blob: 48aaf7b9f253e6ea68587caa1e7e3e254905936a [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 Hiroyukia5d76b52007-10-16 01:26:11 -070051#include <linux/page-isolation.h>
Joonsoo Kimeefa8642014-12-12 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
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -0700128/*
129 * A cached value of the page's pageblock's migratetype, used when the page is
130 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
131 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
132 * Also the migratetype set in the page does not necessarily match the pcplist
133 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
134 * other index - this ensures that it will be put on the correct CMA freelist.
135 */
136static inline int get_pcppage_migratetype(struct page *page)
137{
138 return page->index;
139}
140
141static inline void set_pcppage_migratetype(struct page *page, int migratetype)
142{
143 page->index = migratetype;
144}
145
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800146#ifdef CONFIG_PM_SLEEP
147/*
148 * The following functions are used by the suspend/hibernate code to temporarily
149 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
150 * while devices are suspended. To avoid races with the suspend/hibernate code,
151 * they should always be called with pm_mutex held (gfp_allowed_mask also should
152 * only be modified with pm_mutex held, unless the suspend/hibernate code is
153 * guaranteed not to run in parallel with that modification).
154 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100155
156static gfp_t saved_gfp_mask;
157
158void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800159{
160 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100161 if (saved_gfp_mask) {
162 gfp_allowed_mask = saved_gfp_mask;
163 saved_gfp_mask = 0;
164 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800165}
166
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100167void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800168{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800169 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100170 WARN_ON(saved_gfp_mask);
171 saved_gfp_mask = gfp_allowed_mask;
172 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800173}
Mel Gormanf90ac392012-01-10 15:07:15 -0800174
175bool pm_suspended_storage(void)
176{
177 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
178 return false;
179 return true;
180}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800181#endif /* CONFIG_PM_SLEEP */
182
Mel Gormand9c23402007-10-16 01:26:01 -0700183#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
184int pageblock_order __read_mostly;
185#endif
186
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800187static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189/*
190 * results with 256, 32 in the lowmem_reserve sysctl:
191 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
192 * 1G machine -> (16M dma, 784M normal, 224M high)
193 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
194 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
Yaowei Bai84109e12015-02-12 15:00:22 -0800195 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100196 *
197 * TBD: should special case ZONE_DMA32 machines here - in those we normally
198 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700200int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800201#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700202 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800203#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700204#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700205 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700206#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700207#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700208 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700209#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700210 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700211};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
213EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214
Helge Deller15ad7cd2006-12-06 20:40:36 -0800215static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800216#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700217 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800218#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700219#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700220 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700221#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700222 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700223#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700224 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700225#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700226 "Movable",
Dan Williams033fbae2015-08-09 15:29:06 -0400227#ifdef CONFIG_ZONE_DEVICE
228 "Device",
229#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700230};
231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232int min_free_kbytes = 1024;
Han Pingtian42aa83c2014-01-23 15:53:28 -0800233int user_min_free_kbytes = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
Jan Beulich2c85f512009-09-21 17:03:07 -0700235static unsigned long __meminitdata nr_kernel_pages;
236static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700237static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Tejun Heo0ee332c2011-12-08 10:22:09 -0800239#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
240static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
241static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
242static unsigned long __initdata required_kernelcore;
243static unsigned long __initdata required_movablecore;
244static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gormanc7132162006-09-27 01:49:43 -0700245
Tejun Heo0ee332c2011-12-08 10:22:09 -0800246/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
247int movable_zone;
248EXPORT_SYMBOL(movable_zone);
249#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700250
Miklos Szeredi418508c2007-05-23 13:57:55 -0700251#if MAX_NUMNODES > 1
252int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700253int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700254EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700255EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700256#endif
257
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700258int page_group_by_mobility_disabled __read_mostly;
259
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700260#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
261static inline void reset_deferred_meminit(pg_data_t *pgdat)
262{
263 pgdat->first_deferred_pfn = ULONG_MAX;
264}
265
266/* Returns true if the struct page for the pfn is uninitialised */
Mel Gorman0e1cc952015-06-30 14:57:27 -0700267static inline bool __meminit early_page_uninitialised(unsigned long pfn)
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700268{
Mel Gormanae026b22015-07-17 16:23:48 -0700269 if (pfn >= NODE_DATA(early_pfn_to_nid(pfn))->first_deferred_pfn)
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700270 return true;
271
272 return false;
273}
274
Mel Gorman7e18adb2015-06-30 14:57:05 -0700275static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
276{
277 if (pfn >= NODE_DATA(nid)->first_deferred_pfn)
278 return true;
279
280 return false;
281}
282
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700283/*
284 * Returns false when the remaining initialisation should be deferred until
285 * later in the boot cycle when it can be parallelised.
286 */
287static inline bool update_defer_init(pg_data_t *pgdat,
288 unsigned long pfn, unsigned long zone_end,
289 unsigned long *nr_initialised)
290{
291 /* Always populate low zones for address-contrained allocations */
292 if (zone_end < pgdat_end_pfn(pgdat))
293 return true;
294
295 /* Initialise at least 2G of the highest zone */
296 (*nr_initialised)++;
297 if (*nr_initialised > (2UL << (30 - PAGE_SHIFT)) &&
298 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
299 pgdat->first_deferred_pfn = pfn;
300 return false;
301 }
302
303 return true;
304}
305#else
306static inline void reset_deferred_meminit(pg_data_t *pgdat)
307{
308}
309
310static inline bool early_page_uninitialised(unsigned long pfn)
311{
312 return false;
313}
314
Mel Gorman7e18adb2015-06-30 14:57:05 -0700315static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
316{
317 return false;
318}
319
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700320static inline bool update_defer_init(pg_data_t *pgdat,
321 unsigned long pfn, unsigned long zone_end,
322 unsigned long *nr_initialised)
323{
324 return true;
325}
326#endif
327
328
Minchan Kimee6f5092012-07-31 16:43:50 -0700329void set_pageblock_migratetype(struct page *page, int migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700330{
KOSAKI Motohiro5d0f3f72013-11-12 15:08:18 -0800331 if (unlikely(page_group_by_mobility_disabled &&
332 migratetype < MIGRATE_PCPTYPES))
Mel Gorman49255c62009-06-16 15:31:58 -0700333 migratetype = MIGRATE_UNMOVABLE;
334
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700335 set_pageblock_flags_group(page, (unsigned long)migratetype,
336 PB_migrate, PB_migrate_end);
337}
338
Nick Piggin13e74442006-01-06 00:10:58 -0800339#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700340static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700342 int ret = 0;
343 unsigned seq;
344 unsigned long pfn = page_to_pfn(page);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800345 unsigned long sp, start_pfn;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700346
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700347 do {
348 seq = zone_span_seqbegin(zone);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800349 start_pfn = zone->zone_start_pfn;
350 sp = zone->spanned_pages;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800351 if (!zone_spans_pfn(zone, pfn))
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700352 ret = 1;
353 } while (zone_span_seqretry(zone, seq));
354
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800355 if (ret)
Dave Hansen613813e2014-06-04 16:07:27 -0700356 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
357 pfn, zone_to_nid(zone), zone->name,
358 start_pfn, start_pfn + sp);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800359
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700360 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700361}
362
363static int page_is_consistent(struct zone *zone, struct page *page)
364{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700365 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700366 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700368 return 0;
369
370 return 1;
371}
372/*
373 * Temporary debugging check for pages not lying within a given zone.
374 */
375static int bad_range(struct zone *zone, struct page *page)
376{
377 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700379 if (!page_is_consistent(zone, page))
380 return 1;
381
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 return 0;
383}
Nick Piggin13e74442006-01-06 00:10:58 -0800384#else
385static inline int bad_range(struct zone *zone, struct page *page)
386{
387 return 0;
388}
389#endif
390
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700391static void bad_page(struct page *page, const char *reason,
392 unsigned long bad_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800394 static unsigned long resume;
395 static unsigned long nr_shown;
396 static unsigned long nr_unshown;
397
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200398 /* Don't complain about poisoned pages */
399 if (PageHWPoison(page)) {
Mel Gorman22b751c2013-02-22 16:34:59 -0800400 page_mapcount_reset(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200401 return;
402 }
403
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800404 /*
405 * Allow a burst of 60 reports, then keep quiet for that minute;
406 * or allow a steady drip of one report per second.
407 */
408 if (nr_shown == 60) {
409 if (time_before(jiffies, resume)) {
410 nr_unshown++;
411 goto out;
412 }
413 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800414 printk(KERN_ALERT
415 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800416 nr_unshown);
417 nr_unshown = 0;
418 }
419 nr_shown = 0;
420 }
421 if (nr_shown++ == 0)
422 resume = jiffies + 60 * HZ;
423
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800424 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800425 current->comm, page_to_pfn(page));
Dave Hansenf0b791a2014-01-23 15:52:49 -0800426 dump_page_badflags(page, reason, bad_flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700427
Dave Jones4f318882011-10-31 17:07:24 -0700428 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800430out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800431 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800432 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030433 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434}
435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436/*
437 * Higher-order pages are called "compound pages". They are structured thusly:
438 *
439 * The first PAGE_SIZE page is called the "head page".
440 *
441 * The remaining PAGE_SIZE pages are called "tail pages".
442 *
Wang Sheng-Hui6416b9f2011-11-17 10:53:50 +0100443 * All pages have PG_compound set. All tail pages have their ->first_page
444 * pointing at the head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800446 * The first tail page's ->lru.next holds the address of the compound page's
447 * put_page() function. Its ->lru.prev holds the order of allocation.
448 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800450
451static void free_compound_page(struct page *page)
452{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700453 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800454}
455
Andi Kleen01ad1c02008-07-23 21:27:46 -0700456void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
458 int i;
459 int nr_pages = 1 << order;
460
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800461 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700462 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700463 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800464 for (i = 1; i < nr_pages; i++) {
465 struct page *p = page + i;
Youquan Song58a84aa2011-12-08 14:34:18 -0800466 set_page_count(p, 0);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700467 p->first_page = page;
David Rientjes668f9abb2014-03-03 15:38:18 -0800468 /* Make sure p->first_page is always valid for PageTail() */
469 smp_wmb();
470 __SetPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 }
472}
473
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800474#ifdef CONFIG_DEBUG_PAGEALLOC
475unsigned int _debug_guardpage_minorder;
Joonsoo Kim031bc572014-12-12 16:55:52 -0800476bool _debug_pagealloc_enabled __read_mostly;
Joonsoo Kime30825f2014-12-12 16:55:49 -0800477bool _debug_guardpage_enabled __read_mostly;
478
Joonsoo Kim031bc572014-12-12 16:55:52 -0800479static int __init early_debug_pagealloc(char *buf)
480{
481 if (!buf)
482 return -EINVAL;
483
484 if (strcmp(buf, "on") == 0)
485 _debug_pagealloc_enabled = true;
486
487 return 0;
488}
489early_param("debug_pagealloc", early_debug_pagealloc);
490
Joonsoo Kime30825f2014-12-12 16:55:49 -0800491static bool need_debug_guardpage(void)
492{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800493 /* If we don't use debug_pagealloc, we don't need guard page */
494 if (!debug_pagealloc_enabled())
495 return false;
496
Joonsoo Kime30825f2014-12-12 16:55:49 -0800497 return true;
498}
499
500static void init_debug_guardpage(void)
501{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800502 if (!debug_pagealloc_enabled())
503 return;
504
Joonsoo Kime30825f2014-12-12 16:55:49 -0800505 _debug_guardpage_enabled = true;
506}
507
508struct page_ext_operations debug_guardpage_ops = {
509 .need = need_debug_guardpage,
510 .init = init_debug_guardpage,
511};
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800512
513static int __init debug_guardpage_minorder_setup(char *buf)
514{
515 unsigned long res;
516
517 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
518 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
519 return 0;
520 }
521 _debug_guardpage_minorder = res;
522 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
523 return 0;
524}
525__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
526
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800527static inline void set_page_guard(struct zone *zone, struct page *page,
528 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800529{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800530 struct page_ext *page_ext;
531
532 if (!debug_guardpage_enabled())
533 return;
534
535 page_ext = lookup_page_ext(page);
536 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
537
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800538 INIT_LIST_HEAD(&page->lru);
539 set_page_private(page, order);
540 /* Guard pages are not available for any usage */
541 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800542}
543
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800544static inline void clear_page_guard(struct zone *zone, struct page *page,
545 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800546{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800547 struct page_ext *page_ext;
548
549 if (!debug_guardpage_enabled())
550 return;
551
552 page_ext = lookup_page_ext(page);
553 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
554
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800555 set_page_private(page, 0);
556 if (!is_migrate_isolate(migratetype))
557 __mod_zone_freepage_state(zone, (1 << order), migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800558}
559#else
Joonsoo Kime30825f2014-12-12 16:55:49 -0800560struct page_ext_operations debug_guardpage_ops = { NULL, };
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800561static inline void set_page_guard(struct zone *zone, struct page *page,
562 unsigned int order, int migratetype) {}
563static inline void clear_page_guard(struct zone *zone, struct page *page,
564 unsigned int order, int migratetype) {}
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800565#endif
566
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700567static inline void set_page_order(struct page *page, unsigned int order)
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700568{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700569 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000570 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571}
572
573static inline void rmv_page_order(struct page *page)
574{
Nick Piggin676165a2006-04-10 11:21:48 +1000575 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700576 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577}
578
579/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 * This function checks whether a page is free && is the buddy
581 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800582 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000583 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700584 * (c) a page and its buddy have the same order &&
585 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 *
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700587 * For recording whether a page is in the buddy system, we set ->_mapcount
588 * PAGE_BUDDY_MAPCOUNT_VALUE.
