blob: 5f9394df19bf4708329aacbedb8e3bcb927315c3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
KOSAKI Motohiro10ed2732008-03-04 14:28:32 -080022#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/bootmem.h>
Yinghai Luedbe7d22010-08-25 13:39:16 -070024#include <linux/memblock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070026#include <linux/kernel.h>
Vegard Nossumb1eeab62008-11-25 16:55:53 +010027#include <linux/kmemcheck.h>
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -080028#include <linux/kasan.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/module.h>
30#include <linux/suspend.h>
31#include <linux/pagevec.h>
32#include <linux/blkdev.h>
33#include <linux/slab.h>
Dave Hansena238ab52011-05-24 17:12:16 -070034#include <linux/ratelimit.h>
David Rientjes5a3135c22007-10-16 23:25:53 -070035#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/notifier.h>
37#include <linux/topology.h>
38#include <linux/sysctl.h>
39#include <linux/cpu.h>
40#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070041#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/nodemask.h>
43#include <linux/vmalloc.h>
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -070044#include <linux/vmstat.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080045#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070046#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070047#include <linux/sort.h>
48#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070049#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080050#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -070051#include <linux/page-isolation.h>
Joonsoo Kimeefa864b2014-12-12 16:55:46 -080052#include <linux/page_ext.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070053#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010054#include <linux/kmemleak.h>
Mel Gorman56de7262010-05-24 14:32:30 -070055#include <linux/compaction.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070056#include <trace/events/kmem.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070057#include <linux/prefetch.h>
Lisa Du6e543d52013-09-11 14:22:36 -070058#include <linux/mm_inline.h>
Michal Nazarewicz041d3a82011-12-29 13:09:50 +010059#include <linux/migrate.h>
Joonsoo Kime30825f2014-12-12 16:55:49 -080060#include <linux/page_ext.h>
David Rientjes949f7ec2013-04-29 15:07:48 -070061#include <linux/hugetlb.h>
Clark Williams8bd75c72013-02-07 09:47:07 -060062#include <linux/sched/rt.h>
Joonsoo Kim48c96a32014-12-12 16:56:01 -080063#include <linux/page_owner.h>
Mel Gorman0e1cc952015-06-30 14:57:27 -070064#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Jiang Liu7ee3d4e2013-07-03 15:03:41 -070066#include <asm/sections.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070068#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070069#include "internal.h"
70
Cody P Schaferc8e251f2013-07-03 15:01:29 -070071/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
72static DEFINE_MUTEX(pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -070073#define MIN_PERCPU_PAGELIST_FRACTION (8)
Cody P Schaferc8e251f2013-07-03 15:01:29 -070074
Lee Schermerhorn72812012010-05-26 14:44:56 -070075#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
76DEFINE_PER_CPU(int, numa_node);
77EXPORT_PER_CPU_SYMBOL(numa_node);
78#endif
79
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070080#ifdef CONFIG_HAVE_MEMORYLESS_NODES
81/*
82 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
83 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
84 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
85 * defined in <linux/topology.h>.
86 */
87DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
88EXPORT_PER_CPU_SYMBOL(_numa_mem_);
Joonsoo Kimad2c8142014-10-09 15:26:13 -070089int _node_numa_mem_[MAX_NUMNODES];
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070090#endif
91
Linus Torvalds1da177e2005-04-16 15:20:36 -070092/*
Christoph Lameter13808912007-10-16 01:25:27 -070093 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 */
Christoph Lameter13808912007-10-16 01:25:27 -070095nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
96 [N_POSSIBLE] = NODE_MASK_ALL,
97 [N_ONLINE] = { { [0] = 1UL } },
98#ifndef CONFIG_NUMA
99 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
100#ifdef CONFIG_HIGHMEM
101 [N_HIGH_MEMORY] = { { [0] = 1UL } },
102#endif
Lai Jiangshan20b2f522012-12-12 13:52:00 -0800103#ifdef CONFIG_MOVABLE_NODE
104 [N_MEMORY] = { { [0] = 1UL } },
105#endif
Christoph Lameter13808912007-10-16 01:25:27 -0700106 [N_CPU] = { { [0] = 1UL } },
107#endif /* NUMA */
108};
109EXPORT_SYMBOL(node_states);
110
Jiang Liuc3d5f5f2013-07-03 15:03:14 -0700111/* Protect totalram_pages and zone->managed_pages */
112static DEFINE_SPINLOCK(managed_page_count_lock);
113
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700114unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -0700115unsigned long totalreserve_pages __read_mostly;
Pintu Kumare48322a2014-12-18 16:17:15 -0800116unsigned long totalcma_pages __read_mostly;
Johannes Weinerab8fabd2012-01-10 15:07:42 -0800117/*
118 * When calculating the number of globally allowed dirty pages, there
119 * is a certain number of per-zone reserves that should not be
120 * considered dirtyable memory. This is the sum of those reserves
121 * over all existing zones that contribute dirtyable memory.
122 */
123unsigned long dirty_balance_reserve __read_mostly;
124
Hugh Dickins1b76b022012-05-11 01:00:07 -0700125int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000126gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800128#ifdef CONFIG_PM_SLEEP
129/*
130 * The following functions are used by the suspend/hibernate code to temporarily
131 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
132 * while devices are suspended. To avoid races with the suspend/hibernate code,
133 * they should always be called with pm_mutex held (gfp_allowed_mask also should
134 * only be modified with pm_mutex held, unless the suspend/hibernate code is
135 * guaranteed not to run in parallel with that modification).
136 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100137
138static gfp_t saved_gfp_mask;
139
140void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800141{
142 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100143 if (saved_gfp_mask) {
144 gfp_allowed_mask = saved_gfp_mask;
145 saved_gfp_mask = 0;
146 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800147}
148
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100149void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800150{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800151 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100152 WARN_ON(saved_gfp_mask);
153 saved_gfp_mask = gfp_allowed_mask;
154 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800155}
Mel Gormanf90ac392012-01-10 15:07:15 -0800156
157bool pm_suspended_storage(void)
158{
159 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
160 return false;
161 return true;
162}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800163#endif /* CONFIG_PM_SLEEP */
164
Mel Gormand9c23402007-10-16 01:26:01 -0700165#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
166int pageblock_order __read_mostly;
167#endif
168
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800169static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800170
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171/*
172 * results with 256, 32 in the lowmem_reserve sysctl:
173 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
174 * 1G machine -> (16M dma, 784M normal, 224M high)
175 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
176 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
Yaowei Bai84109e12015-02-12 15:00:22 -0800177 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100178 *
179 * TBD: should special case ZONE_DMA32 machines here - in those we normally
180 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700182int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800183#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700184 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800185#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700186#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700187 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700188#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700189#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700190 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700191#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700192 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700193};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
195EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
Helge Deller15ad7cd2006-12-06 20:40:36 -0800197static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800198#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700199 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800200#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700201#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700202 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700203#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700204 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700205#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700206 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700207#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700208 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700209};
210
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211int min_free_kbytes = 1024;
Han Pingtian42aa83c2014-01-23 15:53:28 -0800212int user_min_free_kbytes = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
Jan Beulich2c85f512009-09-21 17:03:07 -0700214static unsigned long __meminitdata nr_kernel_pages;
215static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700216static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
Tejun Heo0ee332c2011-12-08 10:22:09 -0800218#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
219static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
220static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
221static unsigned long __initdata required_kernelcore;
222static unsigned long __initdata required_movablecore;
223static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gormanc7132162006-09-27 01:49:43 -0700224
Tejun Heo0ee332c2011-12-08 10:22:09 -0800225/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
226int movable_zone;
227EXPORT_SYMBOL(movable_zone);
228#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700229
Miklos Szeredi418508c2007-05-23 13:57:55 -0700230#if MAX_NUMNODES > 1
231int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700232int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700233EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700234EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700235#endif
236
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700237int page_group_by_mobility_disabled __read_mostly;
238
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700239#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
240static inline void reset_deferred_meminit(pg_data_t *pgdat)
241{
242 pgdat->first_deferred_pfn = ULONG_MAX;
243}
244
245/* Returns true if the struct page for the pfn is uninitialised */
Mel Gorman0e1cc952015-06-30 14:57:27 -0700246static inline bool __meminit early_page_uninitialised(unsigned long pfn)
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700247{
Mel Gormanae026b22015-07-17 16:23:48 -0700248 if (pfn >= NODE_DATA(early_pfn_to_nid(pfn))->first_deferred_pfn)
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700249 return true;
250
251 return false;
252}
253
Mel Gorman7e18adb2015-06-30 14:57:05 -0700254static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
255{
256 if (pfn >= NODE_DATA(nid)->first_deferred_pfn)
257 return true;
258
259 return false;
260}
261
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700262/*
263 * Returns false when the remaining initialisation should be deferred until
264 * later in the boot cycle when it can be parallelised.
265 */
266static inline bool update_defer_init(pg_data_t *pgdat,
267 unsigned long pfn, unsigned long zone_end,
268 unsigned long *nr_initialised)
269{
270 /* Always populate low zones for address-contrained allocations */
271 if (zone_end < pgdat_end_pfn(pgdat))
272 return true;
273
274 /* Initialise at least 2G of the highest zone */
275 (*nr_initialised)++;
276 if (*nr_initialised > (2UL << (30 - PAGE_SHIFT)) &&
277 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
278 pgdat->first_deferred_pfn = pfn;
279 return false;
280 }
281
282 return true;
283}
284#else
285static inline void reset_deferred_meminit(pg_data_t *pgdat)
286{
287}
288
289static inline bool early_page_uninitialised(unsigned long pfn)
290{
291 return false;
292}
293
Mel Gorman7e18adb2015-06-30 14:57:05 -0700294static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
295{
296 return false;
297}
298
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700299static inline bool update_defer_init(pg_data_t *pgdat,
300 unsigned long pfn, unsigned long zone_end,
301 unsigned long *nr_initialised)
302{
303 return true;
304}
305#endif
306
307
Minchan Kimee6f5092012-07-31 16:43:50 -0700308void set_pageblock_migratetype(struct page *page, int migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700309{
KOSAKI Motohiro5d0f3f72013-11-12 15:08:18 -0800310 if (unlikely(page_group_by_mobility_disabled &&
311 migratetype < MIGRATE_PCPTYPES))
Mel Gorman49255c62009-06-16 15:31:58 -0700312 migratetype = MIGRATE_UNMOVABLE;
313
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700314 set_pageblock_flags_group(page, (unsigned long)migratetype,
315 PB_migrate, PB_migrate_end);
316}
317
Nick Piggin13e74442006-01-06 00:10:58 -0800318#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700319static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700321 int ret = 0;
322 unsigned seq;
323 unsigned long pfn = page_to_pfn(page);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800324 unsigned long sp, start_pfn;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700325
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700326 do {
327 seq = zone_span_seqbegin(zone);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800328 start_pfn = zone->zone_start_pfn;
329 sp = zone->spanned_pages;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800330 if (!zone_spans_pfn(zone, pfn))
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700331 ret = 1;
332 } while (zone_span_seqretry(zone, seq));
333
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800334 if (ret)
Dave Hansen613813e2014-06-04 16:07:27 -0700335 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
336 pfn, zone_to_nid(zone), zone->name,
337 start_pfn, start_pfn + sp);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800338
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700339 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700340}
341
342static int page_is_consistent(struct zone *zone, struct page *page)
343{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700344 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700345 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700347 return 0;
348
349 return 1;
350}
351/*
352 * Temporary debugging check for pages not lying within a given zone.
353 */
354static int bad_range(struct zone *zone, struct page *page)
355{
356 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700358 if (!page_is_consistent(zone, page))
359 return 1;
360
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 return 0;
362}
Nick Piggin13e74442006-01-06 00:10:58 -0800363#else
364static inline int bad_range(struct zone *zone, struct page *page)
365{
366 return 0;
367}
368#endif
369
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700370static void bad_page(struct page *page, const char *reason,
371 unsigned long bad_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800373 static unsigned long resume;
374 static unsigned long nr_shown;
375 static unsigned long nr_unshown;
376
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200377 /* Don't complain about poisoned pages */
378 if (PageHWPoison(page)) {
Mel Gorman22b751c2013-02-22 16:34:59 -0800379 page_mapcount_reset(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200380 return;
381 }
382
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800383 /*
384 * Allow a burst of 60 reports, then keep quiet for that minute;
385 * or allow a steady drip of one report per second.
386 */
387 if (nr_shown == 60) {
388 if (time_before(jiffies, resume)) {
389 nr_unshown++;
390 goto out;
391 }
392 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800393 printk(KERN_ALERT
394 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800395 nr_unshown);
396 nr_unshown = 0;
397 }
398 nr_shown = 0;
399 }
400 if (nr_shown++ == 0)
401 resume = jiffies + 60 * HZ;
402
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800403 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800404 current->comm, page_to_pfn(page));
Dave Hansenf0b791a2014-01-23 15:52:49 -0800405 dump_page_badflags(page, reason, bad_flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700406
Dave Jones4f318882011-10-31 17:07:24 -0700407 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800409out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800410 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800411 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030412 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413}
414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415/*
416 * Higher-order pages are called "compound pages". They are structured thusly:
417 *
418 * The first PAGE_SIZE page is called the "head page".
419 *
420 * The remaining PAGE_SIZE pages are called "tail pages".
421 *
Wang Sheng-Hui6416b9fa2011-11-17 10:53:50 +0100422 * All pages have PG_compound set. All tail pages have their ->first_page
423 * pointing at the head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800425 * The first tail page's ->lru.next holds the address of the compound page's
426 * put_page() function. Its ->lru.prev holds the order of allocation.
427 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800429
430static void free_compound_page(struct page *page)
431{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700432 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800433}
434
Andi Kleen01ad1c02008-07-23 21:27:46 -0700435void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436{
437 int i;
438 int nr_pages = 1 << order;
439
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800440 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700441 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700442 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800443 for (i = 1; i < nr_pages; i++) {
444 struct page *p = page + i;
Youquan Song58a84aa2011-12-08 14:34:18 -0800445 set_page_count(p, 0);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700446 p->first_page = page;
David Rientjes668f9abb2014-03-03 15:38:18 -0800447 /* Make sure p->first_page is always valid for PageTail() */
448 smp_wmb();
449 __SetPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 }
451}
452
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800453#ifdef CONFIG_DEBUG_PAGEALLOC
454unsigned int _debug_guardpage_minorder;
Joonsoo Kim031bc572014-12-12 16:55:52 -0800455bool _debug_pagealloc_enabled __read_mostly;
Joonsoo Kime30825f2014-12-12 16:55:49 -0800456bool _debug_guardpage_enabled __read_mostly;
457
Joonsoo Kim031bc572014-12-12 16:55:52 -0800458static int __init early_debug_pagealloc(char *buf)
459{
460 if (!buf)
461 return -EINVAL;
462
463 if (strcmp(buf, "on") == 0)
464 _debug_pagealloc_enabled = true;
465
466 return 0;
467}
468early_param("debug_pagealloc", early_debug_pagealloc);
469
Joonsoo Kime30825f2014-12-12 16:55:49 -0800470static bool need_debug_guardpage(void)
471{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800472 /* If we don't use debug_pagealloc, we don't need guard page */
473 if (!debug_pagealloc_enabled())
474 return false;
475
Joonsoo Kime30825f2014-12-12 16:55:49 -0800476 return true;
477}
478
479static void init_debug_guardpage(void)
480{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800481 if (!debug_pagealloc_enabled())
482 return;
483
Joonsoo Kime30825f2014-12-12 16:55:49 -0800484 _debug_guardpage_enabled = true;
485}
486
487struct page_ext_operations debug_guardpage_ops = {
488 .need = need_debug_guardpage,
489 .init = init_debug_guardpage,
490};
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800491
492static int __init debug_guardpage_minorder_setup(char *buf)
493{
494 unsigned long res;
495
496 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
497 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
498 return 0;
499 }
500 _debug_guardpage_minorder = res;
501 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
502 return 0;
503}
504__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
505
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800506static inline void set_page_guard(struct zone *zone, struct page *page,
507 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800508{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800509 struct page_ext *page_ext;
510
511 if (!debug_guardpage_enabled())
512 return;
513
514 page_ext = lookup_page_ext(page);
515 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
516
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800517 INIT_LIST_HEAD(&page->lru);
518 set_page_private(page, order);
519 /* Guard pages are not available for any usage */
520 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800521}
522
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800523static inline void clear_page_guard(struct zone *zone, struct page *page,
524 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800525{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800526 struct page_ext *page_ext;
527
528 if (!debug_guardpage_enabled())
529 return;
530
531 page_ext = lookup_page_ext(page);
532 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
533
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800534 set_page_private(page, 0);
535 if (!is_migrate_isolate(migratetype))
536 __mod_zone_freepage_state(zone, (1 << order), migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800537}
538#else
Joonsoo Kime30825f2014-12-12 16:55:49 -0800539struct page_ext_operations debug_guardpage_ops = { NULL, };
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800540static inline void set_page_guard(struct zone *zone, struct page *page,
541 unsigned int order, int migratetype) {}
542static inline void clear_page_guard(struct zone *zone, struct page *page,
543 unsigned int order, int migratetype) {}
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800544#endif
545
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700546static inline void set_page_order(struct page *page, unsigned int order)
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700547{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700548 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000549 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550}
551
552static inline void rmv_page_order(struct page *page)
553{
Nick Piggin676165a2006-04-10 11:21:48 +1000554 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700555 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556}
557
558/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 * This function checks whether a page is free && is the buddy
560 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800561 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000562 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700563 * (c) a page and its buddy have the same order &&
564 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 *
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700566 * For recording whether a page is in the buddy system, we set ->_mapcount
567 * PAGE_BUDDY_MAPCOUNT_VALUE.
568 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
569 * serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000570 *
571 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700573static inline int page_is_buddy(struct page *page, struct page *buddy,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700574 unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700576 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800577 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800578
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800579 if (page_is_guard(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700580 if (page_zone_id(page) != page_zone_id(buddy))
581 return 0;
582
Weijie Yang4c5018c2015-02-10 14:11:39 -0800583 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
584
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800585 return 1;
586 }
587
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700588 if (PageBuddy(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700589 /*
590 * zone check is done late to avoid uselessly
591 * calculating zone/node ids for pages that could
592 * never merge.
593 */
594 if (page_zone_id(page) != page_zone_id(buddy))
595 return 0;
596
Weijie Yang4c5018c2015-02-10 14:11:39 -0800597 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
598
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700599 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000600 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700601 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602}
603
604/*
605 * Freeing function for a buddy system allocator.
606 *
607 * The concept of a buddy system is to maintain direct-mapped table
608 * (containing bit values) for memory blocks of various "orders".
609 * The bottom level table contains the map for the smallest allocatable
610 * units of memory (here, pages), and each level above it describes
611 * pairs of units from the levels below, hence, "buddies".
612 * At a high level, all that happens here is marking the table entry
613 * at the bottom level available, and propagating the changes upward
614 * as necessary, plus some accounting needed to play nicely with other
615 * parts of the VM system.
616 * At each level, we keep a list of pages, which are heads of continuous
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700617 * free pages of length of (1 << order) and marked with _mapcount
618 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
619 * field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100621 * other. That is, if we allocate a small block, and both were
622 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100624 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100626 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 */
628
Nick Piggin48db57f2006-01-08 01:00:42 -0800629static inline void __free_one_page(struct page *page,
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700630 unsigned long pfn,
Mel Gormaned0ae212009-06-16 15:32:07 -0700631 struct zone *zone, unsigned int order,
632 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633{
634 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700635 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800636 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700637 struct page *buddy;
Joonsoo Kim3c605092014-11-13 15:19:21 -0800638 int max_order = MAX_ORDER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
Cody P Schaferd29bb972013-02-22 16:35:25 -0800640 VM_BUG_ON(!zone_is_initialized(zone));
Kirill A. Shutemov6e9f0d52015-02-11 15:25:50 -0800641 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
Mel Gormaned0ae212009-06-16 15:32:07 -0700643 VM_BUG_ON(migratetype == -1);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800644 if (is_migrate_isolate(migratetype)) {
645 /*
646 * We restrict max order of merging to prevent merge
647 * between freepages on isolate pageblock and normal
648 * pageblock. Without this, pageblock isolation
649 * could cause incorrect freepage accounting.
650 */
651 max_order = min(MAX_ORDER, pageblock_order + 1);
652 } else {
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800653 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800654 }
Mel Gormaned0ae212009-06-16 15:32:07 -0700655
Joonsoo Kim3c605092014-11-13 15:19:21 -0800656 page_idx = pfn & ((1 << max_order) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
Sasha Levin309381fea2014-01-23 15:52:54 -0800658 VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
659 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
Joonsoo Kim3c605092014-11-13 15:19:21 -0800661 while (order < max_order - 1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800662 buddy_idx = __find_buddy_index(page_idx, order);
663 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700664 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700665 break;
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800666 /*
667 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
668 * merge with it and move up one order.
669 */
670 if (page_is_guard(buddy)) {
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800671 clear_page_guard(zone, buddy, order, migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800672 } else {
673 list_del(&buddy->lru);
674 zone->free_area[order].nr_free--;
675 rmv_page_order(buddy);
676 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800677 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 page = page + (combined_idx - page_idx);
679 page_idx = combined_idx;
680 order++;
681 }
682 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700683
684 /*
685 * If this is not the largest possible page, check if the buddy
686 * of the next-highest order is free. If it is, it's possible
687 * that pages are being freed that will coalesce soon. In case,
688 * that is happening, add the free page to the tail of the list
689 * so it's less likely to be used soon and more likely to be merged
690 * as a higher order page
691 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700692 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700693 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800694 combined_idx = buddy_idx & page_idx;
695 higher_page = page + (combined_idx - page_idx);
696 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700697 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700698 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
699 list_add_tail(&page->lru,
700 &zone->free_area[order].free_list[migratetype]);
701 goto out;
702 }
703 }
704
705 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
706out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 zone->free_area[order].nr_free++;
708}
709
Nick Piggin224abf92006-01-06 00:11:11 -0800710static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700712 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800713 unsigned long bad_flags = 0;
714
715 if (unlikely(page_mapcount(page)))
716 bad_reason = "nonzero mapcount";
717 if (unlikely(page->mapping != NULL))
718 bad_reason = "non-NULL mapping";
719 if (unlikely(atomic_read(&page->_count) != 0))
720 bad_reason = "nonzero _count";
721 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
722 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
723 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
724 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -0800725#ifdef CONFIG_MEMCG
726 if (unlikely(page->mem_cgroup))
727 bad_reason = "page still charged to cgroup";
728#endif
Dave Hansenf0b791a2014-01-23 15:52:49 -0800729 if (unlikely(bad_reason)) {
730 bad_page(page, bad_reason, bad_flags);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800731 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800732 }
Peter Zijlstra90572892013-10-07 11:29:20 +0100733 page_cpupid_reset_last(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800734 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
735 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
736 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737}
738
739/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700740 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700742 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 *
744 * If the zone was previously in an "all pages pinned" state then look to
745 * see if this freeing clears that state.
746 *
747 * And clear the zone's pages_scanned counter, to hold off the "all pages are
748 * pinned" detection logic.
749 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700750static void free_pcppages_bulk(struct zone *zone, int count,
751 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700753 int migratetype = 0;
Mel Gormana6f9edd62009-09-21 17:03:20 -0700754 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700755 int to_free = count;
Mel Gorman0d5d8232014-08-06 16:07:16 -0700756 unsigned long nr_scanned;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700757
Nick Pigginc54ad302006-01-06 00:10:56 -0800758 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700759 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
760 if (nr_scanned)
761 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700762
Mel Gorman72853e22010-09-09 16:38:16 -0700763 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800764 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700765 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800766
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700767 /*
Mel Gormana6f9edd62009-09-21 17:03:20 -0700768 * Remove pages from lists in a round-robin fashion. A
769 * batch_free count is maintained that is incremented when an
770 * empty list is encountered. This is so more pages are freed
771 * off fuller lists instead of spinning excessively around empty
772 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700773 */
774 do {
Mel Gormana6f9edd62009-09-21 17:03:20 -0700775 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700776 if (++migratetype == MIGRATE_PCPTYPES)
777 migratetype = 0;
778 list = &pcp->lists[migratetype];
779 } while (list_empty(list));
780
Namhyung Kim1d168712011-03-22 16:32:45 -0700781 /* This is the only non-empty list. Free them all. */
782 if (batch_free == MIGRATE_PCPTYPES)
783 batch_free = to_free;
784
Mel Gormana6f9edd62009-09-21 17:03:20 -0700785 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700786 int mt; /* migratetype of the to-be-freed page */
787
Mel Gormana6f9edd62009-09-21 17:03:20 -0700788 page = list_entry(list->prev, struct page, lru);
789 /* must delete as __free_one_page list manipulates */
790 list_del(&page->lru);
Minchan Kimb12c4ad2012-10-08 16:32:08 -0700791 mt = get_freepage_migratetype(page);
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800792 if (unlikely(has_isolate_pageblock(zone)))
Joonsoo Kim51bb1a42014-11-13 15:19:14 -0800793 mt = get_pageblock_migratetype(page);
Joonsoo Kim51bb1a42014-11-13 15:19:14 -0800794
Hugh Dickinsa7016232010-01-29 17:46:34 +0000795 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700796 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700797 trace_mm_page_pcpu_drain(page, 0, mt);
Mel Gorman72853e22010-09-09 16:38:16 -0700798 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800800 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801}
802
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700803static void free_one_page(struct zone *zone,
804 struct page *page, unsigned long pfn,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700805 unsigned int order,
Mel Gormaned0ae212009-06-16 15:32:07 -0700806 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800807{
Mel Gorman0d5d8232014-08-06 16:07:16 -0700808 unsigned long nr_scanned;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700809 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700810 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
811 if (nr_scanned)
812 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700813
Joonsoo Kimad53f922014-11-13 15:19:11 -0800814 if (unlikely(has_isolate_pageblock(zone) ||
815 is_migrate_isolate(migratetype))) {
816 migratetype = get_pfnblock_migratetype(page, pfn);
Joonsoo Kimad53f922014-11-13 15:19:11 -0800817 }
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700818 __free_one_page(page, pfn, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700819 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800820}
821
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800822static int free_tail_pages_check(struct page *head_page, struct page *page)
823{
824 if (!IS_ENABLED(CONFIG_DEBUG_VM))
825 return 0;
826 if (unlikely(!PageTail(page))) {
827 bad_page(page, "PageTail not set", 0);
828 return 1;
829 }
830 if (unlikely(page->first_page != head_page)) {
831 bad_page(page, "first_page not consistent", 0);
832 return 1;
833 }
834 return 0;
835}
836
Robin Holt1e8ce832015-06-30 14:56:45 -0700837static void __meminit __init_single_page(struct page *page, unsigned long pfn,
838 unsigned long zone, int nid)
839{
Robin Holt1e8ce832015-06-30 14:56:45 -0700840 set_page_links(page, zone, nid, pfn);
Robin Holt1e8ce832015-06-30 14:56:45 -0700841 init_page_count(page);
842 page_mapcount_reset(page);
843 page_cpupid_reset_last(page);
Robin Holt1e8ce832015-06-30 14:56:45 -0700844
Robin Holt1e8ce832015-06-30 14:56:45 -0700845 INIT_LIST_HEAD(&page->lru);
846#ifdef WANT_PAGE_VIRTUAL
847 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
848 if (!is_highmem_idx(zone))
849 set_page_address(page, __va(pfn << PAGE_SHIFT));
850#endif
851}
852
853static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
854 int nid)
855{
856 return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
857}
858
Mel Gorman7e18adb2015-06-30 14:57:05 -0700859#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
860static void init_reserved_page(unsigned long pfn)
861{
862 pg_data_t *pgdat;
863 int nid, zid;
864
865 if (!early_page_uninitialised(pfn))
866 return;
867
868 nid = early_pfn_to_nid(pfn);
869 pgdat = NODE_DATA(nid);
870
871 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
872 struct zone *zone = &pgdat->node_zones[zid];
873
874 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
875 break;
876 }
877 __init_single_pfn(pfn, zid, nid);
878}
879#else
880static inline void init_reserved_page(unsigned long pfn)
881{
882}
883#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
884
Nathan Zimmer92923ca32015-06-30 14:56:48 -0700885/*
886 * Initialised pages do not have PageReserved set. This function is
887 * called for each range allocated by the bootmem allocator and
888 * marks the pages PageReserved. The remaining valid pages are later
889 * sent to the buddy page allocator.
