zones coming right after the local ones are those from
4229 * node N+1 (modulo N)
4230 */
4231 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4232 if (!node_online(node))
4233 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004234 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004235 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07004236 for (node = 0; node < local_node; node++) {
4237 if (!node_online(node))
4238 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004239 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004240 }
4241
Mel Gormandd1a2392008-04-28 02:12:17 -07004242 zonelist->_zonerefs[j].zone = NULL;
4243 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004244}
4245
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004246/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004247static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004248{
Mel Gorman54a6eb52008-04-28 02:12:16 -07004249 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004250}
4251
Linus Torvalds1da177e2005-04-16 15:20:36 -07004252#endif /* CONFIG_NUMA */
4253
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004254/*
4255 * Boot pageset table. One per cpu which is going to be used for all
4256 * zones and all nodes. The parameters will be set in such a way
4257 * that an item put on a list will immediately be handed over to
4258 * the buddy list. This is safe since pageset manipulation is done
4259 * with interrupts disabled.
4260 *
4261 * The boot_pagesets must be kept even after bootup is complete for
4262 * unused processors and/or zones. They do play a role for bootstrapping
4263 * hotplugged processors.
4264 *
4265 * zoneinfo_show() and maybe other functions do
4266 * not check if the processor is online before following the pageset pointer.
4267 * Other parts of the kernel may not check if the zone is available.
4268 */
4269static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4270static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07004271static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004272
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004273/*
4274 * Global mutex to protect against size modification of zonelists
4275 * as well as to serialize pageset setup for the new populated zone.
4276 */
4277DEFINE_MUTEX(zonelists_mutex);
4278
Rusty Russell9b1a4d32008-07-28 12:16:30 -05004279/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07004280static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004281{
Yasunori Goto68113782006-06-23 02:03:11 -07004282 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004283 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07004284 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004285
Bo Liu7f9cfb32009-08-18 14:11:19 -07004286#ifdef CONFIG_NUMA
4287 memset(node_load, 0, sizeof(node_load));
4288#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07004289
4290 if (self && !node_online(self->node_id)) {
4291 build_zonelists(self);
4292 build_zonelist_cache(self);
4293 }
4294
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004295 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07004296 pg_data_t *pgdat = NODE_DATA(nid);
4297
4298 build_zonelists(pgdat);
4299 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004300 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004301
4302 /*
4303 * Initialize the boot_pagesets that are going to be used
4304 * for bootstrapping processors. The real pagesets for
4305 * each zone will be allocated later when the per cpu
4306 * allocator is available.
4307 *
4308 * boot_pagesets are used also for bootstrapping offline
4309 * cpus if the system is already booted because the pagesets
4310 * are needed to initialize allocators on a specific cpu too.
4311 * F.e. the percpu allocator needs the page allocator which
4312 * needs the percpu allocator in order to allocate its pagesets
4313 * (a chicken-egg dilemma).
4314 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004315 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004316 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4317
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004318#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4319 /*
4320 * We now know the "local memory node" for each node--
4321 * i.e., the node of the first zone in the generic zonelist.
4322 * Set up numa_mem percpu variable for on-line cpus. During
4323 * boot, only the boot cpu should be on-line; we'll init the
4324 * secondary cpus' numa_mem as they come on-line. During
4325 * node/memory hotplug, we'll fixup all on-line cpus.
4326 */
4327 if (cpu_online(cpu))
4328 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4329#endif
4330 }
4331
Yasunori Goto68113782006-06-23 02:03:11 -07004332 return 0;
4333}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004334
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004335static noinline void __init
4336build_all_zonelists_init(void)
4337{
4338 __build_all_zonelists(NULL);
4339 mminit_verify_zonelist();
4340 cpuset_init_current_mems_allowed();
4341}
4342
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004343/*
4344 * Called with zonelists_mutex held always
4345 * unless system_state == SYSTEM_BOOTING.
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004346 *
4347 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4348 * [we're only called with non-NULL zone through __meminit paths] and
4349 * (2) call of __init annotated helper build_all_zonelists_init
4350 * [protected by SYSTEM_BOOTING].
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004351 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07004352void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07004353{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004354 set_zonelist_order();
4355
Yasunori Goto68113782006-06-23 02:03:11 -07004356 if (system_state == SYSTEM_BOOTING) {
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004357 build_all_zonelists_init();
Yasunori Goto68113782006-06-23 02:03:11 -07004358 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08004359#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07004360 if (zone)
4361 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08004362#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07004363 /* we have to stop all cpus to guarantee there is no user
4364 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07004365 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07004366 /* cpuset refresh routine should be here */
4367 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07004368 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004369 /*
4370 * Disable grouping by mobility if the number of pages in the
4371 * system is too low to allow the mechanism to work. It would be
4372 * more accurate, but expensive to check per-zone. This check is
4373 * made on memory-hotadd so a system can start with mobility
4374 * disabled and enable it later
4375 */
Mel Gormand9c23402007-10-16 01:26:01 -07004376 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004377 page_group_by_mobility_disabled = 1;
4378 else
4379 page_group_by_mobility_disabled = 0;
4380
Anton Blanchardf88dfff2014-12-10 15:42:53 -08004381 pr_info("Built %i zonelists in %s order, mobility grouping %s. "
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004382 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004383 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004384 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004385 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004386 vm_total_pages);
4387#ifdef CONFIG_NUMA
Anton Blanchardf88dfff2014-12-10 15:42:53 -08004388 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004389#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004390}
4391
4392/*
4393 * Helper functions to size the waitqueue hash table.
4394 * Essentially these want to choose hash table sizes sufficiently
4395 * large so that collisions trying to wait on pages are rare.
4396 * But in fact, the number of active page waitqueues on typical
4397 * systems is ridiculously low, less than 200. So this is even
4398 * conservative, even though it seems large.
4399 *
4400 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4401 * waitqueues, i.e. the size of the waitq table given the number of pages.
4402 */
4403#define PAGES_PER_WAITQUEUE 256
4404
Yasunori Gotocca448f2006-06-23 02:03:10 -07004405#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07004406static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004407{
4408 unsigned long size = 1;
4409
4410 pages /= PAGES_PER_WAITQUEUE;
4411
4412 while (size < pages)
4413 size <<= 1;
4414
4415 /*
4416 * Once we have dozens or even hundreds of threads sleeping
4417 * on IO we've got bigger problems than wait queue collision.
4418 * Limit the size of the wait table to a reasonable size.
4419 */
4420 size = min(size, 4096UL);
4421
4422 return max(size, 4UL);
4423}
Yasunori Gotocca448f2006-06-23 02:03:10 -07004424#else
4425/*
4426 * A zone's size might be changed by hot-add, so it is not possible to determine
4427 * a suitable size for its wait_table. So we use the maximum size now.
4428 *
4429 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4430 *
4431 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4432 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4433 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4434 *
4435 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4436 * or more by the traditional way. (See above). It equals:
4437 *
4438 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4439 * ia64(16K page size) : = ( 8G + 4M)byte.
4440 * powerpc (64K page size) : = (32G +16M)byte.
4441 */
4442static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4443{
4444 return 4096UL;
4445}
4446#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004447
4448/*
4449 * This is an integer logarithm so that shifts can be used later
4450 * to extract the more random high bits from the multiplicative
4451 * hash function before the remainder is taken.
4452 */
4453static inline unsigned long wait_table_bits(unsigned long size)
4454{
4455 return ffz(~size);
4456}
4457
Mel Gorman56fd56b2007-10-16 01:25:58 -07004458/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07004459 * Check if a pageblock contains reserved pages
4460 */
4461static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
4462{
4463 unsigned long pfn;
4464
4465 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4466 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
4467 return 1;
4468 }
4469 return 0;
4470}
4471
4472/*
Mel Gormand9c23402007-10-16 01:26:01 -07004473 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07004474 * of blocks reserved is based on min_wmark_pages(zone). The memory within
4475 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07004476 * higher will lead to a bigger reserve which will get freed as contiguous
4477 * blocks as reclaim kicks in
4478 */
4479static void setup_zone_migrate_reserve(struct zone *zone)
4480{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07004481 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004482 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07004483 unsigned long block_migratetype;
4484 int reserve;
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004485 int old_reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004486
Michal Hockod02156382011-12-08 14:34:27 -08004487 /*
4488 * Get the start pfn, end pfn and the number of blocks to reserve
4489 * We have to be careful to be aligned to pageblock_nr_pages to
4490 * make sure that we always check pfn_valid for the first page in
4491 * the block.
4492 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07004493 start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08004494 end_pfn = zone_end_pfn(zone);
Michal Hockod02156382011-12-08 14:34:27 -08004495 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07004496 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07004497 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004498
Mel Gorman78986a62009-09-21 17:03:02 -07004499 /*
4500 * Reserve blocks are generally in place to help high-order atomic
4501 * allocations that are short-lived. A min_free_kbytes value that
4502 * would result in more than 2 reserve blocks for atomic allocations
4503 * is assumed to be in place to help anti-fragmentation for the
4504 * future allocation of hugepages at runtime.
4505 */
4506 reserve = min(2, reserve);
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004507 old_reserve = zone->nr_migrate_reserve_block;
4508
4509 /* When memory hot-add, we almost always need to do nothing */
4510 if (reserve == old_reserve)
4511 return;
4512 zone->nr_migrate_reserve_block = reserve;
Mel Gorman78986a62009-09-21 17:03:02 -07004513
Mel Gormand9c23402007-10-16 01:26:01 -07004514 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman7e18adb2015-06-30 14:57:05 -07004515 if (!early_page_nid_uninitialised(pfn, zone_to_nid(zone)))
4516 return;
4517
Mel Gorman56fd56b2007-10-16 01:25:58 -07004518 if (!pfn_valid(pfn))
4519 continue;
4520 page = pfn_to_page(pfn);
4521
Adam Litke344c7902008-09-02 14:35:38 -07004522 /* Watch out for overlapping nodes */
4523 if (page_to_nid(page) != zone_to_nid(zone))
4524 continue;
4525
Mel Gorman56fd56b2007-10-16 01:25:58 -07004526 block_migratetype = get_pageblock_migratetype(page);
4527
Mel Gorman938929f2012-01-10 15:07:14 -08004528 /* Only test what is necessary when the reserves are not met */
4529 if (reserve > 0) {
4530 /*
4531 * Blocks with reserved pages will never free, skip
4532 * them.
4533 */
4534 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4535 if (pageblock_is_reserved(pfn, block_end_pfn))
4536 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004537
Mel Gorman938929f2012-01-10 15:07:14 -08004538 /* If this block is reserved, account for it */
4539 if (block_migratetype == MIGRATE_RESERVE) {
4540 reserve--;
4541 continue;
4542 }
4543
4544 /* Suitable for reserving if this block is movable */
4545 if (block_migratetype == MIGRATE_MOVABLE) {
4546 set_pageblock_migratetype(page,
4547 MIGRATE_RESERVE);
4548 move_freepages_block(zone, page,
4549 MIGRATE_RESERVE);
4550 reserve--;
4551 continue;
4552 }
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004553 } else if (!old_reserve) {
4554 /*
4555 * At boot time we don't need to scan the whole zone
4556 * for turning off MIGRATE_RESERVE.