589 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
590 * serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000591 *
592 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700594static inline int page_is_buddy(struct page *page, struct page *buddy,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700595 unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700597 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800598 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800599
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800600 if (page_is_guard(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700601 if (page_zone_id(page) != page_zone_id(buddy))
602 return 0;
603
Weijie Yang4c5018c2015-02-10 14:11:39 -0800604 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
605
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800606 return 1;
607 }
608
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700609 if (PageBuddy(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700610 /*
611 * zone check is done late to avoid uselessly
612 * calculating zone/node ids for pages that could
613 * never merge.
614 */
615 if (page_zone_id(page) != page_zone_id(buddy))
616 return 0;
617
Weijie Yang4c5018c2015-02-10 14:11:39 -0800618 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
619
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700620 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000621 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700622 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623}
624
625/*
626 * Freeing function for a buddy system allocator.
627 *
628 * The concept of a buddy system is to maintain direct-mapped table
629 * (containing bit values) for memory blocks of various "orders".
630 * The bottom level table contains the map for the smallest allocatable
631 * units of memory (here, pages), and each level above it describes
632 * pairs of units from the levels below, hence, "buddies".
633 * At a high level, all that happens here is marking the table entry
634 * at the bottom level available, and propagating the changes upward
635 * as necessary, plus some accounting needed to play nicely with other
636 * parts of the VM system.
637 * At each level, we keep a list of pages, which are heads of continuous
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700638 * free pages of length of (1 << order) and marked with _mapcount
639 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
640 * field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100642 * other. That is, if we allocate a small block, and both were
643 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100645 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100647 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 */
649
Nick Piggin48db57f2006-01-08 01:00:42 -0800650static inline void __free_one_page(struct page *page,
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700651 unsigned long pfn,
Mel Gormaned0ae212009-06-16 15:32:07 -0700652 struct zone *zone, unsigned int order,
653 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654{
655 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700656 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800657 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700658 struct page *buddy;
Joonsoo Kim3c605092014-11-13 15:19:21 -0800659 int max_order = MAX_ORDER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
Cody P Schaferd29bb972013-02-22 16:35:25 -0800661 VM_BUG_ON(!zone_is_initialized(zone));
Kirill A. Shutemov6e9f0d52015-02-11 15:25:50 -0800662 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Mel Gormaned0ae212009-06-16 15:32:07 -0700664 VM_BUG_ON(migratetype == -1);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800665 if (is_migrate_isolate(migratetype)) {
666 /*
667 * We restrict max order of merging to prevent merge
668 * between freepages on isolate pageblock and normal
669 * pageblock. Without this, pageblock isolation
670 * could cause incorrect freepage accounting.
671 */
672 max_order = min(MAX_ORDER, pageblock_order + 1);
673 } else {
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800674 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800675 }
Mel Gormaned0ae212009-06-16 15:32:07 -0700676
Joonsoo Kim3c605092014-11-13 15:19:21 -0800677 page_idx = pfn & ((1 << max_order) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Sasha Levin309381fea2014-01-23 15:52:54 -0800679 VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
680 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
Joonsoo Kim3c605092014-11-13 15:19:21 -0800682 while (order < max_order - 1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800683 buddy_idx = __find_buddy_index(page_idx, order);
684 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700685 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700686 break;
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800687 /*
688 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
689 * merge with it and move up one order.
690 */
691 if (page_is_guard(buddy)) {
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800692 clear_page_guard(zone, buddy, order, migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800693 } else {
694 list_del(&buddy->lru);
695 zone->free_area[order].nr_free--;
696 rmv_page_order(buddy);
697 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800698 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 page = page + (combined_idx - page_idx);
700 page_idx = combined_idx;
701 order++;
702 }
703 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700704
705 /*
706 * If this is not the largest possible page, check if the buddy
707 * of the next-highest order is free. If it is, it's possible
708 * that pages are being freed that will coalesce soon. In case,
709 * that is happening, add the free page to the tail of the list
710 * so it's less likely to be used soon and more likely to be merged
711 * as a higher order page
712 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700713 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700714 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800715 combined_idx = buddy_idx & page_idx;
716 higher_page = page + (combined_idx - page_idx);
717 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700718 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700719 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
720 list_add_tail(&page->lru,
721 &zone->free_area[order].free_list[migratetype]);
722 goto out;
723 }
724 }
725
726 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
727out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 zone->free_area[order].nr_free++;
729}
730
Nick Piggin224abf92006-01-06 00:11:11 -0800731static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700733 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800734 unsigned long bad_flags = 0;
735
736 if (unlikely(page_mapcount(page)))
737 bad_reason = "nonzero mapcount";
738 if (unlikely(page->mapping != NULL))
739 bad_reason = "non-NULL mapping";
740 if (unlikely(atomic_read(&page->_count) != 0))
741 bad_reason = "nonzero _count";
742 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
743 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
744 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
745 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -0800746#ifdef CONFIG_MEMCG
747 if (unlikely(page->mem_cgroup))
748 bad_reason = "page still charged to cgroup";
749#endif
Dave Hansenf0b791a2014-01-23 15:52:49 -0800750 if (unlikely(bad_reason)) {
751 bad_page(page, bad_reason, bad_flags);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800752 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800753 }
Peter Zijlstra90572892013-10-07 11:29:20 +0100754 page_cpupid_reset_last(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800755 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
756 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
757 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758}
759
760/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700761 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700763 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 *
765 * If the zone was previously in an "all pages pinned" state then look to
766 * see if this freeing clears that state.
767 *
768 * And clear the zone's pages_scanned counter, to hold off the "all pages are
769 * pinned" detection logic.
770 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700771static void free_pcppages_bulk(struct zone *zone, int count,
772 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700774 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700775 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700776 int to_free = count;
Mel Gorman0d5d8232014-08-06 16:07:16 -0700777 unsigned long nr_scanned;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700778
Nick Pigginc54ad302006-01-06 00:10:56 -0800779 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700780 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
781 if (nr_scanned)
782 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700783
Mel Gorman72853e22010-09-09 16:38:16 -0700784 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800785 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700786 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800787
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700788 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700789 * Remove pages from lists in a round-robin fashion. A
790 * batch_free count is maintained that is incremented when an
791 * empty list is encountered. This is so more pages are freed
792 * off fuller lists instead of spinning excessively around empty
793 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700794 */
795 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700796 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700797 if (++migratetype == MIGRATE_PCPTYPES)
798 migratetype = 0;
799 list = &pcp->lists[migratetype];
800 } while (list_empty(list));
801
Namhyung Kim1d168712011-03-22 16:32:45 -0700802 /* This is the only non-empty list. Free them all. */
803 if (batch_free == MIGRATE_PCPTYPES)
804 batch_free = to_free;
805
Mel Gormana6f9edd2009-09-21 17:03:20 -0700806 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700807 int mt; /* migratetype of the to-be-freed page */
808
Mel Gormana6f9edd2009-09-21 17:03:20 -0700809 page = list_entry(list->prev, struct page, lru);
810 /* must delete as __free_one_page list manipulates */
811 list_del(&page->lru);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -0700812
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -0700813 mt = get_pcppage_migratetype(page);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -0700814 /* MIGRATE_ISOLATE page should not go to pcplists */
815 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
816 /* Pageblock could have been isolated meanwhile */
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800817 if (unlikely(has_isolate_pageblock(zone)))
Joonsoo Kim51bb1a42014-11-13 15:19:14 -0800818 mt = get_pageblock_migratetype(page);
Joonsoo Kim51bb1a42014-11-13 15:19:14 -0800819
Hugh Dickinsa7016232010-01-29 17:46:34 +0000820 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700821 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700822 trace_mm_page_pcpu_drain(page, 0, mt);
Mel Gorman72853e22010-09-09 16:38:16 -0700823 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800825 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826}
827
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700828static void free_one_page(struct zone *zone,
829 struct page *page, unsigned long pfn,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700830 unsigned int order,
Mel Gormaned0ae212009-06-16 15:32:07 -0700831 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800832{
Mel Gorman0d5d8232014-08-06 16:07:16 -0700833 unsigned long nr_scanned;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700834 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700835 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
836 if (nr_scanned)
837 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700838
Joonsoo Kimad53f922014-11-13 15:19:11 -0800839 if (unlikely(has_isolate_pageblock(zone) ||
840 is_migrate_isolate(migratetype))) {
841 migratetype = get_pfnblock_migratetype(page, pfn);
Joonsoo Kimad53f922014-11-13 15:19:11 -0800842 }
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700843 __free_one_page(page, pfn, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700844 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800845}
846
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800847static int free_tail_pages_check(struct page *head_page, struct page *page)
848{
849 if (!IS_ENABLED(CONFIG_DEBUG_VM))
850 return 0;
851 if (unlikely(!PageTail(page))) {
852 bad_page(page, "PageTail not set", 0);
853 return 1;
854 }
855 if (unlikely(page->first_page != head_page)) {
856 bad_page(page, "first_page not consistent", 0);
857 return 1;
858 }
859 return 0;
860}
861
Robin Holt1e8ce832015-06-30 14:56:45 -0700862static void __meminit __init_single_page(struct page *page, unsigned long pfn,
863 unsigned long zone, int nid)
864{
Robin Holt1e8ce832015-06-30 14:56:45 -0700865 set_page_links(page, zone, nid, pfn);
Robin Holt1e8ce832015-06-30 14:56:45 -0700866 init_page_count(page);
867 page_mapcount_reset(page);
868 page_cpupid_reset_last(page);
Robin Holt1e8ce832015-06-30 14:56:45 -0700869
Robin Holt1e8ce832015-06-30 14:56:45 -0700870 INIT_LIST_HEAD(&page->lru);
871#ifdef WANT_PAGE_VIRTUAL
872 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
873 if (!is_highmem_idx(zone))
874 set_page_address(page, __va(pfn << PAGE_SHIFT));
875#endif
876}
877
878static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
879 int nid)
880{
881 return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
882}
883
Mel Gorman7e18adb2015-06-30 14:57:05 -0700884#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
885static void init_reserved_page(unsigned long pfn)
886{
887 pg_data_t *pgdat;
888 int nid, zid;
889
890 if (!early_page_uninitialised(pfn))
891 return;
892
893 nid = early_pfn_to_nid(pfn);
894 pgdat = NODE_DATA(nid);
895
896 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
897 struct zone *zone = &pgdat->node_zones[zid];
898
899 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
900 break;
901 }
902 __init_single_pfn(pfn, zid, nid);
903}
904#else
905static inline void init_reserved_page(unsigned long pfn)
906{
907}
908#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
909
Nathan Zimmer92923ca2015-06-30 14:56:48 -0700910/*
911 * Initialised pages do not have PageReserved set. This function is
912 * called for each range allocated by the bootmem allocator and
913 * marks the pages PageReserved. The remaining valid pages are later
914 * sent to the buddy page allocator.