890 */
Mel Gorman7e18adb2015-06-30 14:57:05 -0700891void __meminit reserve_bootmem_region(unsigned long start, unsigned long end)
Nathan Zimmer92923ca32015-06-30 14:56:48 -0700892{
893 unsigned long start_pfn = PFN_DOWN(start);
894 unsigned long end_pfn = PFN_UP(end);
895
Mel Gorman7e18adb2015-06-30 14:57:05 -0700896 for (; start_pfn < end_pfn; start_pfn++) {
897 if (pfn_valid(start_pfn)) {
898 struct page *page = pfn_to_page(start_pfn);
899
900 init_reserved_page(start_pfn);
901 SetPageReserved(page);
902 }
903 }
Nathan Zimmer92923ca32015-06-30 14:56:48 -0700904}
905
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700906static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907{
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800908 bool compound = PageCompound(page);
909 int i, bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
Yu Zhaoab1f3062014-12-10 15:43:17 -0800911 VM_BUG_ON_PAGE(PageTail(page), page);
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800912 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
Yu Zhaoab1f3062014-12-10 15:43:17 -0800913
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -0800914 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100915 kmemcheck_free_shadow(page, order);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -0800916 kasan_free_pages(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100917
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800918 if (PageAnon(page))
919 page->mapping = NULL;
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800920 bad += free_pages_check(page);
921 for (i = 1; i < (1 << order); i++) {
922 if (compound)
923 bad += free_tail_pages_check(page, page + i);
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800924 bad += free_pages_check(page + i);
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800925 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800926 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700927 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800928
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800929 reset_page_owner(page, order);
930
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700931 if (!PageHighMem(page)) {
Pintu Kumarb8af2942013-09-11 14:20:34 -0700932 debug_check_no_locks_freed(page_address(page),
933 PAGE_SIZE << order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700934 debug_check_no_obj_freed(page_address(page),
935 PAGE_SIZE << order);
936 }
Nick Piggindafb1362006-10-11 01:21:30 -0700937 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800938 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700939
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700940 return true;
941}
942
943static void __free_pages_ok(struct page *page, unsigned int order)
944{
945 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -0700946 int migratetype;
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700947 unsigned long pfn = page_to_pfn(page);
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700948
949 if (!free_pages_prepare(page, order))
950 return;
951
Mel Gormancfc47a22014-06-04 16:10:19 -0700952 migratetype = get_pfnblock_migratetype(page, pfn);
Nick Pigginc54ad302006-01-06 00:10:56 -0800953 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700954 __count_vm_events(PGFREE, 1 << order);
Minchan Kim95e34412012-10-08 16:32:11 -0700955 set_freepage_migratetype(page, migratetype);
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700956 free_one_page(page_zone(page), page, pfn, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -0800957 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958}
959
Mel Gorman0e1cc952015-06-30 14:57:27 -0700960static void __init __free_pages_boot_core(struct page *page,
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700961 unsigned long pfn, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800962{
Johannes Weinerc3993072012-01-10 15:08:10 -0800963 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700964 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -0800965 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -0800966
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700967 prefetchw(p);
968 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
969 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -0800970 __ClearPageReserved(p);
971 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -0800972 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700973 __ClearPageReserved(p);
974 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -0800975
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700976 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -0800977 set_page_refcounted(page);
978 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800979}
980
Mel Gorman75a592a2015-06-30 14:56:59 -0700981#if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
982 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
Mel Gorman7ace9912015-08-06 15:46:13 -0700983
Mel Gorman75a592a2015-06-30 14:56:59 -0700984static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
985
986int __meminit early_pfn_to_nid(unsigned long pfn)
987{
Mel Gorman7ace9912015-08-06 15:46:13 -0700988 static DEFINE_SPINLOCK(early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -0700989 int nid;
990
Mel Gorman7ace9912015-08-06 15:46:13 -0700991 spin_lock(&early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -0700992 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
Mel Gorman7ace9912015-08-06 15:46:13 -0700993 if (nid < 0)
994 nid = 0;
995 spin_unlock(&early_pfn_lock);
996
997 return nid;
Mel Gorman75a592a2015-06-30 14:56:59 -0700998}
999#endif
1000
1001#ifdef CONFIG_NODES_SPAN_OTHER_NODES
1002static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1003 struct mminit_pfnnid_cache *state)
1004{
1005 int nid;
1006
1007 nid = __early_pfn_to_nid(pfn, state);
1008 if (nid >= 0 && nid != node)
1009 return false;
1010 return true;
1011}
1012
1013/* Only safe to use early in boot when initialisation is single-threaded */
1014static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1015{
1016 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1017}
1018
1019#else
1020
1021static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1022{
1023 return true;
1024}
1025static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1026 struct mminit_pfnnid_cache *state)
1027{
1028 return true;
1029}
1030#endif
1031
1032
Mel Gorman0e1cc952015-06-30 14:57:27 -07001033void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001034 unsigned int order)
1035{
1036 if (early_page_uninitialised(pfn))
1037 return;
1038 return __free_pages_boot_core(page, pfn, order);
1039}
1040
Mel Gorman7e18adb2015-06-30 14:57:05 -07001041#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
Mel Gorman0e1cc952015-06-30 14:57:27 -07001042static void __init deferred_free_range(struct page *page,
Mel Gormana4de83d2015-06-30 14:57:16 -07001043 unsigned long pfn, int nr_pages)
1044{
1045 int i;
1046
1047 if (!page)
1048 return;
1049
1050 /* Free a large naturally-aligned chunk if possible */
1051 if (nr_pages == MAX_ORDER_NR_PAGES &&
1052 (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
Mel Gormanac5d2532015-06-30 14:57:20 -07001053 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormana4de83d2015-06-30 14:57:16 -07001054 __free_pages_boot_core(page, pfn, MAX_ORDER-1);
1055 return;
1056 }
1057
1058 for (i = 0; i < nr_pages; i++, page++, pfn++)
1059 __free_pages_boot_core(page, pfn, 0);
1060}
1061
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001062/* Completion tracking for deferred_init_memmap() threads */
1063static atomic_t pgdat_init_n_undone __initdata;
1064static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1065
1066static inline void __init pgdat_init_report_one_done(void)
1067{
1068 if (atomic_dec_and_test(&pgdat_init_n_undone))
1069 complete(&pgdat_init_all_done_comp);
1070}
Mel Gorman0e1cc952015-06-30 14:57:27 -07001071
Mel Gorman7e18adb2015-06-30 14:57:05 -07001072/* Initialise remaining memory on a node */
Mel Gorman0e1cc952015-06-30 14:57:27 -07001073static int __init deferred_init_memmap(void *data)
Mel Gorman7e18adb2015-06-30 14:57:05 -07001074{
Mel Gorman0e1cc952015-06-30 14:57:27 -07001075 pg_data_t *pgdat = data;
1076 int nid = pgdat->node_id;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001077 struct mminit_pfnnid_cache nid_init_state = { };
1078 unsigned long start = jiffies;
1079 unsigned long nr_pages = 0;
1080 unsigned long walk_start, walk_end;
1081 int i, zid;
1082 struct zone *zone;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001083 unsigned long first_init_pfn = pgdat->first_deferred_pfn;
Mel Gorman0e1cc952015-06-30 14:57:27 -07001084 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001085
Mel Gorman0e1cc952015-06-30 14:57:27 -07001086 if (first_init_pfn == ULONG_MAX) {
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001087 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001088 return 0;
1089 }
1090
1091 /* Bind memory initialisation thread to a local node if possible */
1092 if (!cpumask_empty(cpumask))
1093 set_cpus_allowed_ptr(current, cpumask);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001094
1095 /* Sanity check boundaries */
1096 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1097 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1098 pgdat->first_deferred_pfn = ULONG_MAX;
1099
1100 /* Only the highest zone is deferred so find it */
1101 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1102 zone = pgdat->node_zones + zid;
1103 if (first_init_pfn < zone_end_pfn(zone))
1104 break;
1105 }
1106
1107 for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1108 unsigned long pfn, end_pfn;
Mel Gorman54608c32015-06-30 14:57:09 -07001109 struct page *page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001110 struct page *free_base_page = NULL;
1111 unsigned long free_base_pfn = 0;
1112 int nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001113
1114 end_pfn = min(walk_end, zone_end_pfn(zone));
1115 pfn = first_init_pfn;
1116 if (pfn < walk_start)
1117 pfn = walk_start;
1118 if (pfn < zone->zone_start_pfn)
1119 pfn = zone->zone_start_pfn;
1120
1121 for (; pfn < end_pfn; pfn++) {
Mel Gorman54608c32015-06-30 14:57:09 -07001122 if (!pfn_valid_within(pfn))
Mel Gormana4de83d2015-06-30 14:57:16 -07001123 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001124
Mel Gorman54608c32015-06-30 14:57:09 -07001125 /*
1126 * Ensure pfn_valid is checked every
1127 * MAX_ORDER_NR_PAGES for memory holes
1128 */
1129 if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1130 if (!pfn_valid(pfn)) {
1131 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001132 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001133 }
1134 }
1135
1136 if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1137 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001138 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001139 }
1140
1141 /* Minimise pfn page lookups and scheduler checks */
1142 if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1143 page++;
1144 } else {
Mel Gormana4de83d2015-06-30 14:57:16 -07001145 nr_pages += nr_to_free;
1146 deferred_free_range(free_base_page,
1147 free_base_pfn, nr_to_free);
1148 free_base_page = NULL;
1149 free_base_pfn = nr_to_free = 0;
1150
Mel Gorman54608c32015-06-30 14:57:09 -07001151 page = pfn_to_page(pfn);
1152 cond_resched();
1153 }
Mel Gorman7e18adb2015-06-30 14:57:05 -07001154
1155 if (page->flags) {
1156 VM_BUG_ON(page_zone(page) != zone);
Mel Gormana4de83d2015-06-30 14:57:16 -07001157 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001158 }
1159
1160 __init_single_page(page, pfn, zid, nid);
Mel Gormana4de83d2015-06-30 14:57:16 -07001161 if (!free_base_page) {
1162 free_base_page = page;
1163 free_base_pfn = pfn;
1164 nr_to_free = 0;
1165 }
1166 nr_to_free++;
1167
1168 /* Where possible, batch up pages for a single free */
1169 continue;
1170free_range:
1171 /* Free the current block of pages to allocator */
1172 nr_pages += nr_to_free;
1173 deferred_free_range(free_base_page, free_base_pfn,
1174 nr_to_free);
1175 free_base_page = NULL;
1176 free_base_pfn = nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001177 }
Mel Gormana4de83d2015-06-30 14:57:16 -07001178
Mel Gorman7e18adb2015-06-30 14:57:05 -07001179 first_init_pfn = max(end_pfn, first_init_pfn);
1180 }
1181
1182 /* Sanity check that the next zone really is unpopulated */
1183 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1184
Mel Gorman0e1cc952015-06-30 14:57:27 -07001185 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
Mel Gorman7e18adb2015-06-30 14:57:05 -07001186 jiffies_to_msecs(jiffies - start));
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001187
1188 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001189 return 0;
1190}
1191
1192void __init page_alloc_init_late(void)
1193{
1194 int nid;
1195
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001196 /* There will be num_node_state(N_MEMORY) threads */
1197 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
Mel Gorman0e1cc952015-06-30 14:57:27 -07001198 for_each_node_state(nid, N_MEMORY) {
Mel Gorman0e1cc952015-06-30 14:57:27 -07001199 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1200 }
1201
1202 /* Block until all are initialised */
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001203 wait_for_completion(&pgdat_init_all_done_comp);
Mel Gorman4248b0d2015-08-06 15:46:20 -07001204
1205 /* Reinit limits that are based on free pages after the kernel is up */
1206 files_maxfiles_init();
Mel Gorman7e18adb2015-06-30 14:57:05 -07001207}
1208#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1209
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001210#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +08001211/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001212void __init init_cma_reserved_pageblock(struct page *page)
1213{
1214 unsigned i = pageblock_nr_pages;
1215 struct page *p = page;
1216
1217 do {
1218 __ClearPageReserved(p);
1219 set_page_count(p, 0);
1220 } while (++p, --i);
1221
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001222 set_pageblock_migratetype(page, MIGRATE_CMA);
Michal Nazarewiczdc783272014-07-02 15:22:35 -07001223
1224 if (pageblock_order >= MAX_ORDER) {
1225 i = pageblock_nr_pages;
1226 p = page;
1227 do {
1228 set_page_refcounted(p);
1229 __free_pages(p, MAX_ORDER - 1);
1230 p += MAX_ORDER_NR_PAGES;
1231 } while (i -= MAX_ORDER_NR_PAGES);
1232 } else {
1233 set_page_refcounted(page);
1234 __free_pages(page, pageblock_order);
1235 }
1236
Jiang Liu3dcc0572013-07-03 15:03:21 -07001237 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001238}
1239#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
1241/*
1242 * The order of subdivision here is critical for the IO subsystem.
1243 * Please do not alter this order without good reasons and regression
1244 * testing. Specifically, as large blocks of memory are subdivided,
1245 * the order in which smaller blocks are delivered depends on the order
1246 * they're subdivided in this function. This is the primary factor
1247 * influencing the order in which pages are delivered to the IO
1248 * subsystem according to empirical testing, and this is also justified
1249 * by considering the behavior of a buddy system containing a single
1250 * large block of memory acted on by a series of small allocations.
1251 * This behavior is a critical factor in sglist merging's success.
1252 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +01001253 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 */
Nick Piggin085cc7d2006-01-06 00:11:01 -08001255static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001256 int low, int high, struct free_area *area,
1257 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258{
1259 unsigned long size = 1 << high;
1260
1261 while (high > low) {
1262 area--;
1263 high--;
1264 size >>= 1;
Sasha Levin309381fea2014-01-23 15:52:54 -08001265 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08001266
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001267 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
Joonsoo Kime30825f2014-12-12 16:55:49 -08001268 debug_guardpage_enabled() &&
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001269 high < debug_guardpage_minorder()) {
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08001270 /*
1271 * Mark as guard pages (or page), that will allow to
1272 * merge back to allocator when buddy will be freed.
1273 * Corresponding page table entries will not be touched,
1274 * pages will stay not present in virtual address space
1275 */
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001276 set_page_guard(zone, &page[size], high, migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08001277 continue;
1278 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001279 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 area->nr_free++;
1281 set_page_order(&page[size], high);
1282 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283}
1284
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285/*
1286 * This page is about to be returned from the page allocator
1287 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001288static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -07001290 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -08001291 unsigned long bad_flags = 0;
1292
1293 if (unlikely(page_mapcount(page)))
1294 bad_reason = "nonzero mapcount";
1295 if (unlikely(page->mapping != NULL))
1296 bad_reason = "non-NULL mapping";
1297 if (unlikely(atomic_read(&page->_count) != 0))
1298 bad_reason = "nonzero _count";
Naoya Horiguchif4c18e62015-08-06 15:47:08 -07001299 if (unlikely(page->flags & __PG_HWPOISON)) {
1300 bad_reason = "HWPoisoned (hardware-corrupted)";
1301 bad_flags = __PG_HWPOISON;
1302 }
Dave Hansenf0b791a2014-01-23 15:52:49 -08001303 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1304 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1305 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1306 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -08001307#ifdef CONFIG_MEMCG
1308 if (unlikely(page->mem_cgroup))
1309 bad_reason = "page still charged to cgroup";
1310#endif
Dave Hansenf0b791a2014-01-23 15:52:49 -08001311 if (unlikely(bad_reason)) {
1312 bad_page(page, bad_reason, bad_flags);
Hugh Dickins689bceb2005-11-21 21:32:20 -08001313 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -08001314 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001315 return 0;
1316}
1317
Vlastimil Babka75379192015-02-11 15:25:38 -08001318static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1319 int alloc_flags)
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001320{
1321 int i;
1322
1323 for (i = 0; i < (1 << order); i++) {
1324 struct page *p = page + i;
1325 if (unlikely(check_new_page(p)))
1326 return 1;
1327 }
Hugh Dickins689bceb2005-11-21 21:32:20 -08001328
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001329 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -08001330 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -08001331
1332 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 kernel_map_pages(page, 1 << order, 1);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -08001334 kasan_alloc_pages(page, order);
Nick Piggin17cf4402006-03-22 00:08:41 -08001335
1336 if (gfp_flags & __GFP_ZERO)
Anisse Astierf4d28972015-06-24 16:56:36 -07001337 for (i = 0; i < (1 << order); i++)
1338 clear_highpage(page + i);
Nick Piggin17cf4402006-03-22 00:08:41 -08001339
1340 if (order && (gfp_flags & __GFP_COMP))
1341 prep_compound_page(page, order);
1342
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001343 set_page_owner(page, order, gfp_flags);
1344
Vlastimil Babka75379192015-02-11 15:25:38 -08001345 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001346 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
Vlastimil Babka75379192015-02-11 15:25:38 -08001347 * allocate the page. The expectation is that the caller is taking
1348 * steps that will free more memory. The caller should avoid the page
1349 * being used for !PFMEMALLOC purposes.
1350 */
Michal Hocko2f064f32015-08-21 14:11:51 -07001351 if (alloc_flags & ALLOC_NO_WATERMARKS)
1352 set_page_pfmemalloc(page);
1353 else
1354 clear_page_pfmemalloc(page);
Vlastimil Babka75379192015-02-11 15:25:38 -08001355
Hugh Dickins689bceb2005-11-21 21:32:20 -08001356 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357}
1358
Mel Gorman56fd56b2007-10-16 01:25:58 -07001359/*
1360 * Go through the free lists for the given migratetype and remove
1361 * the smallest available page from the freelists
1362 */
Mel Gorman728ec982009-06-16 15:32:04 -07001363static inline
1364struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -07001365 int migratetype)
1366{
1367 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -07001368 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001369 struct page *page;
1370
1371 /* Find a page of the appropriate size in the preferred list */
1372 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1373 area = &(zone->free_area[current_order]);
1374 if (list_empty(&area->free_list[migratetype]))
1375 continue;
1376
1377 page = list_entry(area->free_list[migratetype].next,
1378 struct page, lru);
1379 list_del(&page->lru);
1380 rmv_page_order(page);
1381 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001382 expand(zone, page, order, current_order, area, migratetype);
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001383 set_freepage_migratetype(page, migratetype);
Mel Gorman56fd56b2007-10-16 01:25:58 -07001384 return page;
1385 }
1386
1387 return NULL;
1388}
1389
1390
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001391/*
1392 * This array describes the order lists are fallen back to when
1393 * the free lists for the desirable migrate type are depleted
1394 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001395static int fallbacks[MIGRATE_TYPES][4] = {
1396 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
1397 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001398 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
Joonsoo Kimdc676472015-04-14 15:45:15 -07001399#ifdef CONFIG_CMA
1400 [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001401#endif
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001402 [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001403#ifdef CONFIG_MEMORY_ISOLATION
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001404 [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001405#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001406};
1407
Joonsoo Kimdc676472015-04-14 15:45:15 -07001408#ifdef CONFIG_CMA
1409static struct page *__rmqueue_cma_fallback(struct zone *zone,
1410 unsigned int order)
1411{
1412 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1413}
1414#else
1415static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1416 unsigned int order) { return NULL; }
1417#endif
1418
Mel Gormanc361be52007-10-16 01:25:51 -07001419/*
1420 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -07001421 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -07001422 * boundary. If alignment is required, use move_freepages_block()
1423 */
Minchan Kim435b4052012-10-08 16:32:16 -07001424int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -07001425 struct page *start_page, struct page *end_page,
1426 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001427{
1428 struct page *page;
1429 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -07001430 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -07001431
1432#ifndef CONFIG_HOLES_IN_ZONE
1433 /*
1434 * page_zone is not safe to call in this context when
1435 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1436 * anyway as we check zone boundaries in move_freepages_block().
1437 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -07001438 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -07001439 */
Mel Gorman97ee4ba2014-10-09 15:28:28 -07001440 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
Mel Gormanc361be52007-10-16 01:25:51 -07001441#endif
1442
1443 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -07001444 /* Make sure we are not inadvertently changing nodes */
Sasha Levin309381fea2014-01-23 15:52:54 -08001445 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
Adam Litke344c7902008-09-02 14:35:38 -07001446
Mel Gormanc361be52007-10-16 01:25:51 -07001447 if (!pfn_valid_within(page_to_pfn(page))) {
1448 page++;
1449 continue;
1450 }
1451
1452 if (!PageBuddy(page)) {
1453 page++;
1454 continue;
1455 }
1456
1457 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001458 list_move(&page->lru,
1459 &zone->free_area[order].free_list[migratetype]);
Minchan Kim95e34412012-10-08 16:32:11 -07001460 set_freepage_migratetype(page, migratetype);
Mel Gormanc361be52007-10-16 01:25:51 -07001461 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001462 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001463 }
1464
Mel Gormand1003132007-10-16 01:26:00 -07001465 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001466}
1467
Minchan Kimee6f5092012-07-31 16:43:50 -07001468int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001469 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001470{
1471 unsigned long start_pfn, end_pfn;
1472 struct page *start_page, *end_page;
1473
1474 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001475 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001476 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001477 end_page = start_page + pageblock_nr_pages - 1;
1478 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001479
1480 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001481 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001482 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001483 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001484 return 0;
1485
1486 return move_freepages(zone, start_page, end_page, migratetype);
1487}
1488
Mel Gorman2f66a682009-09-21 17:02:31 -07001489static void change_pageblock_range(struct page *pageblock_page,
1490 int start_order, int migratetype)
1491{
1492 int nr_pageblocks = 1 << (start_order - pageblock_order);
1493
1494 while (nr_pageblocks--) {
1495 set_pageblock_migratetype(pageblock_page, migratetype);
1496 pageblock_page += pageblock_nr_pages;
1497 }
1498}
1499
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001500/*
Vlastimil Babka9c0415e2015-02-11 15:28:21 -08001501 * When we are falling back to another migratetype during allocation, try to
1502 * steal extra free pages from the same pageblocks to satisfy further
1503 * allocations, instead of polluting multiple pageblocks.
1504 *
1505 * If we are stealing a relatively large buddy page, it is likely there will
1506 * be more free pages in the pageblock, so try to steal them all. For
1507 * reclaimable and unmovable allocations, we steal regardless of page size,
1508 * as fragmentation caused by those allocations polluting movable pageblocks
1509 * is worse than movable allocations stealing from unmovable and reclaimable
1510 * pageblocks.
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001511 */
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001512static bool can_steal_fallback(unsigned int order, int start_mt)
1513{
1514 /*
1515 * Leaving this order check is intended, although there is
1516 * relaxed order check in next check. The reason is that
1517 * we can actually steal whole pageblock if this condition met,
1518 * but, below check doesn't guarantee it and that is just heuristic
1519 * so could be changed anytime.
1520 */
1521 if (order >= pageblock_order)
1522 return true;
1523
1524 if (order >= pageblock_order / 2 ||
1525 start_mt == MIGRATE_RECLAIMABLE ||
1526 start_mt == MIGRATE_UNMOVABLE ||
1527 page_group_by_mobility_disabled)
1528 return true;
1529
1530 return false;
1531}
1532
1533/*
1534 * This function implements actual steal behaviour. If order is large enough,
1535 * we can steal whole pageblock. If not, we first move freepages in this
1536 * pageblock and check whether half of pages are moved or not. If half of
1537 * pages are moved, we can change migratetype of pageblock and permanently
1538 * use it's pages as requested migratetype in the future.
1539 */
1540static void steal_suitable_fallback(struct zone *zone, struct page *page,
1541 int start_type)
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001542{
1543 int current_order = page_order(page);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001544 int pages;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001545
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001546 /* Take ownership for orders >= pageblock_order */
1547 if (current_order >= pageblock_order) {
1548 change_pageblock_range(page, current_order, start_type);
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001549 return;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001550 }
1551
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001552 pages = move_freepages_block(zone, page, start_type);
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001553
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001554 /* Claim the whole block if over half of it is free */
1555 if (pages >= (1 << (pageblock_order-1)) ||
1556 page_group_by_mobility_disabled)
1557 set_pageblock_migratetype(page, start_type);
1558}
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001559
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001560/*
1561 * Check whether there is a suitable fallback freepage with requested order.
1562 * If only_stealable is true, this function returns fallback_mt only if
1563 * we can steal other freepages all together. This would help to reduce
1564 * fragmentation due to mixed migratetype pages in one pageblock.
1565 */
1566int find_suitable_fallback(struct free_area *area, unsigned int order,
1567 int migratetype, bool only_stealable, bool *can_steal)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001568{
1569 int i;
1570 int fallback_mt;
1571
1572 if (area->nr_free == 0)
1573 return -1;
1574
1575 *can_steal = false;
1576 for (i = 0;; i++) {
1577 fallback_mt = fallbacks[migratetype][i];
1578 if (fallback_mt == MIGRATE_RESERVE)
1579 break;
1580
1581 if (list_empty(&area->free_list[fallback_mt]))
1582 continue;
1583
1584 if (can_steal_fallback(order, migratetype))
1585 *can_steal = true;
1586
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001587 if (!only_stealable)
1588 return fallback_mt;
1589
1590 if (*can_steal)
1591 return fallback_mt;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001592 }
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001593
1594 return -1;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001595}
1596
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001597/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001598static inline struct page *
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001599__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001600{
Pintu Kumarb8af2942013-09-11 14:20:34 -07001601 struct free_area *area;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001602 unsigned int current_order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001603 struct page *page;
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001604 int fallback_mt;
1605 bool can_steal;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001606
1607 /* Find the largest possible block of pages in the other list */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001608 for (current_order = MAX_ORDER-1;
1609 current_order >= order && current_order <= MAX_ORDER-1;
1610 --current_order) {
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001611 area = &(zone->free_area[current_order]);
1612 fallback_mt = find_suitable_fallback(area, current_order,
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001613 start_migratetype, false, &can_steal);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001614 if (fallback_mt == -1)
1615 continue;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001616
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001617 page = list_entry(area->free_list[fallback_mt].next,
1618 struct page, lru);
1619 if (can_steal)
1620 steal_suitable_fallback(zone, page, start_migratetype);
Mel Gormane0104872007-10-16 01:25:53 -07001621
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001622 /* Remove the page from the freelists */
1623 area->nr_free--;
1624 list_del(&page->lru);
1625 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001626
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001627 expand(zone, page, order, current_order, area,
1628 start_migratetype);
1629 /*
1630 * The freepage_migratetype may differ from pageblock's
1631 * migratetype depending on the decisions in
1632 * try_to_steal_freepages(). This is OK as long as it
1633 * does not differ for MIGRATE_CMA pageblocks. For CMA
1634 * we need to make sure unallocated pages flushed from
1635 * pcp lists are returned to the correct freelist.
1636 */
1637 set_freepage_migratetype(page, start_migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001638
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001639 trace_mm_page_alloc_extfrag(page, order, current_order,
1640 start_migratetype, fallback_mt);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001641
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001642 return page;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001643 }
1644
Mel Gorman728ec982009-06-16 15:32:04 -07001645 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001646}
1647
Mel Gorman56fd56b2007-10-16 01:25:58 -07001648/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 * Do the hard work of removing an element from the buddy allocator.
1650 * Call me with the zone->lock already held.
1651 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001652static struct page *__rmqueue(struct zone *zone, unsigned int order,
1653 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 struct page *page;
1656
Mel Gorman728ec982009-06-16 15:32:04 -07001657retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001658 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659
Mel Gorman728ec982009-06-16 15:32:04 -07001660 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Joonsoo Kimdc676472015-04-14 15:45:15 -07001661 if (migratetype == MIGRATE_MOVABLE)
1662 page = __rmqueue_cma_fallback(zone, order);
1663
1664 if (!page)
1665 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001666
Mel Gorman728ec982009-06-16 15:32:04 -07001667 /*
1668 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1669 * is used because __rmqueue_smallest is an inline function
1670 * and we want just one call site
1671 */
1672 if (!page) {
1673 migratetype = MIGRATE_RESERVE;
1674 goto retry_reserve;
1675 }
1676 }
1677
Mel Gorman0d3d0622009-09-21 17:02:44 -07001678 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001679 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680}
1681
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001682/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 * Obtain a specified number of elements from the buddy allocator, all under
1684 * a single hold of the lock, for efficiency. Add them to the supplied list.
1685 * Returns the number of new pages which were placed at *list.
1686 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001687static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001688 unsigned long count, struct list_head *list,
Mel Gormanb745bc82014-06-04 16:10:22 -07001689 int migratetype, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690{
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001691 int i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001692
Nick Pigginc54ad302006-01-06 00:10:56 -08001693 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001695 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001696 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001698
1699 /*
1700 * Split buddy pages returned by expand() are received here
1701 * in physical page order. The page is added to the callers and
1702 * list and the list head then moves forward. From the callers
1703 * perspective, the linked list is ordered by page number in
1704 * some conditions. This is useful for IO devices that can
1705 * merge IO requests if the physical pages are ordered
1706 * properly.