4557 */
4558 break;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004559 }
4560
4561 /*
4562 * If the reserve is met and this is a previous reserved block,
4563 * take it back
4564 */
4565 if (block_migratetype == MIGRATE_RESERVE) {
4566 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4567 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4568 }
4569 }
4570}
Mel Gormanac0e5b72007-10-16 01:25:58 -07004571
Linus Torvalds1da177e2005-04-16 15:20:36 -07004572/*
4573 * Initially all pages are reserved - free ones are freed
4574 * up by free_all_bootmem() once the early boot process is
4575 * done. Non-atomic initialization, single-pass.
4576 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004577void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004578 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004579{
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004580 pg_data_t *pgdat = NODE_DATA(nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -07004581 unsigned long end_pfn = start_pfn + size;
4582 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004583 struct zone *z;
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004584 unsigned long nr_initialised = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004585
Hugh Dickins22b31ee2009-01-06 14:40:09 -08004586 if (highest_memmap_pfn < end_pfn - 1)
4587 highest_memmap_pfn = end_pfn - 1;
4588
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004589 z = &pgdat->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08004590 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08004591 /*
4592 * There can be holes in boot-time mem_map[]s
4593 * handed to this function. They do not
4594 * exist on hotplugged memory.
4595 */
4596 if (context == MEMMAP_EARLY) {
4597 if (!early_pfn_valid(pfn))
4598 continue;
4599 if (!early_pfn_in_nid(pfn, nid))
4600 continue;
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004601 if (!update_defer_init(pgdat, pfn, end_pfn,
4602 &nr_initialised))
4603 break;
Dave Hansena2f3aa022007-01-10 23:15:30 -08004604 }
Mel Gormanac5d2532015-06-30 14:57:20 -07004605
4606 /*
4607 * Mark the block movable so that blocks are reserved for
4608 * movable at startup. This will force kernel allocations
4609 * to reserve their blocks rather than leaking throughout
4610 * the address space during boot when many long-lived
4611 * kernel allocations are made. Later some blocks near
4612 * the start are marked MIGRATE_RESERVE by
4613 * setup_zone_migrate_reserve()
4614 *
4615 * bitmap is created for zone's valid pfn range. but memmap
4616 * can be created for invalid pages (for alignment)
4617 * check here not to call set_pageblock_migratetype() against
4618 * pfn out of zone.
4619 */
4620 if (!(pfn & (pageblock_nr_pages - 1))) {
4621 struct page *page = pfn_to_page(pfn);
4622
4623 __init_single_page(page, pfn, zone, nid);
4624 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4625 } else {
4626 __init_single_pfn(pfn, zone, nid);
4627 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004628 }
4629}
4630
Andi Kleen1e548de2008-02-04 22:29:26 -08004631static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004632{
Mel Gorman7aeb09f2014-06-04 16:10:21 -07004633 unsigned int order, t;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004634 for_each_migratetype_order(order, t) {
4635 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004636 zone->free_area[order].nr_free = 0;
4637 }
4638}
4639
4640#ifndef __HAVE_ARCH_MEMMAP_INIT
4641#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08004642 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004643#endif
4644
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004645static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004646{
David Howells3a6be872009-05-06 16:03:03 -07004647#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004648 int batch;
4649
4650 /*
4651 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07004652 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004653 *
4654 * OK, so we don't know how big the cache is. So guess.
4655 */
Jiang Liub40da042013-02-22 16:33:52 -08004656 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07004657 if (batch * PAGE_SIZE > 512 * 1024)
4658 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004659 batch /= 4; /* We effectively *= 4 below */
4660 if (batch < 1)
4661 batch = 1;
4662
4663 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004664 * Clamp the batch to a 2^n - 1 value. Having a power
4665 * of 2 value was found to be more likely to have
4666 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004667 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004668 * For example if 2 tasks are alternately allocating
4669 * batches of pages, one task can end up with a lot
4670 * of pages of one half of the possible page colors
4671 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004672 */
David Howells91552032009-05-06 16:03:02 -07004673 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004674
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004675 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004676
4677#else
4678 /* The deferral and batching of frees should be suppressed under NOMMU
4679 * conditions.
4680 *
4681 * The problem is that NOMMU needs to be able to allocate large chunks
4682 * of contiguous memory as there's no hardware page translation to
4683 * assemble apparent contiguous memory from discontiguous pages.
4684 *
4685 * Queueing large contiguous runs of pages for batching, however,
4686 * causes the pages to actually be freed in smaller chunks. As there
4687 * can be a significant delay between the individual batches being
4688 * recycled, this leads to the once large chunks of space being
4689 * fragmented and becoming unavailable for high-order allocations.
4690 */
4691 return 0;
4692#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004693}
4694
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004695/*
4696 * pcp->high and pcp->batch values are related and dependent on one another:
4697 * ->batch must never be higher then ->high.
4698 * The following function updates them in a safe manner without read side
4699 * locking.
4700 *
4701 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4702 * those fields changing asynchronously (acording the the above rule).
4703 *
4704 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4705 * outside of boot time (or some other assurance that no concurrent updaters
4706 * exist).
4707 */
4708static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4709 unsigned long batch)
4710{
4711 /* start with a fail safe value for batch */
4712 pcp->batch = 1;
4713 smp_wmb();
4714
4715 /* Update high, then batch, in order */
4716 pcp->high = high;
4717 smp_wmb();
4718
4719 pcp->batch = batch;
4720}
4721
Cody P Schafer36640332013-07-03 15:01:40 -07004722/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07004723static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4724{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004725 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07004726}
4727
Cody P Schafer88c90db2013-07-03 15:01:35 -07004728static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004729{
4730 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004731 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004732
Magnus Damm1c6fe942005-10-26 01:58:59 -07004733 memset(p, 0, sizeof(*p));
4734
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004735 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004736 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004737 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4738 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004739}
4740
Cody P Schafer88c90db2013-07-03 15:01:35 -07004741static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4742{
4743 pageset_init(p);
4744 pageset_set_batch(p, batch);
4745}
4746
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004747/*
Cody P Schafer36640332013-07-03 15:01:40 -07004748 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004749 * to the value high for the pageset p.
4750 */
Cody P Schafer36640332013-07-03 15:01:40 -07004751static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004752 unsigned long high)
4753{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004754 unsigned long batch = max(1UL, high / 4);
4755 if ((high / 4) > (PAGE_SHIFT * 8))
4756 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004757
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004758 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004759}
4760
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004761static void pageset_set_high_and_batch(struct zone *zone,
4762 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004763{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004764 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07004765 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004766 (zone->managed_pages /
4767 percpu_pagelist_fraction));
4768 else
4769 pageset_set_batch(pcp, zone_batchsize(zone));
4770}
4771
Cody P Schafer169f6c12013-07-03 15:01:41 -07004772static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4773{
4774 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4775
4776 pageset_init(pcp);
4777 pageset_set_high_and_batch(zone, pcp);
4778}
4779
Jiang Liu4ed7e022012-07-31 16:43:35 -07004780static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004781{
4782 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07004783 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004784 for_each_possible_cpu(cpu)
4785 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07004786}
4787
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004788/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004789 * Allocate per cpu pagesets and initialize them.
4790 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004791 */
Al Viro78d99552005-12-15 09:18:25 +00004792void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004793{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004794 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004795
Wu Fengguang319774e2010-05-24 14:32:49 -07004796 for_each_populated_zone(zone)
4797 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004798}
4799
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004800static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004801int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004802{
4803 int i;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004804 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004805
4806 /*
4807 * The per-page waitqueue mechanism uses hashed waitqueues
4808 * per zone.
4809 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004810 zone->wait_table_hash_nr_entries =
4811 wait_table_hash_nr_entries(zone_size_pages);
4812 zone->wait_table_bits =
4813 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004814 alloc_size = zone->wait_table_hash_nr_entries
4815 * sizeof(wait_queue_head_t);
4816
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004817 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004818 zone->wait_table = (wait_queue_head_t *)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004819 memblock_virt_alloc_node_nopanic(
4820 alloc_size, zone->zone_pgdat->node_id);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004821 } else {
4822 /*
4823 * This case means that a zone whose size was 0 gets new memory
4824 * via memory hot-add.
4825 * But it may be the case that a new node was hot-added. In
4826 * this case vmalloc() will not be able to use this new node's
4827 * memory - this wait_table must be initialized to use this new
4828 * node itself as well.
4829 * To use this new node's memory, further consideration will be
4830 * necessary.
4831 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004832 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004833 }
4834 if (!zone->wait_table)
4835 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004836
Pintu Kumarb8af2942013-09-11 14:20:34 -07004837 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004838 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004839
4840 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004841}
4842
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004843static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004844{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004845 /*
4846 * per cpu subsystem is not up at this point. The following code
4847 * relies on the ability of the linker to provide the
4848 * offset of a (static) per cpu variable into the per cpu area.