915 */
Mel Gorman7e18adb2015-06-30 14:57:05 -0700916void __meminit reserve_bootmem_region(unsigned long start, unsigned long end)
Nathan Zimmer92923ca2015-06-30 14:56:48 -0700917{
918 unsigned long start_pfn = PFN_DOWN(start);
919 unsigned long end_pfn = PFN_UP(end);
920
Mel Gorman7e18adb2015-06-30 14:57:05 -0700921 for (; start_pfn < end_pfn; start_pfn++) {
922 if (pfn_valid(start_pfn)) {
923 struct page *page = pfn_to_page(start_pfn);
924
925 init_reserved_page(start_pfn);
926 SetPageReserved(page);
927 }
928 }
Nathan Zimmer92923ca2015-06-30 14:56:48 -0700929}
930
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700931static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932{
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800933 bool compound = PageCompound(page);
934 int i, bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
Yu Zhaoab1f3062014-12-10 15:43:17 -0800936 VM_BUG_ON_PAGE(PageTail(page), page);
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800937 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
Yu Zhaoab1f3062014-12-10 15:43:17 -0800938
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -0800939 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100940 kmemcheck_free_shadow(page, order);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -0800941 kasan_free_pages(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100942
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800943 if (PageAnon(page))
944 page->mapping = NULL;
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800945 bad += free_pages_check(page);
946 for (i = 1; i < (1 << order); i++) {
947 if (compound)
948 bad += free_tail_pages_check(page, page + i);
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800949 bad += free_pages_check(page + i);
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800950 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800951 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700952 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800953
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800954 reset_page_owner(page, order);
955
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700956 if (!PageHighMem(page)) {
Pintu Kumarb8af2942013-09-11 14:20:34 -0700957 debug_check_no_locks_freed(page_address(page),
958 PAGE_SIZE << order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700959 debug_check_no_obj_freed(page_address(page),
960 PAGE_SIZE << order);
961 }
Nick Piggindafb1362006-10-11 01:21:30 -0700962 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800963 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700964
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700965 return true;
966}
967
968static void __free_pages_ok(struct page *page, unsigned int order)
969{
970 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -0700971 int migratetype;
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700972 unsigned long pfn = page_to_pfn(page);
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700973
974 if (!free_pages_prepare(page, order))
975 return;
976
Mel Gormancfc47a22014-06-04 16:10:19 -0700977 migratetype = get_pfnblock_migratetype(page, pfn);
Nick Pigginc54ad302006-01-06 00:10:56 -0800978 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700979 __count_vm_events(PGFREE, 1 << order);
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700980 free_one_page(page_zone(page), page, pfn, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -0800981 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982}
983
Mel Gorman0e1cc952015-06-30 14:57:27 -0700984static void __init __free_pages_boot_core(struct page *page,
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700985 unsigned long pfn, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800986{
Johannes Weinerc3993072012-01-10 15:08:10 -0800987 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700988 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -0800989 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -0800990
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700991 prefetchw(p);
992 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
993 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -0800994 __ClearPageReserved(p);
995 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -0800996 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -0700997 __ClearPageReserved(p);
998 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -0800999
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001000 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -08001001 set_page_refcounted(page);
1002 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -08001003}
1004
Mel Gorman75a592a2015-06-30 14:56:59 -07001005#if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1006 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
Mel Gorman7ace9912015-08-06 15:46:13 -07001007
Mel Gorman75a592a2015-06-30 14:56:59 -07001008static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1009
1010int __meminit early_pfn_to_nid(unsigned long pfn)
1011{
Mel Gorman7ace9912015-08-06 15:46:13 -07001012 static DEFINE_SPINLOCK(early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -07001013 int nid;
1014
Mel Gorman7ace9912015-08-06 15:46:13 -07001015 spin_lock(&early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -07001016 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
Mel Gorman7ace9912015-08-06 15:46:13 -07001017 if (nid < 0)
1018 nid = 0;
1019 spin_unlock(&early_pfn_lock);
1020
1021 return nid;
Mel Gorman75a592a2015-06-30 14:56:59 -07001022}
1023#endif
1024
1025#ifdef CONFIG_NODES_SPAN_OTHER_NODES
1026static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1027 struct mminit_pfnnid_cache *state)
1028{
1029 int nid;
1030
1031 nid = __early_pfn_to_nid(pfn, state);
1032 if (nid >= 0 && nid != node)
1033 return false;
1034 return true;
1035}
1036
1037/* Only safe to use early in boot when initialisation is single-threaded */
1038static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1039{
1040 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1041}
1042
1043#else
1044
1045static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1046{
1047 return true;
1048}
1049static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1050 struct mminit_pfnnid_cache *state)
1051{
1052 return true;
1053}
1054#endif
1055
1056
Mel Gorman0e1cc952015-06-30 14:57:27 -07001057void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001058 unsigned int order)
1059{
1060 if (early_page_uninitialised(pfn))
1061 return;
1062 return __free_pages_boot_core(page, pfn, order);
1063}
1064
Mel Gorman7e18adb2015-06-30 14:57:05 -07001065#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
Mel Gorman0e1cc952015-06-30 14:57:27 -07001066static void __init deferred_free_range(struct page *page,
Mel Gormana4de83d2015-06-30 14:57:16 -07001067 unsigned long pfn, int nr_pages)
1068{
1069 int i;
1070
1071 if (!page)
1072 return;
1073
1074 /* Free a large naturally-aligned chunk if possible */
1075 if (nr_pages == MAX_ORDER_NR_PAGES &&
1076 (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
Mel Gormanac5d2532015-06-30 14:57:20 -07001077 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormana4de83d2015-06-30 14:57:16 -07001078 __free_pages_boot_core(page, pfn, MAX_ORDER-1);
1079 return;
1080 }
1081
1082 for (i = 0; i < nr_pages; i++, page++, pfn++)
1083 __free_pages_boot_core(page, pfn, 0);
1084}
1085
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001086/* Completion tracking for deferred_init_memmap() threads */
1087static atomic_t pgdat_init_n_undone __initdata;
1088static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1089
1090static inline void __init pgdat_init_report_one_done(void)
1091{
1092 if (atomic_dec_and_test(&pgdat_init_n_undone))
1093 complete(&pgdat_init_all_done_comp);
1094}
Mel Gorman0e1cc952015-06-30 14:57:27 -07001095
Mel Gorman7e18adb2015-06-30 14:57:05 -07001096/* Initialise remaining memory on a node */
Mel Gorman0e1cc952015-06-30 14:57:27 -07001097static int __init deferred_init_memmap(void *data)
Mel Gorman7e18adb2015-06-30 14:57:05 -07001098{
Mel Gorman0e1cc952015-06-30 14:57:27 -07001099 pg_data_t *pgdat = data;
1100 int nid = pgdat->node_id;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001101 struct mminit_pfnnid_cache nid_init_state = { };
1102 unsigned long start = jiffies;
1103 unsigned long nr_pages = 0;
1104 unsigned long walk_start, walk_end;
1105 int i, zid;
1106 struct zone *zone;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001107 unsigned long first_init_pfn = pgdat->first_deferred_pfn;
Mel Gorman0e1cc952015-06-30 14:57:27 -07001108 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001109
Mel Gorman0e1cc952015-06-30 14:57:27 -07001110 if (first_init_pfn == ULONG_MAX) {
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001111 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001112 return 0;
1113 }
1114
1115 /* Bind memory initialisation thread to a local node if possible */
1116 if (!cpumask_empty(cpumask))
1117 set_cpus_allowed_ptr(current, cpumask);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001118
1119 /* Sanity check boundaries */
1120 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1121 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1122 pgdat->first_deferred_pfn = ULONG_MAX;
1123
1124 /* Only the highest zone is deferred so find it */
1125 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1126 zone = pgdat->node_zones + zid;
1127 if (first_init_pfn < zone_end_pfn(zone))
1128 break;
1129 }
1130
1131 for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1132 unsigned long pfn, end_pfn;
Mel Gorman54608c32015-06-30 14:57:09 -07001133 struct page *page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001134 struct page *free_base_page = NULL;
1135 unsigned long free_base_pfn = 0;
1136 int nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001137
1138 end_pfn = min(walk_end, zone_end_pfn(zone));
1139 pfn = first_init_pfn;
1140 if (pfn < walk_start)
1141 pfn = walk_start;
1142 if (pfn < zone->zone_start_pfn)
1143 pfn = zone->zone_start_pfn;
1144
1145 for (; pfn < end_pfn; pfn++) {
Mel Gorman54608c32015-06-30 14:57:09 -07001146 if (!pfn_valid_within(pfn))
Mel Gormana4de83d2015-06-30 14:57:16 -07001147 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001148
Mel Gorman54608c32015-06-30 14:57:09 -07001149 /*
1150 * Ensure pfn_valid is checked every
1151 * MAX_ORDER_NR_PAGES for memory holes
1152 */
1153 if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1154 if (!pfn_valid(pfn)) {
1155 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001156 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001157 }
1158 }
1159
1160 if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1161 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001162 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001163 }
1164
1165 /* Minimise pfn page lookups and scheduler checks */
1166 if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1167 page++;
1168 } else {
Mel Gormana4de83d2015-06-30 14:57:16 -07001169 nr_pages += nr_to_free;
1170 deferred_free_range(free_base_page,
1171 free_base_pfn, nr_to_free);
1172 free_base_page = NULL;
1173 free_base_pfn = nr_to_free = 0;
1174
Mel Gorman54608c32015-06-30 14:57:09 -07001175 page = pfn_to_page(pfn);
1176 cond_resched();
1177 }
Mel Gorman7e18adb2015-06-30 14:57:05 -07001178
1179 if (page->flags) {
1180 VM_BUG_ON(page_zone(page) != zone);
Mel Gormana4de83d2015-06-30 14:57:16 -07001181 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001182 }
1183
1184 __init_single_page(page, pfn, zid, nid);
Mel Gormana4de83d2015-06-30 14:57:16 -07001185 if (!free_base_page) {
1186 free_base_page = page;
1187 free_base_pfn = pfn;
1188 nr_to_free = 0;
1189 }
1190 nr_to_free++;
1191
1192 /* Where possible, batch up pages for a single free */
1193 continue;
1194free_range:
1195 /* Free the current block of pages to allocator */
1196 nr_pages += nr_to_free;
1197 deferred_free_range(free_base_page, free_base_pfn,
1198 nr_to_free);
1199 free_base_page = NULL;
1200 free_base_pfn = nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001201 }
Mel Gormana4de83d2015-06-30 14:57:16 -07001202
Mel Gorman7e18adb2015-06-30 14:57:05 -07001203 first_init_pfn = max(end_pfn, first_init_pfn);
1204 }
1205
1206 /* Sanity check that the next zone really is unpopulated */
1207 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1208
Mel Gorman0e1cc952015-06-30 14:57:27 -07001209 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
Mel Gorman7e18adb2015-06-30 14:57:05 -07001210 jiffies_to_msecs(jiffies - start));
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001211
1212 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001213 return 0;
1214}
1215
1216void __init page_alloc_init_late(void)
1217{
1218 int nid;
1219
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001220 /* There will be num_node_state(N_MEMORY) threads */
1221 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
Mel Gorman0e1cc952015-06-30 14:57:27 -07001222 for_each_node_state(nid, N_MEMORY) {
Mel Gorman0e1cc952015-06-30 14:57:27 -07001223 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1224 }
1225
1226 /* Block until all are initialised */
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001227 wait_for_completion(&pgdat_init_all_done_comp);
Mel Gorman4248b0d2015-08-06 15:46:20 -07001228
1229 /* Reinit limits that are based on free pages after the kernel is up */
1230 files_maxfiles_init();
Mel Gorman7e18adb2015-06-30 14:57:05 -07001231}
1232#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1233
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001234#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +08001235/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001236void __init init_cma_reserved_pageblock(struct page *page)
1237{
1238 unsigned i = pageblock_nr_pages;
1239 struct page *p = page;
1240
1241 do {
1242 __ClearPageReserved(p);
1243 set_page_count(p, 0);
1244 } while (++p, --i);
1245
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001246 set_pageblock_migratetype(page, MIGRATE_CMA);
Michal Nazarewiczdc783272014-07-02 15:22:35 -07001247
1248 if (pageblock_order >= MAX_ORDER) {
1249 i = pageblock_nr_pages;
1250 p = page;
1251 do {
1252 set_page_refcounted(p);
1253 __free_pages(p, MAX_ORDER - 1);
1254 p += MAX_ORDER_NR_PAGES;
1255 } while (i -= MAX_ORDER_NR_PAGES);
1256 } else {
1257 set_page_refcounted(page);
1258 __free_pages(page, pageblock_order);
1259 }
1260
Jiang Liu3dcc0572013-07-03 15:03:21 -07001261 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001262}
1263#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
1265/*
1266 * The order of subdivision here is critical for the IO subsystem.
1267 * Please do not alter this order without good reasons and regression
1268 * testing. Specifically, as large blocks of memory are subdivided,
1269 * the order in which smaller blocks are delivered depends on the order
1270 * they're subdivided in this function. This is the primary factor
1271 * influencing the order in which pages are delivered to the IO
1272 * subsystem according to empirical testing, and this is also justified
1273 * by considering the behavior of a buddy system containing a single
1274 * large block of memory acted on by a series of small allocations.
1275 * This behavior is a critical factor in sglist merging's success.
1276 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +01001277 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 */
Nick Piggin085cc7d2006-01-06 00:11:01 -08001279static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001280 int low, int high, struct free_area *area,
1281 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282{
1283 unsigned long size = 1 << high;
1284
1285 while (high > low) {
1286 area--;
1287 high--;
1288 size >>= 1;
Sasha Levin309381fea2014-01-23 15:52:54 -08001289 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08001290
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001291 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
Joonsoo Kime30825f2014-12-12 16:55:49 -08001292 debug_guardpage_enabled() &&
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001293 high < debug_guardpage_minorder()) {
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08001294 /*
1295 * Mark as guard pages (or page), that will allow to
1296 * merge back to allocator when buddy will be freed.