1707 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001708 if (likely(!cold))
Mel Gormane084b2d2009-07-29 15:02:04 -07001709 list_add(&page->lru, list);
1710 else
1711 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001712 list = &page->lru;
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001713 if (is_migrate_cma(get_freepage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001714 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1715 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001717 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001718 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001719 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720}
1721
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001722#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001723/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001724 * Called from the vmstat counter updater to drain pagesets of this
1725 * currently executing processor on remote nodes after they have
1726 * expired.
1727 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001728 * Note that this function must be called with the thread pinned to
1729 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001730 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001731void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001732{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001733 unsigned long flags;
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001734 int to_drain, batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001735
Christoph Lameter4037d452007-05-09 02:35:14 -07001736 local_irq_save(flags);
Jason Low4db0c3c2015-04-15 16:14:08 -07001737 batch = READ_ONCE(pcp->batch);
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001738 to_drain = min(pcp->count, batch);
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07001739 if (to_drain > 0) {
1740 free_pcppages_bulk(zone, to_drain, pcp);
1741 pcp->count -= to_drain;
1742 }
Christoph Lameter4037d452007-05-09 02:35:14 -07001743 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001744}
1745#endif
1746
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001747/*
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001748 * Drain pcplists of the indicated processor and zone.
1749 *
1750 * The processor must either be the current processor and the
1751 * thread pinned to the current processor or a processor that
1752 * is not online.
1753 */
1754static void drain_pages_zone(unsigned int cpu, struct zone *zone)
1755{
1756 unsigned long flags;
1757 struct per_cpu_pageset *pset;
1758 struct per_cpu_pages *pcp;
1759
1760 local_irq_save(flags);
1761 pset = per_cpu_ptr(zone->pageset, cpu);
1762
1763 pcp = &pset->pcp;
1764 if (pcp->count) {
1765 free_pcppages_bulk(zone, pcp->count, pcp);
1766 pcp->count = 0;
1767 }
1768 local_irq_restore(flags);
1769}
1770
1771/*
1772 * Drain pcplists of all zones on the indicated processor.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001773 *
1774 * The processor must either be the current processor and the
1775 * thread pinned to the current processor or a processor that
1776 * is not online.
1777 */
1778static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779{
1780 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001782 for_each_populated_zone(zone) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001783 drain_pages_zone(cpu, zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 }
1785}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001787/*
1788 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001789 *
1790 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1791 * the single zone's pages.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001792 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001793void drain_local_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001794{
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001795 int cpu = smp_processor_id();
1796
1797 if (zone)
1798 drain_pages_zone(cpu, zone);
1799 else
1800 drain_pages(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001801}
1802
1803/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001804 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1805 *
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001806 * When zone parameter is non-NULL, spill just the single zone's pages.
1807 *
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001808 * Note that this code is protected against sending an IPI to an offline
1809 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1810 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1811 * nothing keeps CPUs from showing up after we populated the cpumask and
1812 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001813 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001814void drain_all_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001815{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001816 int cpu;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001817
1818 /*
1819 * Allocate in the BSS so we wont require allocation in
1820 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1821 */
1822 static cpumask_t cpus_with_pcps;
1823
1824 /*
1825 * We don't care about racing with CPU hotplug event
1826 * as offline notification will cause the notified
1827 * cpu to drain that CPU pcps and on_each_cpu_mask
1828 * disables preemption as part of its processing
1829 */
1830 for_each_online_cpu(cpu) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001831 struct per_cpu_pageset *pcp;
1832 struct zone *z;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001833 bool has_pcps = false;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001834
1835 if (zone) {
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001836 pcp = per_cpu_ptr(zone->pageset, cpu);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001837 if (pcp->pcp.count)
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001838 has_pcps = true;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001839 } else {
1840 for_each_populated_zone(z) {
1841 pcp = per_cpu_ptr(z->pageset, cpu);
1842 if (pcp->pcp.count) {
1843 has_pcps = true;
1844 break;
1845 }
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001846 }
1847 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001848
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001849 if (has_pcps)
1850 cpumask_set_cpu(cpu, &cpus_with_pcps);
1851 else
1852 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1853 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001854 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
1855 zone, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001856}
1857
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001858#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859
1860void mark_free_pages(struct zone *zone)
1861{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001862 unsigned long pfn, max_zone_pfn;
1863 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001864 unsigned int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 struct list_head *curr;
1866
Xishi Qiu8080fc02013-09-11 14:21:45 -07001867 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 return;
1869
1870 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001871
Cody P Schafer108bcc92013-02-22 16:35:23 -08001872 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001873 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1874 if (pfn_valid(pfn)) {
1875 struct page *page = pfn_to_page(pfn);
1876
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001877 if (!swsusp_page_is_forbidden(page))
1878 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001879 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001881 for_each_migratetype_order(order, t) {
1882 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001883 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001885 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1886 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001887 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001888 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001889 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 spin_unlock_irqrestore(&zone->lock, flags);
1891}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001892#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
1894/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 * Free a 0-order page
Mel Gormanb745bc82014-06-04 16:10:22 -07001896 * cold == true ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001898void free_hot_cold_page(struct page *page, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899{
1900 struct zone *zone = page_zone(page);
1901 struct per_cpu_pages *pcp;
1902 unsigned long flags;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001903 unsigned long pfn = page_to_pfn(page);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001904 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001906 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001907 return;
1908
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001909 migratetype = get_pfnblock_migratetype(page, pfn);
Minchan Kimb12c4ad2012-10-08 16:32:08 -07001910 set_freepage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001912 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001913
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001914 /*
1915 * We only track unmovable, reclaimable and movable on pcp lists.
1916 * Free ISOLATE pages back to the allocator because they are being
1917 * offlined but treat RESERVE as movable pages so we can get those
1918 * areas back if necessary. Otherwise, we may have to free
1919 * excessively into the page allocator
1920 */
1921 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08001922 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001923 free_one_page(zone, page, pfn, 0, migratetype);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001924 goto out;
1925 }
1926 migratetype = MIGRATE_MOVABLE;
1927 }
1928
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001929 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Mel Gormanb745bc82014-06-04 16:10:22 -07001930 if (!cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001931 list_add(&page->lru, &pcp->lists[migratetype]);
Mel Gormanb745bc82014-06-04 16:10:22 -07001932 else
1933 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001935 if (pcp->count >= pcp->high) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001936 unsigned long batch = READ_ONCE(pcp->batch);
Cody P Schafer998d39cb2013-07-03 15:01:32 -07001937 free_pcppages_bulk(zone, batch, pcp);
1938 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08001939 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001940
1941out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943}
1944
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001945/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001946 * Free a list of 0-order pages
1947 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001948void free_hot_cold_page_list(struct list_head *list, bool cold)
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001949{
1950 struct page *page, *next;
1951
1952 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001953 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001954 free_hot_cold_page(page, cold);
1955 }
1956}
1957
1958/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001959 * split_page takes a non-compound higher-order page, and splits it into
1960 * n (1<<order) sub-pages: page[0..n]
1961 * Each sub-page must be freed individually.
1962 *
1963 * Note: this is probably too low level an operation for use in drivers.
1964 * Please consult with lkml before using this in your driver.
1965 */
1966void split_page(struct page *page, unsigned int order)
1967{
1968 int i;
Joonsoo Kime2cfc912015-07-17 16:24:18 -07001969 gfp_t gfp_mask;
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001970
Sasha Levin309381fea2014-01-23 15:52:54 -08001971 VM_BUG_ON_PAGE(PageCompound(page), page);
1972 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001973
1974#ifdef CONFIG_KMEMCHECK
1975 /*
1976 * Split shadow pages too, because free(page[0]) would
1977 * otherwise free the whole shadow.
1978 */
1979 if (kmemcheck_page_is_tracked(page))
1980 split_page(virt_to_page(page[0].shadow), order);
1981#endif
1982
Joonsoo Kime2cfc912015-07-17 16:24:18 -07001983 gfp_mask = get_page_owner_gfp(page);
1984 set_page_owner(page, 0, gfp_mask);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001985 for (i = 1; i < (1 << order); i++) {
Nick Piggin7835e982006-03-22 00:08:40 -08001986 set_page_refcounted(page + i);
Joonsoo Kime2cfc912015-07-17 16:24:18 -07001987 set_page_owner(page + i, 0, gfp_mask);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001988 }
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001989}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07001990EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001991
Joonsoo Kim3c605092014-11-13 15:19:21 -08001992int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07001993{
Mel Gorman748446b2010-05-24 14:32:27 -07001994 unsigned long watermark;
1995 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07001996 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07001997
1998 BUG_ON(!PageBuddy(page));
1999
2000 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002001 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07002002
Minchan Kim194159f2013-02-22 16:33:58 -08002003 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002004 /* Obey watermarks as if the page was being allocated */
2005 watermark = low_wmark_pages(zone) + (1 << order);
2006 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2007 return 0;
2008
Mel Gorman8fb74b92013-01-11 14:32:16 -08002009 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002010 }
Mel Gorman748446b2010-05-24 14:32:27 -07002011
2012 /* Remove page from free list */
2013 list_del(&page->lru);
2014 zone->free_area[order].nr_free--;
2015 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002016
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002017 set_page_owner(page, order, __GFP_MOVABLE);
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002018
Mel Gorman8fb74b92013-01-11 14:32:16 -08002019 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07002020 if (order >= pageblock_order - 1) {
2021 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002022 for (; page < endpage; page += pageblock_nr_pages) {
2023 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08002024 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002025 set_pageblock_migratetype(page,
2026 MIGRATE_MOVABLE);
2027 }
Mel Gorman748446b2010-05-24 14:32:27 -07002028 }
2029
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002030
Mel Gorman8fb74b92013-01-11 14:32:16 -08002031 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002032}
2033
2034/*
2035 * Similar to split_page except the page is already free. As this is only
2036 * being used for migration, the migratetype of the block also changes.
2037 * As this is called with interrupts disabled, the caller is responsible
2038 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2039 * are enabled.
2040 *
2041 * Note: this is probably too low level an operation for use in drivers.
2042 * Please consult with lkml before using this in your driver.
2043 */
2044int split_free_page(struct page *page)
2045{
2046 unsigned int order;
2047 int nr_pages;
2048
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002049 order = page_order(page);
2050
Mel Gorman8fb74b92013-01-11 14:32:16 -08002051 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002052 if (!nr_pages)
2053 return 0;
2054
2055 /* Split into individual pages */
2056 set_page_refcounted(page);
2057 split_page(page, order);
2058 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07002059}
2060
2061/*
Vlastimil Babka75379192015-02-11 15:25:38 -08002062 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07002064static inline
2065struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002066 struct zone *zone, unsigned int order,
2067 gfp_t gfp_flags, int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068{
2069 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08002070 struct page *page;
Mel Gormanb745bc82014-06-04 16:10:22 -07002071 bool cold = ((gfp_flags & __GFP_COLD) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072
Nick Piggin48db57f2006-01-08 01:00:42 -08002073 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002075 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002078 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2079 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002080 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07002081 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002082 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07002083 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002084 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07002085 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07002086 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07002087
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002088 if (cold)
2089 page = list_entry(list->prev, struct page, lru);
2090 else
2091 page = list_entry(list->next, struct page, lru);
2092
Mel Gormanb92a6ed2007-10-16 01:25:50 -07002093 list_del(&page->lru);
2094 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002095 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07002096 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
2097 /*
2098 * __GFP_NOFAIL is not to be used in new code.
2099 *
2100 * All __GFP_NOFAIL callers should be fixed so that they
2101 * properly detect and handle allocation failures.
2102 *
2103 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07002104 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07002105 * __GFP_NOFAIL.
2106 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07002107 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07002108 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002110 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08002111 spin_unlock(&zone->lock);
2112 if (!page)
2113 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07002114 __mod_zone_freepage_state(zone, -(1 << order),
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07002115 get_freepage_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 }
2117
Johannes Weiner3a025762014-04-07 15:37:48 -07002118 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Johannes Weinerabe5f972014-10-02 16:21:10 -07002119 if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
Johannes Weiner57054652014-10-09 15:28:17 -07002120 !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
2121 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Johannes Weiner27329362014-03-03 15:38:41 -08002122
Christoph Lameterf8891e52006-06-30 01:55:45 -07002123 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07002124 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002125 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126
Sasha Levin309381fea2014-01-23 15:52:54 -08002127 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08002129
2130failed:
2131 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002132 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133}
2134
Akinobu Mita933e3122006-12-08 02:39:45 -08002135#ifdef CONFIG_FAIL_PAGE_ALLOC
2136
Akinobu Mitab2588c42011-07-26 16:09:03 -07002137static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08002138 struct fault_attr attr;
2139
2140 u32 ignore_gfp_highmem;
2141 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07002142 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08002143} fail_page_alloc = {
2144 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08002145 .ignore_gfp_wait = 1,
2146 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07002147 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08002148};
2149
2150static int __init setup_fail_page_alloc(char *str)
2151{
2152 return setup_fault_attr(&fail_page_alloc.attr, str);
2153}
2154__setup("fail_page_alloc=", setup_fail_page_alloc);
2155
Gavin Shandeaf3862012-07-31 16:41:51 -07002156static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002157{
Akinobu Mita54114992007-07-15 23:40:23 -07002158 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07002159 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002160 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07002161 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002162 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002163 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002164 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
Gavin Shandeaf3862012-07-31 16:41:51 -07002165 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002166
2167 return should_fail(&fail_page_alloc.attr, 1 << order);
2168}
2169
2170#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2171
2172static int __init fail_page_alloc_debugfs(void)
2173{
Al Virof4ae40a62011-07-24 04:33:43 -04002174 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08002175 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08002176
Akinobu Mitadd48c082011-08-03 16:21:01 -07002177 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2178 &fail_page_alloc.attr);
2179 if (IS_ERR(dir))
2180 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002181
Akinobu Mitab2588c42011-07-26 16:09:03 -07002182 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
2183 &fail_page_alloc.ignore_gfp_wait))
2184 goto fail;
2185 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2186 &fail_page_alloc.ignore_gfp_highmem))
2187 goto fail;
2188 if (!debugfs_create_u32("min-order", mode, dir,
2189 &fail_page_alloc.min_order))
2190 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08002191
Akinobu Mitab2588c42011-07-26 16:09:03 -07002192 return 0;
2193fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07002194 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002195
Akinobu Mitab2588c42011-07-26 16:09:03 -07002196 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08002197}
2198
2199late_initcall(fail_page_alloc_debugfs);
2200
2201#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2202
2203#else /* CONFIG_FAIL_PAGE_ALLOC */
2204
Gavin Shandeaf3862012-07-31 16:41:51 -07002205static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002206{
Gavin Shandeaf3862012-07-31 16:41:51 -07002207 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002208}
2209
2210#endif /* CONFIG_FAIL_PAGE_ALLOC */
2211
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08002213 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 * of the allocation.
2215 */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002216static bool __zone_watermark_ok(struct zone *z, unsigned int order,
2217 unsigned long mark, int classzone_idx, int alloc_flags,
2218 long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219{
Wei Yuan26086de2014-12-10 15:44:44 -08002220 /* free_pages may go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08002221 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 int o;
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002223 long free_cma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224
Michal Hockodf0a6da2012-01-10 15:08:02 -08002225 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002226 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002228 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 min -= min / 4;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002230#ifdef CONFIG_CMA
2231 /* If allocation can't use CMA areas don't use free CMA pages */
2232 if (!(alloc_flags & ALLOC_CMA))
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002233 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002234#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002235
Mel Gorman3484b2d2014-08-06 16:07:14 -07002236 if (free_pages - free_cma <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08002237 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 for (o = 0; o < order; o++) {
2239 /* At the next order, this order's pages become unavailable */
2240 free_pages -= z->free_area[o].nr_free << o;
2241
2242 /* Require fewer higher order pages to be free */
2243 min >>= 1;
2244
2245 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002246 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08002248 return true;
2249}
2250
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002251bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002252 int classzone_idx, int alloc_flags)
2253{
2254 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2255 zone_page_state(z, NR_FREE_PAGES));
2256}
2257
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002258bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
2259 unsigned long mark, int classzone_idx, int alloc_flags)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002260{
2261 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2262
2263 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2264 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2265
2266 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2267 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268}
2269
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002270#ifdef CONFIG_NUMA
2271/*
2272 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
2273 * skip over zones that are not allowed by the cpuset, or that have
2274 * been recently (in last second) found to be nearly full. See further
2275 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02002276 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002277 *
Zhi Yong Wua1aeb652013-11-12 15:08:29 -08002278 * If the zonelist cache is present in the passed zonelist, then
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002279 * returns a pointer to the allowed node mask (either the current
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08002280 * tasks mems_allowed, or node_states[N_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002281 *
2282 * If the zonelist cache is not available for this zonelist, does
2283 * nothing and returns NULL.
2284 *
2285 * If the fullzones BITMAP in the zonelist cache is stale (more than
2286 * a second since last zap'd) then we zap it out (clear its bits.)
2287 *
2288 * We hold off even calling zlc_setup, until after we've checked the
2289 * first zone in the zonelist, on the theory that most allocations will
2290 * be satisfied from that first zone, so best to examine that zone as
2291 * quickly as we can.
2292 */
2293static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
2294{
2295 struct zonelist_cache *zlc; /* cached zonelist speedup info */
2296 nodemask_t *allowednodes; /* zonelist_cache approximation */
2297
2298 zlc = zonelist->zlcache_ptr;
2299 if (!zlc)
2300 return NULL;
2301
S.Caglar Onurf05111f2008-04-28 02:12:38 -07002302 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002303 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2304 zlc->last_full_zap = jiffies;
2305 }
2306
2307 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
2308 &cpuset_current_mems_allowed :
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08002309 &node_states[N_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002310 return allowednodes;
2311}
2312
2313/*
2314 * Given 'z' scanning a zonelist, run a couple of quick checks to see
2315 * if it is worth looking at further for free memory:
2316 * 1) Check that the zone isn't thought to be full (doesn't have its
2317 * bit set in the zonelist_cache fullzones BITMAP).
2318 * 2) Check that the zones node (obtained from the zonelist_cache
2319 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
2320 * Return true (non-zero) if zone is worth looking at further, or
2321 * else return false (zero) if it is not.
2322 *
2323 * This check -ignores- the distinction between various watermarks,
2324 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
2325 * found to be full for any variation of these watermarks, it will
2326 * be considered full for up to one second by all requests, unless
2327 * we are so low on memory on all allowed nodes that we are forced
2328 * into the second scan of the zonelist.
2329 *
2330 * In the second scan we ignore this zonelist cache and exactly
2331 * apply the watermarks to all zones, even it is slower to do so.
2332 * We are low on memory in the second scan, and should leave no stone
2333 * unturned looking for a free page.
2334 */
Mel Gormandd1a2392008-04-28 02:12:17 -07002335static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002336 nodemask_t *allowednodes)
2337{
2338 struct zonelist_cache *zlc; /* cached zonelist speedup info */
2339 int i; /* index of *z in zonelist zones */
2340 int n; /* node that zone *z is on */
2341
2342 zlc = zonelist->zlcache_ptr;
2343 if (!zlc)
2344 return 1;
2345
Mel Gormandd1a2392008-04-28 02:12:17 -07002346 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002347 n = zlc->z_to_n[i];
2348
2349 /* This zone is worth trying if it is allowed but not full */
2350 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
2351}
2352
2353/*
2354 * Given 'z' scanning a zonelist, set the corresponding bit in
2355 * zlc->fullzones, so that subsequent attempts to allocate a page
2356 * from that zone don't waste time re-examining it.
2357 */
Mel Gormandd1a2392008-04-28 02:12:17 -07002358static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002359{
2360 struct zonelist_cache *zlc; /* cached zonelist speedup info */
2361 int i; /* index of *z in zonelist zones */
2362
2363 zlc = zonelist->zlcache_ptr;
2364 if (!zlc)
2365 return;
2366
Mel Gormandd1a2392008-04-28 02:12:17 -07002367 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002368
2369 set_bit(i, zlc->fullzones);
2370}
2371
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002372/*
2373 * clear all zones full, called after direct reclaim makes progress so that
2374 * a zone that was recently full is not skipped over for up to a second
2375 */
2376static void zlc_clear_zones_full(struct zonelist *zonelist)
2377{
2378 struct zonelist_cache *zlc; /* cached zonelist speedup info */
2379
2380 zlc = zonelist->zlcache_ptr;
2381 if (!zlc)
2382 return;
2383
2384 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2385}
2386
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002387static bool zone_local(struct zone *local_zone, struct zone *zone)
2388{
Johannes Weinerfff4068c2013-12-20 14:54:12 +00002389 return local_zone->node == zone->node;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002390}
2391
David Rientjes957f8222012-10-08 16:33:24 -07002392static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2393{
Mel Gorman5f7a75a2014-06-04 16:07:15 -07002394 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2395 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07002396}
2397
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002398#else /* CONFIG_NUMA */
2399
2400static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
2401{
2402 return NULL;
2403}
2404
Mel Gormandd1a2392008-04-28 02:12:17 -07002405static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002406 nodemask_t *allowednodes)
2407{
2408 return 1;
2409}
2410
Mel Gormandd1a2392008-04-28 02:12:17 -07002411static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002412{
2413}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002414
2415static void zlc_clear_zones_full(struct zonelist *zonelist)
2416{
2417}
David Rientjes957f8222012-10-08 16:33:24 -07002418
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002419static bool zone_local(struct zone *local_zone, struct zone *zone)
2420{
2421 return true;
2422}
2423
David Rientjes957f8222012-10-08 16:33:24 -07002424static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2425{
2426 return true;
2427}
2428
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002429#endif /* CONFIG_NUMA */
2430
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002431static void reset_alloc_batches(struct zone *preferred_zone)
2432{
2433 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2434
2435 do {
2436 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2437 high_wmark_pages(zone) - low_wmark_pages(zone) -
2438 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner57054652014-10-09 15:28:17 -07002439 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002440 } while (zone++ != preferred_zone);
2441}
2442
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002443/*
Paul Jackson0798e512006-12-06 20:31:38 -08002444 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002445 * a page.
2446 */
2447static struct page *
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002448get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2449 const struct alloc_context *ac)
Martin Hicks753ee722005-06-21 17:14:41 -07002450{
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002451 struct zonelist *zonelist = ac->zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07002452 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002453 struct page *page = NULL;
Mel Gorman5117f452009-06-16 15:31:59 -07002454 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002455 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
2456 int zlc_active = 0; /* set if using zonelist_cache */
2457 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gormana6e21b142014-06-04 16:10:12 -07002458 bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) &&
2459 (gfp_mask & __GFP_WRITE);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002460 int nr_fair_skipped = 0;
2461 bool zonelist_rescan;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002462
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002463zonelist_scan:
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002464 zonelist_rescan = false;
2465
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002466 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002467 * Scan zonelist, looking for a zone with enough free.
Vladimir Davydov344736f2014-10-20 15:50:30 +04002468 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002469 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002470 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2471 ac->nodemask) {
Johannes Weinere085dbc2013-09-11 14:20:46 -07002472 unsigned long mark;
2473
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002474 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002475 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2476 continue;
Mel Gorman664eedd2014-06-04 16:10:08 -07002477 if (cpusets_enabled() &&
2478 (alloc_flags & ALLOC_CPUSET) &&
Vladimir Davydov344736f2014-10-20 15:50:30 +04002479 !cpuset_zone_allowed(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07002480 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08002481 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002482 * Distribute pages in proportion to the individual
2483 * zone size to ensure fair page aging. The zone a
2484 * page was allocated in should have no effect on the
2485 * time the page has in memory before being reclaimed.
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002486 */
Johannes Weiner3a025762014-04-07 15:37:48 -07002487 if (alloc_flags & ALLOC_FAIR) {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002488 if (!zone_local(ac->preferred_zone, zone))
Mel Gormanf7b5d642014-08-06 16:07:20 -07002489 break;
Johannes Weiner57054652014-10-09 15:28:17 -07002490 if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002491 nr_fair_skipped++;
Johannes Weiner3a025762014-04-07 15:37:48 -07002492 continue;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002493 }
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002494 }
2495 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08002496 * When allocating a page cache page for writing, we
2497 * want to get it from a zone that is within its dirty
2498 * limit, such that no single zone holds more than its
2499 * proportional share of globally allowed dirty pages.
2500 * The dirty limits take into account the zone's
2501 * lowmem reserves and high watermark so that kswapd
2502 * should be able to balance it without having to
2503 * write pages from its LRU list.
2504 *
2505 * This may look like it could increase pressure on
2506 * lower zones by failing allocations in higher zones
2507 * before they are full. But the pages that do spill
2508 * over are limited as the lower zones are protected
2509 * by this very same mechanism. It should not become
2510 * a practical burden to them.
2511 *
2512 * XXX: For now, allow allocations to potentially
2513 * exceed the per-zone dirty limit in the slowpath
2514 * (ALLOC_WMARK_LOW unset) before going into reclaim,
2515 * which is important when on a NUMA setup the allowed
2516 * zones are together not big enough to reach the
2517 * global limit. The proper fix for these situations
2518 * will require awareness of zones in the
2519 * dirty-throttling and the flusher threads.
2520 */
Mel Gormana6e21b142014-06-04 16:10:12 -07002521 if (consider_zone_dirty && !zone_dirty_ok(zone))
Mel Gorman800a1e72014-06-04 16:10:06 -07002522 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002523
Johannes Weinere085dbc2013-09-11 14:20:46 -07002524 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2525 if (!zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002526 ac->classzone_idx, alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07002527 int ret;
2528
Mel Gorman5dab2912014-06-04 16:10:14 -07002529 /* Checked here to keep the fast path fast */
2530 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2531 if (alloc_flags & ALLOC_NO_WATERMARKS)
2532 goto try_this_zone;
2533
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002534 if (IS_ENABLED(CONFIG_NUMA) &&
2535 !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormancd38b112011-07-25 17:12:29 -07002536 /*
2537 * we do zlc_setup if there are multiple nodes
2538 * and before considering the first zone allowed
2539 * by the cpuset.
2540 */
2541 allowednodes = zlc_setup(zonelist, alloc_flags);
2542 zlc_active = 1;
2543 did_zlc_setup = 1;
2544 }
2545
David Rientjes957f8222012-10-08 16:33:24 -07002546 if (zone_reclaim_mode == 0 ||
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002547 !zone_allows_reclaim(ac->preferred_zone, zone))
Mel Gormanfa5e0842009-06-16 15:33:22 -07002548 goto this_zone_full;
2549
Mel Gormancd38b112011-07-25 17:12:29 -07002550 /*
2551 * As we may have just activated ZLC, check if the first
2552 * eligible zone has failed zone_reclaim recently.
2553 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002554 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Mel Gormancd38b112011-07-25 17:12:29 -07002555 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2556 continue;
2557
Mel Gormanfa5e0842009-06-16 15:33:22 -07002558 ret = zone_reclaim(zone, gfp_mask, order);
2559 switch (ret) {
2560 case ZONE_RECLAIM_NOSCAN:
2561 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07002562 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002563 case ZONE_RECLAIM_FULL:
2564 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002565 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002566 default:
2567 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002568 if (zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002569 ac->classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002570 goto try_this_zone;
2571
2572 /*
2573 * Failed to reclaim enough to meet watermark.
2574 * Only mark the zone full if checking the min
2575 * watermark or if we failed to reclaim just
2576 * 1<<order pages or else the page allocator
2577 * fastpath will prematurely mark zones full
2578 * when the watermark is between the low and
2579 * min watermarks.
2580 */
2581 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2582 ret == ZONE_RECLAIM_SOME)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002583 goto this_zone_full;
Mel Gormanfed27192013-04-29 15:07:57 -07002584
2585 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002586 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002587 }
2588
Mel Gormanfa5e0842009-06-16 15:33:22 -07002589try_this_zone:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002590 page = buffered_rmqueue(ac->preferred_zone, zone, order,
2591 gfp_mask, ac->migratetype);
Vlastimil Babka75379192015-02-11 15:25:38 -08002592 if (page) {
2593 if (prep_new_page(page, order, gfp_mask, alloc_flags))
2594 goto try_this_zone;
2595 return page;
2596 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002597this_zone_full:
Mel Gorman65bb3712014-06-04 16:10:05 -07002598 if (IS_ENABLED(CONFIG_NUMA) && zlc_active)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002599 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002600 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002601
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002602 /*
2603 * The first pass makes sure allocations are spread fairly within the
2604 * local node. However, the local node might have free pages left
2605 * after the fairness batches are exhausted, and remote zones haven't
2606 * even been considered yet. Try once more without fairness, and
2607 * include remote zones now, before entering the slowpath and waking
2608 * kswapd: prefer spilling to a remote zone over swapping locally.