4849 */
4850 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004851
Xishi Qiub38a8722013-11-12 15:07:20 -08004852 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004853 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4854 zone->name, zone->present_pages,
4855 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004856}
4857
Jiang Liu4ed7e022012-07-31 16:43:35 -07004858int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004859 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004860 unsigned long size,
4861 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07004862{
4863 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004864 int ret;
4865 ret = zone_wait_table_init(zone, size);
4866 if (ret)
4867 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004868 pgdat->nr_zones = zone_idx(zone) + 1;
4869
Dave Hansened8ece22005-10-29 18:16:50 -07004870 zone->zone_start_pfn = zone_start_pfn;
4871
Mel Gorman708614e2008-07-23 21:26:51 -07004872 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4873 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4874 pgdat->node_id,
4875 (unsigned long)zone_idx(zone),
4876 zone_start_pfn, (zone_start_pfn + size));
4877
Andi Kleen1e548de2008-02-04 22:29:26 -08004878 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004879
4880 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004881}
4882
Tejun Heo0ee332c2011-12-08 10:22:09 -08004883#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004884#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
Mel Gorman8a942fd2015-06-30 14:56:55 -07004885
Mel Gormanc7132162006-09-27 01:49:43 -07004886/*
4887 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
Mel Gormanc7132162006-09-27 01:49:43 -07004888 */
Mel Gorman8a942fd2015-06-30 14:56:55 -07004889int __meminit __early_pfn_to_nid(unsigned long pfn,
4890 struct mminit_pfnnid_cache *state)
Mel Gormanc7132162006-09-27 01:49:43 -07004891{
Tejun Heoc13291a2011-07-12 10:46:30 +02004892 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004893 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004894
Mel Gorman8a942fd2015-06-30 14:56:55 -07004895 if (state->last_start <= pfn && pfn < state->last_end)
4896 return state->last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004897
Yinghai Lue76b63f2013-09-11 14:22:17 -07004898 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4899 if (nid != -1) {
Mel Gorman8a942fd2015-06-30 14:56:55 -07004900 state->last_start = start_pfn;
4901 state->last_end = end_pfn;
4902 state->last_nid = nid;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004903 }
4904
4905 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004906}
4907#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4908
Mel Gormanc7132162006-09-27 01:49:43 -07004909/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08004910 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004911 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08004912 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07004913 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004914 * If an architecture guarantees that all ranges registered contain no holes
4915 * and may be freed, this this function may be used instead of calling
4916 * memblock_free_early_nid() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004917 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004918void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004919{
Tejun Heoc13291a2011-07-12 10:46:30 +02004920 unsigned long start_pfn, end_pfn;
4921 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004922
Tejun Heoc13291a2011-07-12 10:46:30 +02004923 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4924 start_pfn = min(start_pfn, max_low_pfn);
4925 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004926
Tejun Heoc13291a2011-07-12 10:46:30 +02004927 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004928 memblock_free_early_nid(PFN_PHYS(start_pfn),
4929 (end_pfn - start_pfn) << PAGE_SHIFT,
4930 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07004931 }
4932}
4933
4934/**
4935 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004936 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004937 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004938 * If an architecture guarantees that all ranges registered contain no holes and may
4939 * be freed, this function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004940 */
4941void __init sparse_memory_present_with_active_regions(int nid)
4942{
Tejun Heoc13291a2011-07-12 10:46:30 +02004943 unsigned long start_pfn, end_pfn;
4944 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004945
Tejun Heoc13291a2011-07-12 10:46:30 +02004946 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4947 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004948}
4949
4950/**
4951 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004952 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4953 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4954 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004955 *
4956 * It returns the start and end page frame of a node based on information
Zhang Zhen7d018172014-06-04 16:10:53 -07004957 * provided by memblock_set_node(). If called for a node
Mel Gormanc7132162006-09-27 01:49:43 -07004958 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004959 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004960 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004961void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004962 unsigned long *start_pfn, unsigned long *end_pfn)
4963{
Tejun Heoc13291a2011-07-12 10:46:30 +02004964 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004965 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02004966
Mel Gormanc7132162006-09-27 01:49:43 -07004967 *start_pfn = -1UL;
4968 *end_pfn = 0;
4969
Tejun Heoc13291a2011-07-12 10:46:30 +02004970 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4971 *start_pfn = min(*start_pfn, this_start_pfn);
4972 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004973 }
4974
Christoph Lameter633c0662007-10-16 01:25:37 -07004975 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004976 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004977}
4978
4979/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07004980 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4981 * assumption is made that zones within a node are ordered in monotonic
4982 * increasing memory addresses so that the "highest" populated zone is used
4983 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004984static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004985{
4986 int zone_index;
4987 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4988 if (zone_index == ZONE_MOVABLE)
4989 continue;
4990
4991 if (arch_zone_highest_possible_pfn[zone_index] >
4992 arch_zone_lowest_possible_pfn[zone_index])
4993 break;
4994 }
4995
4996 VM_BUG_ON(zone_index == -1);
4997 movable_zone = zone_index;
4998}
4999
5000/*
5001 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03005002 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005003 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5004 * in each node depending on the size of each node and how evenly kernelcore
5005 * is distributed. This helper function adjusts the zone ranges
5006 * provided by the architecture for a given node by using the end of the
5007 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5008 * zones within a node are in order of monotonic increases memory addresses
5009 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005010static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005011 unsigned long zone_type,
5012 unsigned long node_start_pfn,
5013 unsigned long node_end_pfn,
5014 unsigned long *zone_start_pfn,
5015 unsigned long *zone_end_pfn)
5016{
5017 /* Only adjust if ZONE_MOVABLE is on this node */
5018 if (zone_movable_pfn[nid]) {
5019 /* Size ZONE_MOVABLE */
5020 if (zone_type == ZONE_MOVABLE) {
5021 *zone_start_pfn = zone_movable_pfn[nid];
5022 *zone_end_pfn = min(node_end_pfn,
5023 arch_zone_highest_possible_pfn[movable_zone]);
5024
5025 /* Adjust for ZONE_MOVABLE starting within this range */
5026 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
5027 *zone_end_pfn > zone_movable_pfn[nid]) {
5028 *zone_end_pfn = zone_movable_pfn[nid];
5029
5030 /* Check if this whole range is within ZONE_MOVABLE */
5031 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5032 *zone_start_pfn = *zone_end_pfn;
5033 }
5034}
5035
5036/*
Mel Gormanc7132162006-09-27 01:49:43 -07005037 * Return the number of pages a zone spans in a node, including holes
5038 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5039 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005040static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005041 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005042 unsigned long node_start_pfn,
5043 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005044 unsigned long *ignored)
5045{
Mel Gormanc7132162006-09-27 01:49:43 -07005046 unsigned long zone_start_pfn, zone_end_pfn;
5047
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005048 /* Get the start and end of the zone */
Mel Gormanc7132162006-09-27 01:49:43 -07005049 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5050 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07005051 adjust_zone_range_for_zone_movable(nid, zone_type,
5052 node_start_pfn, node_end_pfn,
5053 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005054
5055 /* Check that this node has pages within the zone's required range */
5056 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
5057 return 0;
5058
5059 /* Move the zone boundaries inside the node if necessary */
5060 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
5061 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
5062
5063 /* Return the spanned pages */
5064 return zone_end_pfn - zone_start_pfn;
5065}
5066
5067/*
5068 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005069 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07005070 */
Yinghai Lu32996252009-12-15 17:59:02 -08005071unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005072 unsigned long range_start_pfn,
5073 unsigned long range_end_pfn)
5074{
Tejun Heo96e907d2011-07-12 10:46:29 +02005075 unsigned long nr_absent = range_end_pfn - range_start_pfn;
5076 unsigned long start_pfn, end_pfn;
5077 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07005078
Tejun Heo96e907d2011-07-12 10:46:29 +02005079 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5080 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5081 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5082 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005083 }
Tejun Heo96e907d2011-07-12 10:46:29 +02005084 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07005085}
5086
5087/**
5088 * absent_pages_in_range - Return number of page frames in holes within a range
5089 * @start_pfn: The start PFN to start searching for holes
5090 * @end_pfn: The end PFN to stop searching for holes
5091 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005092 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07005093 */
5094unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5095 unsigned long end_pfn)
5096{
5097 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5098}
5099
5100/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005101static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005102 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005103 unsigned long node_start_pfn,
5104 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005105 unsigned long *ignored)
5106{
Tejun Heo96e907d2011-07-12 10:46:29 +02005107 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5108 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07005109 unsigned long zone_start_pfn, zone_end_pfn;
5110
Tejun Heo96e907d2011-07-12 10:46:29 +02005111 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5112 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07005113
Mel Gorman2a1e2742007-07-17 04:03:12 -07005114 adjust_zone_range_for_zone_movable(nid, zone_type,
5115 node_start_pfn, node_end_pfn,
5116 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07005117 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005118}
Mel Gorman0e0b8642006-09-27 01:49:56 -07005119
Tejun Heo0ee332c2011-12-08 10:22:09 -08005120#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005121static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005122 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005123 unsigned long node_start_pfn,
5124 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005125 unsigned long *zones_size)
5126{
5127 return zones_size[zone_type];
5128}
5129
Paul Mundt6ea6e682007-07-15 23:38:20 -07005130static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005131 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005132 unsigned long node_start_pfn,
5133 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005134 unsigned long *zholes_size)
5135{
5136 if (!zholes_size)
5137 return 0;
5138
5139 return zholes_size[zone_type];
5140}
Yinghai Lu20e69262013-03-01 14:51:27 -08005141
Tejun Heo0ee332c2011-12-08 10:22:09 -08005142#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005143
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005144static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005145 unsigned long node_start_pfn,
5146 unsigned long node_end_pfn,
5147 unsigned long *zones_size,
5148 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07005149{
Gu Zhengfebd5942015-06-24 16:57:02 -07005150 unsigned long realtotalpages = 0, totalpages = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005151 enum zone_type i;
5152
Gu Zhengfebd5942015-06-24 16:57:02 -07005153 for (i = 0; i < MAX_NR_ZONES; i++) {
5154 struct zone *zone = pgdat->node_zones + i;
5155 unsigned long size, real_size;
Mel Gormanc7132162006-09-27 01:49:43 -07005156
Gu Zhengfebd5942015-06-24 16:57:02 -07005157 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5158 node_start_pfn,
5159 node_end_pfn,
5160 zones_size);
5161 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005162 node_start_pfn, node_end_pfn,
5163 zholes_size);
Gu Zhengfebd5942015-06-24 16:57:02 -07005164 zone->spanned_pages = size;
5165 zone->present_pages = real_size;
5166
5167 totalpages += size;
5168 realtotalpages += real_size;
5169 }
5170
5171 pgdat->node_spanned_pages = totalpages;
Mel Gormanc7132162006-09-27 01:49:43 -07005172 pgdat->node_present_pages = realtotalpages;
5173 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5174 realtotalpages);
5175}
5176
Mel Gorman835c1342007-10-16 01:25:47 -07005177#ifndef CONFIG_SPARSEMEM
5178/*
5179 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07005180 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5181 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07005182 * round what is now in bits to nearest long in bits, then return it in
5183 * bytes.
5184 */
Linus Torvalds7c455122013-02-18 09:58:02 -08005185static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005186{
5187 unsigned long usemapsize;
5188
Linus Torvalds7c455122013-02-18 09:58:02 -08005189 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005190 usemapsize = roundup(zonesize, pageblock_nr_pages);
5191 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07005192 usemapsize *= NR_PAGEBLOCK_BITS;
5193 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5194
5195 return usemapsize / 8;
5196}
5197
5198static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08005199 struct zone *zone,
5200 unsigned long zone_start_pfn,
5201 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005202{
Linus Torvalds7c455122013-02-18 09:58:02 -08005203 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07005204 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08005205 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005206 zone->pageblock_flags =
5207 memblock_virt_alloc_node_nopanic(usemapsize,
5208 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07005209}
5210#else
Linus Torvalds7c455122013-02-18 09:58:02 -08005211static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5212 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07005213#endif /* CONFIG_SPARSEMEM */
5214
Mel Gormand9c23402007-10-16 01:26:01 -07005215#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08005216
Mel Gormand9c23402007-10-16 01:26:01 -07005217/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07005218void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07005219{
Andrew Morton955c1cd2012-05-29 15:06:31 -07005220 unsigned int order;
5221
Mel Gormand9c23402007-10-16 01:26:01 -07005222 /* Check that pageblock_nr_pages has not already been setup */
5223 if (pageblock_order)
5224 return;
5225
Andrew Morton955c1cd2012-05-29 15:06:31 -07005226 if (HPAGE_SHIFT > PAGE_SHIFT)
5227 order = HUGETLB_PAGE_ORDER;
5228 else
5229 order = MAX_ORDER - 1;
5230
Mel Gormand9c23402007-10-16 01:26:01 -07005231 /*
5232 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07005233 * This value may be variable depending on boot parameters on IA64 and
5234 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07005235 */
5236 pageblock_order = order;
5237}
5238#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5239
Mel Gormanba72cb82007-11-28 16:21:13 -08005240/*
5241 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07005242 * is unused as pageblock_order is set at compile-time. See
5243 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5244 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08005245 */
Chen Gang15ca2202013-09-11 14:20:27 -07005246void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08005247{
Mel Gormanba72cb82007-11-28 16:21:13 -08005248}
Mel Gormand9c23402007-10-16 01:26:01 -07005249
5250#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5251
Jiang Liu01cefae2012-12-12 13:52:19 -08005252static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5253 unsigned long present_pages)
5254{
5255 unsigned long pages = spanned_pages;
5256
5257 /*
5258 * Provide a more accurate estimation if there are holes within
5259 * the zone and SPARSEMEM is in use. If there are holes within the
5260 * zone, each populated memory region may cost us one or two extra
5261 * memmap pages due to alignment because memmap pages for each
5262 * populated regions may not naturally algined on page boundary.