1297 * Corresponding page table entries will not be touched,
1298 * pages will stay not present in virtual address space
1299 */
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001300 set_page_guard(zone, &page[size], high, migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08001301 continue;
1302 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001303 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 area->nr_free++;
1305 set_page_order(&page[size], high);
1306 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307}
1308
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309/*
1310 * This page is about to be returned from the page allocator
1311 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001312static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -07001314 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -08001315 unsigned long bad_flags = 0;
1316
1317 if (unlikely(page_mapcount(page)))
1318 bad_reason = "nonzero mapcount";
1319 if (unlikely(page->mapping != NULL))
1320 bad_reason = "non-NULL mapping";
1321 if (unlikely(atomic_read(&page->_count) != 0))
1322 bad_reason = "nonzero _count";
Naoya Horiguchif4c18e62015-08-06 15:47:08 -07001323 if (unlikely(page->flags & __PG_HWPOISON)) {
1324 bad_reason = "HWPoisoned (hardware-corrupted)";
1325 bad_flags = __PG_HWPOISON;
1326 }
Dave Hansenf0b791a2014-01-23 15:52:49 -08001327 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1328 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1329 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1330 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -08001331#ifdef CONFIG_MEMCG
1332 if (unlikely(page->mem_cgroup))
1333 bad_reason = "page still charged to cgroup";
1334#endif
Dave Hansenf0b791a2014-01-23 15:52:49 -08001335 if (unlikely(bad_reason)) {
1336 bad_page(page, bad_reason, bad_flags);
Hugh Dickins689bceb2005-11-21 21:32:20 -08001337 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -08001338 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001339 return 0;
1340}
1341
Vlastimil Babka75379192015-02-11 15:25:38 -08001342static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1343 int alloc_flags)
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001344{
1345 int i;
1346
1347 for (i = 0; i < (1 << order); i++) {
1348 struct page *p = page + i;
1349 if (unlikely(check_new_page(p)))
1350 return 1;
1351 }
Hugh Dickins689bceb2005-11-21 21:32:20 -08001352
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001353 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -08001354 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -08001355
1356 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 kernel_map_pages(page, 1 << order, 1);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -08001358 kasan_alloc_pages(page, order);
Nick Piggin17cf4402006-03-22 00:08:41 -08001359
1360 if (gfp_flags & __GFP_ZERO)
Anisse Astierf4d28972015-06-24 16:56:36 -07001361 for (i = 0; i < (1 << order); i++)
1362 clear_highpage(page + i);
Nick Piggin17cf4402006-03-22 00:08:41 -08001363
1364 if (order && (gfp_flags & __GFP_COMP))
1365 prep_compound_page(page, order);
1366
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001367 set_page_owner(page, order, gfp_flags);
1368
Vlastimil Babka75379192015-02-11 15:25:38 -08001369 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001370 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
Vlastimil Babka75379192015-02-11 15:25:38 -08001371 * allocate the page. The expectation is that the caller is taking
1372 * steps that will free more memory. The caller should avoid the page
1373 * being used for !PFMEMALLOC purposes.
1374 */
Michal Hocko2f064f32015-08-21 14:11:51 -07001375 if (alloc_flags & ALLOC_NO_WATERMARKS)
1376 set_page_pfmemalloc(page);
1377 else
1378 clear_page_pfmemalloc(page);
Vlastimil Babka75379192015-02-11 15:25:38 -08001379
Hugh Dickins689bceb2005-11-21 21:32:20 -08001380 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381}
1382
Mel Gorman56fd56b2007-10-16 01:25:58 -07001383/*
1384 * Go through the free lists for the given migratetype and remove
1385 * the smallest available page from the freelists
1386 */
Mel Gorman728ec982009-06-16 15:32:04 -07001387static inline
1388struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -07001389 int migratetype)
1390{
1391 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -07001392 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001393 struct page *page;
1394
1395 /* Find a page of the appropriate size in the preferred list */
1396 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1397 area = &(zone->free_area[current_order]);
1398 if (list_empty(&area->free_list[migratetype]))
1399 continue;
1400
1401 page = list_entry(area->free_list[migratetype].next,
1402 struct page, lru);
1403 list_del(&page->lru);
1404 rmv_page_order(page);
1405 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001406 expand(zone, page, order, current_order, area, migratetype);
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001407 set_pcppage_migratetype(page, migratetype);
Mel Gorman56fd56b2007-10-16 01:25:58 -07001408 return page;
1409 }
1410
1411 return NULL;
1412}
1413
1414
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001415/*
1416 * This array describes the order lists are fallen back to when
1417 * the free lists for the desirable migrate type are depleted
1418 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001419static int fallbacks[MIGRATE_TYPES][4] = {
1420 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
1421 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001422 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
Joonsoo Kimdc676472015-04-14 15:45:15 -07001423#ifdef CONFIG_CMA
1424 [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001425#endif
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001426 [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001427#ifdef CONFIG_MEMORY_ISOLATION
Michal Nazarewicz6d4a4912012-01-11 15:31:33 +01001428 [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001429#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001430};
1431
Joonsoo Kimdc676472015-04-14 15:45:15 -07001432#ifdef CONFIG_CMA
1433static struct page *__rmqueue_cma_fallback(struct zone *zone,
1434 unsigned int order)
1435{
1436 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1437}
1438#else
1439static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1440 unsigned int order) { return NULL; }
1441#endif
1442
Mel Gormanc361be52007-10-16 01:25:51 -07001443/*
1444 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -07001445 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -07001446 * boundary. If alignment is required, use move_freepages_block()
1447 */
Minchan Kim435b4052012-10-08 16:32:16 -07001448int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -07001449 struct page *start_page, struct page *end_page,
1450 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001451{
1452 struct page *page;
1453 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -07001454 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -07001455
1456#ifndef CONFIG_HOLES_IN_ZONE
1457 /*
1458 * page_zone is not safe to call in this context when
1459 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1460 * anyway as we check zone boundaries in move_freepages_block().
1461 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -07001462 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -07001463 */
Mel Gorman97ee4ba2014-10-09 15:28:28 -07001464 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
Mel Gormanc361be52007-10-16 01:25:51 -07001465#endif
1466
1467 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -07001468 /* Make sure we are not inadvertently changing nodes */
Sasha Levin309381fea2014-01-23 15:52:54 -08001469 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
Adam Litke344c7902008-09-02 14:35:38 -07001470
Mel Gormanc361be52007-10-16 01:25:51 -07001471 if (!pfn_valid_within(page_to_pfn(page))) {
1472 page++;
1473 continue;
1474 }
1475
1476 if (!PageBuddy(page)) {
1477 page++;
1478 continue;
1479 }
1480
1481 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001482 list_move(&page->lru,
1483 &zone->free_area[order].free_list[migratetype]);
Mel Gormanc361be52007-10-16 01:25:51 -07001484 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001485 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001486 }
1487
Mel Gormand1003132007-10-16 01:26:00 -07001488 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001489}
1490
Minchan Kimee6f5092012-07-31 16:43:50 -07001491int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001492 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001493{
1494 unsigned long start_pfn, end_pfn;
1495 struct page *start_page, *end_page;
1496
1497 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001498 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001499 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001500 end_page = start_page + pageblock_nr_pages - 1;
1501 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001502
1503 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001504 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001505 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001506 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001507 return 0;
1508
1509 return move_freepages(zone, start_page, end_page, migratetype);
1510}
1511
Mel Gorman2f66a682009-09-21 17:02:31 -07001512static void change_pageblock_range(struct page *pageblock_page,
1513 int start_order, int migratetype)
1514{
1515 int nr_pageblocks = 1 << (start_order - pageblock_order);
1516
1517 while (nr_pageblocks--) {
1518 set_pageblock_migratetype(pageblock_page, migratetype);
1519 pageblock_page += pageblock_nr_pages;
1520 }
1521}
1522
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001523/*
Vlastimil Babka9c0415e2015-02-11 15:28:21 -08001524 * When we are falling back to another migratetype during allocation, try to
1525 * steal extra free pages from the same pageblocks to satisfy further
1526 * allocations, instead of polluting multiple pageblocks.
1527 *
1528 * If we are stealing a relatively large buddy page, it is likely there will
1529 * be more free pages in the pageblock, so try to steal them all. For
1530 * reclaimable and unmovable allocations, we steal regardless of page size,
1531 * as fragmentation caused by those allocations polluting movable pageblocks
1532 * is worse than movable allocations stealing from unmovable and reclaimable
1533 * pageblocks.
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001534 */
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001535static bool can_steal_fallback(unsigned int order, int start_mt)
1536{
1537 /*
1538 * Leaving this order check is intended, although there is
1539 * relaxed order check in next check. The reason is that
1540 * we can actually steal whole pageblock if this condition met,
1541 * but, below check doesn't guarantee it and that is just heuristic
1542 * so could be changed anytime.
1543 */
1544 if (order >= pageblock_order)
1545 return true;
1546
1547 if (order >= pageblock_order / 2 ||
1548 start_mt == MIGRATE_RECLAIMABLE ||
1549 start_mt == MIGRATE_UNMOVABLE ||
1550 page_group_by_mobility_disabled)
1551 return true;
1552
1553 return false;
1554}
1555
1556/*
1557 * This function implements actual steal behaviour. If order is large enough,
1558 * we can steal whole pageblock. If not, we first move freepages in this
1559 * pageblock and check whether half of pages are moved or not. If half of
1560 * pages are moved, we can change migratetype of pageblock and permanently
1561 * use it's pages as requested migratetype in the future.
1562 */
1563static void steal_suitable_fallback(struct zone *zone, struct page *page,
1564 int start_type)
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001565{
1566 int current_order = page_order(page);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001567 int pages;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001568
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001569 /* Take ownership for orders >= pageblock_order */
1570 if (current_order >= pageblock_order) {
1571 change_pageblock_range(page, current_order, start_type);
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001572 return;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001573 }
1574
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001575 pages = move_freepages_block(zone, page, start_type);
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001576
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001577 /* Claim the whole block if over half of it is free */
1578 if (pages >= (1 << (pageblock_order-1)) ||
1579 page_group_by_mobility_disabled)
1580 set_pageblock_migratetype(page, start_type);
1581}
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001582
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001583/*
1584 * Check whether there is a suitable fallback freepage with requested order.
1585 * If only_stealable is true, this function returns fallback_mt only if
1586 * we can steal other freepages all together. This would help to reduce
1587 * fragmentation due to mixed migratetype pages in one pageblock.
1588 */
1589int find_suitable_fallback(struct free_area *area, unsigned int order,
1590 int migratetype, bool only_stealable, bool *can_steal)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001591{
1592 int i;
1593 int fallback_mt;
1594
1595 if (area->nr_free == 0)
1596 return -1;
1597
1598 *can_steal = false;
1599 for (i = 0;; i++) {
1600 fallback_mt = fallbacks[migratetype][i];
1601 if (fallback_mt == MIGRATE_RESERVE)
1602 break;
1603
1604 if (list_empty(&area->free_list[fallback_mt]))
1605 continue;
1606
1607 if (can_steal_fallback(order, migratetype))
1608 *can_steal = true;
1609
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001610 if (!only_stealable)
1611 return fallback_mt;
1612
1613 if (*can_steal)
1614 return fallback_mt;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001615 }
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001616
1617 return -1;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001618}
1619
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001620/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001621static inline struct page *
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001622__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001623{
Pintu Kumarb8af2942013-09-11 14:20:34 -07001624 struct free_area *area;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001625 unsigned int current_order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001626 struct page *page;
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001627 int fallback_mt;
1628 bool can_steal;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001629
1630 /* Find the largest possible block of pages in the other list */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001631 for (current_order = MAX_ORDER-1;
1632 current_order >= order && current_order <= MAX_ORDER-1;
1633 --current_order) {
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001634 area = &(zone->free_area[current_order]);
1635 fallback_mt = find_suitable_fallback(area, current_order,
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001636 start_migratetype, false, &can_steal);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001637 if (fallback_mt == -1)
1638 continue;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001639
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001640 page = list_entry(area->free_list[fallback_mt].next,
1641 struct page, lru);
1642 if (can_steal)
1643 steal_suitable_fallback(zone, page, start_migratetype);
Mel Gormane0104872007-10-16 01:25:53 -07001644
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001645 /* Remove the page from the freelists */
1646 area->nr_free--;
1647 list_del(&page->lru);
1648 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001649
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001650 expand(zone, page, order, current_order, area,
1651 start_migratetype);
1652 /*
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001653 * The pcppage_migratetype may differ from pageblock's
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001654 * migratetype depending on the decisions in
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001655 * find_suitable_fallback(). This is OK as long as it does not
1656 * differ for MIGRATE_CMA pageblocks. Those can be used as
1657 * fallback only via special __rmqueue_cma_fallback() function
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001658 */
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001659 set_pcppage_migratetype(page, start_migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001660
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001661 trace_mm_page_alloc_extfrag(page, order, current_order,
1662 start_migratetype, fallback_mt);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001663
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001664 return page;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001665 }
1666
Mel Gorman728ec982009-06-16 15:32:04 -07001667 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001668}
1669
Mel Gorman56fd56b2007-10-16 01:25:58 -07001670/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 * Do the hard work of removing an element from the buddy allocator.