2609 */
2610 if (alloc_flags & ALLOC_FAIR) {
2611 alloc_flags &= ~ALLOC_FAIR;
2612 if (nr_fair_skipped) {
2613 zonelist_rescan = true;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002614 reset_alloc_batches(ac->preferred_zone);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002615 }
2616 if (nr_online_nodes > 1)
2617 zonelist_rescan = true;
2618 }
2619
2620 if (unlikely(IS_ENABLED(CONFIG_NUMA) && zlc_active)) {
2621 /* Disable zlc cache for second zonelist scan */
2622 zlc_active = 0;
2623 zonelist_rescan = true;
2624 }
2625
2626 if (zonelist_rescan)
2627 goto zonelist_scan;
2628
2629 return NULL;
Martin Hicks753ee722005-06-21 17:14:41 -07002630}
2631
David Rientjes29423e772011-03-22 16:30:47 -07002632/*
2633 * Large machines with many possible nodes should not always dump per-node
2634 * meminfo in irq context.
2635 */
2636static inline bool should_suppress_show_mem(void)
2637{
2638 bool ret = false;
2639
2640#if NODES_SHIFT > 8
2641 ret = in_interrupt();
2642#endif
2643 return ret;
2644}
2645
Dave Hansena238ab52011-05-24 17:12:16 -07002646static DEFINE_RATELIMIT_STATE(nopage_rs,
2647 DEFAULT_RATELIMIT_INTERVAL,
2648 DEFAULT_RATELIMIT_BURST);
2649
2650void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2651{
Dave Hansena238ab52011-05-24 17:12:16 -07002652 unsigned int filter = SHOW_MEM_FILTER_NODES;
2653
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08002654 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2655 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002656 return;
2657
2658 /*
2659 * This documents exceptions given to allocations in certain
2660 * contexts that are allowed to allocate outside current's set
2661 * of allowed nodes.
2662 */
2663 if (!(gfp_mask & __GFP_NOMEMALLOC))
2664 if (test_thread_flag(TIF_MEMDIE) ||
2665 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2666 filter &= ~SHOW_MEM_FILTER_NODES;
2667 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2668 filter &= ~SHOW_MEM_FILTER_NODES;
2669
2670 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002671 struct va_format vaf;
2672 va_list args;
2673
Dave Hansena238ab52011-05-24 17:12:16 -07002674 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002675
2676 vaf.fmt = fmt;
2677 vaf.va = &args;
2678
2679 pr_warn("%pV", &vaf);
2680
Dave Hansena238ab52011-05-24 17:12:16 -07002681 va_end(args);
2682 }
2683
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002684 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2685 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002686
2687 dump_stack();
2688 if (!should_suppress_show_mem())
2689 show_mem(filter);
2690}
2691
Mel Gorman11e33f62009-06-16 15:31:57 -07002692static inline struct page *
2693__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002694 const struct alloc_context *ac, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07002695{
David Rientjes6e0fc462015-09-08 15:00:36 -07002696 struct oom_control oc = {
2697 .zonelist = ac->zonelist,
2698 .nodemask = ac->nodemask,
2699 .gfp_mask = gfp_mask,
2700 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07002701 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703
Johannes Weiner9879de72015-01-26 12:58:32 -08002704 *did_some_progress = 0;
2705
Johannes Weiner9879de72015-01-26 12:58:32 -08002706 /*
Johannes Weinerdc564012015-06-24 16:57:19 -07002707 * Acquire the oom lock. If that fails, somebody else is
2708 * making progress for us.
Johannes Weiner9879de72015-01-26 12:58:32 -08002709 */
Johannes Weinerdc564012015-06-24 16:57:19 -07002710 if (!mutex_trylock(&oom_lock)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002711 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002712 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 return NULL;
2714 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002715
Mel Gorman11e33f62009-06-16 15:31:57 -07002716 /*
2717 * Go through the zonelist yet one more time, keep very high watermark
2718 * here, this is only to catch a parallel oom killing, we must fail if
2719 * we're still under heavy pressure.
2720 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002721 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2722 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002723 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002724 goto out;
2725
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002726 if (!(gfp_mask & __GFP_NOFAIL)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002727 /* Coredumps can quickly deplete all memory reserves */
2728 if (current->flags & PF_DUMPCORE)
2729 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002730 /* The OOM killer will not help higher order allocs */
2731 if (order > PAGE_ALLOC_COSTLY_ORDER)
2732 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002733 /* The OOM killer does not needlessly kill tasks for lowmem */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002734 if (ac->high_zoneidx < ZONE_NORMAL)
David Rientjes03668b32010-08-09 17:18:54 -07002735 goto out;
Johannes Weiner90839052015-06-24 16:57:21 -07002736 /* The OOM killer does not compensate for IO-less reclaim */
Johannes Weinercc873172015-02-27 15:52:09 -08002737 if (!(gfp_mask & __GFP_FS)) {
2738 /*
2739 * XXX: Page reclaim didn't yield anything,
2740 * and the OOM killer can't be invoked, but
Johannes Weiner90839052015-06-24 16:57:21 -07002741 * keep looping as per tradition.
Johannes Weinercc873172015-02-27 15:52:09 -08002742 */
2743 *did_some_progress = 1;
Johannes Weiner9879de72015-01-26 12:58:32 -08002744 goto out;
Johannes Weinercc873172015-02-27 15:52:09 -08002745 }
Johannes Weiner90839052015-06-24 16:57:21 -07002746 if (pm_suspended_storage())
2747 goto out;
David Rientjes4167e9b2015-04-14 15:46:55 -07002748 /* The OOM killer may not free memory on a specific node */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002749 if (gfp_mask & __GFP_THISNODE)
2750 goto out;
2751 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002752 /* Exhausted what can be done so it's blamo time */
David Rientjes6e0fc462015-09-08 15:00:36 -07002753 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL))
Michal Hockoc32b3cb2015-02-11 15:26:24 -08002754 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002755out:
Johannes Weinerdc564012015-06-24 16:57:19 -07002756 mutex_unlock(&oom_lock);
Mel Gorman11e33f62009-06-16 15:31:57 -07002757 return page;
2758}
2759
Mel Gorman56de7262010-05-24 14:32:30 -07002760#ifdef CONFIG_COMPACTION
2761/* Try memory compaction for high-order allocations before reclaim */
2762static struct page *
2763__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002764 int alloc_flags, const struct alloc_context *ac,
2765 enum migrate_mode mode, int *contended_compaction,
2766 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002767{
Vlastimil Babka53853e22014-10-09 15:27:02 -07002768 unsigned long compact_result;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002769 struct page *page;
Vlastimil Babka53853e22014-10-09 15:27:02 -07002770
Mel Gorman66199712012-01-12 17:19:41 -08002771 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002772 return NULL;
2773
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002774 current->flags |= PF_MEMALLOC;
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08002775 compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2776 mode, contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002777 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002778
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002779 switch (compact_result) {
2780 case COMPACT_DEFERRED:
Vlastimil Babka53853e22014-10-09 15:27:02 -07002781 *deferred_compaction = true;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002782 /* fall-through */
2783 case COMPACT_SKIPPED:
2784 return NULL;
2785 default:
2786 break;
Mel Gorman56de7262010-05-24 14:32:30 -07002787 }
2788
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002789 /*
2790 * At least in one zone compaction wasn't deferred or skipped, so let's
2791 * count a compaction stall
2792 */
2793 count_vm_event(COMPACTSTALL);
2794
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002795 page = get_page_from_freelist(gfp_mask, order,
2796 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002797
2798 if (page) {
2799 struct zone *zone = page_zone(page);
2800
2801 zone->compact_blockskip_flush = false;
2802 compaction_defer_reset(zone, order, true);
2803 count_vm_event(COMPACTSUCCESS);
2804 return page;
2805 }
2806
2807 /*
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002808 * It's bad if compaction run occurs and fails. The most likely reason
2809 * is that pages exist, but not enough to satisfy watermarks.
2810 */
2811 count_vm_event(COMPACTFAIL);
2812
2813 cond_resched();
2814
Mel Gorman56de7262010-05-24 14:32:30 -07002815 return NULL;
2816}
2817#else
2818static inline struct page *
2819__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002820 int alloc_flags, const struct alloc_context *ac,
2821 enum migrate_mode mode, int *contended_compaction,
2822 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002823{
2824 return NULL;
2825}
2826#endif /* CONFIG_COMPACTION */
2827
Marek Szyprowskibba90712012-01-25 12:09:52 +01002828/* Perform direct synchronous page reclaim */
2829static int
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002830__perform_reclaim(gfp_t gfp_mask, unsigned int order,
2831 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002832{
Mel Gorman11e33f62009-06-16 15:31:57 -07002833 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002834 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002835
2836 cond_resched();
2837
2838 /* We now go into synchronous reclaim */
2839 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002840 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002841 lockdep_set_current_reclaim_state(gfp_mask);
2842 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002843 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002844
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002845 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
2846 ac->nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002847
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002848 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002849 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002850 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002851
2852 cond_resched();
2853
Marek Szyprowskibba90712012-01-25 12:09:52 +01002854 return progress;
2855}
2856
2857/* The really slow allocator path where we enter direct reclaim */
2858static inline struct page *
2859__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002860 int alloc_flags, const struct alloc_context *ac,
2861 unsigned long *did_some_progress)
Marek Szyprowskibba90712012-01-25 12:09:52 +01002862{
2863 struct page *page = NULL;
2864 bool drained = false;
2865
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002866 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002867 if (unlikely(!(*did_some_progress)))
2868 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002869
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002870 /* After successful reclaim, reconsider all zones for allocation */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002871 if (IS_ENABLED(CONFIG_NUMA))
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002872 zlc_clear_zones_full(ac->zonelist);
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002873
Mel Gorman9ee493c2010-09-09 16:38:18 -07002874retry:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002875 page = get_page_from_freelist(gfp_mask, order,
2876 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002877
2878 /*
2879 * If an allocation failed after direct reclaim, it could be because
2880 * pages are pinned on the per-cpu lists. Drain them and try again
2881 */
2882 if (!page && !drained) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002883 drain_all_pages(NULL);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002884 drained = true;
2885 goto retry;
2886 }
2887
Mel Gorman11e33f62009-06-16 15:31:57 -07002888 return page;
2889}
2890
Mel Gorman11e33f62009-06-16 15:31:57 -07002891/*
2892 * This is called in the allocator slow-path if the allocation request is of
2893 * sufficient urgency to ignore watermarks and take other desperate measures
2894 */
2895static inline struct page *
2896__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002897 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002898{
2899 struct page *page;
2900
2901 do {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002902 page = get_page_from_freelist(gfp_mask, order,
2903 ALLOC_NO_WATERMARKS, ac);
Mel Gorman11e33f62009-06-16 15:31:57 -07002904
2905 if (!page && gfp_mask & __GFP_NOFAIL)
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002906 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC,
2907 HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002908 } while (!page && (gfp_mask & __GFP_NOFAIL));
2909
2910 return page;
2911}
2912
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002913static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002914{
2915 struct zoneref *z;
2916 struct zone *zone;
2917
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002918 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
2919 ac->high_zoneidx, ac->nodemask)
2920 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
Mel Gorman11e33f62009-06-16 15:31:57 -07002921}
2922
Peter Zijlstra341ce062009-06-16 15:32:02 -07002923static inline int
2924gfp_to_alloc_flags(gfp_t gfp_mask)
2925{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002926 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
David Rientjesb104a352014-07-30 16:08:24 -07002927 const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD));
Peter Zijlstra341ce062009-06-16 15:32:02 -07002928
Mel Gormana56f57f2009-06-16 15:32:02 -07002929 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002930 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002931
Peter Zijlstra341ce062009-06-16 15:32:02 -07002932 /*
2933 * The caller may dip into page reserves a bit more if the caller
2934 * cannot run direct reclaim, or if the caller has realtime scheduling
2935 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
David Rientjesb104a352014-07-30 16:08:24 -07002936 * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
Peter Zijlstra341ce062009-06-16 15:32:02 -07002937 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002938 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002939
David Rientjesb104a352014-07-30 16:08:24 -07002940 if (atomic) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002941 /*
David Rientjesb104a352014-07-30 16:08:24 -07002942 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2943 * if it can't schedule.
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002944 */
David Rientjesb104a352014-07-30 16:08:24 -07002945 if (!(gfp_mask & __GFP_NOMEMALLOC))
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002946 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002947 /*
David Rientjesb104a352014-07-30 16:08:24 -07002948 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
Vladimir Davydov344736f2014-10-20 15:50:30 +04002949 * comment for __cpuset_node_allowed().
Peter Zijlstra341ce062009-06-16 15:32:02 -07002950 */
2951 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002952 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002953 alloc_flags |= ALLOC_HARDER;
2954
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002955 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2956 if (gfp_mask & __GFP_MEMALLOC)
2957 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07002958 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2959 alloc_flags |= ALLOC_NO_WATERMARKS;
2960 else if (!in_interrupt() &&
2961 ((current->flags & PF_MEMALLOC) ||
2962 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07002963 alloc_flags |= ALLOC_NO_WATERMARKS;
2964 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002965#ifdef CONFIG_CMA
David Rientjes43e7a342014-10-09 15:27:25 -07002966 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002967 alloc_flags |= ALLOC_CMA;
2968#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002969 return alloc_flags;
2970}
2971
Mel Gorman072bb0a2012-07-31 16:43:58 -07002972bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2973{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002974 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002975}
2976
Mel Gorman11e33f62009-06-16 15:31:57 -07002977static inline struct page *
2978__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002979 struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002980{
2981 const gfp_t wait = gfp_mask & __GFP_WAIT;
2982 struct page *page = NULL;
2983 int alloc_flags;
2984 unsigned long pages_reclaimed = 0;
2985 unsigned long did_some_progress;
David Rientjese0b9dae2014-06-04 16:08:28 -07002986 enum migrate_mode migration_mode = MIGRATE_ASYNC;
Mel Gorman66199712012-01-12 17:19:41 -08002987 bool deferred_compaction = false;
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07002988 int contended_compaction = COMPACT_CONTENDED_NONE;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002989
Christoph Lameter952f3b52006-12-06 20:33:26 -08002990 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002991 * In the slowpath, we sanity check order to avoid ever trying to
2992 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2993 * be using allocators in order of preference for an area that is
2994 * too large.
2995 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002996 if (order >= MAX_ORDER) {
2997 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002998 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002999 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003000
Christoph Lameter952f3b52006-12-06 20:33:26 -08003001 /*
David Rientjes4167e9b2015-04-14 15:46:55 -07003002 * If this allocation cannot block and it is for a specific node, then
3003 * fail early. There's no need to wakeup kswapd or retry for a
3004 * speculative node-specific allocation.
Christoph Lameter952f3b52006-12-06 20:33:26 -08003005 */
David Rientjes4167e9b2015-04-14 15:46:55 -07003006 if (IS_ENABLED(CONFIG_NUMA) && (gfp_mask & __GFP_THISNODE) && !wait)
Christoph Lameter952f3b52006-12-06 20:33:26 -08003007 goto nopage;
3008
Johannes Weiner9879de72015-01-26 12:58:32 -08003009retry:
Johannes Weiner3a025762014-04-07 15:37:48 -07003010 if (!(gfp_mask & __GFP_NO_KSWAPD))
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003011 wake_all_kswapds(order, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003012
Paul Jackson9bf22292005-09-06 15:18:12 -07003013 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003014 * OK, we're below the kswapd watermark and have kicked background
3015 * reclaim. Now things get more complex, so set up alloc_flags according
3016 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07003017 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07003018 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019
David Rientjesf33261d2011-01-25 15:07:20 -08003020 /*
3021 * Find the true preferred zone if the allocation is unconstrained by
3022 * cpusets.
3023 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003024 if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
Mel Gormand8846372014-06-04 16:10:33 -07003025 struct zoneref *preferred_zoneref;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003026 preferred_zoneref = first_zones_zonelist(ac->zonelist,
3027 ac->high_zoneidx, NULL, &ac->preferred_zone);
3028 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gormand8846372014-06-04 16:10:33 -07003029 }
David Rientjesf33261d2011-01-25 15:07:20 -08003030
Peter Zijlstra341ce062009-06-16 15:32:02 -07003031 /* This is the last chance, in general, before the goto nopage. */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003032 page = get_page_from_freelist(gfp_mask, order,
3033 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003034 if (page)
3035 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036
Mel Gorman11e33f62009-06-16 15:31:57 -07003037 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07003038 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07003039 /*
3040 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3041 * the allocation is high priority and these type of
3042 * allocations are system rather than user orientated
3043 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003044 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
Mel Gorman183f6372012-07-31 16:44:12 -07003045
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003046 page = __alloc_pages_high_priority(gfp_mask, order, ac);
3047
Mel Gormancfd19c52012-07-31 16:44:10 -07003048 if (page) {
Peter Zijlstra341ce062009-06-16 15:32:02 -07003049 goto got_pg;
Mel Gormancfd19c52012-07-31 16:44:10 -07003050 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051 }
3052
3053 /* Atomic allocations - we can't balance anything */
David Rientjesaed0a0e2014-01-21 15:51:12 -08003054 if (!wait) {
3055 /*
3056 * All existing users of the deprecated __GFP_NOFAIL are
3057 * blockable, so warn of any new users that actually allow this
3058 * type of allocation to fail.
3059 */
3060 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 goto nopage;
David Rientjesaed0a0e2014-01-21 15:51:12 -08003062 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063
Peter Zijlstra341ce062009-06-16 15:32:02 -07003064 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003065 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07003066 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067
David Rientjes6583bb62009-07-29 15:02:06 -07003068 /* Avoid allocations with no watermarks from looping endlessly */
3069 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3070 goto nopage;
3071
Mel Gorman77f1fe62011-01-13 15:45:57 -08003072 /*
3073 * Try direct compaction. The first pass is asynchronous. Subsequent
3074 * attempts after direct reclaim are synchronous
3075 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003076 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3077 migration_mode,
3078 &contended_compaction,
Vlastimil Babka53853e22014-10-09 15:27:02 -07003079 &deferred_compaction);
Mel Gorman56de7262010-05-24 14:32:30 -07003080 if (page)
3081 goto got_pg;
David Rientjes75f30862014-06-04 16:08:30 -07003082
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003083 /* Checks for THP-specific high-order allocations */
3084 if ((gfp_mask & GFP_TRANSHUGE) == GFP_TRANSHUGE) {
3085 /*
3086 * If compaction is deferred for high-order allocations, it is
3087 * because sync compaction recently failed. If this is the case
3088 * and the caller requested a THP allocation, we do not want
3089 * to heavily disrupt the system, so we fail the allocation
3090 * instead of entering direct reclaim.
3091 */
3092 if (deferred_compaction)
3093 goto nopage;
3094
3095 /*
3096 * In all zones where compaction was attempted (and not
3097 * deferred or skipped), lock contention has been detected.
3098 * For THP allocation we do not want to disrupt the others
3099 * so we fallback to base pages instead.
3100 */
3101 if (contended_compaction == COMPACT_CONTENDED_LOCK)
3102 goto nopage;
3103
3104 /*
3105 * If compaction was aborted due to need_resched(), we do not
3106 * want to further increase allocation latency, unless it is
3107 * khugepaged trying to collapse.
3108 */
3109 if (contended_compaction == COMPACT_CONTENDED_SCHED
3110 && !(current->flags & PF_KTHREAD))
3111 goto nopage;
3112 }
Mel Gorman66199712012-01-12 17:19:41 -08003113
David Rientjes8fe78042014-08-06 16:07:54 -07003114 /*
3115 * It can become very expensive to allocate transparent hugepages at
3116 * fault, so use asynchronous memory compaction for THP unless it is
3117 * khugepaged trying to collapse.
3118 */
3119 if ((gfp_mask & GFP_TRANSHUGE) != GFP_TRANSHUGE ||
3120 (current->flags & PF_KTHREAD))
3121 migration_mode = MIGRATE_SYNC_LIGHT;
3122
Mel Gorman11e33f62009-06-16 15:31:57 -07003123 /* Try direct reclaim and then allocating */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003124 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3125 &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07003126 if (page)
3127 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128
Johannes Weiner90839052015-06-24 16:57:21 -07003129 /* Do not loop if specifically requested */
3130 if (gfp_mask & __GFP_NORETRY)
3131 goto noretry;
3132
3133 /* Keep reclaiming pages as long as there is reasonable progress */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07003134 pages_reclaimed += did_some_progress;
Johannes Weiner90839052015-06-24 16:57:21 -07003135 if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) ||
3136 ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) {
Mel Gorman11e33f62009-06-16 15:31:57 -07003137 /* Wait for some write requests to complete then retry */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003138 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
Johannes Weiner9879de72015-01-26 12:58:32 -08003139 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 }
3141
Johannes Weiner90839052015-06-24 16:57:21 -07003142 /* Reclaim has failed us, start killing things */
3143 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3144 if (page)
3145 goto got_pg;
3146
3147 /* Retry as long as the OOM killer is making progress */
3148 if (did_some_progress)
3149 goto retry;
3150
3151noretry:
3152 /*
3153 * High-order allocations do not necessarily loop after
3154 * direct reclaim and reclaim/compaction depends on compaction
3155 * being called after reclaim so call directly if necessary
3156 */
3157 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags,
3158 ac, migration_mode,
3159 &contended_compaction,
3160 &deferred_compaction);
3161 if (page)
3162 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07003164 warn_alloc_failed(gfp_mask, order, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165got_pg:
Mel Gorman072bb0a2012-07-31 16:43:58 -07003166 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167}
Mel Gorman11e33f62009-06-16 15:31:57 -07003168
3169/*
3170 * This is the 'heart' of the zoned buddy allocator.
3171 */
3172struct page *
3173__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3174 struct zonelist *zonelist, nodemask_t *nodemask)
3175{
Mel Gormand8846372014-06-04 16:10:33 -07003176 struct zoneref *preferred_zoneref;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003177 struct page *page = NULL;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003178 unsigned int cpuset_mems_cookie;
Johannes Weiner3a025762014-04-07 15:37:48 -07003179 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
Andrew Morton91fbdc02015-02-11 15:25:04 -08003180 gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003181 struct alloc_context ac = {
3182 .high_zoneidx = gfp_zone(gfp_mask),
3183 .nodemask = nodemask,
3184 .migratetype = gfpflags_to_migratetype(gfp_mask),
3185 };
Mel Gorman11e33f62009-06-16 15:31:57 -07003186
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003187 gfp_mask &= gfp_allowed_mask;
3188
Mel Gorman11e33f62009-06-16 15:31:57 -07003189 lockdep_trace_alloc(gfp_mask);
3190
3191 might_sleep_if(gfp_mask & __GFP_WAIT);
3192
3193 if (should_fail_alloc_page(gfp_mask, order))
3194 return NULL;
3195
3196 /*
3197 * Check the zones suitable for the gfp_mask contain at least one
3198 * valid zone. It's possible to have an empty zonelist as a result
David Rientjes4167e9b2015-04-14 15:46:55 -07003199 * of __GFP_THISNODE and a memoryless node
Mel Gorman11e33f62009-06-16 15:31:57 -07003200 */
3201 if (unlikely(!zonelist->_zonerefs->zone))
3202 return NULL;
3203
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003204 if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
Vlastimil Babka21bb9bd2014-10-09 15:26:51 -07003205 alloc_flags |= ALLOC_CMA;
3206
Mel Gormancc9a6c82012-03-21 16:34:11 -07003207retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07003208 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003209
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003210 /* We set it here, as __alloc_pages_slowpath might have changed it */
3211 ac.zonelist = zonelist;
Mel Gorman5117f452009-06-16 15:31:59 -07003212 /* The preferred zone is used for statistics later */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003213 preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
3214 ac.nodemask ? : &cpuset_current_mems_allowed,
3215 &ac.preferred_zone);
3216 if (!ac.preferred_zone)
Mel Gormancc9a6c82012-03-21 16:34:11 -07003217 goto out;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003218 ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gorman5117f452009-06-16 15:31:59 -07003219
3220 /* First allocation attempt */
Andrew Morton91fbdc02015-02-11 15:25:04 -08003221 alloc_mask = gfp_mask|__GFP_HARDWALL;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003222 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08003223 if (unlikely(!page)) {
3224 /*
3225 * Runtime PM, block IO and its error handling path
3226 * can deadlock because I/O on the device might not
3227 * complete.
3228 */
Andrew Morton91fbdc02015-02-11 15:25:04 -08003229 alloc_mask = memalloc_noio_flags(gfp_mask);
3230
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003231 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08003232 }
Mel Gorman11e33f62009-06-16 15:31:57 -07003233
Xishi Qiu23f086f2015-02-11 15:25:07 -08003234 if (kmemcheck_enabled && page)
3235 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3236
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003237 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07003238
3239out:
3240 /*
3241 * When updating a task's mems_allowed, it is possible to race with
3242 * parallel threads in such a way that an allocation can fail while
3243 * the mask is being updated. If a page allocation is about to fail,
3244 * check if the cpuset changed during allocation and if so, retry.
3245 */
Mel Gormand26914d2014-04-03 14:47:24 -07003246 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07003247 goto retry_cpuset;
3248
Mel Gorman11e33f62009-06-16 15:31:57 -07003249 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003250}
Mel Gormand2391712009-06-16 15:31:52 -07003251EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252
3253/*
3254 * Common helper functions.
3255 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08003256unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257{
Akinobu Mita945a1112009-09-21 17:01:47 -07003258 struct page *page;
3259
3260 /*
3261 * __get_free_pages() returns a 32-bit address, which cannot represent
3262 * a highmem page
3263 */
3264 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3265
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266 page = alloc_pages(gfp_mask, order);
3267 if (!page)
3268 return 0;
3269 return (unsigned long) page_address(page);
3270}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271EXPORT_SYMBOL(__get_free_pages);
3272
Harvey Harrison920c7a52008-02-04 22:29:26 -08003273unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274{
Akinobu Mita945a1112009-09-21 17:01:47 -07003275 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277EXPORT_SYMBOL(get_zeroed_page);
3278
Harvey Harrison920c7a52008-02-04 22:29:26 -08003279void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280{
Nick Pigginb5810032005-10-29 18:16:12 -07003281 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003282 if (order == 0)
Mel Gormanb745bc82014-06-04 16:10:22 -07003283 free_hot_cold_page(page, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003284 else
3285 __free_pages_ok(page, order);
3286 }
3287}
3288
3289EXPORT_SYMBOL(__free_pages);
3290
Harvey Harrison920c7a52008-02-04 22:29:26 -08003291void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292{
3293 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07003294 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295 __free_pages(virt_to_page((void *)addr), order);
3296 }
3297}
3298
3299EXPORT_SYMBOL(free_pages);
3300
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003301/*
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003302 * Page Fragment:
3303 * An arbitrary-length arbitrary-offset area of memory which resides
3304 * within a 0 or higher order page. Multiple fragments within that page
3305 * are individually refcounted, in the page's reference counter.
3306 *
3307 * The page_frag functions below provide a simple allocation framework for
3308 * page fragments. This is used by the network stack and network device
3309 * drivers to provide a backing region of memory for use as either an
3310 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3311 */
3312static struct page *__page_frag_refill(struct page_frag_cache *nc,
3313 gfp_t gfp_mask)
3314{
3315 struct page *page = NULL;
3316 gfp_t gfp = gfp_mask;
3317
3318#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3319 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3320 __GFP_NOMEMALLOC;
3321 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3322 PAGE_FRAG_CACHE_MAX_ORDER);
3323 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3324#endif
3325 if (unlikely(!page))
3326 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3327
3328 nc->va = page ? page_address(page) : NULL;
3329
3330 return page;
3331}
3332
3333void *__alloc_page_frag(struct page_frag_cache *nc,
3334 unsigned int fragsz, gfp_t gfp_mask)
3335{
3336 unsigned int size = PAGE_SIZE;
3337 struct page *page;
3338 int offset;
3339
3340 if (unlikely(!nc->va)) {
3341refill:
3342 page = __page_frag_refill(nc, gfp_mask);
3343 if (!page)
3344 return NULL;
3345
3346#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3347 /* if size can vary use size else just use PAGE_SIZE */
3348 size = nc->size;
3349#endif
3350 /* Even if we own the page, we do not use atomic_set().