5263 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5264 */
5265 if (spanned_pages > present_pages + (present_pages >> 4) &&
5266 IS_ENABLED(CONFIG_SPARSEMEM))
5267 pages = present_pages;
5268
5269 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5270}
5271
Linus Torvalds1da177e2005-04-16 15:20:36 -07005272/*
5273 * Set up the zone data structures:
5274 * - mark all pages reserved
5275 * - mark all memory queues empty
5276 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07005277 *
5278 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005279 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08005280static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Gu Zhengfebd5942015-06-24 16:57:02 -07005281 unsigned long node_start_pfn, unsigned long node_end_pfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005282{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07005283 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07005284 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005285 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07005286 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005287
Dave Hansen208d54e2005-10-29 18:16:52 -07005288 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01005289#ifdef CONFIG_NUMA_BALANCING
5290 spin_lock_init(&pgdat->numabalancing_migrate_lock);
5291 pgdat->numabalancing_migrate_nr_pages = 0;
5292 pgdat->numabalancing_migrate_next_window = jiffies;
5293#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005294 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07005295 init_waitqueue_head(&pgdat->pfmemalloc_wait);
Joonsoo Kimeefa8642014-12-12 16:55:46 -08005296 pgdat_page_ext_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01005297
Linus Torvalds1da177e2005-04-16 15:20:36 -07005298 for (j = 0; j < MAX_NR_ZONES; j++) {
5299 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08005300 unsigned long size, realsize, freesize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005301
Gu Zhengfebd5942015-06-24 16:57:02 -07005302 size = zone->spanned_pages;
5303 realsize = freesize = zone->present_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005304
Mel Gorman0e0b8642006-09-27 01:49:56 -07005305 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08005306 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07005307 * is used by this zone for memmap. This affects the watermark
5308 * and per-cpu initialisations
5309 */
Jiang Liu01cefae2012-12-12 13:52:19 -08005310 memmap_pages = calc_memmap_size(size, realsize);
Zhong Hongboba914f42014-12-12 16:56:21 -08005311 if (!is_highmem_idx(j)) {
5312 if (freesize >= memmap_pages) {
5313 freesize -= memmap_pages;
5314 if (memmap_pages)
5315 printk(KERN_DEBUG
5316 " %s zone: %lu pages used for memmap\n",
5317 zone_names[j], memmap_pages);
5318 } else
5319 printk(KERN_WARNING
5320 " %s zone: %lu pages exceeds freesize %lu\n",
5321 zone_names[j], memmap_pages, freesize);
5322 }
Mel Gorman0e0b8642006-09-27 01:49:56 -07005323
Christoph Lameter62672762007-02-10 01:43:07 -08005324 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08005325 if (j == 0 && freesize > dma_reserve) {
5326 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07005327 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08005328 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07005329 }
5330
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07005331 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08005332 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08005333 /* Charge for highmem memmap if there are enough kernel pages */
5334 else if (nr_kernel_pages > memmap_pages * 2)
5335 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08005336 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005337
Jiang Liu9feedc92012-12-12 13:52:12 -08005338 /*
5339 * Set an approximate value for lowmem here, it will be adjusted
5340 * when the bootmem allocator frees pages into the buddy system.
5341 * And all highmem pages will be managed by the buddy system.
5342 */
5343 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07005344#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07005345 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08005346 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07005347 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08005348 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07005349#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005350 zone->name = zone_names[j];
5351 spin_lock_init(&zone->lock);
5352 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07005353 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005354 zone->zone_pgdat = pgdat;
Dave Hansened8ece22005-10-29 18:16:50 -07005355 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005356
5357 /* For bootup, initialized properly in watermark setup */
5358 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
5359
Hugh Dickinsbea8c152012-11-16 14:14:54 -08005360 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005361 if (!size)
5362 continue;
5363
Andrew Morton955c1cd2012-05-29 15:06:31 -07005364 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08005365 setup_usemap(pgdat, zone, zone_start_pfn, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08005366 ret = init_currently_empty_zone(zone, zone_start_pfn,
5367 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07005368 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07005369 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005370 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005371 }
5372}
5373
Sam Ravnborg577a32f2007-05-17 23:29:25 +02005374static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005375{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005376 /* Skip empty nodes */
5377 if (!pgdat->node_spanned_pages)
5378 return;
5379
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005380#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07005381 /* ia64 gets its own node_mem_map, before this, without bootmem */
5382 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07005383 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005384 struct page *map;
5385
Bob Piccoe984bb42006-05-20 15:00:31 -07005386 /*
5387 * The zone's endpoints aren't required to be MAX_ORDER
5388 * aligned but the node_mem_map endpoints must be in order
5389 * for the buddy allocator to function correctly.
5390 */
5391 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
Cody P Schafer108bcc92013-02-22 16:35:23 -08005392 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07005393 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5394 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07005395 map = alloc_remap(pgdat->node_id, size);
5396 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005397 map = memblock_virt_alloc_node_nopanic(size,
5398 pgdat->node_id);
Bob Piccoe984bb42006-05-20 15:00:31 -07005399 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005400 }
Roman Zippel12d810c2007-05-31 00:40:54 -07005401#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07005402 /*
5403 * With no DISCONTIG, the global mem_map is just set as node 0's
5404 */
Mel Gormanc7132162006-09-27 01:49:43 -07005405 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005406 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08005407#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07005408 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08005409 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08005410#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005411 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005412#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005413#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005414}
5415
Johannes Weiner9109fb72008-07-23 21:27:20 -07005416void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5417 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005418{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005419 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005420 unsigned long start_pfn = 0;
5421 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07005422
Minchan Kim88fdf752012-07-31 16:46:14 -07005423 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07005424 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07005425
Mel Gorman3a80a7f2015-06-30 14:57:02 -07005426 reset_deferred_meminit(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005427 pgdat->node_id = nid;
5428 pgdat->node_start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005429#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5430 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
Juergen Gross8d29e182015-02-11 15:26:01 -08005431 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
5432 (u64)start_pfn << PAGE_SHIFT, ((u64)end_pfn << PAGE_SHIFT) - 1);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005433#endif
5434 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5435 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005436
5437 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07005438#ifdef CONFIG_FLAT_NODE_MEM_MAP
5439 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5440 nid, (unsigned long)pgdat,
5441 (unsigned long)pgdat->node_mem_map);
5442#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005443
Gu Zhengfebd5942015-06-24 16:57:02 -07005444 free_area_init_core(pgdat, start_pfn, end_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005445}
5446
Tejun Heo0ee332c2011-12-08 10:22:09 -08005447#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07005448
5449#if MAX_NUMNODES > 1
5450/*
5451 * Figure out the number of possible node ids.
5452 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07005453void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07005454{
5455 unsigned int node;
5456 unsigned int highest = 0;
5457
5458 for_each_node_mask(node, node_possible_map)
5459 highest = node;
5460 nr_node_ids = highest + 1;
5461}
Miklos Szeredi418508c2007-05-23 13:57:55 -07005462#endif
5463
Mel Gormanc7132162006-09-27 01:49:43 -07005464/**
Tejun Heo1e019792011-07-12 09:45:34 +02005465 * node_map_pfn_alignment - determine the maximum internode alignment
5466 *
5467 * This function should be called after node map is populated and sorted.
5468 * It calculates the maximum power of two alignment which can distinguish
5469 * all the nodes.
5470 *
5471 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5472 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5473 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5474 * shifted, 1GiB is enough and this function will indicate so.
5475 *
5476 * This is used to test whether pfn -> nid mapping of the chosen memory
5477 * model has fine enough granularity to avoid incorrect mapping for the
5478 * populated node map.
5479 *
5480 * Returns the determined alignment in pfn's. 0 if there is no alignment
5481 * requirement (single node).
5482 */
5483unsigned long __init node_map_pfn_alignment(void)
5484{
5485 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005486 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02005487 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02005488 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02005489
Tejun Heoc13291a2011-07-12 10:46:30 +02005490 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02005491 if (!start || last_nid < 0 || last_nid == nid) {
5492 last_nid = nid;
5493 last_end = end;
5494 continue;
5495 }
5496
5497 /*
5498 * Start with a mask granular enough to pin-point to the
5499 * start pfn and tick off bits one-by-one until it becomes
5500 * too coarse to separate the current node from the last.
5501 */
5502 mask = ~((1 << __ffs(start)) - 1);
5503 while (mask && last_end <= (start & (mask << 1)))
5504 mask <<= 1;
5505
5506 /* accumulate all internode masks */
5507 accl_mask |= mask;
5508 }
5509
5510 /* convert mask to number of pages */
5511 return ~accl_mask + 1;
5512}
5513
Mel Gormana6af2bc2007-02-10 01:42:57 -08005514/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005515static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07005516{
Mel Gormana6af2bc2007-02-10 01:42:57 -08005517 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02005518 unsigned long start_pfn;
5519 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00005520
Tejun Heoc13291a2011-07-12 10:46:30 +02005521 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5522 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005523
Mel Gormana6af2bc2007-02-10 01:42:57 -08005524 if (min_pfn == ULONG_MAX) {
5525 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07005526 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08005527 return 0;
5528 }
5529
5530 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005531}
5532
5533/**
5534 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5535 *
5536 * It returns the minimum PFN based on information provided via
Zhang Zhen7d018172014-06-04 16:10:53 -07005537 * memblock_set_node().
Mel Gormanc7132162006-09-27 01:49:43 -07005538 */
5539unsigned long __init find_min_pfn_with_active_regions(void)
5540{
5541 return find_min_pfn_for_node(MAX_NUMNODES);
5542}
5543
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005544/*
5545 * early_calculate_totalpages()
5546 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005547 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005548 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07005549static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07005550{
Mel Gorman7e63efe2007-07-17 04:03:15 -07005551 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005552 unsigned long start_pfn, end_pfn;
5553 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005554
Tejun Heoc13291a2011-07-12 10:46:30 +02005555 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5556 unsigned long pages = end_pfn - start_pfn;
5557
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005558 totalpages += pages;
5559 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005560 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005561 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07005562 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005563}
5564
Mel Gorman2a1e2742007-07-17 04:03:12 -07005565/*
5566 * Find the PFN the Movable zone begins in each node. Kernel memory
5567 * is spread evenly between nodes as long as the nodes have enough
5568 * memory. When they don't, some nodes will have more kernelcore than
5569 * others
5570 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07005571static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005572{
5573 int i, nid;
5574 unsigned long usable_startpfn;
5575 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07005576 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005577 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005578 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005579 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07005580 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005581
5582 /* Need to find movable_zone earlier when movable_node is specified. */
5583 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07005584
Mel Gorman7e63efe2007-07-17 04:03:15 -07005585 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005586 * If movable_node is specified, ignore kernelcore and movablecore
5587 * options.