1672 * Call me with the zone->lock already held.
1673 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001674static struct page *__rmqueue(struct zone *zone, unsigned int order,
1675 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 struct page *page;
1678
Mel Gorman728ec982009-06-16 15:32:04 -07001679retry_reserve:
Mel Gorman56fd56b2007-10-16 01:25:58 -07001680 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681
Mel Gorman728ec982009-06-16 15:32:04 -07001682 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Joonsoo Kimdc676472015-04-14 15:45:15 -07001683 if (migratetype == MIGRATE_MOVABLE)
1684 page = __rmqueue_cma_fallback(zone, order);
1685
1686 if (!page)
1687 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001688
Mel Gorman728ec982009-06-16 15:32:04 -07001689 /*
1690 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1691 * is used because __rmqueue_smallest is an inline function
1692 * and we want just one call site
1693 */
1694 if (!page) {
1695 migratetype = MIGRATE_RESERVE;
1696 goto retry_reserve;
1697 }
1698 }
1699
Mel Gorman0d3d0622009-09-21 17:02:44 -07001700 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001701 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702}
1703
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001704/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 * Obtain a specified number of elements from the buddy allocator, all under
1706 * a single hold of the lock, for efficiency. Add them to the supplied list.
1707 * Returns the number of new pages which were placed at *list.
1708 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001709static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001710 unsigned long count, struct list_head *list,
Mel Gormanb745bc82014-06-04 16:10:22 -07001711 int migratetype, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712{
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001713 int i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001714
Nick Pigginc54ad302006-01-06 00:10:56 -08001715 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001717 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001718 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001720
1721 /*
1722 * Split buddy pages returned by expand() are received here
1723 * in physical page order. The page is added to the callers and
1724 * list and the list head then moves forward. From the callers
1725 * perspective, the linked list is ordered by page number in
1726 * some conditions. This is useful for IO devices that can
1727 * merge IO requests if the physical pages are ordered
1728 * properly.
1729 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001730 if (likely(!cold))
Mel Gormane084b2d2009-07-29 15:02:04 -07001731 list_add(&page->lru, list);
1732 else
1733 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001734 list = &page->lru;
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001735 if (is_migrate_cma(get_pcppage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001736 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1737 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001739 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001740 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001741 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742}
1743
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001744#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001745/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001746 * Called from the vmstat counter updater to drain pagesets of this
1747 * currently executing processor on remote nodes after they have
1748 * expired.
1749 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001750 * Note that this function must be called with the thread pinned to
1751 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001752 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001753void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001754{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001755 unsigned long flags;
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001756 int to_drain, batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001757
Christoph Lameter4037d452007-05-09 02:35:14 -07001758 local_irq_save(flags);
Jason Low4db0c3c2015-04-15 16:14:08 -07001759 batch = READ_ONCE(pcp->batch);
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001760 to_drain = min(pcp->count, batch);
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07001761 if (to_drain > 0) {
1762 free_pcppages_bulk(zone, to_drain, pcp);
1763 pcp->count -= to_drain;
1764 }
Christoph Lameter4037d452007-05-09 02:35:14 -07001765 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001766}
1767#endif
1768
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001769/*
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001770 * Drain pcplists of the indicated processor and zone.
1771 *
1772 * The processor must either be the current processor and the
1773 * thread pinned to the current processor or a processor that
1774 * is not online.
1775 */
1776static void drain_pages_zone(unsigned int cpu, struct zone *zone)
1777{
1778 unsigned long flags;
1779 struct per_cpu_pageset *pset;
1780 struct per_cpu_pages *pcp;
1781
1782 local_irq_save(flags);
1783 pset = per_cpu_ptr(zone->pageset, cpu);
1784
1785 pcp = &pset->pcp;
1786 if (pcp->count) {
1787 free_pcppages_bulk(zone, pcp->count, pcp);
1788 pcp->count = 0;
1789 }
1790 local_irq_restore(flags);
1791}
1792
1793/*
1794 * Drain pcplists of all zones on the indicated processor.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001795 *
1796 * The processor must either be the current processor and the
1797 * thread pinned to the current processor or a processor that
1798 * is not online.
1799 */
1800static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801{
1802 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001804 for_each_populated_zone(zone) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001805 drain_pages_zone(cpu, zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 }
1807}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001809/*
1810 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001811 *
1812 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1813 * the single zone's pages.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001814 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001815void drain_local_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001816{
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001817 int cpu = smp_processor_id();
1818
1819 if (zone)
1820 drain_pages_zone(cpu, zone);
1821 else
1822 drain_pages(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001823}
1824
1825/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001826 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1827 *
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001828 * When zone parameter is non-NULL, spill just the single zone's pages.
1829 *
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001830 * Note that this code is protected against sending an IPI to an offline
1831 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1832 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1833 * nothing keeps CPUs from showing up after we populated the cpumask and
1834 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001835 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001836void drain_all_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001837{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001838 int cpu;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001839
1840 /*
1841 * Allocate in the BSS so we wont require allocation in
1842 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1843 */
1844 static cpumask_t cpus_with_pcps;
1845
1846 /*
1847 * We don't care about racing with CPU hotplug event
1848 * as offline notification will cause the notified
1849 * cpu to drain that CPU pcps and on_each_cpu_mask
1850 * disables preemption as part of its processing
1851 */
1852 for_each_online_cpu(cpu) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001853 struct per_cpu_pageset *pcp;
1854 struct zone *z;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001855 bool has_pcps = false;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001856
1857 if (zone) {
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001858 pcp = per_cpu_ptr(zone->pageset, cpu);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001859 if (pcp->pcp.count)
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001860 has_pcps = true;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001861 } else {
1862 for_each_populated_zone(z) {
1863 pcp = per_cpu_ptr(z->pageset, cpu);
1864 if (pcp->pcp.count) {
1865 has_pcps = true;
1866 break;
1867 }
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001868 }
1869 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001870
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001871 if (has_pcps)
1872 cpumask_set_cpu(cpu, &cpus_with_pcps);
1873 else
1874 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1875 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001876 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
1877 zone, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001878}
1879
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001880#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881
1882void mark_free_pages(struct zone *zone)
1883{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001884 unsigned long pfn, max_zone_pfn;
1885 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001886 unsigned int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 struct list_head *curr;
1888
Xishi Qiu8080fc02013-09-11 14:21:45 -07001889 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 return;
1891
1892 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001893
Cody P Schafer108bcc92013-02-22 16:35:23 -08001894 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001895 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1896 if (pfn_valid(pfn)) {
1897 struct page *page = pfn_to_page(pfn);
1898
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001899 if (!swsusp_page_is_forbidden(page))
1900 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001901 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001903 for_each_migratetype_order(order, t) {
1904 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001905 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001907 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1908 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001909 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001910 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001911 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 spin_unlock_irqrestore(&zone->lock, flags);
1913}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001914#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915
1916/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 * Free a 0-order page
Mel Gormanb745bc82014-06-04 16:10:22 -07001918 * cold == true ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001920void free_hot_cold_page(struct page *page, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921{
1922 struct zone *zone = page_zone(page);
1923 struct per_cpu_pages *pcp;
1924 unsigned long flags;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001925 unsigned long pfn = page_to_pfn(page);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001926 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001928 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001929 return;
1930
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001931 migratetype = get_pfnblock_migratetype(page, pfn);
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001932 set_pcppage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001934 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001935
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001936 /*
1937 * We only track unmovable, reclaimable and movable on pcp lists.
1938 * Free ISOLATE pages back to the allocator because they are being
1939 * offlined but treat RESERVE as movable pages so we can get those
1940 * areas back if necessary. Otherwise, we may have to free
1941 * excessively into the page allocator
1942 */
1943 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08001944 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001945 free_one_page(zone, page, pfn, 0, migratetype);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001946 goto out;
1947 }
1948 migratetype = MIGRATE_MOVABLE;
1949 }
1950
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001951 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Mel Gormanb745bc82014-06-04 16:10:22 -07001952 if (!cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001953 list_add(&page->lru, &pcp->lists[migratetype]);
Mel Gormanb745bc82014-06-04 16:10:22 -07001954 else
1955 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001957 if (pcp->count >= pcp->high) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001958 unsigned long batch = READ_ONCE(pcp->batch);
Cody P Schafer998d39c2013-07-03 15:01:32 -07001959 free_pcppages_bulk(zone, batch, pcp);
1960 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08001961 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001962
1963out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965}
1966
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001967/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001968 * Free a list of 0-order pages
1969 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001970void free_hot_cold_page_list(struct list_head *list, bool cold)
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001971{
1972 struct page *page, *next;
1973
1974 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001975 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001976 free_hot_cold_page(page, cold);
1977 }
1978}
1979
1980/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001981 * split_page takes a non-compound higher-order page, and splits it into
1982 * n (1<<order) sub-pages: page[0..n]
1983 * Each sub-page must be freed individually.
1984 *
1985 * Note: this is probably too low level an operation for use in drivers.
1986 * Please consult with lkml before using this in your driver.
1987 */
1988void split_page(struct page *page, unsigned int order)
1989{
1990 int i;
Joonsoo Kime2cfc912015-07-17 16:24:18 -07001991 gfp_t gfp_mask;
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001992
Sasha Levin309381fea2014-01-23 15:52:54 -08001993 VM_BUG_ON_PAGE(PageCompound(page), page);
1994 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001995
1996#ifdef CONFIG_KMEMCHECK
1997 /*
1998 * Split shadow pages too, because free(page[0]) would
1999 * otherwise free the whole shadow.
2000 */
2001 if (kmemcheck_page_is_tracked(page))
2002 split_page(virt_to_page(page[0].shadow), order);
2003#endif
2004
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002005 gfp_mask = get_page_owner_gfp(page);
2006 set_page_owner(page, 0, gfp_mask);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08002007 for (i = 1; i < (1 << order); i++) {
Nick Piggin7835e982006-03-22 00:08:40 -08002008 set_page_refcounted(page + i);
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002009 set_page_owner(page + i, 0, gfp_mask);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08002010 }
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002011}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07002012EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002013
Joonsoo Kim3c605092014-11-13 15:19:21 -08002014int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07002015{
Mel Gorman748446b2010-05-24 14:32:27 -07002016 unsigned long watermark;
2017 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002018 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07002019
2020 BUG_ON(!PageBuddy(page));
2021
2022 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002023 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07002024
Minchan Kim194159f2013-02-22 16:33:58 -08002025 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002026 /* Obey watermarks as if the page was being allocated */
2027 watermark = low_wmark_pages(zone) + (1 << order);
2028 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2029 return 0;
2030
Mel Gorman8fb74b92013-01-11 14:32:16 -08002031 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002032 }
Mel Gorman748446b2010-05-24 14:32:27 -07002033
2034 /* Remove page from free list */
2035 list_del(&page->lru);
2036 zone->free_area[order].nr_free--;
2037 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002038
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002039 set_page_owner(page, order, __GFP_MOVABLE);
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002040
Mel Gorman8fb74b92013-01-11 14:32:16 -08002041 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07002042 if (order >= pageblock_order - 1) {
2043 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002044 for (; page < endpage; page += pageblock_nr_pages) {
2045 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08002046 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002047 set_pageblock_migratetype(page,
2048 MIGRATE_MOVABLE);
2049 }
Mel Gorman748446b2010-05-24 14:32:27 -07002050 }
2051
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002052
Mel Gorman8fb74b92013-01-11 14:32:16 -08002053 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002054}
2055
2056/*
2057 * Similar to split_page except the page is already free. As this is only
2058 * being used for migration, the migratetype of the block also changes.
2059 * As this is called with interrupts disabled, the caller is responsible
2060 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2061 * are enabled.
2062 *
2063 * Note: this is probably too low level an operation for use in drivers.
2064 * Please consult with lkml before using this in your driver.