3351 * This would break get_page_unless_zero() users.
3352 */
3353 atomic_add(size - 1, &page->_count);
3354
3355 /* reset page count bias and offset to start of new frag */
Michal Hocko2f064f32015-08-21 14:11:51 -07003356 nc->pfmemalloc = page_is_pfmemalloc(page);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003357 nc->pagecnt_bias = size;
3358 nc->offset = size;
3359 }
3360
3361 offset = nc->offset - fragsz;
3362 if (unlikely(offset < 0)) {
3363 page = virt_to_page(nc->va);
3364
3365 if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count))
3366 goto refill;
3367
3368#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3369 /* if size can vary use size else just use PAGE_SIZE */
3370 size = nc->size;
3371#endif
3372 /* OK, page count is 0, we can safely set it */
3373 atomic_set(&page->_count, size);
3374
3375 /* reset page count bias and offset to start of new frag */
3376 nc->pagecnt_bias = size;
3377 offset = size - fragsz;
3378 }
3379
3380 nc->pagecnt_bias--;
3381 nc->offset = offset;
3382
3383 return nc->va + offset;
3384}
3385EXPORT_SYMBOL(__alloc_page_frag);
3386
3387/*
3388 * Frees a page fragment allocated out of either a compound or order 0 page.
3389 */
3390void __free_page_frag(void *addr)
3391{
3392 struct page *page = virt_to_head_page(addr);
3393
3394 if (unlikely(put_page_testzero(page)))
3395 __free_pages_ok(page, compound_order(page));
3396}
3397EXPORT_SYMBOL(__free_page_frag);
3398
3399/*
Vladimir Davydov52383432014-06-04 16:06:39 -07003400 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3401 * of the current memory cgroup.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003402 *
Vladimir Davydov52383432014-06-04 16:06:39 -07003403 * It should be used when the caller would like to use kmalloc, but since the
3404 * allocation is large, it has to fall back to the page allocator.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003405 */
Vladimir Davydov52383432014-06-04 16:06:39 -07003406struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
3407{
3408 struct page *page;
3409 struct mem_cgroup *memcg = NULL;
3410
3411 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
3412 return NULL;
3413 page = alloc_pages(gfp_mask, order);
3414 memcg_kmem_commit_charge(page, memcg, order);
3415 return page;
3416}
3417
3418struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
3419{
3420 struct page *page;
3421 struct mem_cgroup *memcg = NULL;
3422
3423 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
3424 return NULL;
3425 page = alloc_pages_node(nid, gfp_mask, order);
3426 memcg_kmem_commit_charge(page, memcg, order);
3427 return page;
3428}
3429
3430/*
3431 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3432 * alloc_kmem_pages.
3433 */
3434void __free_kmem_pages(struct page *page, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003435{
3436 memcg_kmem_uncharge_pages(page, order);
3437 __free_pages(page, order);
3438}
3439
Vladimir Davydov52383432014-06-04 16:06:39 -07003440void free_kmem_pages(unsigned long addr, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003441{
3442 if (addr != 0) {
3443 VM_BUG_ON(!virt_addr_valid((void *)addr));
Vladimir Davydov52383432014-06-04 16:06:39 -07003444 __free_kmem_pages(virt_to_page((void *)addr), order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003445 }
3446}
3447
Andi Kleenee85c2e2011-05-11 15:13:34 -07003448static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
3449{
3450 if (addr) {
3451 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3452 unsigned long used = addr + PAGE_ALIGN(size);
3453
3454 split_page(virt_to_page((void *)addr), order);
3455 while (used < alloc_end) {
3456 free_page(used);
3457 used += PAGE_SIZE;
3458 }
3459 }
3460 return (void *)addr;
3461}
3462
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003463/**
3464 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3465 * @size: the number of bytes to allocate
3466 * @gfp_mask: GFP flags for the allocation
3467 *
3468 * This function is similar to alloc_pages(), except that it allocates the
3469 * minimum number of pages to satisfy the request. alloc_pages() can only
3470 * allocate memory in power-of-two pages.
3471 *
3472 * This function is also limited by MAX_ORDER.
3473 *
3474 * Memory allocated by this function must be released by free_pages_exact().
3475 */
3476void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3477{
3478 unsigned int order = get_order(size);
3479 unsigned long addr;
3480
3481 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07003482 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003483}
3484EXPORT_SYMBOL(alloc_pages_exact);
3485
3486/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07003487 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3488 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07003489 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07003490 * @size: the number of bytes to allocate
3491 * @gfp_mask: GFP flags for the allocation
3492 *
3493 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3494 * back.
3495 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
3496 * but is not exact.
3497 */
Fabian Fredericke1931812014-08-06 16:04:59 -07003498void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
Andi Kleenee85c2e2011-05-11 15:13:34 -07003499{
3500 unsigned order = get_order(size);
3501 struct page *p = alloc_pages_node(nid, gfp_mask, order);
3502 if (!p)
3503 return NULL;
3504 return make_alloc_exact((unsigned long)page_address(p), order, size);
3505}
Andi Kleenee85c2e2011-05-11 15:13:34 -07003506
3507/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003508 * free_pages_exact - release memory allocated via alloc_pages_exact()
3509 * @virt: the value returned by alloc_pages_exact.
3510 * @size: size of allocation, same value as passed to alloc_pages_exact().
3511 *
3512 * Release the memory allocated by a previous call to alloc_pages_exact.
3513 */
3514void free_pages_exact(void *virt, size_t size)
3515{
3516 unsigned long addr = (unsigned long)virt;
3517 unsigned long end = addr + PAGE_ALIGN(size);
3518
3519 while (addr < end) {
3520 free_page(addr);
3521 addr += PAGE_SIZE;
3522 }
3523}
3524EXPORT_SYMBOL(free_pages_exact);
3525
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003526/**
3527 * nr_free_zone_pages - count number of pages beyond high watermark
3528 * @offset: The zone index of the highest zone
3529 *
3530 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3531 * high watermark within all zones at or below a given zone index. For each
3532 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07003533 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003534 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003535static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003536{
Mel Gormandd1a2392008-04-28 02:12:17 -07003537 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003538 struct zone *zone;
3539
Martin J. Blighe310fd42005-07-29 22:59:18 -07003540 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003541 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542
Mel Gorman0e884602008-04-28 02:12:14 -07003543 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003544
Mel Gorman54a6eb52008-04-28 02:12:16 -07003545 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08003546 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003547 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07003548 if (size > high)
3549 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550 }
3551
3552 return sum;
3553}
3554
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003555/**
3556 * nr_free_buffer_pages - count number of pages beyond high watermark
3557 *
3558 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3559 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003561unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562{
Al Viroaf4ca452005-10-21 02:55:38 -04003563 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003564}
Meelap Shahc2f1a552007-07-17 04:04:39 -07003565EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003566
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003567/**
3568 * nr_free_pagecache_pages - count number of pages beyond high watermark
3569 *
3570 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3571 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003573unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003574{
Mel Gorman2a1e2742007-07-17 04:03:12 -07003575 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07003577
3578static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08003580 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08003581 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003583
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584void si_meminfo(struct sysinfo *val)
3585{
3586 val->totalram = totalram_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003587 val->sharedram = global_page_state(NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003588 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003589 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590 val->totalhigh = totalhigh_pages;
3591 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003592 val->mem_unit = PAGE_SIZE;
3593}
3594
3595EXPORT_SYMBOL(si_meminfo);
3596
3597#ifdef CONFIG_NUMA
3598void si_meminfo_node(struct sysinfo *val, int nid)
3599{
Jiang Liucdd91a72013-07-03 15:03:27 -07003600 int zone_type; /* needs to be signed */
3601 unsigned long managed_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003602 pg_data_t *pgdat = NODE_DATA(nid);
3603
Jiang Liucdd91a72013-07-03 15:03:27 -07003604 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3605 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3606 val->totalram = managed_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003607 val->sharedram = node_page_state(nid, NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003608 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003609#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08003610 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08003611 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3612 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003613#else
3614 val->totalhigh = 0;
3615 val->freehigh = 0;
3616#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003617 val->mem_unit = PAGE_SIZE;
3618}
3619#endif
3620
David Rientjesddd588b2011-03-22 16:30:46 -07003621/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07003622 * Determine whether the node should be displayed or not, depending on whether
3623 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07003624 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003625bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07003626{
3627 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003628 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07003629
3630 if (!(flags & SHOW_MEM_FILTER_NODES))
3631 goto out;
3632
Mel Gormancc9a6c82012-03-21 16:34:11 -07003633 do {
Mel Gormand26914d2014-04-03 14:47:24 -07003634 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003635 ret = !node_isset(nid, cpuset_current_mems_allowed);
Mel Gormand26914d2014-04-03 14:47:24 -07003636 } while (read_mems_allowed_retry(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07003637out:
3638 return ret;
3639}
3640
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641#define K(x) ((x) << (PAGE_SHIFT-10))
3642
Rabin Vincent377e4f12012-12-11 16:00:24 -08003643static void show_migration_types(unsigned char type)
3644{
3645 static const char types[MIGRATE_TYPES] = {
3646 [MIGRATE_UNMOVABLE] = 'U',
3647 [MIGRATE_RECLAIMABLE] = 'E',
3648 [MIGRATE_MOVABLE] = 'M',
3649 [MIGRATE_RESERVE] = 'R',
3650#ifdef CONFIG_CMA
3651 [MIGRATE_CMA] = 'C',
3652#endif
Minchan Kim194159f2013-02-22 16:33:58 -08003653#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08003654 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08003655#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08003656 };
3657 char tmp[MIGRATE_TYPES + 1];
3658 char *p = tmp;
3659 int i;
3660
3661 for (i = 0; i < MIGRATE_TYPES; i++) {
3662 if (type & (1 << i))
3663 *p++ = types[i];
3664 }
3665
3666 *p = '\0';
3667 printk("(%s) ", tmp);
3668}
3669
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670/*
3671 * Show free area list (used inside shift_scroll-lock stuff)
3672 * We also calculate the percentage fragmentation. We do this by counting the
3673 * memory on each free list with the exception of the first item on the list.
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003674 *
3675 * Bits in @filter:
3676 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3677 * cpuset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003679void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680{
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003681 unsigned long free_pcp = 0;
Jes Sorensenc7241912006-09-27 01:50:05 -07003682 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003683 struct zone *zone;
3684
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003685 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07003686 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003687 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003688
Konstantin Khlebnikov761b0672015-04-14 15:45:32 -07003689 for_each_online_cpu(cpu)
3690 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003691 }
3692
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003693 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3694 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003695 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3696 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003697 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003698 " free:%lu free_pcp:%lu free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003699 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003700 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003701 global_page_state(NR_ISOLATED_ANON),
3702 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003703 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003704 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003705 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003706 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003707 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003708 global_page_state(NR_UNSTABLE_NFS),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003709 global_page_state(NR_SLAB_RECLAIMABLE),
3710 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003711 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003712 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003713 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003714 global_page_state(NR_BOUNCE),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003715 global_page_state(NR_FREE_PAGES),
3716 free_pcp,
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003717 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003718
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003719 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003720 int i;
3721
David Rientjes7bf02ea2011-05-24 17:11:16 -07003722 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003723 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003724
3725 free_pcp = 0;
3726 for_each_online_cpu(cpu)
3727 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3728
Linus Torvalds1da177e2005-04-16 15:20:36 -07003729 show_node(zone);
3730 printk("%s"
3731 " free:%lukB"
3732 " min:%lukB"
3733 " low:%lukB"
3734 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003735 " active_anon:%lukB"
3736 " inactive_anon:%lukB"
3737 " active_file:%lukB"
3738 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003739 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003740 " isolated(anon):%lukB"
3741 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003743 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003744 " mlocked:%lukB"
3745 " dirty:%lukB"
3746 " writeback:%lukB"
3747 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003748 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003749 " slab_reclaimable:%lukB"
3750 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003751 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003752 " pagetables:%lukB"
3753 " unstable:%lukB"
3754 " bounce:%lukB"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003755 " free_pcp:%lukB"
3756 " local_pcp:%ukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003757 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003758 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003759 " pages_scanned:%lu"
3760 " all_unreclaimable? %s"
3761 "\n",
3762 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003763 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003764 K(min_wmark_pages(zone)),
3765 K(low_wmark_pages(zone)),
3766 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003767 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3768 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3769 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3770 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003771 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003772 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3773 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003774 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003775 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003776 K(zone_page_state(zone, NR_MLOCK)),
3777 K(zone_page_state(zone, NR_FILE_DIRTY)),
3778 K(zone_page_state(zone, NR_WRITEBACK)),
3779 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003780 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003781 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3782 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003783 zone_page_state(zone, NR_KERNEL_STACK) *
3784 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003785 K(zone_page_state(zone, NR_PAGETABLE)),
3786 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3787 K(zone_page_state(zone, NR_BOUNCE)),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003788 K(free_pcp),
3789 K(this_cpu_read(zone->pageset->pcp.count)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003790 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003791 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Mel Gorman0d5d8232014-08-06 16:07:16 -07003792 K(zone_page_state(zone, NR_PAGES_SCANNED)),
Lisa Du6e543d52013-09-11 14:22:36 -07003793 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794 );
3795 printk("lowmem_reserve[]:");
3796 for (i = 0; i < MAX_NR_ZONES; i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07003797 printk(" %ld", zone->lowmem_reserve[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 printk("\n");
3799 }
3800
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003801 for_each_populated_zone(zone) {
Pintu Kumarb8af2942013-09-11 14:20:34 -07003802 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003803 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804
David Rientjes7bf02ea2011-05-24 17:11:16 -07003805 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003806 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807 show_node(zone);
3808 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809
3810 spin_lock_irqsave(&zone->lock, flags);
3811 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08003812 struct free_area *area = &zone->free_area[order];
3813 int type;
3814
3815 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003816 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003817
3818 types[order] = 0;
3819 for (type = 0; type < MIGRATE_TYPES; type++) {
3820 if (!list_empty(&area->free_list[type]))
3821 types[order] |= 1 << type;
3822 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823 }
3824 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003825 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003826 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003827 if (nr[order])
3828 show_migration_types(types[order]);
3829 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830 printk("= %lukB\n", K(total));
3831 }
3832
David Rientjes949f7ec2013-04-29 15:07:48 -07003833 hugetlb_show_meminfo();
3834
Larry Woodmane6f36022008-02-04 22:29:30 -08003835 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3836
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837 show_swap_cache_info();
3838}
3839
Mel Gorman19770b32008-04-28 02:12:18 -07003840static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3841{
3842 zoneref->zone = zone;
3843 zoneref->zone_idx = zone_idx(zone);
3844}
3845
Linus Torvalds1da177e2005-04-16 15:20:36 -07003846/*
3847 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003848 *
3849 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003850 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003851static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003852 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853{
Christoph Lameter1a932052006-01-06 00:11:16 -08003854 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003855 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003856
3857 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003858 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003859 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003860 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003861 zoneref_set_zone(zone,
3862 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003863 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003864 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003865 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003866
Christoph Lameter070f8032006-01-06 00:11:19 -08003867 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868}
3869
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003870
3871/*
3872 * zonelist_order:
3873 * 0 = automatic detection of better ordering.
3874 * 1 = order by ([node] distance, -zonetype)
3875 * 2 = order by (-zonetype, [node] distance)
3876 *
3877 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3878 * the same zonelist. So only NUMA can configure this param.
3879 */
3880#define ZONELIST_ORDER_DEFAULT 0
3881#define ZONELIST_ORDER_NODE 1
3882#define ZONELIST_ORDER_ZONE 2
3883
3884/* zonelist order in the kernel.
3885 * set_zonelist_order() will set this to NODE or ZONE.
3886 */
3887static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3888static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3889
3890
Linus Torvalds1da177e2005-04-16 15:20:36 -07003891#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003892/* The value user specified ....changed by config */
3893static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3894/* string for sysctl */
3895#define NUMA_ZONELIST_ORDER_LEN 16
3896char numa_zonelist_order[16] = "default";
3897
3898/*
3899 * interface for configure zonelist ordering.
3900 * command line option "numa_zonelist_order"
3901 * = "[dD]efault - default, automatic configuration.
3902 * = "[nN]ode - order by node locality, then by zone within node
3903 * = "[zZ]one - order by zone, then by locality within zone
3904 */
3905
3906static int __parse_numa_zonelist_order(char *s)
3907{
3908 if (*s == 'd' || *s == 'D') {
3909 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3910 } else if (*s == 'n' || *s == 'N') {
3911 user_zonelist_order = ZONELIST_ORDER_NODE;
3912 } else if (*s == 'z' || *s == 'Z') {
3913 user_zonelist_order = ZONELIST_ORDER_ZONE;
3914 } else {
3915 printk(KERN_WARNING
3916 "Ignoring invalid numa_zonelist_order value: "
3917 "%s\n", s);
3918 return -EINVAL;
3919 }
3920 return 0;
3921}
3922
3923static __init int setup_numa_zonelist_order(char *s)
3924{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003925 int ret;
3926
3927 if (!s)
3928 return 0;
3929
3930 ret = __parse_numa_zonelist_order(s);
3931 if (ret == 0)
3932 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3933
3934 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003935}
3936early_param("numa_zonelist_order", setup_numa_zonelist_order);
3937
3938/*
3939 * sysctl handler for numa_zonelist_order
3940 */
Joe Perchescccad5b2014-06-06 14:38:09 -07003941int numa_zonelist_order_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003942 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003943 loff_t *ppos)
3944{
3945 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3946 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003947 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003948
Andi Kleen443c6f12009-12-23 21:00:47 +01003949 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07003950 if (write) {
3951 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3952 ret = -EINVAL;
3953 goto out;
3954 }
3955 strcpy(saved_string, (char *)table->data);
3956 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003957 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003958 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003959 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003960 if (write) {
3961 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07003962
3963 ret = __parse_numa_zonelist_order((char *)table->data);
3964 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003965 /*
3966 * bogus value. restore saved string
3967 */
Chen Gangdacbde02013-07-03 15:02:35 -07003968 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003969 NUMA_ZONELIST_ORDER_LEN);
3970 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003971 } else if (oldval != user_zonelist_order) {
3972 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07003973 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003974 mutex_unlock(&zonelists_mutex);
3975 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003976 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003977out:
3978 mutex_unlock(&zl_order_mutex);
3979 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003980}
3981
3982
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003983#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003984static int node_load[MAX_NUMNODES];
3985
Linus Torvalds1da177e2005-04-16 15:20:36 -07003986/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07003987 * 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 -07003988 * @node: node whose fallback list we're appending
3989 * @used_node_mask: nodemask_t of already used nodes
3990 *
3991 * We use a number of factors to determine which is the next node that should
3992 * appear on a given node's fallback list. The node should not have appeared
3993 * already in @node's fallback list, and it should be the next closest node
3994 * according to the distance array (which contains arbitrary distance values
3995 * from each node to each node in the system), and should also prefer nodes
3996 * with no CPUs, since presumably they'll have very little allocation pressure
3997 * on them otherwise.
3998 * It returns -1 if no node is found.
3999 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004000static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001{
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01004002 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004003 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08004004 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10304005 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01004007 /* Use the local node if we haven't already */
4008 if (!node_isset(node, *used_node_mask)) {
4009 node_set(node, *used_node_mask);
4010 return node;
4011 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004012
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004013 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004014
4015 /* Don't want a node to appear more than once */
4016 if (node_isset(n, *used_node_mask))
4017 continue;
4018
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019 /* Use the distance array to find the distance */
4020 val = node_distance(node, n);
4021
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01004022 /* Penalize nodes under us ("prefer the next node") */
4023 val += (n < node);
4024
Linus Torvalds1da177e2005-04-16 15:20:36 -07004025 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10304026 tmp = cpumask_of_node(n);
4027 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028 val += PENALTY_FOR_NODE_WITH_CPUS;
4029
4030 /* Slight preference for less loaded node */
4031 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4032 val += node_load[n];
4033
4034 if (val < min_val) {
4035 min_val = val;
4036 best_node = n;
4037 }
4038 }
4039
4040 if (best_node >= 0)
4041 node_set(best_node, *used_node_mask);
4042
4043 return best_node;
4044}
4045
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004046
4047/*
4048 * Build zonelists ordered by node and zones within node.
4049 * This results in maximum locality--normal zone overflows into local
4050 * DMA zone, if any--but risks exhausting DMA zone.
4051 */
4052static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004053{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004054 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004055 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004056
Mel Gorman54a6eb52008-04-28 02:12:16 -07004057 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07004058 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07004059 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004060 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07004061 zonelist->_zonerefs[j].zone = NULL;
4062 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004063}
4064
4065/*
Christoph Lameter523b9452007-10-16 01:25:37 -07004066 * Build gfp_thisnode zonelists
4067 */
4068static void build_thisnode_zonelists(pg_data_t *pgdat)
4069{
Christoph Lameter523b9452007-10-16 01:25:37 -07004070 int j;
4071 struct zonelist *zonelist;
4072
Mel Gorman54a6eb52008-04-28 02:12:16 -07004073 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004074 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07004075 zonelist->_zonerefs[j].zone = NULL;
4076 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07004077}
4078
4079/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004080 * Build zonelists ordered by zone and nodes within zones.
4081 * This results in conserving DMA zone[s] until all Normal memory is
4082 * exhausted, but results in overflowing to remote node while memory
4083 * may still exist in local DMA zone.
4084 */
4085static int node_order[MAX_NUMNODES];
4086
4087static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4088{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004089 int pos, j, node;
4090 int zone_type; /* needs to be signed */
4091 struct zone *z;
4092 struct zonelist *zonelist;
4093
Mel Gorman54a6eb52008-04-28 02:12:16 -07004094 zonelist = &pgdat->node_zonelists[0];
4095 pos = 0;
4096 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4097 for (j = 0; j < nr_nodes; j++) {
4098 node = node_order[j];
4099 z = &NODE_DATA(node)->node_zones[zone_type];
4100 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07004101 zoneref_set_zone(z,
4102 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004103 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004104 }
4105 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004106 }
Mel Gormandd1a2392008-04-28 02:12:17 -07004107 zonelist->_zonerefs[pos].zone = NULL;
4108 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004109}
4110
Mel Gorman31939132014-10-09 15:28:30 -07004111#if defined(CONFIG_64BIT)
4112/*
4113 * Devices that require DMA32/DMA are relatively rare and do not justify a
4114 * penalty to every machine in case the specialised case applies. Default
4115 * to Node-ordering on 64-bit NUMA machines
4116 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004117static int default_zonelist_order(void)
4118{
Mel Gorman31939132014-10-09 15:28:30 -07004119 return ZONELIST_ORDER_NODE;
4120}
4121#else
4122/*
4123 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4124 * by the kernel. If processes running on node 0 deplete the low memory zone
4125 * then reclaim will occur more frequency increasing stalls and potentially
4126 * be easier to OOM if a large percentage of the zone is under writeback or
4127 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4128 * Hence, default to zone ordering on 32-bit.
4129 */
4130static int default_zonelist_order(void)
4131{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004132 return ZONELIST_ORDER_ZONE;
4133}
Mel Gorman31939132014-10-09 15:28:30 -07004134#endif /* CONFIG_64BIT */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004135
4136static void set_zonelist_order(void)
4137{
4138 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4139 current_zonelist_order = default_zonelist_order();
4140 else
4141 current_zonelist_order = user_zonelist_order;
4142}
4143
4144static void build_zonelists(pg_data_t *pgdat)
4145{
4146 int j, node, load;
4147 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004148 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004149 int local_node, prev_node;
4150 struct zonelist *zonelist;
4151 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004152
4153 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07004154 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004155 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07004156 zonelist->_zonerefs[0].zone = NULL;
4157 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158 }
4159
4160 /* NUMA-aware ordering of nodes */
4161 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004162 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004163 prev_node = local_node;
4164 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004165
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004166 memset(node_order, 0, sizeof(node_order));
4167 j = 0;
4168
Linus Torvalds1da177e2005-04-16 15:20:36 -07004169 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4170 /*
4171 * We don't want to pressure a particular node.
4172 * So adding penalty to the first node in same
4173 * distance group to make it round-robin.
4174 */
David Rientjes957f8222012-10-08 16:33:24 -07004175 if (node_distance(local_node, node) !=
4176 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004177 node_load[node] = load;
4178
Linus Torvalds1da177e2005-04-16 15:20:36 -07004179 prev_node = node;
4180 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004181 if (order == ZONELIST_ORDER_NODE)
4182 build_zonelists_in_node_order(pgdat, node);
4183 else
4184 node_order[j++] = node; /* remember order */
4185 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004186
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004187 if (order == ZONELIST_ORDER_ZONE) {
4188 /* calculate node order -- i.e., DMA last! */
4189 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004190 }
Christoph Lameter523b9452007-10-16 01:25:37 -07004191
4192 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004193}
4194
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004195/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004196static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004197{
Mel Gorman54a6eb52008-04-28 02:12:16 -07004198 struct zonelist *zonelist;
4199 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07004200 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004201
Mel Gorman54a6eb52008-04-28 02:12:16 -07004202 zonelist = &pgdat->node_zonelists[0];
4203 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
4204 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07004205 for (z = zonelist->_zonerefs; z->zone; z++)
4206 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004207}
4208
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004209#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4210/*
4211 * Return node id of node used for "local" allocations.
4212 * I.e., first node id of first zone in arg node's generic zonelist.
4213 * Used for initializing percpu 'numa_mem', which is used primarily
4214 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4215 */
4216int local_memory_node(int node)
4217{
4218 struct zone *zone;
4219
4220 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
4221 gfp_zone(GFP_KERNEL),
4222 NULL,
4223 &zone);
4224 return zone->node;
4225}
4226#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004227
Linus Torvalds1da177e2005-04-16 15:20:36 -07004228#else /* CONFIG_NUMA */
4229
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004230static void set_zonelist_order(void)
4231{
4232 current_zonelist_order = ZONELIST_ORDER_ZONE;
4233}
4234
4235static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004236{
Christoph Lameter19655d32006-09-25 23:31:19 -07004237 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07004238 enum zone_type j;
4239 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004240
4241 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004242
Mel Gorman54a6eb52008-04-28 02:12:16 -07004243 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004244 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004245
Mel Gorman54a6eb52008-04-28 02:12:16 -07004246 /*
4247 * Now we build the zonelist so that it contains the zones
4248 * of all the other nodes.
4249 * We don't want to pressure a particular node, so when
4250 * building the zones for node N, we make sure that the
4251 * zones coming right after the local ones are those from
4252 * node N+1 (modulo N)
4253 */
4254 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4255 if (!node_online(node))
4256 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004257 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004258 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07004259 for (node = 0; node < local_node; node++) {
4260 if (!node_online(node))
4261 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004262 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004263 }
4264
Mel Gormandd1a2392008-04-28 02:12:17 -07004265 zonelist->_zonerefs[j].zone = NULL;
4266 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004267}
4268
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004269/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004270static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004271{
Mel Gorman54a6eb52008-04-28 02:12:16 -07004272 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004273}
4274
Linus Torvalds1da177e2005-04-16 15:20:36 -07004275#endif /* CONFIG_NUMA */
4276
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004277/*
4278 * Boot pageset table. One per cpu which is going to be used for all
4279 * zones and all nodes. The parameters will be set in such a way
4280 * that an item put on a list will immediately be handed over to
4281 * the buddy list. This is safe since pageset manipulation is done
4282 * with interrupts disabled.
4283 *
4284 * The boot_pagesets must be kept even after bootup is complete for
4285 * unused processors and/or zones. They do play a role for bootstrapping
4286 * hotplugged processors.
4287 *
4288 * zoneinfo_show() and maybe other functions do
4289 * not check if the processor is online before following the pageset pointer.
4290 * Other parts of the kernel may not check if the zone is available.
4291 */
4292static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4293static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07004294static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004295
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004296/*
4297 * Global mutex to protect against size modification of zonelists
4298 * as well as to serialize pageset setup for the new populated zone.
4299 */
4300DEFINE_MUTEX(zonelists_mutex);
4301
Rusty Russell9b1a4d32008-07-28 12:16:30 -05004302/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07004303static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004304{
Yasunori Goto68113782006-06-23 02:03:11 -07004305 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004306 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07004307 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004308
Bo Liu7f9cfb32009-08-18 14:11:19 -07004309#ifdef CONFIG_NUMA
4310 memset(node_load, 0, sizeof(node_load));
4311#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07004312
4313 if (self && !node_online(self->node_id)) {
4314 build_zonelists(self);
4315 build_zonelist_cache(self);
4316 }
4317
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004318 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07004319 pg_data_t *pgdat = NODE_DATA(nid);
4320
4321 build_zonelists(pgdat);
4322 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004323 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004324
4325 /*
4326 * Initialize the boot_pagesets that are going to be used
4327 * for bootstrapping processors. The real pagesets for
4328 * each zone will be allocated later when the per cpu
4329 * allocator is available.