5588 */
5589 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07005590 for_each_memblock(memory, r) {
5591 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005592 continue;
5593
Emil Medve136199f2014-04-07 15:37:52 -07005594 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005595
Emil Medve136199f2014-04-07 15:37:52 -07005596 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005597 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5598 min(usable_startpfn, zone_movable_pfn[nid]) :
5599 usable_startpfn;
5600 }
5601
5602 goto out2;
5603 }
5604
5605 /*
5606 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07005607 * kernelcore that corresponds so that memory usable for
5608 * any allocation type is evenly spread. If both kernelcore
5609 * and movablecore are specified, then the value of kernelcore
5610 * will be used for required_kernelcore if it's greater than
5611 * what movablecore would have allowed.
5612 */
5613 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07005614 unsigned long corepages;
5615
5616 /*
5617 * Round-up so that ZONE_MOVABLE is at least as large as what
5618 * was requested by the user
5619 */
5620 required_movablecore =
5621 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5622 corepages = totalpages - required_movablecore;
5623
5624 required_kernelcore = max(required_kernelcore, corepages);
5625 }
5626
Yinghai Lu20e69262013-03-01 14:51:27 -08005627 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5628 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07005629 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005630
5631 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005632 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5633
5634restart:
5635 /* Spread kernelcore memory as evenly as possible throughout nodes */
5636 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005637 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02005638 unsigned long start_pfn, end_pfn;
5639
Mel Gorman2a1e2742007-07-17 04:03:12 -07005640 /*
5641 * Recalculate kernelcore_node if the division per node
5642 * now exceeds what is necessary to satisfy the requested
5643 * amount of memory for the kernel
5644 */
5645 if (required_kernelcore < kernelcore_node)
5646 kernelcore_node = required_kernelcore / usable_nodes;
5647
5648 /*
5649 * As the map is walked, we track how much memory is usable
5650 * by the kernel using kernelcore_remaining. When it is
5651 * 0, the rest of the node is usable by ZONE_MOVABLE
5652 */
5653 kernelcore_remaining = kernelcore_node;
5654
5655 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02005656 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005657 unsigned long size_pages;
5658
Tejun Heoc13291a2011-07-12 10:46:30 +02005659 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005660 if (start_pfn >= end_pfn)
5661 continue;
5662
5663 /* Account for what is only usable for kernelcore */
5664 if (start_pfn < usable_startpfn) {
5665 unsigned long kernel_pages;
5666 kernel_pages = min(end_pfn, usable_startpfn)
5667 - start_pfn;
5668
5669 kernelcore_remaining -= min(kernel_pages,
5670 kernelcore_remaining);
5671 required_kernelcore -= min(kernel_pages,
5672 required_kernelcore);
5673
5674 /* Continue if range is now fully accounted */
5675 if (end_pfn <= usable_startpfn) {
5676
5677 /*
5678 * Push zone_movable_pfn to the end so
5679 * that if we have to rebalance
5680 * kernelcore across nodes, we will
5681 * not double account here
5682 */
5683 zone_movable_pfn[nid] = end_pfn;
5684 continue;
5685 }
5686 start_pfn = usable_startpfn;
5687 }
5688
5689 /*
5690 * The usable PFN range for ZONE_MOVABLE is from
5691 * start_pfn->end_pfn. Calculate size_pages as the
5692 * number of pages used as kernelcore
5693 */
5694 size_pages = end_pfn - start_pfn;
5695 if (size_pages > kernelcore_remaining)
5696 size_pages = kernelcore_remaining;
5697 zone_movable_pfn[nid] = start_pfn + size_pages;
5698
5699 /*
5700 * Some kernelcore has been met, update counts and
5701 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07005702 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005703 */
5704 required_kernelcore -= min(required_kernelcore,
5705 size_pages);
5706 kernelcore_remaining -= size_pages;
5707 if (!kernelcore_remaining)
5708 break;
5709 }
5710 }
5711
5712 /*
5713 * If there is still required_kernelcore, we do another pass with one
5714 * less node in the count. This will push zone_movable_pfn[nid] further
5715 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07005716 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005717 */
5718 usable_nodes--;
5719 if (usable_nodes && required_kernelcore > usable_nodes)
5720 goto restart;
5721
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005722out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07005723 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5724 for (nid = 0; nid < MAX_NUMNODES; nid++)
5725 zone_movable_pfn[nid] =
5726 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005727
Yinghai Lu20e69262013-03-01 14:51:27 -08005728out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005729 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005730 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005731}
5732
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005733/* Any regular or high memory on that node ? */
5734static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005735{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005736 enum zone_type zone_type;
5737
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005738 if (N_MEMORY == N_NORMAL_MEMORY)
5739 return;
5740
5741 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005742 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08005743 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005744 node_set_state(nid, N_HIGH_MEMORY);
5745 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5746 zone_type <= ZONE_NORMAL)
5747 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005748 break;
5749 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005750 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005751}
5752
Mel Gormanc7132162006-09-27 01:49:43 -07005753/**
5754 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005755 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005756 *
5757 * This will call free_area_init_node() for each active node in the system.
Zhang Zhen7d018172014-06-04 16:10:53 -07005758 * Using the page ranges provided by memblock_set_node(), the size of each
Mel Gormanc7132162006-09-27 01:49:43 -07005759 * zone in each node and their holes is calculated. If the maximum PFN
5760 * between two adjacent zones match, it is assumed that the zone is empty.
5761 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5762 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5763 * starts where the previous one ended. For example, ZONE_DMA32 starts
5764 * at arch_max_dma_pfn.
5765 */
5766void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5767{
Tejun Heoc13291a2011-07-12 10:46:30 +02005768 unsigned long start_pfn, end_pfn;
5769 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005770
Mel Gormanc7132162006-09-27 01:49:43 -07005771 /* Record where the zone boundaries are */
5772 memset(arch_zone_lowest_possible_pfn, 0,
5773 sizeof(arch_zone_lowest_possible_pfn));
5774 memset(arch_zone_highest_possible_pfn, 0,
5775 sizeof(arch_zone_highest_possible_pfn));
5776 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5777 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5778 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005779 if (i == ZONE_MOVABLE)
5780 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005781 arch_zone_lowest_possible_pfn[i] =
5782 arch_zone_highest_possible_pfn[i-1];
5783 arch_zone_highest_possible_pfn[i] =
5784 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5785 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005786 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5787 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5788
5789 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5790 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005791 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005792
Mel Gormanc7132162006-09-27 01:49:43 -07005793 /* Print out the zone ranges */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005794 pr_info("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005795 for (i = 0; i < MAX_NR_ZONES; i++) {
5796 if (i == ZONE_MOVABLE)
5797 continue;
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005798 pr_info(" %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08005799 if (arch_zone_lowest_possible_pfn[i] ==
5800 arch_zone_highest_possible_pfn[i])
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005801 pr_cont("empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08005802 else
Juergen Gross8d29e182015-02-11 15:26:01 -08005803 pr_cont("[mem %#018Lx-%#018Lx]\n",
5804 (u64)arch_zone_lowest_possible_pfn[i]
5805 << PAGE_SHIFT,
5806 ((u64)arch_zone_highest_possible_pfn[i]
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005807 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005808 }
5809
5810 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005811 pr_info("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005812 for (i = 0; i < MAX_NUMNODES; i++) {
5813 if (zone_movable_pfn[i])
Juergen Gross8d29e182015-02-11 15:26:01 -08005814 pr_info(" Node %d: %#018Lx\n", i,
5815 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005816 }
Mel Gormanc7132162006-09-27 01:49:43 -07005817
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07005818 /* Print out the early node map */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005819 pr_info("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02005820 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Juergen Gross8d29e182015-02-11 15:26:01 -08005821 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
5822 (u64)start_pfn << PAGE_SHIFT,
5823 ((u64)end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07005824
5825 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005826 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005827 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005828 for_each_online_node(nid) {
5829 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005830 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005831 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005832
5833 /* Any memory on that node */
5834 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005835 node_set_state(nid, N_MEMORY);
5836 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005837 }
5838}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005839
Mel Gorman7e63efe2007-07-17 04:03:15 -07005840static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005841{
5842 unsigned long long coremem;
5843 if (!p)
5844 return -EINVAL;
5845
5846 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005847 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005848
Mel Gorman7e63efe2007-07-17 04:03:15 -07005849 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005850 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5851
5852 return 0;
5853}
Mel Gormaned7ed362007-07-17 04:03:14 -07005854
Mel Gorman7e63efe2007-07-17 04:03:15 -07005855/*
5856 * kernelcore=size sets the amount of memory for use for allocations that
5857 * cannot be reclaimed or migrated.
5858 */
5859static int __init cmdline_parse_kernelcore(char *p)
5860{
5861 return cmdline_parse_core(p, &required_kernelcore);
5862}
5863
5864/*
5865 * movablecore=size sets the amount of memory for use for allocations that
5866 * can be reclaimed or migrated.
5867 */
5868static int __init cmdline_parse_movablecore(char *p)
5869{
5870 return cmdline_parse_core(p, &required_movablecore);
5871}
5872
Mel Gormaned7ed362007-07-17 04:03:14 -07005873early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005874early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005875
Tejun Heo0ee332c2011-12-08 10:22:09 -08005876#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005877
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005878void adjust_managed_page_count(struct page *page, long count)
5879{
5880 spin_lock(&managed_page_count_lock);
5881 page_zone(page)->managed_pages += count;
5882 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07005883#ifdef CONFIG_HIGHMEM
5884 if (PageHighMem(page))
5885 totalhigh_pages += count;
5886#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005887 spin_unlock(&managed_page_count_lock);
5888}
Jiang Liu3dcc0572013-07-03 15:03:21 -07005889EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005890
Jiang Liu11199692013-07-03 15:02:48 -07005891unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07005892{
Jiang Liu11199692013-07-03 15:02:48 -07005893 void *pos;
5894 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07005895
Jiang Liu11199692013-07-03 15:02:48 -07005896 start = (void *)PAGE_ALIGN((unsigned long)start);
5897 end = (void *)((unsigned long)end & PAGE_MASK);
5898 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07005899 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07005900 memset(pos, poison, PAGE_SIZE);
5901 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07005902 }
5903
5904 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07005905 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07005906 s, pages << (PAGE_SHIFT - 10), start, end);
5907
5908 return pages;
5909}
Jiang Liu11199692013-07-03 15:02:48 -07005910EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07005911
Jiang Liucfa11e02013-04-29 15:07:00 -07005912#ifdef CONFIG_HIGHMEM
5913void free_highmem_page(struct page *page)
5914{
5915 __free_reserved_page(page);
5916 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07005917 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07005918 totalhigh_pages++;
5919}
5920#endif
5921
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005922
5923void __init mem_init_print_info(const char *str)
5924{
5925 unsigned long physpages, codesize, datasize, rosize, bss_size;
5926 unsigned long init_code_size, init_data_size;
5927
5928 physpages = get_num_physpages();
5929 codesize = _etext - _stext;
5930 datasize = _edata - _sdata;
5931 rosize = __end_rodata - __start_rodata;
5932 bss_size = __bss_stop - __bss_start;
5933 init_data_size = __init_end - __init_begin;
5934 init_code_size = _einittext - _sinittext;
5935
5936 /*
5937 * Detect special cases and adjust section sizes accordingly:
5938 * 1) .init.* may be embedded into .data sections
5939 * 2) .init.text.* may be out of [__init_begin, __init_end],
5940 * please refer to arch/tile/kernel/vmlinux.lds.S.