2065 */
2066int split_free_page(struct page *page)
2067{
2068 unsigned int order;
2069 int nr_pages;
2070
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002071 order = page_order(page);
2072
Mel Gorman8fb74b92013-01-11 14:32:16 -08002073 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002074 if (!nr_pages)
2075 return 0;
2076
2077 /* Split into individual pages */
2078 set_page_refcounted(page);
2079 split_page(page, order);
2080 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07002081}
2082
2083/*
Vlastimil Babka75379192015-02-11 15:25:38 -08002084 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07002086static inline
2087struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002088 struct zone *zone, unsigned int order,
2089 gfp_t gfp_flags, int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090{
2091 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08002092 struct page *page;
Mel Gormanb745bc82014-06-04 16:10:22 -07002093 bool cold = ((gfp_flags & __GFP_COLD) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094
Nick Piggin48db57f2006-01-08 01:00:42 -08002095 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002097 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002100 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2101 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002102 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07002103 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002104 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07002105 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002106 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07002107 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07002108 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07002109
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002110 if (cold)
2111 page = list_entry(list->prev, struct page, lru);
2112 else
2113 page = list_entry(list->next, struct page, lru);
2114
Mel Gormanb92a6ed2007-10-16 01:25:50 -07002115 list_del(&page->lru);
2116 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002117 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07002118 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
2119 /*
2120 * __GFP_NOFAIL is not to be used in new code.
2121 *
2122 * All __GFP_NOFAIL callers should be fixed so that they
2123 * properly detect and handle allocation failures.
2124 *
2125 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07002126 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07002127 * __GFP_NOFAIL.
2128 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07002129 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07002130 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002132 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08002133 spin_unlock(&zone->lock);
2134 if (!page)
2135 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07002136 __mod_zone_freepage_state(zone, -(1 << order),
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002137 get_pcppage_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 }
2139
Johannes Weiner3a025762014-04-07 15:37:48 -07002140 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Johannes Weinerabe5f972014-10-02 16:21:10 -07002141 if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
Johannes Weiner57054652014-10-09 15:28:17 -07002142 !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
2143 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Johannes Weiner27329362014-03-03 15:38:41 -08002144
Christoph Lameterf8891e52006-06-30 01:55:45 -07002145 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07002146 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002147 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148
Sasha Levin309381fea2014-01-23 15:52:54 -08002149 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08002151
2152failed:
2153 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002154 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155}
2156
Akinobu Mita933e3122006-12-08 02:39:45 -08002157#ifdef CONFIG_FAIL_PAGE_ALLOC
2158
Akinobu Mitab2588c42011-07-26 16:09:03 -07002159static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08002160 struct fault_attr attr;
2161
2162 u32 ignore_gfp_highmem;
2163 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07002164 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08002165} fail_page_alloc = {
2166 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08002167 .ignore_gfp_wait = 1,
2168 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07002169 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08002170};
2171
2172static int __init setup_fail_page_alloc(char *str)
2173{
2174 return setup_fault_attr(&fail_page_alloc.attr, str);
2175}
2176__setup("fail_page_alloc=", setup_fail_page_alloc);
2177
Gavin Shandeaf3862012-07-31 16:41:51 -07002178static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002179{
Akinobu Mita54114992007-07-15 23:40:23 -07002180 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07002181 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002182 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07002183 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002184 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002185 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002186 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
Gavin Shandeaf3862012-07-31 16:41:51 -07002187 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002188
2189 return should_fail(&fail_page_alloc.attr, 1 << order);
2190}
2191
2192#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2193
2194static int __init fail_page_alloc_debugfs(void)
2195{
Al Virof4ae40a2011-07-24 04:33:43 -04002196 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08002197 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08002198
Akinobu Mitadd48c082011-08-03 16:21:01 -07002199 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2200 &fail_page_alloc.attr);
2201 if (IS_ERR(dir))
2202 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002203
Akinobu Mitab2588c42011-07-26 16:09:03 -07002204 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
2205 &fail_page_alloc.ignore_gfp_wait))
2206 goto fail;
2207 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2208 &fail_page_alloc.ignore_gfp_highmem))
2209 goto fail;
2210 if (!debugfs_create_u32("min-order", mode, dir,
2211 &fail_page_alloc.min_order))
2212 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08002213
Akinobu Mitab2588c42011-07-26 16:09:03 -07002214 return 0;
2215fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07002216 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002217
Akinobu Mitab2588c42011-07-26 16:09:03 -07002218 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08002219}
2220
2221late_initcall(fail_page_alloc_debugfs);
2222
2223#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2224
2225#else /* CONFIG_FAIL_PAGE_ALLOC */
2226
Gavin Shandeaf3862012-07-31 16:41:51 -07002227static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002228{
Gavin Shandeaf3862012-07-31 16:41:51 -07002229 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002230}
2231
2232#endif /* CONFIG_FAIL_PAGE_ALLOC */
2233
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08002235 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 * of the allocation.
2237 */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002238static bool __zone_watermark_ok(struct zone *z, unsigned int order,
2239 unsigned long mark, int classzone_idx, int alloc_flags,
2240 long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241{
Wei Yuan26086de2014-12-10 15:44:44 -08002242 /* free_pages may go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08002243 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 int o;
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002245 long free_cma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246
Michal Hockodf0a6da2012-01-10 15:08:02 -08002247 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002248 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002250 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 min -= min / 4;
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002252#ifdef CONFIG_CMA
2253 /* If allocation can't use CMA areas don't use free CMA pages */
2254 if (!(alloc_flags & ALLOC_CMA))
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002255 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002256#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002257
Mel Gorman3484b2d2014-08-06 16:07:14 -07002258 if (free_pages - free_cma <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08002259 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 for (o = 0; o < order; o++) {
2261 /* At the next order, this order's pages become unavailable */
2262 free_pages -= z->free_area[o].nr_free << o;
2263
2264 /* Require fewer higher order pages to be free */
2265 min >>= 1;
2266
2267 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002268 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08002270 return true;
2271}
2272
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002273bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002274 int classzone_idx, int alloc_flags)
2275{
2276 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2277 zone_page_state(z, NR_FREE_PAGES));
2278}
2279
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002280bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
2281 unsigned long mark, int classzone_idx, int alloc_flags)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002282{
2283 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2284
2285 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2286 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2287
2288 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2289 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290}
2291
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002292#ifdef CONFIG_NUMA
2293/*
2294 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
2295 * skip over zones that are not allowed by the cpuset, or that have
2296 * been recently (in last second) found to be nearly full. See further
2297 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02002298 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002299 *
Zhi Yong Wua1aeb652013-11-12 15:08:29 -08002300 * If the zonelist cache is present in the passed zonelist, then
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002301 * returns a pointer to the allowed node mask (either the current
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08002302 * tasks mems_allowed, or node_states[N_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002303 *
2304 * If the zonelist cache is not available for this zonelist, does
2305 * nothing and returns NULL.
2306 *
2307 * If the fullzones BITMAP in the zonelist cache is stale (more than
2308 * a second since last zap'd) then we zap it out (clear its bits.)
2309 *
2310 * We hold off even calling zlc_setup, until after we've checked the
2311 * first zone in the zonelist, on the theory that most allocations will
2312 * be satisfied from that first zone, so best to examine that zone as
2313 * quickly as we can.
2314 */
2315static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
2316{
2317 struct zonelist_cache *zlc; /* cached zonelist speedup info */
2318 nodemask_t *allowednodes; /* zonelist_cache approximation */
2319
2320 zlc = zonelist->zlcache_ptr;
2321 if (!zlc)
2322 return NULL;
2323
S.Caglar Onurf05111f2008-04-28 02:12:38 -07002324 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002325 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2326 zlc->last_full_zap = jiffies;
2327 }
2328
2329 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
2330 &cpuset_current_mems_allowed :
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08002331 &node_states[N_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002332 return allowednodes;
2333}
2334
2335/*
2336 * Given 'z' scanning a zonelist, run a couple of quick checks to see
2337 * if it is worth looking at further for free memory:
2338 * 1) Check that the zone isn't thought to be full (doesn't have its
2339 * bit set in the zonelist_cache fullzones BITMAP).
2340 * 2) Check that the zones node (obtained from the zonelist_cache
2341 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
2342 * Return true (non-zero) if zone is worth looking at further, or
2343 * else return false (zero) if it is not.
2344 *
2345 * This check -ignores- the distinction between various watermarks,
2346 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
2347 * found to be full for any variation of these watermarks, it will
2348 * be considered full for up to one second by all requests, unless
2349 * we are so low on memory on all allowed nodes that we are forced
2350 * into the second scan of the zonelist.
2351 *
2352 * In the second scan we ignore this zonelist cache and exactly
2353 * apply the watermarks to all zones, even it is slower to do so.
2354 * We are low on memory in the second scan, and should leave no stone
2355 * unturned looking for a free page.
2356 */
Mel Gormandd1a2392008-04-28 02:12:17 -07002357static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002358 nodemask_t *allowednodes)
2359{
2360 struct zonelist_cache *zlc; /* cached zonelist speedup info */
2361 int i; /* index of *z in zonelist zones */
2362 int n; /* node that zone *z is on */
2363
2364 zlc = zonelist->zlcache_ptr;
2365 if (!zlc)
2366 return 1;
2367
Mel Gormandd1a2392008-04-28 02:12:17 -07002368 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002369 n = zlc->z_to_n[i];
2370
2371 /* This zone is worth trying if it is allowed but not full */
2372 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
2373}
2374
2375/*
2376 * Given 'z' scanning a zonelist, set the corresponding bit in
2377 * zlc->fullzones, so that subsequent attempts to allocate a page
2378 * from that zone don't waste time re-examining it.
2379 */
Mel Gormandd1a2392008-04-28 02:12:17 -07002380static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002381{
2382 struct zonelist_cache *zlc; /* cached zonelist speedup info */
2383 int i; /* index of *z in zonelist zones */
2384
2385 zlc = zonelist->zlcache_ptr;
2386 if (!zlc)
2387 return;
2388
Mel Gormandd1a2392008-04-28 02:12:17 -07002389 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002390
2391 set_bit(i, zlc->fullzones);
2392}
2393
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002394/*
2395 * clear all zones full, called after direct reclaim makes progress so that
2396 * a zone that was recently full is not skipped over for up to a second
2397 */
2398static void zlc_clear_zones_full(struct zonelist *zonelist)
2399{
2400 struct zonelist_cache *zlc; /* cached zonelist speedup info */
2401
2402 zlc = zonelist->zlcache_ptr;
2403 if (!zlc)
2404 return;
2405
2406 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2407}
2408
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002409static bool zone_local(struct zone *local_zone, struct zone *zone)
2410{
Johannes Weinerfff40682013-12-20 14:54:12 +00002411 return local_zone->node == zone->node;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002412}
2413
David Rientjes957f8222012-10-08 16:33:24 -07002414static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2415{
Mel Gorman5f7a75a2014-06-04 16:07:15 -07002416 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2417 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07002418}
2419
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002420#else /* CONFIG_NUMA */
2421
2422static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
2423{
2424 return NULL;
2425}
2426
Mel Gormandd1a2392008-04-28 02:12:17 -07002427static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002428 nodemask_t *allowednodes)
2429{
2430 return 1;
2431}
2432
Mel Gormandd1a2392008-04-28 02:12:17 -07002433static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002434{
2435}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002436
2437static void zlc_clear_zones_full(struct zonelist *zonelist)
2438{
2439}
David Rientjes957f8222012-10-08 16:33:24 -07002440
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002441static bool zone_local(struct zone *local_zone, struct zone *zone)
2442{
2443 return true;
2444}
2445
David Rientjes957f8222012-10-08 16:33:24 -07002446static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2447{
2448 return true;
2449}
2450
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002451#endif /* CONFIG_NUMA */
2452
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002453static void reset_alloc_batches(struct zone *preferred_zone)
2454{
2455 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2456
2457 do {
2458 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2459 high_wmark_pages(zone) - low_wmark_pages(zone) -
2460 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner57054652014-10-09 15:28:17 -07002461 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002462 } while (zone++ != preferred_zone);
2463}
2464
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002465/*
Paul Jackson0798e512006-12-06 20:31:38 -08002466 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002467 * a page.
2468 */
2469static struct page *
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002470get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2471 const struct alloc_context *ac)
Martin Hicks753ee722005-06-21 17:14:41 -07002472{
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002473 struct zonelist *zonelist = ac->zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07002474 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002475 struct page *page = NULL;
Mel Gorman5117f452009-06-16 15:31:59 -07002476 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002477 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
2478 int zlc_active = 0; /* set if using zonelist_cache */
2479 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gormana6e21b142014-06-04 16:10:12 -07002480 bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) &&
2481 (gfp_mask & __GFP_WRITE);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002482 int nr_fair_skipped = 0;
2483 bool zonelist_rescan;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002484
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002485zonelist_scan:
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002486 zonelist_rescan = false;
2487
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002488 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002489 * Scan zonelist, looking for a zone with enough free.