4330 *
4331 * boot_pagesets are used also for bootstrapping offline
4332 * cpus if the system is already booted because the pagesets
4333 * are needed to initialize allocators on a specific cpu too.
4334 * F.e. the percpu allocator needs the page allocator which
4335 * needs the percpu allocator in order to allocate its pagesets
4336 * (a chicken-egg dilemma).
4337 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004338 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004339 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4340
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004341#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4342 /*
4343 * We now know the "local memory node" for each node--
4344 * i.e., the node of the first zone in the generic zonelist.
4345 * Set up numa_mem percpu variable for on-line cpus. During
4346 * boot, only the boot cpu should be on-line; we'll init the
4347 * secondary cpus' numa_mem as they come on-line. During
4348 * node/memory hotplug, we'll fixup all on-line cpus.
4349 */
4350 if (cpu_online(cpu))
4351 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4352#endif
4353 }
4354
Yasunori Goto68113782006-06-23 02:03:11 -07004355 return 0;
4356}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004357
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004358static noinline void __init
4359build_all_zonelists_init(void)
4360{
4361 __build_all_zonelists(NULL);
4362 mminit_verify_zonelist();
4363 cpuset_init_current_mems_allowed();
4364}
4365
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004366/*
4367 * Called with zonelists_mutex held always
4368 * unless system_state == SYSTEM_BOOTING.
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004369 *
4370 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4371 * [we're only called with non-NULL zone through __meminit paths] and
4372 * (2) call of __init annotated helper build_all_zonelists_init
4373 * [protected by SYSTEM_BOOTING].
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004374 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07004375void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07004376{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004377 set_zonelist_order();
4378
Yasunori Goto68113782006-06-23 02:03:11 -07004379 if (system_state == SYSTEM_BOOTING) {
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004380 build_all_zonelists_init();
Yasunori Goto68113782006-06-23 02:03:11 -07004381 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08004382#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07004383 if (zone)
4384 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08004385#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07004386 /* we have to stop all cpus to guarantee there is no user
4387 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07004388 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07004389 /* cpuset refresh routine should be here */
4390 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07004391 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004392 /*
4393 * Disable grouping by mobility if the number of pages in the
4394 * system is too low to allow the mechanism to work. It would be
4395 * more accurate, but expensive to check per-zone. This check is
4396 * made on memory-hotadd so a system can start with mobility
4397 * disabled and enable it later
4398 */
Mel Gormand9c23402007-10-16 01:26:01 -07004399 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004400 page_group_by_mobility_disabled = 1;
4401 else
4402 page_group_by_mobility_disabled = 0;
4403
Anton Blanchardf88dfff2014-12-10 15:42:53 -08004404 pr_info("Built %i zonelists in %s order, mobility grouping %s. "
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004405 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004406 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004407 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004408 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004409 vm_total_pages);
4410#ifdef CONFIG_NUMA
Anton Blanchardf88dfff2014-12-10 15:42:53 -08004411 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004412#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004413}
4414
4415/*
4416 * Helper functions to size the waitqueue hash table.
4417 * Essentially these want to choose hash table sizes sufficiently
4418 * large so that collisions trying to wait on pages are rare.
4419 * But in fact, the number of active page waitqueues on typical
4420 * systems is ridiculously low, less than 200. So this is even
4421 * conservative, even though it seems large.
4422 *
4423 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4424 * waitqueues, i.e. the size of the waitq table given the number of pages.
4425 */
4426#define PAGES_PER_WAITQUEUE 256
4427
Yasunori Gotocca448f2006-06-23 02:03:10 -07004428#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07004429static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004430{
4431 unsigned long size = 1;
4432
4433 pages /= PAGES_PER_WAITQUEUE;
4434
4435 while (size < pages)
4436 size <<= 1;
4437
4438 /*
4439 * Once we have dozens or even hundreds of threads sleeping
4440 * on IO we've got bigger problems than wait queue collision.
4441 * Limit the size of the wait table to a reasonable size.
4442 */
4443 size = min(size, 4096UL);
4444
4445 return max(size, 4UL);
4446}
Yasunori Gotocca448f2006-06-23 02:03:10 -07004447#else
4448/*
4449 * A zone's size might be changed by hot-add, so it is not possible to determine
4450 * a suitable size for its wait_table. So we use the maximum size now.
4451 *
4452 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4453 *
4454 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4455 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4456 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4457 *
4458 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4459 * or more by the traditional way. (See above). It equals:
4460 *
4461 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4462 * ia64(16K page size) : = ( 8G + 4M)byte.
4463 * powerpc (64K page size) : = (32G +16M)byte.
4464 */
4465static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4466{
4467 return 4096UL;
4468}
4469#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004470
4471/*
4472 * This is an integer logarithm so that shifts can be used later
4473 * to extract the more random high bits from the multiplicative
4474 * hash function before the remainder is taken.
4475 */
4476static inline unsigned long wait_table_bits(unsigned long size)
4477{
4478 return ffz(~size);
4479}
4480
Mel Gorman56fd56b2007-10-16 01:25:58 -07004481/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07004482 * Check if a pageblock contains reserved pages
4483 */
4484static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
4485{
4486 unsigned long pfn;
4487
4488 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4489 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
4490 return 1;
4491 }
4492 return 0;
4493}
4494
4495/*
Mel Gormand9c23402007-10-16 01:26:01 -07004496 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07004497 * of blocks reserved is based on min_wmark_pages(zone). The memory within
4498 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07004499 * higher will lead to a bigger reserve which will get freed as contiguous
4500 * blocks as reclaim kicks in
4501 */
4502static void setup_zone_migrate_reserve(struct zone *zone)
4503{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07004504 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004505 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07004506 unsigned long block_migratetype;
4507 int reserve;
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004508 int old_reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004509
Michal Hockod02156382011-12-08 14:34:27 -08004510 /*
4511 * Get the start pfn, end pfn and the number of blocks to reserve
4512 * We have to be careful to be aligned to pageblock_nr_pages to
4513 * make sure that we always check pfn_valid for the first page in
4514 * the block.
4515 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07004516 start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08004517 end_pfn = zone_end_pfn(zone);
Michal Hockod02156382011-12-08 14:34:27 -08004518 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07004519 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07004520 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004521
Mel Gorman78986a62009-09-21 17:03:02 -07004522 /*
4523 * Reserve blocks are generally in place to help high-order atomic
4524 * allocations that are short-lived. A min_free_kbytes value that
4525 * would result in more than 2 reserve blocks for atomic allocations
4526 * is assumed to be in place to help anti-fragmentation for the
4527 * future allocation of hugepages at runtime.
4528 */
4529 reserve = min(2, reserve);
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004530 old_reserve = zone->nr_migrate_reserve_block;
4531
4532 /* When memory hot-add, we almost always need to do nothing */
4533 if (reserve == old_reserve)
4534 return;
4535 zone->nr_migrate_reserve_block = reserve;
Mel Gorman78986a62009-09-21 17:03:02 -07004536
Mel Gormand9c23402007-10-16 01:26:01 -07004537 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman7e18adb2015-06-30 14:57:05 -07004538 if (!early_page_nid_uninitialised(pfn, zone_to_nid(zone)))
4539 return;
4540
Mel Gorman56fd56b2007-10-16 01:25:58 -07004541 if (!pfn_valid(pfn))
4542 continue;
4543 page = pfn_to_page(pfn);
4544
Adam Litke344c7902008-09-02 14:35:38 -07004545 /* Watch out for overlapping nodes */
4546 if (page_to_nid(page) != zone_to_nid(zone))
4547 continue;
4548
Mel Gorman56fd56b2007-10-16 01:25:58 -07004549 block_migratetype = get_pageblock_migratetype(page);
4550
Mel Gorman938929f2012-01-10 15:07:14 -08004551 /* Only test what is necessary when the reserves are not met */
4552 if (reserve > 0) {
4553 /*
4554 * Blocks with reserved pages will never free, skip
4555 * them.
4556 */
4557 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4558 if (pageblock_is_reserved(pfn, block_end_pfn))
4559 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004560
Mel Gorman938929f2012-01-10 15:07:14 -08004561 /* If this block is reserved, account for it */
4562 if (block_migratetype == MIGRATE_RESERVE) {
4563 reserve--;
4564 continue;
4565 }
4566
4567 /* Suitable for reserving if this block is movable */
4568 if (block_migratetype == MIGRATE_MOVABLE) {
4569 set_pageblock_migratetype(page,
4570 MIGRATE_RESERVE);
4571 move_freepages_block(zone, page,
4572 MIGRATE_RESERVE);
4573 reserve--;
4574 continue;
4575 }
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004576 } else if (!old_reserve) {
4577 /*
4578 * At boot time we don't need to scan the whole zone
4579 * for turning off MIGRATE_RESERVE.
4580 */
4581 break;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004582 }
4583
4584 /*
4585 * If the reserve is met and this is a previous reserved block,
4586 * take it back
4587 */
4588 if (block_migratetype == MIGRATE_RESERVE) {
4589 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4590 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4591 }
4592 }
4593}
Mel Gormanac0e5b72007-10-16 01:25:58 -07004594
Linus Torvalds1da177e2005-04-16 15:20:36 -07004595/*
4596 * Initially all pages are reserved - free ones are freed
4597 * up by free_all_bootmem() once the early boot process is
4598 * done. Non-atomic initialization, single-pass.
4599 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004600void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004601 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004602{
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004603 pg_data_t *pgdat = NODE_DATA(nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -07004604 unsigned long end_pfn = start_pfn + size;
4605 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004606 struct zone *z;
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004607 unsigned long nr_initialised = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004608
Hugh Dickins22b31ee2009-01-06 14:40:09 -08004609 if (highest_memmap_pfn < end_pfn - 1)
4610 highest_memmap_pfn = end_pfn - 1;
4611
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004612 z = &pgdat->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08004613 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08004614 /*
4615 * There can be holes in boot-time mem_map[]s
4616 * handed to this function. They do not
4617 * exist on hotplugged memory.
4618 */
4619 if (context == MEMMAP_EARLY) {
4620 if (!early_pfn_valid(pfn))
4621 continue;
4622 if (!early_pfn_in_nid(pfn, nid))
4623 continue;
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004624 if (!update_defer_init(pgdat, pfn, end_pfn,
4625 &nr_initialised))
4626 break;
Dave Hansena2f3aa022007-01-10 23:15:30 -08004627 }
Mel Gormanac5d2532015-06-30 14:57:20 -07004628
4629 /*
4630 * Mark the block movable so that blocks are reserved for
4631 * movable at startup. This will force kernel allocations
4632 * to reserve their blocks rather than leaking throughout
4633 * the address space during boot when many long-lived
4634 * kernel allocations are made. Later some blocks near
4635 * the start are marked MIGRATE_RESERVE by
4636 * setup_zone_migrate_reserve()
4637 *
4638 * bitmap is created for zone's valid pfn range. but memmap
4639 * can be created for invalid pages (for alignment)
4640 * check here not to call set_pageblock_migratetype() against
4641 * pfn out of zone.
4642 */
4643 if (!(pfn & (pageblock_nr_pages - 1))) {
4644 struct page *page = pfn_to_page(pfn);
4645
4646 __init_single_page(page, pfn, zone, nid);
4647 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4648 } else {
4649 __init_single_pfn(pfn, zone, nid);
4650 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004651 }
4652}
4653
Andi Kleen1e548de2008-02-04 22:29:26 -08004654static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004655{
Mel Gorman7aeb09f2014-06-04 16:10:21 -07004656 unsigned int order, t;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004657 for_each_migratetype_order(order, t) {
4658 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004659 zone->free_area[order].nr_free = 0;
4660 }
4661}
4662
4663#ifndef __HAVE_ARCH_MEMMAP_INIT
4664#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08004665 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004666#endif
4667
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004668static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004669{
David Howells3a6be872009-05-06 16:03:03 -07004670#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004671 int batch;
4672
4673 /*
4674 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07004675 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004676 *
4677 * OK, so we don't know how big the cache is. So guess.
4678 */
Jiang Liub40da042013-02-22 16:33:52 -08004679 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07004680 if (batch * PAGE_SIZE > 512 * 1024)
4681 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004682 batch /= 4; /* We effectively *= 4 below */
4683 if (batch < 1)
4684 batch = 1;
4685
4686 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004687 * Clamp the batch to a 2^n - 1 value. Having a power
4688 * of 2 value was found to be more likely to have
4689 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004690 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004691 * For example if 2 tasks are alternately allocating
4692 * batches of pages, one task can end up with a lot
4693 * of pages of one half of the possible page colors
4694 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004695 */
David Howells91552032009-05-06 16:03:02 -07004696 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004697
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004698 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004699
4700#else
4701 /* The deferral and batching of frees should be suppressed under NOMMU
4702 * conditions.
4703 *
4704 * The problem is that NOMMU needs to be able to allocate large chunks
4705 * of contiguous memory as there's no hardware page translation to
4706 * assemble apparent contiguous memory from discontiguous pages.
4707 *
4708 * Queueing large contiguous runs of pages for batching, however,
4709 * causes the pages to actually be freed in smaller chunks. As there
4710 * can be a significant delay between the individual batches being
4711 * recycled, this leads to the once large chunks of space being
4712 * fragmented and becoming unavailable for high-order allocations.
4713 */
4714 return 0;
4715#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004716}
4717
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004718/*
4719 * pcp->high and pcp->batch values are related and dependent on one another:
4720 * ->batch must never be higher then ->high.
4721 * The following function updates them in a safe manner without read side
4722 * locking.
4723 *
4724 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4725 * those fields changing asynchronously (acording the the above rule).
4726 *
4727 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4728 * outside of boot time (or some other assurance that no concurrent updaters
4729 * exist).
4730 */
4731static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4732 unsigned long batch)
4733{
4734 /* start with a fail safe value for batch */
4735 pcp->batch = 1;
4736 smp_wmb();
4737
4738 /* Update high, then batch, in order */
4739 pcp->high = high;
4740 smp_wmb();
4741
4742 pcp->batch = batch;
4743}
4744
Cody P Schafer36640332013-07-03 15:01:40 -07004745/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07004746static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4747{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004748 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07004749}
4750
Cody P Schafer88c90db2013-07-03 15:01:35 -07004751static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004752{
4753 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004754 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004755
Magnus Damm1c6fe942005-10-26 01:58:59 -07004756 memset(p, 0, sizeof(*p));
4757
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004758 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004759 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004760 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4761 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004762}
4763
Cody P Schafer88c90db2013-07-03 15:01:35 -07004764static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4765{
4766 pageset_init(p);
4767 pageset_set_batch(p, batch);
4768}
4769
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004770/*
Cody P Schafer36640332013-07-03 15:01:40 -07004771 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004772 * to the value high for the pageset p.
4773 */
Cody P Schafer36640332013-07-03 15:01:40 -07004774static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004775 unsigned long high)
4776{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004777 unsigned long batch = max(1UL, high / 4);
4778 if ((high / 4) > (PAGE_SHIFT * 8))
4779 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004780
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004781 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004782}
4783
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004784static void pageset_set_high_and_batch(struct zone *zone,
4785 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004786{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004787 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07004788 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004789 (zone->managed_pages /
4790 percpu_pagelist_fraction));
4791 else
4792 pageset_set_batch(pcp, zone_batchsize(zone));
4793}
4794
Cody P Schafer169f6c12013-07-03 15:01:41 -07004795static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4796{
4797 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4798
4799 pageset_init(pcp);
4800 pageset_set_high_and_batch(zone, pcp);
4801}
4802
Jiang Liu4ed7e022012-07-31 16:43:35 -07004803static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004804{
4805 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07004806 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004807 for_each_possible_cpu(cpu)
4808 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07004809}
4810
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004811/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004812 * Allocate per cpu pagesets and initialize them.
4813 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004814 */
Al Viro78d99552005-12-15 09:18:25 +00004815void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004816{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004817 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004818
Wu Fengguang319774e2010-05-24 14:32:49 -07004819 for_each_populated_zone(zone)
4820 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004821}
4822
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004823static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004824int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004825{
4826 int i;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004827 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004828
4829 /*
4830 * The per-page waitqueue mechanism uses hashed waitqueues
4831 * per zone.
4832 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004833 zone->wait_table_hash_nr_entries =
4834 wait_table_hash_nr_entries(zone_size_pages);
4835 zone->wait_table_bits =
4836 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004837 alloc_size = zone->wait_table_hash_nr_entries
4838 * sizeof(wait_queue_head_t);
4839
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004840 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004841 zone->wait_table = (wait_queue_head_t *)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004842 memblock_virt_alloc_node_nopanic(
4843 alloc_size, zone->zone_pgdat->node_id);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004844 } else {
4845 /*
4846 * This case means that a zone whose size was 0 gets new memory
4847 * via memory hot-add.
4848 * But it may be the case that a new node was hot-added. In
4849 * this case vmalloc() will not be able to use this new node's
4850 * memory - this wait_table must be initialized to use this new
4851 * node itself as well.
4852 * To use this new node's memory, further consideration will be
4853 * necessary.
4854 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004855 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004856 }
4857 if (!zone->wait_table)
4858 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004859
Pintu Kumarb8af2942013-09-11 14:20:34 -07004860 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004861 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004862
4863 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004864}
4865
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004866static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004867{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004868 /*
4869 * per cpu subsystem is not up at this point. The following code
4870 * relies on the ability of the linker to provide the
4871 * offset of a (static) per cpu variable into the per cpu area.
4872 */
4873 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004874
Xishi Qiub38a8722013-11-12 15:07:20 -08004875 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004876 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4877 zone->name, zone->present_pages,
4878 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004879}
4880
Jiang Liu4ed7e022012-07-31 16:43:35 -07004881int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004882 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004883 unsigned long size,
4884 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07004885{
4886 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004887 int ret;
4888 ret = zone_wait_table_init(zone, size);
4889 if (ret)
4890 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004891 pgdat->nr_zones = zone_idx(zone) + 1;
4892
Dave Hansened8ece22005-10-29 18:16:50 -07004893 zone->zone_start_pfn = zone_start_pfn;
4894
Mel Gorman708614e2008-07-23 21:26:51 -07004895 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4896 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4897 pgdat->node_id,
4898 (unsigned long)zone_idx(zone),
4899 zone_start_pfn, (zone_start_pfn + size));
4900
Andi Kleen1e548de2008-02-04 22:29:26 -08004901 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004902
4903 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004904}
4905
Tejun Heo0ee332c2011-12-08 10:22:09 -08004906#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004907#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
Mel Gorman8a942fd2015-06-30 14:56:55 -07004908
Mel Gormanc7132162006-09-27 01:49:43 -07004909/*
4910 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
Mel Gormanc7132162006-09-27 01:49:43 -07004911 */
Mel Gorman8a942fd2015-06-30 14:56:55 -07004912int __meminit __early_pfn_to_nid(unsigned long pfn,
4913 struct mminit_pfnnid_cache *state)
Mel Gormanc7132162006-09-27 01:49:43 -07004914{
Tejun Heoc13291a2011-07-12 10:46:30 +02004915 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004916 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004917
Mel Gorman8a942fd2015-06-30 14:56:55 -07004918 if (state->last_start <= pfn && pfn < state->last_end)
4919 return state->last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004920
Yinghai Lue76b63f2013-09-11 14:22:17 -07004921 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4922 if (nid != -1) {
Mel Gorman8a942fd2015-06-30 14:56:55 -07004923 state->last_start = start_pfn;
4924 state->last_end = end_pfn;
4925 state->last_nid = nid;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004926 }
4927
4928 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004929}
4930#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4931
Mel Gormanc7132162006-09-27 01:49:43 -07004932/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08004933 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004934 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08004935 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07004936 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004937 * If an architecture guarantees that all ranges registered contain no holes
4938 * and may be freed, this this function may be used instead of calling
4939 * memblock_free_early_nid() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004940 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004941void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004942{
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 start_pfn = min(start_pfn, max_low_pfn);
4948 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004949
Tejun Heoc13291a2011-07-12 10:46:30 +02004950 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004951 memblock_free_early_nid(PFN_PHYS(start_pfn),
4952 (end_pfn - start_pfn) << PAGE_SHIFT,
4953 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07004954 }
4955}
4956
4957/**
4958 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004959 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004960 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004961 * If an architecture guarantees that all ranges registered contain no holes and may
4962 * be freed, this function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004963 */
4964void __init sparse_memory_present_with_active_regions(int nid)
4965{
Tejun Heoc13291a2011-07-12 10:46:30 +02004966 unsigned long start_pfn, end_pfn;
4967 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004968
Tejun Heoc13291a2011-07-12 10:46:30 +02004969 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4970 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004971}
4972
4973/**
4974 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004975 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4976 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4977 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004978 *
4979 * It returns the start and end page frame of a node based on information
Zhang Zhen7d018172014-06-04 16:10:53 -07004980 * provided by memblock_set_node(). If called for a node
Mel Gormanc7132162006-09-27 01:49:43 -07004981 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004982 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004983 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004984void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004985 unsigned long *start_pfn, unsigned long *end_pfn)
4986{
Tejun Heoc13291a2011-07-12 10:46:30 +02004987 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004988 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02004989
Mel Gormanc7132162006-09-27 01:49:43 -07004990 *start_pfn = -1UL;
4991 *end_pfn = 0;
4992
Tejun Heoc13291a2011-07-12 10:46:30 +02004993 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4994 *start_pfn = min(*start_pfn, this_start_pfn);
4995 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004996 }
4997
Christoph Lameter633c0662007-10-16 01:25:37 -07004998 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004999 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005000}
5001
5002/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07005003 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5004 * assumption is made that zones within a node are ordered in monotonic
5005 * increasing memory addresses so that the "highest" populated zone is used
5006 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005007static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005008{
5009 int zone_index;
5010 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5011 if (zone_index == ZONE_MOVABLE)
5012 continue;
5013
5014 if (arch_zone_highest_possible_pfn[zone_index] >
5015 arch_zone_lowest_possible_pfn[zone_index])
5016 break;
5017 }
5018
5019 VM_BUG_ON(zone_index == -1);
5020 movable_zone = zone_index;
5021}
5022
5023/*
5024 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03005025 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005026 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5027 * in each node depending on the size of each node and how evenly kernelcore
5028 * is distributed. This helper function adjusts the zone ranges
5029 * provided by the architecture for a given node by using the end of the
5030 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5031 * zones within a node are in order of monotonic increases memory addresses
5032 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005033static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005034 unsigned long zone_type,
5035 unsigned long node_start_pfn,
5036 unsigned long node_end_pfn,
5037 unsigned long *zone_start_pfn,
5038 unsigned long *zone_end_pfn)
5039{
5040 /* Only adjust if ZONE_MOVABLE is on this node */
5041 if (zone_movable_pfn[nid]) {
5042 /* Size ZONE_MOVABLE */
5043 if (zone_type == ZONE_MOVABLE) {
5044 *zone_start_pfn = zone_movable_pfn[nid];
5045 *zone_end_pfn = min(node_end_pfn,
5046 arch_zone_highest_possible_pfn[movable_zone]);
5047
5048 /* Adjust for ZONE_MOVABLE starting within this range */
5049 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
5050 *zone_end_pfn > zone_movable_pfn[nid]) {
5051 *zone_end_pfn = zone_movable_pfn[nid];
5052
5053 /* Check if this whole range is within ZONE_MOVABLE */
5054 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5055 *zone_start_pfn = *zone_end_pfn;
5056 }
5057}
5058
5059/*
Mel Gormanc7132162006-09-27 01:49:43 -07005060 * Return the number of pages a zone spans in a node, including holes
5061 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5062 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005063static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005064 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005065 unsigned long node_start_pfn,
5066 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005067 unsigned long *ignored)
5068{
Mel Gormanc7132162006-09-27 01:49:43 -07005069 unsigned long zone_start_pfn, zone_end_pfn;
5070
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005071 /* When hotadd a new node, the node should be empty */
5072 if (!node_start_pfn && !node_end_pfn)
5073 return 0;
5074
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005075 /* Get the start and end of the zone */
Mel Gormanc7132162006-09-27 01:49:43 -07005076 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5077 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07005078 adjust_zone_range_for_zone_movable(nid, zone_type,
5079 node_start_pfn, node_end_pfn,
5080 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005081
5082 /* Check that this node has pages within the zone's required range */
5083 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
5084 return 0;
5085
5086 /* Move the zone boundaries inside the node if necessary */
5087 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
5088 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
5089
5090 /* Return the spanned pages */
5091 return zone_end_pfn - zone_start_pfn;
5092}
5093
5094/*
5095 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005096 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07005097 */
Yinghai Lu32996252009-12-15 17:59:02 -08005098unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005099 unsigned long range_start_pfn,
5100 unsigned long range_end_pfn)
5101{
Tejun Heo96e907d2011-07-12 10:46:29 +02005102 unsigned long nr_absent = range_end_pfn - range_start_pfn;
5103 unsigned long start_pfn, end_pfn;
5104 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07005105
Tejun Heo96e907d2011-07-12 10:46:29 +02005106 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5107 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5108 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5109 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005110 }
Tejun Heo96e907d2011-07-12 10:46:29 +02005111 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07005112}
5113
5114/**
5115 * absent_pages_in_range - Return number of page frames in holes within a range
5116 * @start_pfn: The start PFN to start searching for holes
5117 * @end_pfn: The end PFN to stop searching for holes
5118 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005119 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07005120 */
5121unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5122 unsigned long end_pfn)
5123{
5124 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5125}
5126
5127/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005128static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005129 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005130 unsigned long node_start_pfn,
5131 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005132 unsigned long *ignored)
5133{
Tejun Heo96e907d2011-07-12 10:46:29 +02005134 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5135 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07005136 unsigned long zone_start_pfn, zone_end_pfn;
5137
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005138 /* When hotadd a new node, the node should be empty */
5139 if (!node_start_pfn && !node_end_pfn)
5140 return 0;
5141
Tejun Heo96e907d2011-07-12 10:46:29 +02005142 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5143 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07005144
Mel Gorman2a1e2742007-07-17 04:03:12 -07005145 adjust_zone_range_for_zone_movable(nid, zone_type,
5146 node_start_pfn, node_end_pfn,
5147 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07005148 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005149}
Mel Gorman0e0b8642006-09-27 01:49:56 -07005150
Tejun Heo0ee332c2011-12-08 10:22:09 -08005151#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005152static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005153 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005154 unsigned long node_start_pfn,
5155 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005156 unsigned long *zones_size)
5157{
5158 return zones_size[zone_type];
5159}
5160
Paul Mundt6ea6e682007-07-15 23:38:20 -07005161static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005162 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005163 unsigned long node_start_pfn,
5164 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005165 unsigned long *zholes_size)
5166{
5167 if (!zholes_size)
5168 return 0;
5169
5170 return zholes_size[zone_type];
5171}
Yinghai Lu20e69262013-03-01 14:51:27 -08005172
Tejun Heo0ee332c2011-12-08 10:22:09 -08005173#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005174
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005175static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005176 unsigned long node_start_pfn,
5177 unsigned long node_end_pfn,
5178 unsigned long *zones_size,
5179 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07005180{
Gu Zhengfebd5942015-06-24 16:57:02 -07005181 unsigned long realtotalpages = 0, totalpages = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005182 enum zone_type i;
5183
Gu Zhengfebd5942015-06-24 16:57:02 -07005184 for (i = 0; i < MAX_NR_ZONES; i++) {
5185 struct zone *zone = pgdat->node_zones + i;
5186 unsigned long size, real_size;
Mel Gormanc7132162006-09-27 01:49:43 -07005187
Gu Zhengfebd5942015-06-24 16:57:02 -07005188 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5189 node_start_pfn,
5190 node_end_pfn,
5191 zones_size);
5192 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005193 node_start_pfn, node_end_pfn,
5194 zholes_size);
Gu Zhengfebd5942015-06-24 16:57:02 -07005195 zone->spanned_pages = size;
5196 zone->present_pages = real_size;
5197
5198 totalpages += size;
5199 realtotalpages += real_size;
5200 }
5201
5202 pgdat->node_spanned_pages = totalpages;
Mel Gormanc7132162006-09-27 01:49:43 -07005203 pgdat->node_present_pages = realtotalpages;
5204 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5205 realtotalpages);
5206}
5207
Mel Gorman835c1342007-10-16 01:25:47 -07005208#ifndef CONFIG_SPARSEMEM
5209/*
5210 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07005211 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5212 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07005213 * round what is now in bits to nearest long in bits, then return it in
5214 * bytes.