5941 * 3) .rodata.* may be embedded into .text or .data sections.
5942 */
5943#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07005944 do { \
5945 if (start <= pos && pos < end && size > adj) \
5946 size -= adj; \
5947 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005948
5949 adj_init_size(__init_begin, __init_end, init_data_size,
5950 _sinittext, init_code_size);
5951 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5952 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5953 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5954 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5955
5956#undef adj_init_size
5957
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005958 pr_info("Memory: %luK/%luK available "
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005959 "(%luK kernel code, %luK rwdata, %luK rodata, "
Pintu Kumare48322a2014-12-18 16:17:15 -08005960 "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005961#ifdef CONFIG_HIGHMEM
5962 ", %luK highmem"
5963#endif
5964 "%s%s)\n",
5965 nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5966 codesize >> 10, datasize >> 10, rosize >> 10,
5967 (init_data_size + init_code_size) >> 10, bss_size >> 10,
Pintu Kumare48322a2014-12-18 16:17:15 -08005968 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10),
5969 totalcma_pages << (PAGE_SHIFT-10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005970#ifdef CONFIG_HIGHMEM
5971 totalhigh_pages << (PAGE_SHIFT-10),
5972#endif
5973 str ? ", " : "", str ? str : "");
5974}
5975
Mel Gorman0e0b8642006-09-27 01:49:56 -07005976/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005977 * set_dma_reserve - set the specified number of pages reserved in the first zone
5978 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07005979 *
5980 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5981 * In the DMA zone, a significant percentage may be consumed by kernel image
5982 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005983 * function may optionally be used to account for unfreeable pages in the
5984 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5985 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005986 */
5987void __init set_dma_reserve(unsigned long new_dma_reserve)
5988{
5989 dma_reserve = new_dma_reserve;
5990}
5991
Linus Torvalds1da177e2005-04-16 15:20:36 -07005992void __init free_area_init(unsigned long *zones_size)
5993{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005994 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005995 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5996}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005997
Linus Torvalds1da177e2005-04-16 15:20:36 -07005998static int page_alloc_cpu_notify(struct notifier_block *self,
5999 unsigned long action, void *hcpu)
6000{
6001 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006002
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07006003 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07006004 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006005 drain_pages(cpu);
6006
6007 /*
6008 * Spill the event counters of the dead processor
6009 * into the current processors event counters.
6010 * This artificially elevates the count of the current
6011 * processor.
6012 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07006013 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006014
6015 /*
6016 * Zero the differential counters of the dead processor
6017 * so that the vm statistics are consistent.
6018 *
6019 * This is only okay since the processor is dead and cannot
6020 * race with what we are doing.
6021 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07006022 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006023 }
6024 return NOTIFY_OK;
6025}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006026
6027void __init page_alloc_init(void)
6028{
6029 hotcpu_notifier(page_alloc_cpu_notify, 0);
6030}
6031
6032/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006033 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
6034 * or min_free_kbytes changes.
6035 */
6036static void calculate_totalreserve_pages(void)
6037{
6038 struct pglist_data *pgdat;
6039 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006040 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006041
6042 for_each_online_pgdat(pgdat) {
6043 for (i = 0; i < MAX_NR_ZONES; i++) {
6044 struct zone *zone = pgdat->node_zones + i;
Mel Gorman3484b2d2014-08-06 16:07:14 -07006045 long max = 0;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006046
6047 /* Find valid and maximum lowmem_reserve in the zone */
6048 for (j = i; j < MAX_NR_ZONES; j++) {
6049 if (zone->lowmem_reserve[j] > max)
6050 max = zone->lowmem_reserve[j];
6051 }
6052
Mel Gorman41858962009-06-16 15:32:12 -07006053 /* we treat the high watermark as reserved pages. */
6054 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006055
Jiang Liub40da042013-02-22 16:33:52 -08006056 if (max > zone->managed_pages)
6057 max = zone->managed_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006058 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08006059 /*
6060 * Lowmem reserves are not available to
6061 * GFP_HIGHUSER page cache allocations and
6062 * kswapd tries to balance zones to their high
6063 * watermark. As a result, neither should be
6064 * regarded as dirtyable memory, to prevent a
6065 * situation where reclaim has to clean pages
6066 * in order to balance the zones.
6067 */
6068 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006069 }
6070 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08006071 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006072 totalreserve_pages = reserve_pages;
6073}
6074
6075/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07006076 * setup_per_zone_lowmem_reserve - called whenever
6077 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
6078 * has a correct pages reserved value, so an adequate number of
6079 * pages are left in the zone after a successful __alloc_pages().
6080 */
6081static void setup_per_zone_lowmem_reserve(void)
6082{
6083 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006084 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006085
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08006086 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006087 for (j = 0; j < MAX_NR_ZONES; j++) {
6088 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08006089 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006090
6091 zone->lowmem_reserve[j] = 0;
6092
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006093 idx = j;
6094 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006095 struct zone *lower_zone;
6096
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006097 idx--;
6098
Linus Torvalds1da177e2005-04-16 15:20:36 -07006099 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6100 sysctl_lowmem_reserve_ratio[idx] = 1;
6101
6102 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08006103 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07006104 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08006105 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006106 }
6107 }
6108 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006109
6110 /* update totalreserve_pages */
6111 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006112}
6113
Mel Gormancfd3da12011-04-25 21:36:42 +00006114static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006115{
6116 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6117 unsigned long lowmem_pages = 0;
6118 struct zone *zone;
6119 unsigned long flags;
6120
6121 /* Calculate total number of !ZONE_HIGHMEM pages */
6122 for_each_zone(zone) {
6123 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08006124 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006125 }
6126
6127 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07006128 u64 tmp;
6129
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006130 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08006131 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07006132 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006133 if (is_highmem(zone)) {
6134 /*
Nick Piggin669ed172005-11-13 16:06:45 -08006135 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6136 * need highmem pages, so cap pages_min to a small
6137 * value here.
6138 *
Mel Gorman41858962009-06-16 15:32:12 -07006139 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Yaowei Bai42ff2702015-04-14 15:47:14 -07006140 * deltas control asynch page reclaim, and so should
Nick Piggin669ed172005-11-13 16:06:45 -08006141 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006142 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08006143 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006144
Jiang Liub40da042013-02-22 16:33:52 -08006145 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08006146 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07006147 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006148 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08006149 /*
6150 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07006151 * proportionate to the zone's size.
6152 */
Mel Gorman41858962009-06-16 15:32:12 -07006153 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006154 }
6155
Mel Gorman41858962009-06-16 15:32:12 -07006156 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
6157 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006158
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006159 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
Johannes Weinerabe5f972014-10-02 16:21:10 -07006160 high_wmark_pages(zone) - low_wmark_pages(zone) -
6161 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006162
Mel Gorman56fd56b2007-10-16 01:25:58 -07006163 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006164 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006165 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006166
6167 /* update totalreserve_pages */
6168 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006169}
6170
Mel Gormancfd3da12011-04-25 21:36:42 +00006171/**
6172 * setup_per_zone_wmarks - called when min_free_kbytes changes
6173 * or when memory is hot-{added|removed}
6174 *
6175 * Ensures that the watermark[min,low,high] values for each zone are set
6176 * correctly with respect to min_free_kbytes.
6177 */
6178void setup_per_zone_wmarks(void)
6179{
6180 mutex_lock(&zonelists_mutex);
6181 __setup_per_zone_wmarks();
6182 mutex_unlock(&zonelists_mutex);
6183}
6184
Randy Dunlap55a44622009-09-21 17:01:20 -07006185/*
Rik van Riel556adec2008-10-18 20:26:34 -07006186 * The inactive anon list should be small enough that the VM never has to
6187 * do too much work, but large enough that each inactive page has a chance
6188 * to be referenced again before it is swapped out.
6189 *
6190 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6191 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6192 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6193 * the anonymous pages are kept on the inactive list.
6194 *
6195 * total target max
6196 * memory ratio inactive anon
6197 * -------------------------------------
6198 * 10MB 1 5MB
6199 * 100MB 1 50MB
6200 * 1GB 3 250MB
6201 * 10GB 10 0.9GB
6202 * 100GB 31 3GB
6203 * 1TB 101 10GB
6204 * 10TB 320 32GB
6205 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07006206static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07006207{
6208 unsigned int gb, ratio;
6209
6210 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08006211 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07006212 if (gb)
6213 ratio = int_sqrt(10 * gb);
6214 else
6215 ratio = 1;
6216
6217 zone->inactive_ratio = ratio;
6218}
6219
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07006220static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07006221{
6222 struct zone *zone;
6223
Minchan Kim96cb4df2009-06-16 15:32:49 -07006224 for_each_zone(zone)
6225 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07006226}
6227
Linus Torvalds1da177e2005-04-16 15:20:36 -07006228/*
6229 * Initialise min_free_kbytes.
6230 *
6231 * For small machines we want it small (128k min). For large machines
6232 * we want it large (64MB max). But it is not linear, because network
6233 * bandwidth does not increase linearly with machine size. We use
6234 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006235 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07006236 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6237 *
6238 * which yields
6239 *
6240 * 16MB: 512k
6241 * 32MB: 724k
6242 * 64MB: 1024k
6243 * 128MB: 1448k
6244 * 256MB: 2048k
6245 * 512MB: 2896k
6246 * 1024MB: 4096k
6247 * 2048MB: 5792k
6248 * 4096MB: 8192k
6249 * 8192MB: 11584k
6250 * 16384MB: 16384k
6251 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07006252int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006253{
6254 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07006255 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006256
6257 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07006258 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006259
Michal Hocko5f127332013-07-08 16:00:40 -07006260 if (new_min_free_kbytes > user_min_free_kbytes) {
6261 min_free_kbytes = new_min_free_kbytes;
6262 if (min_free_kbytes < 128)
6263 min_free_kbytes = 128;
6264 if (min_free_kbytes > 65536)
6265 min_free_kbytes = 65536;
6266 } else {
6267 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6268 new_min_free_kbytes, user_min_free_kbytes);
6269 }
Minchan Kimbc75d332009-06-16 15:32:48 -07006270 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07006271 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006272 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07006273 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006274 return 0;
6275}
Minchan Kimbc75d332009-06-16 15:32:48 -07006276module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006277
6278/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07006279 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07006280 * that we can call two helper functions whenever min_free_kbytes
6281 * changes.