Vladimir Davydov344736f2014-10-20 15:50:30 +04002490 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002491 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002492 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2493 ac->nodemask) {
Johannes Weinere085dbc2013-09-11 14:20:46 -07002494 unsigned long mark;
2495
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002496 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002497 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2498 continue;
Mel Gorman664eedd2014-06-04 16:10:08 -07002499 if (cpusets_enabled() &&
2500 (alloc_flags & ALLOC_CPUSET) &&
Vladimir Davydov344736f2014-10-20 15:50:30 +04002501 !cpuset_zone_allowed(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07002502 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08002503 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002504 * Distribute pages in proportion to the individual
2505 * zone size to ensure fair page aging. The zone a
2506 * page was allocated in should have no effect on the
2507 * time the page has in memory before being reclaimed.
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002508 */
Johannes Weiner3a025762014-04-07 15:37:48 -07002509 if (alloc_flags & ALLOC_FAIR) {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002510 if (!zone_local(ac->preferred_zone, zone))
Mel Gormanf7b5d642014-08-06 16:07:20 -07002511 break;
Johannes Weiner57054652014-10-09 15:28:17 -07002512 if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002513 nr_fair_skipped++;
Johannes Weiner3a025762014-04-07 15:37:48 -07002514 continue;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002515 }
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002516 }
2517 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08002518 * When allocating a page cache page for writing, we
2519 * want to get it from a zone that is within its dirty
2520 * limit, such that no single zone holds more than its
2521 * proportional share of globally allowed dirty pages.
2522 * The dirty limits take into account the zone's
2523 * lowmem reserves and high watermark so that kswapd
2524 * should be able to balance it without having to
2525 * write pages from its LRU list.
2526 *
2527 * This may look like it could increase pressure on
2528 * lower zones by failing allocations in higher zones
2529 * before they are full. But the pages that do spill
2530 * over are limited as the lower zones are protected
2531 * by this very same mechanism. It should not become
2532 * a practical burden to them.
2533 *
2534 * XXX: For now, allow allocations to potentially
2535 * exceed the per-zone dirty limit in the slowpath
2536 * (ALLOC_WMARK_LOW unset) before going into reclaim,
2537 * which is important when on a NUMA setup the allowed
2538 * zones are together not big enough to reach the
2539 * global limit. The proper fix for these situations
2540 * will require awareness of zones in the
2541 * dirty-throttling and the flusher threads.
2542 */
Mel Gormana6e21b142014-06-04 16:10:12 -07002543 if (consider_zone_dirty && !zone_dirty_ok(zone))
Mel Gorman800a1e72014-06-04 16:10:06 -07002544 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002545
Johannes Weinere085dbc2013-09-11 14:20:46 -07002546 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2547 if (!zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002548 ac->classzone_idx, alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07002549 int ret;
2550
Mel Gorman5dab2912014-06-04 16:10:14 -07002551 /* Checked here to keep the fast path fast */
2552 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2553 if (alloc_flags & ALLOC_NO_WATERMARKS)
2554 goto try_this_zone;
2555
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002556 if (IS_ENABLED(CONFIG_NUMA) &&
2557 !did_zlc_setup && nr_online_nodes > 1) {
Mel Gormancd38b112011-07-25 17:12:29 -07002558 /*
2559 * we do zlc_setup if there are multiple nodes
2560 * and before considering the first zone allowed
2561 * by the cpuset.
2562 */
2563 allowednodes = zlc_setup(zonelist, alloc_flags);
2564 zlc_active = 1;
2565 did_zlc_setup = 1;
2566 }
2567
David Rientjes957f8222012-10-08 16:33:24 -07002568 if (zone_reclaim_mode == 0 ||
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002569 !zone_allows_reclaim(ac->preferred_zone, zone))
Mel Gormanfa5e0842009-06-16 15:33:22 -07002570 goto this_zone_full;
2571
Mel Gormancd38b112011-07-25 17:12:29 -07002572 /*
2573 * As we may have just activated ZLC, check if the first
2574 * eligible zone has failed zone_reclaim recently.
2575 */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002576 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
Mel Gormancd38b112011-07-25 17:12:29 -07002577 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2578 continue;
2579
Mel Gormanfa5e0842009-06-16 15:33:22 -07002580 ret = zone_reclaim(zone, gfp_mask, order);
2581 switch (ret) {
2582 case ZONE_RECLAIM_NOSCAN:
2583 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07002584 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002585 case ZONE_RECLAIM_FULL:
2586 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002587 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002588 default:
2589 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002590 if (zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002591 ac->classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002592 goto try_this_zone;
2593
2594 /*
2595 * Failed to reclaim enough to meet watermark.
2596 * Only mark the zone full if checking the min
2597 * watermark or if we failed to reclaim just
2598 * 1<<order pages or else the page allocator
2599 * fastpath will prematurely mark zones full
2600 * when the watermark is between the low and
2601 * min watermarks.
2602 */
2603 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2604 ret == ZONE_RECLAIM_SOME)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002605 goto this_zone_full;
Mel Gormanfed27192013-04-29 15:07:57 -07002606
2607 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002608 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002609 }
2610
Mel Gormanfa5e0842009-06-16 15:33:22 -07002611try_this_zone:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002612 page = buffered_rmqueue(ac->preferred_zone, zone, order,
2613 gfp_mask, ac->migratetype);
Vlastimil Babka75379192015-02-11 15:25:38 -08002614 if (page) {
2615 if (prep_new_page(page, order, gfp_mask, alloc_flags))
2616 goto try_this_zone;
2617 return page;
2618 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002619this_zone_full:
Mel Gorman65bb3712014-06-04 16:10:05 -07002620 if (IS_ENABLED(CONFIG_NUMA) && zlc_active)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002621 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002622 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002623
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002624 /*
2625 * The first pass makes sure allocations are spread fairly within the
2626 * local node. However, the local node might have free pages left
2627 * after the fairness batches are exhausted, and remote zones haven't
2628 * even been considered yet. Try once more without fairness, and
2629 * include remote zones now, before entering the slowpath and waking
2630 * kswapd: prefer spilling to a remote zone over swapping locally.
2631 */
2632 if (alloc_flags & ALLOC_FAIR) {
2633 alloc_flags &= ~ALLOC_FAIR;
2634 if (nr_fair_skipped) {
2635 zonelist_rescan = true;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002636 reset_alloc_batches(ac->preferred_zone);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002637 }
2638 if (nr_online_nodes > 1)
2639 zonelist_rescan = true;
2640 }
2641
2642 if (unlikely(IS_ENABLED(CONFIG_NUMA) && zlc_active)) {
2643 /* Disable zlc cache for second zonelist scan */
2644 zlc_active = 0;
2645 zonelist_rescan = true;
2646 }
2647
2648 if (zonelist_rescan)
2649 goto zonelist_scan;
2650
2651 return NULL;
Martin Hicks753ee722005-06-21 17:14:41 -07002652}
2653
David Rientjes29423e772011-03-22 16:30:47 -07002654/*
2655 * Large machines with many possible nodes should not always dump per-node
2656 * meminfo in irq context.
2657 */
2658static inline bool should_suppress_show_mem(void)
2659{
2660 bool ret = false;
2661
2662#if NODES_SHIFT > 8
2663 ret = in_interrupt();
2664#endif
2665 return ret;
2666}
2667
Dave Hansena238ab52011-05-24 17:12:16 -07002668static DEFINE_RATELIMIT_STATE(nopage_rs,
2669 DEFAULT_RATELIMIT_INTERVAL,
2670 DEFAULT_RATELIMIT_BURST);
2671
2672void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2673{
Dave Hansena238ab52011-05-24 17:12:16 -07002674 unsigned int filter = SHOW_MEM_FILTER_NODES;
2675
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08002676 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2677 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002678 return;
2679
2680 /*
2681 * This documents exceptions given to allocations in certain
2682 * contexts that are allowed to allocate outside current's set
2683 * of allowed nodes.
2684 */
2685 if (!(gfp_mask & __GFP_NOMEMALLOC))
2686 if (test_thread_flag(TIF_MEMDIE) ||
2687 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2688 filter &= ~SHOW_MEM_FILTER_NODES;
2689 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2690 filter &= ~SHOW_MEM_FILTER_NODES;
2691
2692 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002693 struct va_format vaf;
2694 va_list args;
2695
Dave Hansena238ab52011-05-24 17:12:16 -07002696 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002697
2698 vaf.fmt = fmt;
2699 vaf.va = &args;
2700
2701 pr_warn("%pV", &vaf);
2702
Dave Hansena238ab52011-05-24 17:12:16 -07002703 va_end(args);
2704 }
2705
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002706 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2707 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002708
2709 dump_stack();
2710 if (!should_suppress_show_mem())
2711 show_mem(filter);
2712}
2713
Mel Gorman11e33f62009-06-16 15:31:57 -07002714static inline struct page *
2715__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002716 const struct alloc_context *ac, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07002717{
David Rientjes6e0fc462015-09-08 15:00:36 -07002718 struct oom_control oc = {
2719 .zonelist = ac->zonelist,
2720 .nodemask = ac->nodemask,
2721 .gfp_mask = gfp_mask,
2722 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07002723 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725
Johannes Weiner9879de72015-01-26 12:58:32 -08002726 *did_some_progress = 0;
2727
Johannes Weiner9879de72015-01-26 12:58:32 -08002728 /*
Johannes Weinerdc564012015-06-24 16:57:19 -07002729 * Acquire the oom lock. If that fails, somebody else is
2730 * making progress for us.
Johannes Weiner9879de72015-01-26 12:58:32 -08002731 */
Johannes Weinerdc564012015-06-24 16:57:19 -07002732 if (!mutex_trylock(&oom_lock)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002733 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002734 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 return NULL;
2736 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002737
Mel Gorman11e33f62009-06-16 15:31:57 -07002738 /*
2739 * Go through the zonelist yet one more time, keep very high watermark
2740 * here, this is only to catch a parallel oom killing, we must fail if
2741 * we're still under heavy pressure.
2742 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002743 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2744 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002745 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002746 goto out;
2747
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002748 if (!(gfp_mask & __GFP_NOFAIL)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002749 /* Coredumps can quickly deplete all memory reserves */
2750 if (current->flags & PF_DUMPCORE)
2751 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002752 /* The OOM killer will not help higher order allocs */
2753 if (order > PAGE_ALLOC_COSTLY_ORDER)
2754 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002755 /* The OOM killer does not needlessly kill tasks for lowmem */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002756 if (ac->high_zoneidx < ZONE_NORMAL)
David Rientjes03668b32010-08-09 17:18:54 -07002757 goto out;
Johannes Weiner90839052015-06-24 16:57:21 -07002758 /* The OOM killer does not compensate for IO-less reclaim */
Johannes Weinercc873172015-02-27 15:52:09 -08002759 if (!(gfp_mask & __GFP_FS)) {
2760 /*
2761 * XXX: Page reclaim didn't yield anything,
2762 * and the OOM killer can't be invoked, but
Johannes Weiner90839052015-06-24 16:57:21 -07002763 * keep looping as per tradition.
Johannes Weinercc873172015-02-27 15:52:09 -08002764 */
2765 *did_some_progress = 1;
Johannes Weiner9879de72015-01-26 12:58:32 -08002766 goto out;
Johannes Weinercc873172015-02-27 15:52:09 -08002767 }
Johannes Weiner90839052015-06-24 16:57:21 -07002768 if (pm_suspended_storage())
2769 goto out;
David Rientjes4167e9b2015-04-14 15:46:55 -07002770 /* The OOM killer may not free memory on a specific node */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002771 if (gfp_mask & __GFP_THISNODE)
2772 goto out;
2773 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002774 /* Exhausted what can be done so it's blamo time */
David Rientjes6e0fc462015-09-08 15:00:36 -07002775 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL))
Michal Hockoc32b3cb2015-02-11 15:26:24 -08002776 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002777out:
Johannes Weinerdc564012015-06-24 16:57:19 -07002778 mutex_unlock(&oom_lock);
Mel Gorman11e33f62009-06-16 15:31:57 -07002779 return page;
2780}
2781
Mel Gorman56de7262010-05-24 14:32:30 -07002782#ifdef CONFIG_COMPACTION
2783/* Try memory compaction for high-order allocations before reclaim */
2784static struct page *
2785__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002786 int alloc_flags, const struct alloc_context *ac,
2787 enum migrate_mode mode, int *contended_compaction,
2788 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002789{
Vlastimil Babka53853e22014-10-09 15:27:02 -07002790 unsigned long compact_result;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002791 struct page *page;
Vlastimil Babka53853e22014-10-09 15:27:02 -07002792
Mel Gorman66199712012-01-12 17:19:41 -08002793 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002794 return NULL;
2795
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002796 current->flags |= PF_MEMALLOC;
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08002797 compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2798 mode, contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002799 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002800
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002801 switch (compact_result) {
2802 case COMPACT_DEFERRED:
Vlastimil Babka53853e22014-10-09 15:27:02 -07002803 *deferred_compaction = true;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002804 /* fall-through */
2805 case COMPACT_SKIPPED:
2806 return NULL;
2807 default:
2808 break;
Mel Gorman56de7262010-05-24 14:32:30 -07002809 }
2810
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002811 /*
2812 * At least in one zone compaction wasn't deferred or skipped, so let's
2813 * count a compaction stall
2814 */
2815 count_vm_event(COMPACTSTALL);
2816
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002817 page = get_page_from_freelist(gfp_mask, order,
2818 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002819
2820 if (page) {
2821 struct zone *zone = page_zone(page);
2822
2823 zone->compact_blockskip_flush = false;
2824 compaction_defer_reset(zone, order, true);
2825 count_vm_event(COMPACTSUCCESS);
2826 return page;
2827 }
2828
2829 /*
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002830 * It's bad if compaction run occurs and fails. The most likely reason
2831 * is that pages exist, but not enough to satisfy watermarks.