5215 */
Linus Torvalds7c455122013-02-18 09:58:02 -08005216static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005217{
5218 unsigned long usemapsize;
5219
Linus Torvalds7c455122013-02-18 09:58:02 -08005220 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005221 usemapsize = roundup(zonesize, pageblock_nr_pages);
5222 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07005223 usemapsize *= NR_PAGEBLOCK_BITS;
5224 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5225
5226 return usemapsize / 8;
5227}
5228
5229static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08005230 struct zone *zone,
5231 unsigned long zone_start_pfn,
5232 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005233{
Linus Torvalds7c455122013-02-18 09:58:02 -08005234 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07005235 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08005236 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005237 zone->pageblock_flags =
5238 memblock_virt_alloc_node_nopanic(usemapsize,
5239 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07005240}
5241#else
Linus Torvalds7c455122013-02-18 09:58:02 -08005242static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5243 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07005244#endif /* CONFIG_SPARSEMEM */
5245
Mel Gormand9c23402007-10-16 01:26:01 -07005246#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08005247
Mel Gormand9c23402007-10-16 01:26:01 -07005248/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07005249void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07005250{
Andrew Morton955c1cd2012-05-29 15:06:31 -07005251 unsigned int order;
5252
Mel Gormand9c23402007-10-16 01:26:01 -07005253 /* Check that pageblock_nr_pages has not already been setup */
5254 if (pageblock_order)
5255 return;
5256
Andrew Morton955c1cd2012-05-29 15:06:31 -07005257 if (HPAGE_SHIFT > PAGE_SHIFT)
5258 order = HUGETLB_PAGE_ORDER;
5259 else
5260 order = MAX_ORDER - 1;
5261
Mel Gormand9c23402007-10-16 01:26:01 -07005262 /*
5263 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07005264 * This value may be variable depending on boot parameters on IA64 and
5265 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07005266 */
5267 pageblock_order = order;
5268}
5269#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5270
Mel Gormanba72cb82007-11-28 16:21:13 -08005271/*
5272 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07005273 * is unused as pageblock_order is set at compile-time. See
5274 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5275 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08005276 */
Chen Gang15ca2202013-09-11 14:20:27 -07005277void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08005278{
Mel Gormanba72cb82007-11-28 16:21:13 -08005279}
Mel Gormand9c23402007-10-16 01:26:01 -07005280
5281#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5282
Jiang Liu01cefae2012-12-12 13:52:19 -08005283static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5284 unsigned long present_pages)
5285{
5286 unsigned long pages = spanned_pages;
5287
5288 /*
5289 * Provide a more accurate estimation if there are holes within
5290 * the zone and SPARSEMEM is in use. If there are holes within the
5291 * zone, each populated memory region may cost us one or two extra
5292 * memmap pages due to alignment because memmap pages for each
5293 * populated regions may not naturally algined on page boundary.
5294 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5295 */
5296 if (spanned_pages > present_pages + (present_pages >> 4) &&
5297 IS_ENABLED(CONFIG_SPARSEMEM))
5298 pages = present_pages;
5299
5300 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5301}
5302
Linus Torvalds1da177e2005-04-16 15:20:36 -07005303/*
5304 * Set up the zone data structures:
5305 * - mark all pages reserved
5306 * - mark all memory queues empty
5307 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07005308 *
5309 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005310 */
Wei Yang7f3eb552015-09-08 14:59:50 -07005311static void __paginginit free_area_init_core(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005312{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07005313 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07005314 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005315 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07005316 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005317
Dave Hansen208d54e2005-10-29 18:16:52 -07005318 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01005319#ifdef CONFIG_NUMA_BALANCING
5320 spin_lock_init(&pgdat->numabalancing_migrate_lock);
5321 pgdat->numabalancing_migrate_nr_pages = 0;
5322 pgdat->numabalancing_migrate_next_window = jiffies;
5323#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005324 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07005325 init_waitqueue_head(&pgdat->pfmemalloc_wait);
Joonsoo Kimeefa864b2014-12-12 16:55:46 -08005326 pgdat_page_ext_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01005327
Linus Torvalds1da177e2005-04-16 15:20:36 -07005328 for (j = 0; j < MAX_NR_ZONES; j++) {
5329 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08005330 unsigned long size, realsize, freesize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005331
Gu Zhengfebd5942015-06-24 16:57:02 -07005332 size = zone->spanned_pages;
5333 realsize = freesize = zone->present_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005334
Mel Gorman0e0b8642006-09-27 01:49:56 -07005335 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08005336 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07005337 * is used by this zone for memmap. This affects the watermark
5338 * and per-cpu initialisations
5339 */
Jiang Liu01cefae2012-12-12 13:52:19 -08005340 memmap_pages = calc_memmap_size(size, realsize);
Zhong Hongboba914f42014-12-12 16:56:21 -08005341 if (!is_highmem_idx(j)) {
5342 if (freesize >= memmap_pages) {
5343 freesize -= memmap_pages;
5344 if (memmap_pages)
5345 printk(KERN_DEBUG
5346 " %s zone: %lu pages used for memmap\n",
5347 zone_names[j], memmap_pages);
5348 } else
5349 printk(KERN_WARNING
5350 " %s zone: %lu pages exceeds freesize %lu\n",
5351 zone_names[j], memmap_pages, freesize);
5352 }
Mel Gorman0e0b8642006-09-27 01:49:56 -07005353
Christoph Lameter62672762007-02-10 01:43:07 -08005354 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08005355 if (j == 0 && freesize > dma_reserve) {
5356 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07005357 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08005358 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07005359 }
5360
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07005361 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08005362 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08005363 /* Charge for highmem memmap if there are enough kernel pages */
5364 else if (nr_kernel_pages > memmap_pages * 2)
5365 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08005366 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005367
Jiang Liu9feedc92012-12-12 13:52:12 -08005368 /*
5369 * Set an approximate value for lowmem here, it will be adjusted
5370 * when the bootmem allocator frees pages into the buddy system.
5371 * And all highmem pages will be managed by the buddy system.
5372 */
5373 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07005374#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07005375 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08005376 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07005377 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08005378 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07005379#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005380 zone->name = zone_names[j];
5381 spin_lock_init(&zone->lock);
5382 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07005383 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005384 zone->zone_pgdat = pgdat;
Dave Hansened8ece22005-10-29 18:16:50 -07005385 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005386
5387 /* For bootup, initialized properly in watermark setup */
5388 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
5389
Hugh Dickinsbea8c152012-11-16 14:14:54 -08005390 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005391 if (!size)
5392 continue;
5393
Andrew Morton955c1cd2012-05-29 15:06:31 -07005394 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08005395 setup_usemap(pgdat, zone, zone_start_pfn, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08005396 ret = init_currently_empty_zone(zone, zone_start_pfn,
5397 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07005398 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07005399 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005400 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005401 }
5402}
5403
Sam Ravnborg577a32f2007-05-17 23:29:25 +02005404static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005405{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005406 /* Skip empty nodes */
5407 if (!pgdat->node_spanned_pages)
5408 return;
5409
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005410#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07005411 /* ia64 gets its own node_mem_map, before this, without bootmem */
5412 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07005413 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005414 struct page *map;
5415
Bob Piccoe984bb42006-05-20 15:00:31 -07005416 /*
5417 * The zone's endpoints aren't required to be MAX_ORDER
5418 * aligned but the node_mem_map endpoints must be in order
5419 * for the buddy allocator to function correctly.
5420 */
5421 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
Cody P Schafer108bcc92013-02-22 16:35:23 -08005422 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07005423 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5424 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07005425 map = alloc_remap(pgdat->node_id, size);
5426 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005427 map = memblock_virt_alloc_node_nopanic(size,
5428 pgdat->node_id);
Bob Piccoe984bb42006-05-20 15:00:31 -07005429 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005430 }
Roman Zippel12d810c2007-05-31 00:40:54 -07005431#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07005432 /*
5433 * With no DISCONTIG, the global mem_map is just set as node 0's
5434 */
Mel Gormanc7132162006-09-27 01:49:43 -07005435 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005436 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08005437#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07005438 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08005439 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08005440#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005442#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005443#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005444}
5445
Johannes Weiner9109fb72008-07-23 21:27:20 -07005446void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5447 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005448{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005449 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005450 unsigned long start_pfn = 0;
5451 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07005452
Minchan Kim88fdf752012-07-31 16:46:14 -07005453 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07005454 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07005455
Mel Gorman3a80a7f2015-06-30 14:57:02 -07005456 reset_deferred_meminit(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005457 pgdat->node_id = nid;
5458 pgdat->node_start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005459#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5460 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
Juergen Gross8d29e182015-02-11 15:26:01 -08005461 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
5462 (u64)start_pfn << PAGE_SHIFT, ((u64)end_pfn << PAGE_SHIFT) - 1);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005463#endif
5464 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5465 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005466
5467 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07005468#ifdef CONFIG_FLAT_NODE_MEM_MAP
5469 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5470 nid, (unsigned long)pgdat,
5471 (unsigned long)pgdat->node_mem_map);
5472#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005473
Wei Yang7f3eb552015-09-08 14:59:50 -07005474 free_area_init_core(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005475}
5476
Tejun Heo0ee332c2011-12-08 10:22:09 -08005477#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07005478
5479#if MAX_NUMNODES > 1
5480/*
5481 * Figure out the number of possible node ids.
5482 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07005483void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07005484{
Wei Yang904a9552015-09-08 14:59:48 -07005485 unsigned int highest;
Miklos Szeredi418508c2007-05-23 13:57:55 -07005486
Wei Yang904a9552015-09-08 14:59:48 -07005487 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
Miklos Szeredi418508c2007-05-23 13:57:55 -07005488 nr_node_ids = highest + 1;
5489}
Miklos Szeredi418508c2007-05-23 13:57:55 -07005490#endif
5491
Mel Gormanc7132162006-09-27 01:49:43 -07005492/**
Tejun Heo1e019792011-07-12 09:45:34 +02005493 * node_map_pfn_alignment - determine the maximum internode alignment
5494 *
5495 * This function should be called after node map is populated and sorted.
5496 * It calculates the maximum power of two alignment which can distinguish
5497 * all the nodes.
5498 *
5499 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5500 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5501 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5502 * shifted, 1GiB is enough and this function will indicate so.
5503 *
5504 * This is used to test whether pfn -> nid mapping of the chosen memory
5505 * model has fine enough granularity to avoid incorrect mapping for the
5506 * populated node map.
5507 *
5508 * Returns the determined alignment in pfn's. 0 if there is no alignment
5509 * requirement (single node).
5510 */
5511unsigned long __init node_map_pfn_alignment(void)
5512{
5513 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005514 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02005515 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02005516 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02005517
Tejun Heoc13291a2011-07-12 10:46:30 +02005518 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02005519 if (!start || last_nid < 0 || last_nid == nid) {
5520 last_nid = nid;
5521 last_end = end;
5522 continue;
5523 }
5524
5525 /*
5526 * Start with a mask granular enough to pin-point to the
5527 * start pfn and tick off bits one-by-one until it becomes
5528 * too coarse to separate the current node from the last.
5529 */
5530 mask = ~((1 << __ffs(start)) - 1);
5531 while (mask && last_end <= (start & (mask << 1)))
5532 mask <<= 1;
5533
5534 /* accumulate all internode masks */
5535 accl_mask |= mask;
5536 }
5537
5538 /* convert mask to number of pages */
5539 return ~accl_mask + 1;
5540}
5541
Mel Gormana6af2bc2007-02-10 01:42:57 -08005542/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005543static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07005544{
Mel Gormana6af2bc2007-02-10 01:42:57 -08005545 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02005546 unsigned long start_pfn;
5547 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00005548
Tejun Heoc13291a2011-07-12 10:46:30 +02005549 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5550 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005551
Mel Gormana6af2bc2007-02-10 01:42:57 -08005552 if (min_pfn == ULONG_MAX) {
5553 printk(KERN_WARNING
Paul Jackson2bc0d2612008-06-22 07:22:17 -07005554 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08005555 return 0;
5556 }
5557
5558 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005559}
5560
5561/**
5562 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5563 *
5564 * It returns the minimum PFN based on information provided via
Zhang Zhen7d018172014-06-04 16:10:53 -07005565 * memblock_set_node().
Mel Gormanc7132162006-09-27 01:49:43 -07005566 */
5567unsigned long __init find_min_pfn_with_active_regions(void)
5568{
5569 return find_min_pfn_for_node(MAX_NUMNODES);
5570}
5571
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005572/*
5573 * early_calculate_totalpages()
5574 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005575 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005576 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07005577static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07005578{
Mel Gorman7e63efe2007-07-17 04:03:15 -07005579 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005580 unsigned long start_pfn, end_pfn;
5581 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005582
Tejun Heoc13291a2011-07-12 10:46:30 +02005583 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5584 unsigned long pages = end_pfn - start_pfn;
5585
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005586 totalpages += pages;
5587 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005588 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005589 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07005590 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005591}
5592
Mel Gorman2a1e2742007-07-17 04:03:12 -07005593/*
5594 * Find the PFN the Movable zone begins in each node. Kernel memory
5595 * is spread evenly between nodes as long as the nodes have enough
5596 * memory. When they don't, some nodes will have more kernelcore than
5597 * others
5598 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07005599static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005600{
5601 int i, nid;
5602 unsigned long usable_startpfn;
5603 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07005604 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005605 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005606 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005607 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07005608 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005609
5610 /* Need to find movable_zone earlier when movable_node is specified. */
5611 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07005612
Mel Gorman7e63efe2007-07-17 04:03:15 -07005613 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005614 * If movable_node is specified, ignore kernelcore and movablecore
5615 * options.
5616 */
5617 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07005618 for_each_memblock(memory, r) {
5619 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005620 continue;
5621
Emil Medve136199f2014-04-07 15:37:52 -07005622 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005623
Emil Medve136199f2014-04-07 15:37:52 -07005624 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005625 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5626 min(usable_startpfn, zone_movable_pfn[nid]) :
5627 usable_startpfn;
5628 }
5629
5630 goto out2;
5631 }
5632
5633 /*
5634 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07005635 * kernelcore that corresponds so that memory usable for
5636 * any allocation type is evenly spread. If both kernelcore
5637 * and movablecore are specified, then the value of kernelcore
5638 * will be used for required_kernelcore if it's greater than
5639 * what movablecore would have allowed.
5640 */
5641 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07005642 unsigned long corepages;
5643
5644 /*
5645 * Round-up so that ZONE_MOVABLE is at least as large as what
5646 * was requested by the user
5647 */
5648 required_movablecore =
5649 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5650 corepages = totalpages - required_movablecore;
5651
5652 required_kernelcore = max(required_kernelcore, corepages);
5653 }
5654
Yinghai Lu20e69262013-03-01 14:51:27 -08005655 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5656 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07005657 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005658
5659 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005660 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5661
5662restart:
5663 /* Spread kernelcore memory as evenly as possible throughout nodes */
5664 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005665 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02005666 unsigned long start_pfn, end_pfn;
5667
Mel Gorman2a1e2742007-07-17 04:03:12 -07005668 /*
5669 * Recalculate kernelcore_node if the division per node
5670 * now exceeds what is necessary to satisfy the requested
5671 * amount of memory for the kernel
5672 */
5673 if (required_kernelcore < kernelcore_node)
5674 kernelcore_node = required_kernelcore / usable_nodes;
5675
5676 /*
5677 * As the map is walked, we track how much memory is usable
5678 * by the kernel using kernelcore_remaining. When it is
5679 * 0, the rest of the node is usable by ZONE_MOVABLE
5680 */
5681 kernelcore_remaining = kernelcore_node;
5682
5683 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02005684 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005685 unsigned long size_pages;
5686
Tejun Heoc13291a2011-07-12 10:46:30 +02005687 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005688 if (start_pfn >= end_pfn)
5689 continue;
5690
5691 /* Account for what is only usable for kernelcore */
5692 if (start_pfn < usable_startpfn) {
5693 unsigned long kernel_pages;
5694 kernel_pages = min(end_pfn, usable_startpfn)
5695 - start_pfn;
5696
5697 kernelcore_remaining -= min(kernel_pages,
5698 kernelcore_remaining);
5699 required_kernelcore -= min(kernel_pages,
5700 required_kernelcore);
5701
5702 /* Continue if range is now fully accounted */
5703 if (end_pfn <= usable_startpfn) {
5704
5705 /*
5706 * Push zone_movable_pfn to the end so
5707 * that if we have to rebalance
5708 * kernelcore across nodes, we will
5709 * not double account here
5710 */
5711 zone_movable_pfn[nid] = end_pfn;
5712 continue;
5713 }
5714 start_pfn = usable_startpfn;
5715 }
5716
5717 /*
5718 * The usable PFN range for ZONE_MOVABLE is from
5719 * start_pfn->end_pfn. Calculate size_pages as the
5720 * number of pages used as kernelcore
5721 */
5722 size_pages = end_pfn - start_pfn;
5723 if (size_pages > kernelcore_remaining)
5724 size_pages = kernelcore_remaining;
5725 zone_movable_pfn[nid] = start_pfn + size_pages;
5726
5727 /*
5728 * Some kernelcore has been met, update counts and
5729 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07005730 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005731 */
5732 required_kernelcore -= min(required_kernelcore,
5733 size_pages);
5734 kernelcore_remaining -= size_pages;
5735 if (!kernelcore_remaining)
5736 break;
5737 }
5738 }
5739
5740 /*
5741 * If there is still required_kernelcore, we do another pass with one
5742 * less node in the count. This will push zone_movable_pfn[nid] further
5743 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07005744 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005745 */
5746 usable_nodes--;
5747 if (usable_nodes && required_kernelcore > usable_nodes)
5748 goto restart;
5749
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005750out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07005751 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5752 for (nid = 0; nid < MAX_NUMNODES; nid++)
5753 zone_movable_pfn[nid] =
5754 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005755
Yinghai Lu20e69262013-03-01 14:51:27 -08005756out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005757 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005758 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005759}
5760
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005761/* Any regular or high memory on that node ? */
5762static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005763{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005764 enum zone_type zone_type;
5765
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005766 if (N_MEMORY == N_NORMAL_MEMORY)
5767 return;
5768
5769 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005770 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08005771 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005772 node_set_state(nid, N_HIGH_MEMORY);
5773 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5774 zone_type <= ZONE_NORMAL)
5775 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005776 break;
5777 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005778 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005779}
5780
Mel Gormanc7132162006-09-27 01:49:43 -07005781/**
5782 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005783 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005784 *
5785 * This will call free_area_init_node() for each active node in the system.
Zhang Zhen7d018172014-06-04 16:10:53 -07005786 * Using the page ranges provided by memblock_set_node(), the size of each
Mel Gormanc7132162006-09-27 01:49:43 -07005787 * zone in each node and their holes is calculated. If the maximum PFN
5788 * between two adjacent zones match, it is assumed that the zone is empty.
5789 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5790 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5791 * starts where the previous one ended. For example, ZONE_DMA32 starts
5792 * at arch_max_dma_pfn.
5793 */
5794void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5795{
Tejun Heoc13291a2011-07-12 10:46:30 +02005796 unsigned long start_pfn, end_pfn;
5797 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005798
Mel Gormanc7132162006-09-27 01:49:43 -07005799 /* Record where the zone boundaries are */
5800 memset(arch_zone_lowest_possible_pfn, 0,
5801 sizeof(arch_zone_lowest_possible_pfn));
5802 memset(arch_zone_highest_possible_pfn, 0,
5803 sizeof(arch_zone_highest_possible_pfn));
5804 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5805 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5806 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005807 if (i == ZONE_MOVABLE)
5808 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005809 arch_zone_lowest_possible_pfn[i] =
5810 arch_zone_highest_possible_pfn[i-1];
5811 arch_zone_highest_possible_pfn[i] =
5812 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5813 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005814 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5815 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5816
5817 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5818 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005819 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005820
Mel Gormanc7132162006-09-27 01:49:43 -07005821 /* Print out the zone ranges */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005822 pr_info("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005823 for (i = 0; i < MAX_NR_ZONES; i++) {
5824 if (i == ZONE_MOVABLE)
5825 continue;
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005826 pr_info(" %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08005827 if (arch_zone_lowest_possible_pfn[i] ==
5828 arch_zone_highest_possible_pfn[i])
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005829 pr_cont("empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08005830 else
Juergen Gross8d29e182015-02-11 15:26:01 -08005831 pr_cont("[mem %#018Lx-%#018Lx]\n",
5832 (u64)arch_zone_lowest_possible_pfn[i]
5833 << PAGE_SHIFT,
5834 ((u64)arch_zone_highest_possible_pfn[i]
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005835 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005836 }
5837
5838 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005839 pr_info("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005840 for (i = 0; i < MAX_NUMNODES; i++) {
5841 if (zone_movable_pfn[i])
Juergen Gross8d29e182015-02-11 15:26:01 -08005842 pr_info(" Node %d: %#018Lx\n", i,
5843 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005844 }
Mel Gormanc7132162006-09-27 01:49:43 -07005845
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07005846 /* Print out the early node map */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005847 pr_info("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02005848 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Juergen Gross8d29e182015-02-11 15:26:01 -08005849 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
5850 (u64)start_pfn << PAGE_SHIFT,
5851 ((u64)end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07005852
5853 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005854 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005855 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005856 for_each_online_node(nid) {
5857 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005858 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005859 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005860
5861 /* Any memory on that node */
5862 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005863 node_set_state(nid, N_MEMORY);
5864 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005865 }
5866}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005867
Mel Gorman7e63efe2007-07-17 04:03:15 -07005868static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005869{
5870 unsigned long long coremem;
5871 if (!p)
5872 return -EINVAL;
5873
5874 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005875 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005876
Mel Gorman7e63efe2007-07-17 04:03:15 -07005877 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005878 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5879
5880 return 0;
5881}
Mel Gormaned7ed362007-07-17 04:03:14 -07005882
Mel Gorman7e63efe2007-07-17 04:03:15 -07005883/*
5884 * kernelcore=size sets the amount of memory for use for allocations that
5885 * cannot be reclaimed or migrated.
5886 */
5887static int __init cmdline_parse_kernelcore(char *p)
5888{
5889 return cmdline_parse_core(p, &required_kernelcore);
5890}
5891
5892/*
5893 * movablecore=size sets the amount of memory for use for allocations that
5894 * can be reclaimed or migrated.
5895 */
5896static int __init cmdline_parse_movablecore(char *p)
5897{
5898 return cmdline_parse_core(p, &required_movablecore);
5899}
5900
Mel Gormaned7ed362007-07-17 04:03:14 -07005901early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005902early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005903
Tejun Heo0ee332c2011-12-08 10:22:09 -08005904#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005905
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005906void adjust_managed_page_count(struct page *page, long count)
5907{
5908 spin_lock(&managed_page_count_lock);
5909 page_zone(page)->managed_pages += count;
5910 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07005911#ifdef CONFIG_HIGHMEM
5912 if (PageHighMem(page))
5913 totalhigh_pages += count;
5914#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005915 spin_unlock(&managed_page_count_lock);
5916}
Jiang Liu3dcc0572013-07-03 15:03:21 -07005917EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005918
Jiang Liu11199692013-07-03 15:02:48 -07005919unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07005920{
Jiang Liu11199692013-07-03 15:02:48 -07005921 void *pos;
5922 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07005923
Jiang Liu11199692013-07-03 15:02:48 -07005924 start = (void *)PAGE_ALIGN((unsigned long)start);
5925 end = (void *)((unsigned long)end & PAGE_MASK);
5926 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07005927 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07005928 memset(pos, poison, PAGE_SIZE);
5929 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07005930 }
5931
5932 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07005933 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07005934 s, pages << (PAGE_SHIFT - 10), start, end);
5935
5936 return pages;
5937}
Jiang Liu11199692013-07-03 15:02:48 -07005938EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07005939
Jiang Liucfa11e02013-04-29 15:07:00 -07005940#ifdef CONFIG_HIGHMEM
5941void free_highmem_page(struct page *page)
5942{
5943 __free_reserved_page(page);
5944 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07005945 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07005946 totalhigh_pages++;
5947}
5948#endif
5949
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005950
5951void __init mem_init_print_info(const char *str)
5952{
5953 unsigned long physpages, codesize, datasize, rosize, bss_size;
5954 unsigned long init_code_size, init_data_size;
5955
5956 physpages = get_num_physpages();
5957 codesize = _etext - _stext;
5958 datasize = _edata - _sdata;
5959 rosize = __end_rodata - __start_rodata;
5960 bss_size = __bss_stop - __bss_start;
5961 init_data_size = __init_end - __init_begin;
5962 init_code_size = _einittext - _sinittext;
5963
5964 /*
5965 * Detect special cases and adjust section sizes accordingly:
5966 * 1) .init.* may be embedded into .data sections
5967 * 2) .init.text.* may be out of [__init_begin, __init_end],
5968 * please refer to arch/tile/kernel/vmlinux.lds.S.
5969 * 3) .rodata.* may be embedded into .text or .data sections.
5970 */
5971#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07005972 do { \
5973 if (start <= pos && pos < end && size > adj) \
5974 size -= adj; \
5975 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005976
5977 adj_init_size(__init_begin, __init_end, init_data_size,
5978 _sinittext, init_code_size);
5979 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5980 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5981 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5982 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5983
5984#undef adj_init_size
5985
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005986 pr_info("Memory: %luK/%luK available "
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005987 "(%luK kernel code, %luK rwdata, %luK rodata, "
Pintu Kumare48322a2014-12-18 16:17:15 -08005988 "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005989#ifdef CONFIG_HIGHMEM
5990 ", %luK highmem"
5991#endif
5992 "%s%s)\n",
5993 nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5994 codesize >> 10, datasize >> 10, rosize >> 10,
5995 (init_data_size + init_code_size) >> 10, bss_size >> 10,
Pintu Kumare48322a2014-12-18 16:17:15 -08005996 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10),
5997 totalcma_pages << (PAGE_SHIFT-10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005998#ifdef CONFIG_HIGHMEM
5999 totalhigh_pages << (PAGE_SHIFT-10),
6000#endif
6001 str ? ", " : "", str ? str : "");
6002}
6003
Mel Gorman0e0b8642006-09-27 01:49:56 -07006004/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006005 * set_dma_reserve - set the specified number of pages reserved in the first zone
6006 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07006007 *
6008 * The per-cpu batchsize and zone watermarks are determined by present_pages.
6009 * In the DMA zone, a significant percentage may be consumed by kernel image
6010 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006011 * function may optionally be used to account for unfreeable pages in the
6012 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6013 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07006014 */
6015void __init set_dma_reserve(unsigned long new_dma_reserve)
6016{
6017 dma_reserve = new_dma_reserve;
6018}
6019
Linus Torvalds1da177e2005-04-16 15:20:36 -07006020void __init free_area_init(unsigned long *zones_size)
6021{
Johannes Weiner9109fb72008-07-23 21:27:20 -07006022 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006023 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
6024}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006025
Linus Torvalds1da177e2005-04-16 15:20:36 -07006026static int page_alloc_cpu_notify(struct notifier_block *self,
6027 unsigned long action, void *hcpu)
6028{
6029 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006030
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07006031 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07006032 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006033 drain_pages(cpu);
6034
6035 /*
6036 * Spill the event counters of the dead processor
6037 * into the current processors event counters.
6038 * This artificially elevates the count of the current
6039 * processor.
6040 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07006041 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006042
6043 /*
6044 * Zero the differential counters of the dead processor
6045 * so that the vm statistics are consistent.
6046 *
6047 * This is only okay since the processor is dead and cannot
6048 * race with what we are doing.
6049 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07006050 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006051 }
6052 return NOTIFY_OK;
6053}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006054
6055void __init page_alloc_init(void)
6056{
6057 hotcpu_notifier(page_alloc_cpu_notify, 0);
6058}
6059
6060/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006061 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
6062 * or min_free_kbytes changes.