6282 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006283int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006284 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006285{
Han Pingtianda8c7572014-01-23 15:53:17 -08006286 int rc;
6287
6288 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6289 if (rc)
6290 return rc;
6291
Michal Hocko5f127332013-07-08 16:00:40 -07006292 if (write) {
6293 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07006294 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07006295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006296 return 0;
6297}
6298
Christoph Lameter96146342006-07-03 00:24:13 -07006299#ifdef CONFIG_NUMA
Joe Perchescccad5b2014-06-06 14:38:09 -07006300int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006301 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07006302{
6303 struct zone *zone;
6304 int rc;
6305
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006306 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07006307 if (rc)
6308 return rc;
6309
6310 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08006311 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07006312 sysctl_min_unmapped_ratio) / 100;
6313 return 0;
6314}
Christoph Lameter0ff38492006-09-25 23:31:52 -07006315
Joe Perchescccad5b2014-06-06 14:38:09 -07006316int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006317 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07006318{
6319 struct zone *zone;
6320 int rc;
6321
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006322 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07006323 if (rc)
6324 return rc;
6325
6326 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08006327 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07006328 sysctl_min_slab_ratio) / 100;
6329 return 0;
6330}
Christoph Lameter96146342006-07-03 00:24:13 -07006331#endif
6332
Linus Torvalds1da177e2005-04-16 15:20:36 -07006333/*
6334 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6335 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6336 * whenever sysctl_lowmem_reserve_ratio changes.
6337 *
6338 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07006339 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07006340 * if in function of the boot time zone sizes.
6341 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006342int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006343 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006344{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006345 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006346 setup_per_zone_lowmem_reserve();
6347 return 0;
6348}
6349
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006350/*
6351 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07006352 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6353 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006354 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006355int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006356 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006357{
6358 struct zone *zone;
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006359 int old_percpu_pagelist_fraction;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006360 int ret;
6361
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006362 mutex_lock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006363 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6364
6365 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
6366 if (!write || ret < 0)
6367 goto out;
6368
6369 /* Sanity checking to avoid pcp imbalance */
6370 if (percpu_pagelist_fraction &&
6371 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6372 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6373 ret = -EINVAL;
6374 goto out;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006375 }
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006376
6377 /* No change? */
6378 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6379 goto out;
6380
6381 for_each_populated_zone(zone) {
6382 unsigned int cpu;
6383
6384 for_each_possible_cpu(cpu)
6385 pageset_set_high_and_batch(zone,
6386 per_cpu_ptr(zone->pageset, cpu));
6387 }
6388out:
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006389 mutex_unlock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006390 return ret;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006391}
6392
Rasmus Villemoesa9919c72015-06-24 16:56:28 -07006393#ifdef CONFIG_NUMA
David S. Millerf034b5d2006-08-24 03:08:07 -07006394int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006395
Linus Torvalds1da177e2005-04-16 15:20:36 -07006396static int __init set_hashdist(char *str)
6397{
6398 if (!str)
6399 return 0;
6400 hashdist = simple_strtoul(str, &str, 0);
6401 return 1;
6402}
6403__setup("hashdist=", set_hashdist);
6404#endif
6405
6406/*
6407 * allocate a large system hash table from bootmem
6408 * - it is assumed that the hash table must contain an exact power-of-2
6409 * quantity of entries
6410 * - limit is the number of hash buckets, not the total allocation size
6411 */
6412void *__init alloc_large_system_hash(const char *tablename,
6413 unsigned long bucketsize,
6414 unsigned long numentries,
6415 int scale,
6416 int flags,
6417 unsigned int *_hash_shift,
6418 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00006419 unsigned long low_limit,
6420 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006421{
Tim Bird31fe62b2012-05-23 13:33:35 +00006422 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006423 unsigned long log2qty, size;
6424 void *table = NULL;
6425
6426 /* allow the kernel cmdline to have a say */
6427 if (!numentries) {
6428 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08006429 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07006430
6431 /* It isn't necessary when PAGE_SIZE >= 1MB */
6432 if (PAGE_SHIFT < 20)
6433 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006434
6435 /* limit to 1 bucket per 2^scale bytes of low memory */
6436 if (scale > PAGE_SHIFT)
6437 numentries >>= (scale - PAGE_SHIFT);
6438 else
6439 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08006440
6441 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07006442 if (unlikely(flags & HASH_SMALL)) {
6443 /* Makes no sense without HASH_EARLY */
6444 WARN_ON(!(flags & HASH_EARLY));
6445 if (!(numentries >> *_hash_shift)) {
6446 numentries = 1UL << *_hash_shift;
6447 BUG_ON(!numentries);
6448 }
6449 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08006450 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006451 }
John Hawkes6e692ed2006-03-25 03:08:02 -08006452 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006453
6454 /* limit allocation size to 1/16 total memory by default */
6455 if (max == 0) {
6456 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6457 do_div(max, bucketsize);
6458 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08006459 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006460
Tim Bird31fe62b2012-05-23 13:33:35 +00006461 if (numentries < low_limit)
6462 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006463 if (numentries > max)
6464 numentries = max;
6465
David Howellsf0d1b0b2006-12-08 02:37:49 -08006466 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006467
6468 do {
6469 size = bucketsize << log2qty;
6470 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08006471 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006472 else if (hashdist)
6473 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6474 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07006475 /*
6476 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07006477 * some pages at the end of hash table which
6478 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07006479 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006480 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07006481 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006482 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6483 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006484 }
6485 } while (!table && size > PAGE_SIZE && --log2qty);
6486
6487 if (!table)
6488 panic("Failed to allocate %s hash table\n", tablename);
6489
Robin Holtf241e6602010-10-07 12:59:26 -07006490 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07006491 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07006492 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08006493 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006494 size);
6495
6496 if (_hash_shift)
6497 *_hash_shift = log2qty;
6498 if (_hash_mask)
6499 *_hash_mask = (1 << log2qty) - 1;
6500
6501 return table;
6502}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08006503
Mel Gorman835c1342007-10-16 01:25:47 -07006504/* Return a pointer to the bitmap storing bits affecting a block of pages */
6505static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6506 unsigned long pfn)
6507{
6508#ifdef CONFIG_SPARSEMEM
6509 return __pfn_to_section(pfn)->pageblock_flags;
6510#else
6511 return zone->pageblock_flags;
6512#endif /* CONFIG_SPARSEMEM */
6513}
Andrew Morton6220ec72006-10-19 23:29:05 -07006514
Mel Gorman835c1342007-10-16 01:25:47 -07006515static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6516{
6517#ifdef CONFIG_SPARSEMEM
6518 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07006519 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006520#else
Laura Abbottc060f942013-01-11 14:31:51 -08006521 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07006522 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006523#endif /* CONFIG_SPARSEMEM */
6524}
6525
6526/**
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006527 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
Mel Gorman835c1342007-10-16 01:25:47 -07006528 * @page: The page within the block of interest
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006529 * @pfn: The target page frame number
6530 * @end_bitidx: The last bit of interest to retrieve
6531 * @mask: mask of bits that the caller is interested in
6532 *
6533 * Return: pageblock_bits flags
Mel Gorman835c1342007-10-16 01:25:47 -07006534 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006535unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006536 unsigned long end_bitidx,
6537 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006538{
6539 struct zone *zone;
6540 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006541 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006542 unsigned long word;
Mel Gorman835c1342007-10-16 01:25:47 -07006543
6544 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006545 bitmap = get_pageblock_bitmap(zone, pfn);
6546 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006547 word_bitidx = bitidx / BITS_PER_LONG;
6548 bitidx &= (BITS_PER_LONG-1);
Mel Gorman835c1342007-10-16 01:25:47 -07006549
Mel Gormane58469b2014-06-04 16:10:16 -07006550 word = bitmap[word_bitidx];
6551 bitidx += end_bitidx;
6552 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
Mel Gorman835c1342007-10-16 01:25:47 -07006553}
6554
6555/**
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006556 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
Mel Gorman835c1342007-10-16 01:25:47 -07006557 * @page: The page within the block of interest
Mel Gorman835c1342007-10-16 01:25:47 -07006558 * @flags: The flags to set
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006559 * @pfn: The target page frame number
6560 * @end_bitidx: The last bit of interest
6561 * @mask: mask of bits that the caller is interested in
Mel Gorman835c1342007-10-16 01:25:47 -07006562 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006563void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6564 unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006565 unsigned long end_bitidx,
6566 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006567{
6568 struct zone *zone;
6569 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006570 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006571 unsigned long old_word, word;
6572
6573 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
Mel Gorman835c1342007-10-16 01:25:47 -07006574
6575 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006576 bitmap = get_pageblock_bitmap(zone, pfn);
6577 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006578 word_bitidx = bitidx / BITS_PER_LONG;
6579 bitidx &= (BITS_PER_LONG-1);
6580
Sasha Levin309381fea2014-01-23 15:52:54 -08006581 VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
Mel Gorman835c1342007-10-16 01:25:47 -07006582
Mel Gormane58469b2014-06-04 16:10:16 -07006583 bitidx += end_bitidx;
6584 mask <<= (BITS_PER_LONG - bitidx - 1);
6585 flags <<= (BITS_PER_LONG - bitidx - 1);
6586
Jason Low4db0c3c2015-04-15 16:14:08 -07006587 word = READ_ONCE(bitmap[word_bitidx]);
Mel Gormane58469b2014-06-04 16:10:16 -07006588 for (;;) {
6589 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6590 if (word == old_word)
6591 break;
6592 word = old_word;
6593 }
Mel Gorman835c1342007-10-16 01:25:47 -07006594}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006595
6596/*
Minchan Kim80934512012-07-31 16:43:01 -07006597 * This function checks whether pageblock includes unmovable pages or not.
6598 * If @count is not zero, it is okay to include less @count unmovable pages
6599 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006600 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07006601 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6602 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006603 */
Wen Congyangb023f462012-12-11 16:00:45 -08006604bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6605 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006606{
6607 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006608 int mt;
6609
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006610 /*
6611 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07006612 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006613 */
6614 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07006615 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006616 mt = get_pageblock_migratetype(page);
6617 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07006618 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006619
6620 pfn = page_to_pfn(page);
6621 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6622 unsigned long check = pfn + iter;
6623
Namhyung Kim29723fc2011-02-25 14:44:25 -08006624 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006625 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08006626
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006627 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07006628
6629 /*
6630 * Hugepages are not in LRU lists, but they're movable.
6631 * We need not scan over tail pages bacause we don't
6632 * handle each tail page individually in migration.
6633 */
6634 if (PageHuge(page)) {
6635 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6636 continue;
6637 }
6638
Minchan Kim97d255c2012-07-31 16:42:59 -07006639 /*
6640 * We can't use page_count without pin a page
6641 * because another CPU can free compound page.
6642 * This check already skips compound tails of THP
6643 * because their page->_count is zero at all time.