6063 */
6064static void calculate_totalreserve_pages(void)
6065{
6066 struct pglist_data *pgdat;
6067 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006068 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006069
6070 for_each_online_pgdat(pgdat) {
6071 for (i = 0; i < MAX_NR_ZONES; i++) {
6072 struct zone *zone = pgdat->node_zones + i;
Mel Gorman3484b2d2014-08-06 16:07:14 -07006073 long max = 0;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006074
6075 /* Find valid and maximum lowmem_reserve in the zone */
6076 for (j = i; j < MAX_NR_ZONES; j++) {
6077 if (zone->lowmem_reserve[j] > max)
6078 max = zone->lowmem_reserve[j];
6079 }
6080
Mel Gorman41858962009-06-16 15:32:12 -07006081 /* we treat the high watermark as reserved pages. */
6082 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006083
Jiang Liub40da042013-02-22 16:33:52 -08006084 if (max > zone->managed_pages)
6085 max = zone->managed_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006086 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08006087 /*
6088 * Lowmem reserves are not available to
6089 * GFP_HIGHUSER page cache allocations and
6090 * kswapd tries to balance zones to their high
6091 * watermark. As a result, neither should be
6092 * regarded as dirtyable memory, to prevent a
6093 * situation where reclaim has to clean pages
6094 * in order to balance the zones.
6095 */
6096 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006097 }
6098 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08006099 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006100 totalreserve_pages = reserve_pages;
6101}
6102
6103/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07006104 * setup_per_zone_lowmem_reserve - called whenever
6105 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
6106 * has a correct pages reserved value, so an adequate number of
6107 * pages are left in the zone after a successful __alloc_pages().
6108 */
6109static void setup_per_zone_lowmem_reserve(void)
6110{
6111 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006112 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006113
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08006114 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006115 for (j = 0; j < MAX_NR_ZONES; j++) {
6116 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08006117 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006118
6119 zone->lowmem_reserve[j] = 0;
6120
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006121 idx = j;
6122 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006123 struct zone *lower_zone;
6124
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006125 idx--;
6126
Linus Torvalds1da177e2005-04-16 15:20:36 -07006127 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6128 sysctl_lowmem_reserve_ratio[idx] = 1;
6129
6130 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08006131 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07006132 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08006133 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006134 }
6135 }
6136 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006137
6138 /* update totalreserve_pages */
6139 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006140}
6141
Mel Gormancfd3da12011-04-25 21:36:42 +00006142static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006143{
6144 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6145 unsigned long lowmem_pages = 0;
6146 struct zone *zone;
6147 unsigned long flags;
6148
6149 /* Calculate total number of !ZONE_HIGHMEM pages */
6150 for_each_zone(zone) {
6151 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08006152 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006153 }
6154
6155 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07006156 u64 tmp;
6157
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006158 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08006159 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07006160 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006161 if (is_highmem(zone)) {
6162 /*
Nick Piggin669ed172005-11-13 16:06:45 -08006163 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6164 * need highmem pages, so cap pages_min to a small
6165 * value here.
6166 *
Mel Gorman41858962009-06-16 15:32:12 -07006167 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Yaowei Bai42ff2702015-04-14 15:47:14 -07006168 * deltas control asynch page reclaim, and so should
Nick Piggin669ed172005-11-13 16:06:45 -08006169 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006170 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08006171 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006172
Jiang Liub40da042013-02-22 16:33:52 -08006173 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08006174 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07006175 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006176 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08006177 /*
6178 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07006179 * proportionate to the zone's size.
6180 */
Mel Gorman41858962009-06-16 15:32:12 -07006181 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006182 }
6183
Mel Gorman41858962009-06-16 15:32:12 -07006184 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
6185 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006186
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006187 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
Johannes Weinerabe5f972014-10-02 16:21:10 -07006188 high_wmark_pages(zone) - low_wmark_pages(zone) -
6189 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006190
Mel Gorman56fd56b2007-10-16 01:25:58 -07006191 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006192 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006193 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006194
6195 /* update totalreserve_pages */
6196 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006197}
6198
Mel Gormancfd3da12011-04-25 21:36:42 +00006199/**
6200 * setup_per_zone_wmarks - called when min_free_kbytes changes
6201 * or when memory is hot-{added|removed}
6202 *
6203 * Ensures that the watermark[min,low,high] values for each zone are set
6204 * correctly with respect to min_free_kbytes.
6205 */
6206void setup_per_zone_wmarks(void)
6207{
6208 mutex_lock(&zonelists_mutex);
6209 __setup_per_zone_wmarks();
6210 mutex_unlock(&zonelists_mutex);
6211}
6212
Randy Dunlap55a44622009-09-21 17:01:20 -07006213/*
Rik van Riel556adec2008-10-18 20:26:34 -07006214 * The inactive anon list should be small enough that the VM never has to
6215 * do too much work, but large enough that each inactive page has a chance
6216 * to be referenced again before it is swapped out.
6217 *
6218 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6219 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6220 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6221 * the anonymous pages are kept on the inactive list.
6222 *
6223 * total target max
6224 * memory ratio inactive anon
6225 * -------------------------------------
6226 * 10MB 1 5MB
6227 * 100MB 1 50MB
6228 * 1GB 3 250MB
6229 * 10GB 10 0.9GB
6230 * 100GB 31 3GB
6231 * 1TB 101 10GB
6232 * 10TB 320 32GB
6233 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07006234static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07006235{
6236 unsigned int gb, ratio;
6237
6238 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08006239 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07006240 if (gb)
6241 ratio = int_sqrt(10 * gb);
6242 else
6243 ratio = 1;
6244
6245 zone->inactive_ratio = ratio;
6246}
6247
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07006248static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07006249{
6250 struct zone *zone;
6251
Minchan Kim96cb4df2009-06-16 15:32:49 -07006252 for_each_zone(zone)
6253 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07006254}
6255
Linus Torvalds1da177e2005-04-16 15:20:36 -07006256/*
6257 * Initialise min_free_kbytes.
6258 *
6259 * For small machines we want it small (128k min). For large machines
6260 * we want it large (64MB max). But it is not linear, because network
6261 * bandwidth does not increase linearly with machine size. We use
6262 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006263 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07006264 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6265 *
6266 * which yields
6267 *
6268 * 16MB: 512k
6269 * 32MB: 724k
6270 * 64MB: 1024k
6271 * 128MB: 1448k
6272 * 256MB: 2048k
6273 * 512MB: 2896k
6274 * 1024MB: 4096k
6275 * 2048MB: 5792k
6276 * 4096MB: 8192k
6277 * 8192MB: 11584k
6278 * 16384MB: 16384k
6279 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07006280int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006281{
6282 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07006283 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006284
6285 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07006286 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006287
Michal Hocko5f127332013-07-08 16:00:40 -07006288 if (new_min_free_kbytes > user_min_free_kbytes) {
6289 min_free_kbytes = new_min_free_kbytes;
6290 if (min_free_kbytes < 128)
6291 min_free_kbytes = 128;
6292 if (min_free_kbytes > 65536)
6293 min_free_kbytes = 65536;
6294 } else {
6295 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6296 new_min_free_kbytes, user_min_free_kbytes);
6297 }
Minchan Kimbc75d332009-06-16 15:32:48 -07006298 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07006299 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006300 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07006301 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006302 return 0;
6303}
Minchan Kimbc75d332009-06-16 15:32:48 -07006304module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006305
6306/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07006307 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07006308 * that we can call two helper functions whenever min_free_kbytes
6309 * changes.
6310 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006311int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006312 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006313{
Han Pingtianda8c7572014-01-23 15:53:17 -08006314 int rc;
6315
6316 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6317 if (rc)
6318 return rc;
6319
Michal Hocko5f127332013-07-08 16:00:40 -07006320 if (write) {
6321 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07006322 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07006323 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006324 return 0;
6325}
6326
Christoph Lameter96146342006-07-03 00:24:13 -07006327#ifdef CONFIG_NUMA
Joe Perchescccad5b2014-06-06 14:38:09 -07006328int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006329 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07006330{
6331 struct zone *zone;
6332 int rc;
6333
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006334 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07006335 if (rc)
6336 return rc;
6337
6338 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08006339 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07006340 sysctl_min_unmapped_ratio) / 100;
6341 return 0;
6342}
Christoph Lameter0ff38492006-09-25 23:31:52 -07006343
Joe Perchescccad5b2014-06-06 14:38:09 -07006344int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006345 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07006346{
6347 struct zone *zone;
6348 int rc;
6349
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006350 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07006351 if (rc)
6352 return rc;
6353
6354 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08006355 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07006356 sysctl_min_slab_ratio) / 100;
6357 return 0;
6358}
Christoph Lameter96146342006-07-03 00:24:13 -07006359#endif
6360
Linus Torvalds1da177e2005-04-16 15:20:36 -07006361/*
6362 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6363 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6364 * whenever sysctl_lowmem_reserve_ratio changes.
6365 *
6366 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07006367 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07006368 * if in function of the boot time zone sizes.
6369 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006370int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006371 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006372{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006373 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006374 setup_per_zone_lowmem_reserve();
6375 return 0;
6376}
6377
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006378/*
6379 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07006380 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6381 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006382 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006383int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006384 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006385{
6386 struct zone *zone;
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006387 int old_percpu_pagelist_fraction;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006388 int ret;
6389
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006390 mutex_lock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006391 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6392
6393 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
6394 if (!write || ret < 0)
6395 goto out;
6396
6397 /* Sanity checking to avoid pcp imbalance */
6398 if (percpu_pagelist_fraction &&
6399 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6400 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6401 ret = -EINVAL;
6402 goto out;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006403 }
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006404
6405 /* No change? */
6406 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6407 goto out;
6408
6409 for_each_populated_zone(zone) {
6410 unsigned int cpu;
6411
6412 for_each_possible_cpu(cpu)
6413 pageset_set_high_and_batch(zone,
6414 per_cpu_ptr(zone->pageset, cpu));
6415 }
6416out:
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006417 mutex_unlock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006418 return ret;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006419}
6420
Rasmus Villemoesa9919c72015-06-24 16:56:28 -07006421#ifdef CONFIG_NUMA
David S. Millerf034b5d2006-08-24 03:08:07 -07006422int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006423
Linus Torvalds1da177e2005-04-16 15:20:36 -07006424static int __init set_hashdist(char *str)
6425{
6426 if (!str)
6427 return 0;
6428 hashdist = simple_strtoul(str, &str, 0);
6429 return 1;
6430}
6431__setup("hashdist=", set_hashdist);
6432#endif
6433
6434/*
6435 * allocate a large system hash table from bootmem
6436 * - it is assumed that the hash table must contain an exact power-of-2
6437 * quantity of entries
6438 * - limit is the number of hash buckets, not the total allocation size
6439 */
6440void *__init alloc_large_system_hash(const char *tablename,
6441 unsigned long bucketsize,
6442 unsigned long numentries,
6443 int scale,
6444 int flags,
6445 unsigned int *_hash_shift,
6446 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00006447 unsigned long low_limit,
6448 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006449{
Tim Bird31fe62b2012-05-23 13:33:35 +00006450 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006451 unsigned long log2qty, size;
6452 void *table = NULL;
6453
6454 /* allow the kernel cmdline to have a say */
6455 if (!numentries) {
6456 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08006457 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07006458
6459 /* It isn't necessary when PAGE_SIZE >= 1MB */
6460 if (PAGE_SHIFT < 20)
6461 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006462
6463 /* limit to 1 bucket per 2^scale bytes of low memory */
6464 if (scale > PAGE_SHIFT)
6465 numentries >>= (scale - PAGE_SHIFT);
6466 else
6467 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08006468
6469 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07006470 if (unlikely(flags & HASH_SMALL)) {
6471 /* Makes no sense without HASH_EARLY */
6472 WARN_ON(!(flags & HASH_EARLY));
6473 if (!(numentries >> *_hash_shift)) {
6474 numentries = 1UL << *_hash_shift;
6475 BUG_ON(!numentries);
6476 }
6477 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08006478 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006479 }
John Hawkes6e692ed2006-03-25 03:08:02 -08006480 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006481
6482 /* limit allocation size to 1/16 total memory by default */
6483 if (max == 0) {
6484 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6485 do_div(max, bucketsize);
6486 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08006487 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006488
Tim Bird31fe62b2012-05-23 13:33:35 +00006489 if (numentries < low_limit)
6490 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006491 if (numentries > max)
6492 numentries = max;
6493
David Howellsf0d1b0b2006-12-08 02:37:49 -08006494 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006495
6496 do {
6497 size = bucketsize << log2qty;
6498 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08006499 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006500 else if (hashdist)
6501 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6502 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07006503 /*
6504 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07006505 * some pages at the end of hash table which
6506 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07006507 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006508 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07006509 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006510 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6511 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006512 }
6513 } while (!table && size > PAGE_SIZE && --log2qty);
6514
6515 if (!table)
6516 panic("Failed to allocate %s hash table\n", tablename);
6517
Robin Holtf241e6602010-10-07 12:59:26 -07006518 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07006519 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07006520 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08006521 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006522 size);
6523
6524 if (_hash_shift)
6525 *_hash_shift = log2qty;
6526 if (_hash_mask)
6527 *_hash_mask = (1 << log2qty) - 1;
6528
6529 return table;
6530}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08006531
Mel Gorman835c1342007-10-16 01:25:47 -07006532/* Return a pointer to the bitmap storing bits affecting a block of pages */
6533static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6534 unsigned long pfn)
6535{
6536#ifdef CONFIG_SPARSEMEM
6537 return __pfn_to_section(pfn)->pageblock_flags;
6538#else
6539 return zone->pageblock_flags;
6540#endif /* CONFIG_SPARSEMEM */
6541}
Andrew Morton6220ec72006-10-19 23:29:05 -07006542
Mel Gorman835c1342007-10-16 01:25:47 -07006543static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6544{
6545#ifdef CONFIG_SPARSEMEM
6546 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07006547 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006548#else
Laura Abbottc060f942013-01-11 14:31:51 -08006549 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07006550 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006551#endif /* CONFIG_SPARSEMEM */
6552}
6553
6554/**
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006555 * 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 -07006556 * @page: The page within the block of interest
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006557 * @pfn: The target page frame number
6558 * @end_bitidx: The last bit of interest to retrieve
6559 * @mask: mask of bits that the caller is interested in
6560 *
6561 * Return: pageblock_bits flags
Mel Gorman835c1342007-10-16 01:25:47 -07006562 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006563unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006564 unsigned long end_bitidx,
6565 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006566{
6567 struct zone *zone;
6568 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006569 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006570 unsigned long word;
Mel Gorman835c1342007-10-16 01:25:47 -07006571
6572 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006573 bitmap = get_pageblock_bitmap(zone, pfn);
6574 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006575 word_bitidx = bitidx / BITS_PER_LONG;
6576 bitidx &= (BITS_PER_LONG-1);
Mel Gorman835c1342007-10-16 01:25:47 -07006577
Mel Gormane58469b2014-06-04 16:10:16 -07006578 word = bitmap[word_bitidx];
6579 bitidx += end_bitidx;
6580 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
Mel Gorman835c1342007-10-16 01:25:47 -07006581}
6582
6583/**
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006584 * 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 -07006585 * @page: The page within the block of interest
Mel Gorman835c1342007-10-16 01:25:47 -07006586 * @flags: The flags to set
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006587 * @pfn: The target page frame number
6588 * @end_bitidx: The last bit of interest
6589 * @mask: mask of bits that the caller is interested in
Mel Gorman835c1342007-10-16 01:25:47 -07006590 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006591void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6592 unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006593 unsigned long end_bitidx,
6594 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006595{
6596 struct zone *zone;
6597 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006598 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006599 unsigned long old_word, word;
6600
6601 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
Mel Gorman835c1342007-10-16 01:25:47 -07006602
6603 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006604 bitmap = get_pageblock_bitmap(zone, pfn);
6605 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006606 word_bitidx = bitidx / BITS_PER_LONG;
6607 bitidx &= (BITS_PER_LONG-1);
6608
Sasha Levin309381fea2014-01-23 15:52:54 -08006609 VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
Mel Gorman835c1342007-10-16 01:25:47 -07006610
Mel Gormane58469b2014-06-04 16:10:16 -07006611 bitidx += end_bitidx;
6612 mask <<= (BITS_PER_LONG - bitidx - 1);
6613 flags <<= (BITS_PER_LONG - bitidx - 1);
6614
Jason Low4db0c3c2015-04-15 16:14:08 -07006615 word = READ_ONCE(bitmap[word_bitidx]);
Mel Gormane58469b2014-06-04 16:10:16 -07006616 for (;;) {
6617 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6618 if (word == old_word)
6619 break;
6620 word = old_word;
6621 }
Mel Gorman835c1342007-10-16 01:25:47 -07006622}
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07006623
6624/*
Minchan Kim80934512012-07-31 16:43:01 -07006625 * This function checks whether pageblock includes unmovable pages or not.
6626 * If @count is not zero, it is okay to include less @count unmovable pages
6627 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006628 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07006629 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6630 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07006631 */
Wen Congyangb023f462012-12-11 16:00:45 -08006632bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6633 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006634{
6635 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006636 int mt;
6637
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006638 /*
6639 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07006640 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006641 */
6642 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07006643 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006644 mt = get_pageblock_migratetype(page);
6645 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07006646 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006647
6648 pfn = page_to_pfn(page);
6649 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6650 unsigned long check = pfn + iter;
6651
Namhyung Kim29723fc2011-02-25 14:44:25 -08006652 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006653 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08006654
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006655 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07006656
6657 /*
6658 * Hugepages are not in LRU lists, but they're movable.
6659 * We need not scan over tail pages bacause we don't
6660 * handle each tail page individually in migration.
6661 */
6662 if (PageHuge(page)) {
6663 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6664 continue;
6665 }
6666
Minchan Kim97d255c2012-07-31 16:42:59 -07006667 /*
6668 * We can't use page_count without pin a page
6669 * because another CPU can free compound page.
6670 * This check already skips compound tails of THP
6671 * because their page->_count is zero at all time.
6672 */
6673 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006674 if (PageBuddy(page))
6675 iter += (1 << page_order(page)) - 1;
6676 continue;
6677 }
Minchan Kim97d255c2012-07-31 16:42:59 -07006678
Wen Congyangb023f462012-12-11 16:00:45 -08006679 /*
6680 * The HWPoisoned page may be not in buddy system, and
6681 * page_count() is not 0.
6682 */
6683 if (skip_hwpoisoned_pages && PageHWPoison(page))
6684 continue;
6685
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006686 if (!PageLRU(page))
6687 found++;
6688 /*
Johannes Weiner6b4f7792014-12-12 16:56:13 -08006689 * If there are RECLAIMABLE pages, we need to check
6690 * it. But now, memory offline itself doesn't call
6691 * shrink_node_slabs() and it still to be fixed.
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006692 */
6693 /*
6694 * If the page is not RAM, page_count()should be 0.
6695 * we don't need more check. This is an _used_ not-movable page.
6696 *
6697 * The problematic thing here is PG_reserved pages. PG_reserved
6698 * is set to both of a memory hole page and a _used_ kernel
6699 * page at boot.
6700 */
6701 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07006702 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006703 }
Minchan Kim80934512012-07-31 16:43:01 -07006704 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006705}
6706
6707bool is_pageblock_removable_nolock(struct page *page)
6708{
Michal Hocko656a0702012-01-20 14:33:58 -08006709 struct zone *zone;
6710 unsigned long pfn;
Michal Hocko687875fb2012-01-20 14:33:55 -08006711
6712 /*
6713 * We have to be careful here because we are iterating over memory
6714 * sections which are not zone aware so we might end up outside of
6715 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08006716 * We have to take care about the node as well. If the node is offline
6717 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875fb2012-01-20 14:33:55 -08006718 */
Michal Hocko656a0702012-01-20 14:33:58 -08006719 if (!node_online(page_to_nid(page)))
6720 return false;
6721
6722 zone = page_zone(page);
6723 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08006724 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875fb2012-01-20 14:33:55 -08006725 return false;
6726
Wen Congyangb023f462012-12-11 16:00:45 -08006727 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07006728}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006729
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006730#ifdef CONFIG_CMA
6731
6732static unsigned long pfn_max_align_down(unsigned long pfn)
6733{
6734 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6735 pageblock_nr_pages) - 1);
6736}
6737
6738static unsigned long pfn_max_align_up(unsigned long pfn)
6739{
6740 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6741 pageblock_nr_pages));
6742}
6743
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006744/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006745static int __alloc_contig_migrate_range(struct compact_control *cc,
6746 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006747{
6748 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006749 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006750 unsigned long pfn = start;
6751 unsigned int tries = 0;
6752 int ret = 0;
6753
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08006754 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006755
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006756 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006757 if (fatal_signal_pending(current)) {
6758 ret = -EINTR;
6759 break;
6760 }
6761
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006762 if (list_empty(&cc->migratepages)) {
6763 cc->nr_migratepages = 0;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07006764 pfn = isolate_migratepages_range(cc, pfn, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006765 if (!pfn) {
6766 ret = -EINTR;
6767 break;
6768 }
6769 tries = 0;
6770 } else if (++tries == 5) {
6771 ret = ret < 0 ? ret : -EBUSY;
6772 break;
6773 }
6774
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006775 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6776 &cc->migratepages);
6777 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07006778
Hugh Dickins9c620e22013-02-22 16:35:14 -08006779 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07006780 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006781 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08006782 if (ret < 0) {
6783 putback_movable_pages(&cc->migratepages);
6784 return ret;
6785 }
6786 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006787}
6788
6789/**
6790 * alloc_contig_range() -- tries to allocate given range of pages
6791 * @start: start PFN to allocate
6792 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006793 * @migratetype: migratetype of the underlaying pageblocks (either
6794 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6795 * in range must have the same migratetype and it must
6796 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006797 *
6798 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6799 * aligned, however it's the caller's responsibility to guarantee that
6800 * we are the only thread that changes migrate type of pageblocks the
6801 * pages fall in.
6802 *
6803 * The PFN range must belong to a single zone.
6804 *
6805 * Returns zero on success or negative error code. On success all
6806 * pages which PFN is in [start, end) are allocated for the caller and
6807 * need to be freed with free_contig_range().
6808 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006809int alloc_contig_range(unsigned long start, unsigned long end,
6810 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006811{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006812 unsigned long outer_start, outer_end;
6813 int ret = 0, order;
6814
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006815 struct compact_control cc = {
6816 .nr_migratepages = 0,
6817 .order = -1,
6818 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07006819 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006820 .ignore_skip_hint = true,
6821 };
6822 INIT_LIST_HEAD(&cc.migratepages);
6823
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006824 /*
6825 * What we do here is we mark all pageblocks in range as
6826 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6827 * have different sizes, and due to the way page allocator
6828 * work, we align the range to biggest of the two pages so
6829 * that page allocator won't try to merge buddies from
6830 * different pageblocks and change MIGRATE_ISOLATE to some
6831 * other migration type.
6832 *
6833 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6834 * migrate the pages from an unaligned range (ie. pages that
6835 * we are interested in). This will put all the pages in
6836 * range back to page allocator as MIGRATE_ISOLATE.
6837 *
6838 * When this is done, we take the pages in range from page
6839 * allocator removing them from the buddy system. This way
6840 * page allocator will never consider using them.
6841 *
6842 * This lets us mark the pageblocks back as
6843 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6844 * aligned range but not in the unaligned, original range are
6845 * put back to page allocator so that buddy can use them.
6846 */
6847
6848 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08006849 pfn_max_align_up(end), migratetype,
6850 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006851 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07006852 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006853
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006854 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006855 if (ret)
6856 goto done;
6857
6858 /*
6859 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6860 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6861 * more, all pages in [start, end) are free in page allocator.
6862 * What we are going to do is to allocate all pages from
6863 * [start, end) (that is remove them from page allocator).
6864 *
6865 * The only problem is that pages at the beginning and at the
6866 * end of interesting range may be not aligned with pages that
6867 * page allocator holds, ie. they can be part of higher order
6868 * pages. Because of this, we reserve the bigger range and
6869 * once this is done free the pages we are not interested in.
6870 *
6871 * We don't have to hold zone->lock here because the pages are
6872 * isolated thus they won't get removed from buddy.
6873 */
6874
6875 lru_add_drain_all();
Vlastimil Babka510f5502014-12-10 15:43:07 -08006876 drain_all_pages(cc.zone);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006877
6878 order = 0;
6879 outer_start = start;
6880 while (!PageBuddy(pfn_to_page(outer_start))) {
6881 if (++order >= MAX_ORDER) {
6882 ret = -EBUSY;
6883 goto done;
6884 }
6885 outer_start &= ~0UL << order;
6886 }
6887
6888 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08006889 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewiczdae803e2014-11-13 15:19:27 -08006890 pr_info("%s: [%lx, %lx) PFNs busy\n",
6891 __func__, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006892 ret = -EBUSY;
6893 goto done;
6894 }
6895
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006896 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006897 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006898 if (!outer_end) {
6899 ret = -EBUSY;
6900 goto done;
6901 }
6902
6903 /* Free head and tail (if any) */
6904 if (start != outer_start)
6905 free_contig_range(outer_start, start - outer_start);
6906 if (end != outer_end)
6907 free_contig_range(end, outer_end - end);
6908
6909done:
6910 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006911 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006912 return ret;
6913}
6914
6915void free_contig_range(unsigned long pfn, unsigned nr_pages)
6916{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08006917 unsigned int count = 0;
6918
6919 for (; nr_pages--; pfn++) {
6920 struct page *page = pfn_to_page(pfn);
6921
6922 count += page_count(page) != 1;
6923 __free_page(page);
6924 }
6925 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006926}
6927#endif
6928
Jiang Liu4ed7e022012-07-31 16:43:35 -07006929#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07006930/*
6931 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6932 * page high values need to be recalulated.
6933 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07006934void __meminit zone_pcp_update(struct zone *zone)
6935{
Cody P Schafer0a647f32013-07-03 15:01:33 -07006936 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006937 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07006938 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07006939 pageset_set_high_and_batch(zone,
6940 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006941 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07006942}
6943#endif
6944
Jiang Liu340175b2012-07-31 16:43:32 -07006945void zone_pcp_reset(struct zone *zone)
6946{
6947 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07006948 int cpu;
6949 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07006950
6951 /* avoid races with drain_pages() */
6952 local_irq_save(flags);
6953 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07006954 for_each_online_cpu(cpu) {
6955 pset = per_cpu_ptr(zone->pageset, cpu);
6956 drain_zonestat(zone, pset);
6957 }
Jiang Liu340175b2012-07-31 16:43:32 -07006958 free_percpu(zone->pageset);
6959 zone->pageset = &boot_pageset;
6960 }
6961 local_irq_restore(flags);
6962}
6963
Wen Congyang6dcd73d2012-12-11 16:01:01 -08006964#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006965/*
6966 * All pages in the range must be isolated before calling this.
6967 */
6968void
6969__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6970{
6971 struct page *page;
6972 struct zone *zone;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006973 unsigned int order, i;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006974 unsigned long pfn;
6975 unsigned long flags;
6976 /* find the first valid pfn */
6977 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6978 if (pfn_valid(pfn))
6979 break;
6980 if (pfn == end_pfn)
6981 return;
6982 zone = page_zone(pfn_to_page(pfn));
6983 spin_lock_irqsave(&zone->lock, flags);
6984 pfn = start_pfn;
6985 while (pfn < end_pfn) {
6986 if (!pfn_valid(pfn)) {
6987 pfn++;
6988 continue;
6989 }
6990 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08006991 /*
6992 * The HWPoisoned page may be not in buddy system, and
6993 * page_count() is not 0.
6994 */
6995 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6996 pfn++;
6997 SetPageReserved(page);
6998 continue;
6999 }
7000
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007001 BUG_ON(page_count(page));
7002 BUG_ON(!PageBuddy(page));
7003 order = page_order(page);
7004#ifdef CONFIG_DEBUG_VM
7005 printk(KERN_INFO "remove from free list %lx %d %lx\n",
7006 pfn, 1 << order, end_pfn);
7007#endif
7008 list_del(&page->lru);
7009 rmv_page_order(page);
7010 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007011 for (i = 0; i < (1 << order); i++)
7012 SetPageReserved((page+i));
7013 pfn += (1 << order);
7014 }
7015 spin_unlock_irqrestore(&zone->lock, flags);
7016}
7017#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007018
7019#ifdef CONFIG_MEMORY_FAILURE
7020bool is_free_buddy_page(struct page *page)
7021{
7022 struct zone *zone = page_zone(page);
7023 unsigned long pfn = page_to_pfn(page);
7024 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07007025 unsigned int order;
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007026
7027 spin_lock_irqsave(&zone->lock, flags);
7028 for (order = 0; order < MAX_ORDER; order++) {
7029 struct page *page_head = page - (pfn & ((1 << order) - 1));
7030
7031 if (PageBuddy(page_head) && page_order(page_head) >= order)
7032 break;
7033 }
7034 spin_unlock_irqrestore(&zone->lock, flags);
7035
7036 return order < MAX_ORDER;
7037}
7038#endif