6644 */
6645 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006646 if (PageBuddy(page))
6647 iter += (1 << page_order(page)) - 1;
6648 continue;
6649 }
Minchan Kim97d255c2012-07-31 16:42:59 -07006650
Wen Congyangb023f462012-12-11 16:00:45 -08006651 /*
6652 * The HWPoisoned page may be not in buddy system, and
6653 * page_count() is not 0.
6654 */
6655 if (skip_hwpoisoned_pages && PageHWPoison(page))
6656 continue;
6657
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006658 if (!PageLRU(page))
6659 found++;
6660 /*
Johannes Weiner6b4f7792014-12-12 16:56:13 -08006661 * If there are RECLAIMABLE pages, we need to check
6662 * it. But now, memory offline itself doesn't call
6663 * shrink_node_slabs() and it still to be fixed.
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006664 */
6665 /*
6666 * If the page is not RAM, page_count()should be 0.
6667 * we don't need more check. This is an _used_ not-movable page.
6668 *
6669 * The problematic thing here is PG_reserved pages. PG_reserved
6670 * is set to both of a memory hole page and a _used_ kernel
6671 * page at boot.
6672 */
6673 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07006674 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006675 }
Minchan Kim80934512012-07-31 16:43:01 -07006676 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006677}
6678
6679bool is_pageblock_removable_nolock(struct page *page)
6680{
Michal Hocko656a0702012-01-20 14:33:58 -08006681 struct zone *zone;
6682 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08006683
6684 /*
6685 * We have to be careful here because we are iterating over memory
6686 * sections which are not zone aware so we might end up outside of
6687 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08006688 * We have to take care about the node as well. If the node is offline
6689 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08006690 */
Michal Hocko656a0702012-01-20 14:33:58 -08006691 if (!node_online(page_to_nid(page)))
6692 return false;
6693
6694 zone = page_zone(page);
6695 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08006696 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08006697 return false;
6698
Wen Congyangb023f462012-12-11 16:00:45 -08006699 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006700}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006701
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006702#ifdef CONFIG_CMA
6703
6704static unsigned long pfn_max_align_down(unsigned long pfn)
6705{
6706 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6707 pageblock_nr_pages) - 1);
6708}
6709
6710static unsigned long pfn_max_align_up(unsigned long pfn)
6711{
6712 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6713 pageblock_nr_pages));
6714}
6715
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006716/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006717static int __alloc_contig_migrate_range(struct compact_control *cc,
6718 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006719{
6720 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006721 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006722 unsigned long pfn = start;
6723 unsigned int tries = 0;
6724 int ret = 0;
6725
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08006726 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006727
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006728 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006729 if (fatal_signal_pending(current)) {
6730 ret = -EINTR;
6731 break;
6732 }
6733
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006734 if (list_empty(&cc->migratepages)) {
6735 cc->nr_migratepages = 0;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07006736 pfn = isolate_migratepages_range(cc, pfn, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006737 if (!pfn) {
6738 ret = -EINTR;
6739 break;
6740 }
6741 tries = 0;
6742 } else if (++tries == 5) {
6743 ret = ret < 0 ? ret : -EBUSY;
6744 break;
6745 }
6746
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006747 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6748 &cc->migratepages);
6749 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07006750
Hugh Dickins9c620e22013-02-22 16:35:14 -08006751 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07006752 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006753 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08006754 if (ret < 0) {
6755 putback_movable_pages(&cc->migratepages);
6756 return ret;
6757 }
6758 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006759}
6760
6761/**
6762 * alloc_contig_range() -- tries to allocate given range of pages
6763 * @start: start PFN to allocate
6764 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006765 * @migratetype: migratetype of the underlaying pageblocks (either
6766 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6767 * in range must have the same migratetype and it must
6768 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006769 *
6770 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6771 * aligned, however it's the caller's responsibility to guarantee that
6772 * we are the only thread that changes migrate type of pageblocks the
6773 * pages fall in.
6774 *
6775 * The PFN range must belong to a single zone.
6776 *
6777 * Returns zero on success or negative error code. On success all
6778 * pages which PFN is in [start, end) are allocated for the caller and
6779 * need to be freed with free_contig_range().
6780 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006781int alloc_contig_range(unsigned long start, unsigned long end,
6782 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006783{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006784 unsigned long outer_start, outer_end;
6785 int ret = 0, order;
6786
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006787 struct compact_control cc = {
6788 .nr_migratepages = 0,
6789 .order = -1,
6790 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07006791 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006792 .ignore_skip_hint = true,
6793 };
6794 INIT_LIST_HEAD(&cc.migratepages);
6795
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006796 /*
6797 * What we do here is we mark all pageblocks in range as
6798 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6799 * have different sizes, and due to the way page allocator
6800 * work, we align the range to biggest of the two pages so
6801 * that page allocator won't try to merge buddies from
6802 * different pageblocks and change MIGRATE_ISOLATE to some
6803 * other migration type.
6804 *
6805 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6806 * migrate the pages from an unaligned range (ie. pages that
6807 * we are interested in). This will put all the pages in
6808 * range back to page allocator as MIGRATE_ISOLATE.
6809 *
6810 * When this is done, we take the pages in range from page
6811 * allocator removing them from the buddy system. This way
6812 * page allocator will never consider using them.
6813 *
6814 * This lets us mark the pageblocks back as
6815 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6816 * aligned range but not in the unaligned, original range are
6817 * put back to page allocator so that buddy can use them.
6818 */
6819
6820 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08006821 pfn_max_align_up(end), migratetype,
6822 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006823 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07006824 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006825
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006826 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006827 if (ret)
6828 goto done;
6829
6830 /*
6831 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6832 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6833 * more, all pages in [start, end) are free in page allocator.
6834 * What we are going to do is to allocate all pages from
6835 * [start, end) (that is remove them from page allocator).
6836 *
6837 * The only problem is that pages at the beginning and at the
6838 * end of interesting range may be not aligned with pages that
6839 * page allocator holds, ie. they can be part of higher order
6840 * pages. Because of this, we reserve the bigger range and
6841 * once this is done free the pages we are not interested in.
6842 *
6843 * We don't have to hold zone->lock here because the pages are
6844 * isolated thus they won't get removed from buddy.
6845 */
6846
6847 lru_add_drain_all();
Vlastimil Babka510f5502014-12-10 15:43:07 -08006848 drain_all_pages(cc.zone);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006849
6850 order = 0;
6851 outer_start = start;
6852 while (!PageBuddy(pfn_to_page(outer_start))) {
6853 if (++order >= MAX_ORDER) {
6854 ret = -EBUSY;
6855 goto done;
6856 }
6857 outer_start &= ~0UL << order;
6858 }
6859
6860 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08006861 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewiczdae803e2014-11-13 15:19:27 -08006862 pr_info("%s: [%lx, %lx) PFNs busy\n",
6863 __func__, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006864 ret = -EBUSY;
6865 goto done;
6866 }
6867
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006868 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006869 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006870 if (!outer_end) {
6871 ret = -EBUSY;
6872 goto done;
6873 }
6874
6875 /* Free head and tail (if any) */
6876 if (start != outer_start)
6877 free_contig_range(outer_start, start - outer_start);
6878 if (end != outer_end)
6879 free_contig_range(end, outer_end - end);
6880
6881done:
6882 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006883 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006884 return ret;
6885}
6886
6887void free_contig_range(unsigned long pfn, unsigned nr_pages)
6888{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08006889 unsigned int count = 0;
6890
6891 for (; nr_pages--; pfn++) {
6892 struct page *page = pfn_to_page(pfn);
6893
6894 count += page_count(page) != 1;
6895 __free_page(page);
6896 }
6897 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006898}
6899#endif
6900
Jiang Liu4ed7e022012-07-31 16:43:35 -07006901#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07006902/*
6903 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6904 * page high values need to be recalulated.
6905 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07006906void __meminit zone_pcp_update(struct zone *zone)
6907{
Cody P Schafer0a647f32013-07-03 15:01:33 -07006908 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006909 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07006910 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07006911 pageset_set_high_and_batch(zone,
6912 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006913 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07006914}
6915#endif
6916
Jiang Liu340175b2012-07-31 16:43:32 -07006917void zone_pcp_reset(struct zone *zone)
6918{
6919 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07006920 int cpu;
6921 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07006922
6923 /* avoid races with drain_pages() */
6924 local_irq_save(flags);
6925 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07006926 for_each_online_cpu(cpu) {
6927 pset = per_cpu_ptr(zone->pageset, cpu);
6928 drain_zonestat(zone, pset);
6929 }
Jiang Liu340175b2012-07-31 16:43:32 -07006930 free_percpu(zone->pageset);
6931 zone->pageset = &boot_pageset;
6932 }
6933 local_irq_restore(flags);
6934}
6935
Wen Congyang6dcd73d2012-12-11 16:01:01 -08006936#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006937/*
6938 * All pages in the range must be isolated before calling this.
6939 */
6940void
6941__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6942{
6943 struct page *page;
6944 struct zone *zone;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006945 unsigned int order, i;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006946 unsigned long pfn;
6947 unsigned long flags;
6948 /* find the first valid pfn */
6949 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6950 if (pfn_valid(pfn))
6951 break;
6952 if (pfn == end_pfn)
6953 return;
6954 zone = page_zone(pfn_to_page(pfn));
6955 spin_lock_irqsave(&zone->lock, flags);
6956 pfn = start_pfn;
6957 while (pfn < end_pfn) {
6958 if (!pfn_valid(pfn)) {
6959 pfn++;
6960 continue;
6961 }
6962 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08006963 /*
6964 * The HWPoisoned page may be not in buddy system, and
6965 * page_count() is not 0.
6966 */
6967 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6968 pfn++;
6969 SetPageReserved(page);
6970 continue;
6971 }
6972
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006973 BUG_ON(page_count(page));
6974 BUG_ON(!PageBuddy(page));
6975 order = page_order(page);
6976#ifdef CONFIG_DEBUG_VM
6977 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6978 pfn, 1 << order, end_pfn);
6979#endif
6980 list_del(&page->lru);
6981 rmv_page_order(page);
6982 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006983 for (i = 0; i < (1 << order); i++)
6984 SetPageReserved((page+i));
6985 pfn += (1 << order);
6986 }
6987 spin_unlock_irqrestore(&zone->lock, flags);
6988}
6989#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006990
6991#ifdef CONFIG_MEMORY_FAILURE
6992bool is_free_buddy_page(struct page *page)
6993{
6994 struct zone *zone = page_zone(page);
6995 unsigned long pfn = page_to_pfn(page);
6996 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006997 unsigned int order;
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006998
6999 spin_lock_irqsave(&zone->lock, flags);
7000 for (order = 0; order < MAX_ORDER; order++) {
7001 struct page *page_head = page - (pfn & ((1 << order) - 1));
7002
7003 if (PageBuddy(page_head) && page_order(page_head) >= order)
7004 break;
7005 }
7006 spin_unlock_irqrestore(&zone->lock, flags);
7007
7008 return order < MAX_ORDER;
7009}
7010#endif