2832 */
2833 count_vm_event(COMPACTFAIL);
2834
2835 cond_resched();
2836
Mel Gorman56de7262010-05-24 14:32:30 -07002837 return NULL;
2838}
2839#else
2840static inline struct page *
2841__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002842 int alloc_flags, const struct alloc_context *ac,
2843 enum migrate_mode mode, int *contended_compaction,
2844 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002845{
2846 return NULL;
2847}
2848#endif /* CONFIG_COMPACTION */
2849
Marek Szyprowskibba90712012-01-25 12:09:52 +01002850/* Perform direct synchronous page reclaim */
2851static int
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002852__perform_reclaim(gfp_t gfp_mask, unsigned int order,
2853 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002854{
Mel Gorman11e33f62009-06-16 15:31:57 -07002855 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002856 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002857
2858 cond_resched();
2859
2860 /* We now go into synchronous reclaim */
2861 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002862 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002863 lockdep_set_current_reclaim_state(gfp_mask);
2864 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002865 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002866
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002867 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
2868 ac->nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002869
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002870 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002871 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002872 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002873
2874 cond_resched();
2875
Marek Szyprowskibba90712012-01-25 12:09:52 +01002876 return progress;
2877}
2878
2879/* The really slow allocator path where we enter direct reclaim */
2880static inline struct page *
2881__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002882 int alloc_flags, const struct alloc_context *ac,
2883 unsigned long *did_some_progress)
Marek Szyprowskibba90712012-01-25 12:09:52 +01002884{
2885 struct page *page = NULL;
2886 bool drained = false;
2887
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002888 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002889 if (unlikely(!(*did_some_progress)))
2890 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002891
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002892 /* After successful reclaim, reconsider all zones for allocation */
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08002893 if (IS_ENABLED(CONFIG_NUMA))
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002894 zlc_clear_zones_full(ac->zonelist);
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002895
Mel Gorman9ee493c2010-09-09 16:38:18 -07002896retry:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002897 page = get_page_from_freelist(gfp_mask, order,
2898 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002899
2900 /*
2901 * If an allocation failed after direct reclaim, it could be because
2902 * pages are pinned on the per-cpu lists. Drain them and try again
2903 */
2904 if (!page && !drained) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002905 drain_all_pages(NULL);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002906 drained = true;
2907 goto retry;
2908 }
2909
Mel Gorman11e33f62009-06-16 15:31:57 -07002910 return page;
2911}
2912
Mel Gorman11e33f62009-06-16 15:31:57 -07002913/*
2914 * This is called in the allocator slow-path if the allocation request is of
2915 * sufficient urgency to ignore watermarks and take other desperate measures
2916 */
2917static inline struct page *
2918__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002919 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002920{
2921 struct page *page;
2922
2923 do {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002924 page = get_page_from_freelist(gfp_mask, order,
2925 ALLOC_NO_WATERMARKS, ac);
Mel Gorman11e33f62009-06-16 15:31:57 -07002926
2927 if (!page && gfp_mask & __GFP_NOFAIL)
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002928 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC,
2929 HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002930 } while (!page && (gfp_mask & __GFP_NOFAIL));
2931
2932 return page;
2933}
2934
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002935static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002936{
2937 struct zoneref *z;
2938 struct zone *zone;
2939
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002940 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
2941 ac->high_zoneidx, ac->nodemask)
2942 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
Mel Gorman11e33f62009-06-16 15:31:57 -07002943}
2944
Peter Zijlstra341ce062009-06-16 15:32:02 -07002945static inline int
2946gfp_to_alloc_flags(gfp_t gfp_mask)
2947{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002948 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
David Rientjesb104a352014-07-30 16:08:24 -07002949 const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD));
Peter Zijlstra341ce062009-06-16 15:32:02 -07002950
Mel Gormana56f57f2009-06-16 15:32:02 -07002951 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002952 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002953
Peter Zijlstra341ce062009-06-16 15:32:02 -07002954 /*
2955 * The caller may dip into page reserves a bit more if the caller
2956 * cannot run direct reclaim, or if the caller has realtime scheduling
2957 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
David Rientjesb104a352014-07-30 16:08:24 -07002958 * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
Peter Zijlstra341ce062009-06-16 15:32:02 -07002959 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002960 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002961
David Rientjesb104a352014-07-30 16:08:24 -07002962 if (atomic) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002963 /*
David Rientjesb104a352014-07-30 16:08:24 -07002964 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2965 * if it can't schedule.
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002966 */
David Rientjesb104a352014-07-30 16:08:24 -07002967 if (!(gfp_mask & __GFP_NOMEMALLOC))
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002968 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002969 /*
David Rientjesb104a352014-07-30 16:08:24 -07002970 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
Vladimir Davydov344736f2014-10-20 15:50:30 +04002971 * comment for __cpuset_node_allowed().
Peter Zijlstra341ce062009-06-16 15:32:02 -07002972 */
2973 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002974 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002975 alloc_flags |= ALLOC_HARDER;
2976
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002977 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2978 if (gfp_mask & __GFP_MEMALLOC)
2979 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07002980 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2981 alloc_flags |= ALLOC_NO_WATERMARKS;
2982 else if (!in_interrupt() &&
2983 ((current->flags & PF_MEMALLOC) ||
2984 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07002985 alloc_flags |= ALLOC_NO_WATERMARKS;
2986 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002987#ifdef CONFIG_CMA
David Rientjes43e7a342014-10-09 15:27:25 -07002988 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002989 alloc_flags |= ALLOC_CMA;
2990#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002991 return alloc_flags;
2992}
2993
Mel Gorman072bb0a2012-07-31 16:43:58 -07002994bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2995{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002996 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002997}
2998
Mel Gorman11e33f62009-06-16 15:31:57 -07002999static inline struct page *
3000__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003001 struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003002{
3003 const gfp_t wait = gfp_mask & __GFP_WAIT;
3004 struct page *page = NULL;
3005 int alloc_flags;
3006 unsigned long pages_reclaimed = 0;
3007 unsigned long did_some_progress;
David Rientjese0b9dae2014-06-04 16:08:28 -07003008 enum migrate_mode migration_mode = MIGRATE_ASYNC;
Mel Gorman66199712012-01-12 17:19:41 -08003009 bool deferred_compaction = false;
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003010 int contended_compaction = COMPACT_CONTENDED_NONE;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003011
Christoph Lameter952f3b52006-12-06 20:33:26 -08003012 /*
Mel Gorman72807a72009-06-16 15:32:18 -07003013 * In the slowpath, we sanity check order to avoid ever trying to
3014 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3015 * be using allocators in order of preference for an area that is
3016 * too large.
3017 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07003018 if (order >= MAX_ORDER) {
3019 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07003020 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07003021 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003022
Christoph Lameter952f3b52006-12-06 20:33:26 -08003023 /*
David Rientjes4167e9b2015-04-14 15:46:55 -07003024 * If this allocation cannot block and it is for a specific node, then
3025 * fail early. There's no need to wakeup kswapd or retry for a
3026 * speculative node-specific allocation.
Christoph Lameter952f3b52006-12-06 20:33:26 -08003027 */
David Rientjes4167e9b2015-04-14 15:46:55 -07003028 if (IS_ENABLED(CONFIG_NUMA) && (gfp_mask & __GFP_THISNODE) && !wait)
Christoph Lameter952f3b52006-12-06 20:33:26 -08003029 goto nopage;
3030
Johannes Weiner9879de72015-01-26 12:58:32 -08003031retry:
Johannes Weiner3a025762014-04-07 15:37:48 -07003032 if (!(gfp_mask & __GFP_NO_KSWAPD))
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003033 wake_all_kswapds(order, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003034
Paul Jackson9bf22292005-09-06 15:18:12 -07003035 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003036 * OK, we're below the kswapd watermark and have kicked background
3037 * reclaim. Now things get more complex, so set up alloc_flags according
3038 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07003039 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07003040 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041
David Rientjesf33261d2011-01-25 15:07:20 -08003042 /*
3043 * Find the true preferred zone if the allocation is unconstrained by
3044 * cpusets.
3045 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003046 if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
Mel Gormand8846372014-06-04 16:10:33 -07003047 struct zoneref *preferred_zoneref;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003048 preferred_zoneref = first_zones_zonelist(ac->zonelist,
3049 ac->high_zoneidx, NULL, &ac->preferred_zone);
3050 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gormand8846372014-06-04 16:10:33 -07003051 }
David Rientjesf33261d2011-01-25 15:07:20 -08003052
Peter Zijlstra341ce062009-06-16 15:32:02 -07003053 /* This is the last chance, in general, before the goto nopage. */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003054 page = get_page_from_freelist(gfp_mask, order,
3055 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003056 if (page)
3057 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058
Mel Gorman11e33f62009-06-16 15:31:57 -07003059 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07003060 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07003061 /*
3062 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3063 * the allocation is high priority and these type of
3064 * allocations are system rather than user orientated
3065 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003066 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
Mel Gorman183f6372012-07-31 16:44:12 -07003067
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003068 page = __alloc_pages_high_priority(gfp_mask, order, ac);
3069
Mel Gormancfd19c52012-07-31 16:44:10 -07003070 if (page) {
Peter Zijlstra341ce062009-06-16 15:32:02 -07003071 goto got_pg;
Mel Gormancfd19c52012-07-31 16:44:10 -07003072 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073 }
3074
3075 /* Atomic allocations - we can't balance anything */
David Rientjesaed0a0e2014-01-21 15:51:12 -08003076 if (!wait) {
3077 /*
3078 * All existing users of the deprecated __GFP_NOFAIL are
3079 * blockable, so warn of any new users that actually allow this
3080 * type of allocation to fail.
3081 */
3082 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083 goto nopage;
David Rientjesaed0a0e2014-01-21 15:51:12 -08003084 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085
Peter Zijlstra341ce062009-06-16 15:32:02 -07003086 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003087 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07003088 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089
David Rientjes6583bb62009-07-29 15:02:06 -07003090 /* Avoid allocations with no watermarks from looping endlessly */
3091 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3092 goto nopage;
3093
Mel Gorman77f1fe62011-01-13 15:45:57 -08003094 /*
3095 * Try direct compaction. The first pass is asynchronous. Subsequent
3096 * attempts after direct reclaim are synchronous
3097 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003098 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3099 migration_mode,
3100 &contended_compaction,
Vlastimil Babka53853e22014-10-09 15:27:02 -07003101 &deferred_compaction);
Mel Gorman56de7262010-05-24 14:32:30 -07003102 if (page)
3103 goto got_pg;
David Rientjes75f30862014-06-04 16:08:30 -07003104
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003105 /* Checks for THP-specific high-order allocations */
3106 if ((gfp_mask & GFP_TRANSHUGE) == GFP_TRANSHUGE) {
3107 /*
3108 * If compaction is deferred for high-order allocations, it is
3109 * because sync compaction recently failed. If this is the case
3110 * and the caller requested a THP allocation, we do not want
3111 * to heavily disrupt the system, so we fail the allocation
3112 * instead of entering direct reclaim.
3113 */
3114 if (deferred_compaction)
3115 goto nopage;
3116
3117 /*
3118 * In all zones where compaction was attempted (and not
3119 * deferred or skipped), lock contention has been detected.
3120 * For THP allocation we do not want to disrupt the others
3121 * so we fallback to base pages instead.
3122 */
3123 if (contended_compaction == COMPACT_CONTENDED_LOCK)
3124 goto nopage;
3125
3126 /*
3127 * If compaction was aborted due to need_resched(), we do not
3128 * want to further increase allocation latency, unless it is
3129 * khugepaged trying to collapse.
3130 */
3131 if (contended_compaction == COMPACT_CONTENDED_SCHED
3132 && !(current->flags & PF_KTHREAD))
3133 goto nopage;
3134 }
Mel Gorman66199712012-01-12 17:19:41 -08003135
David Rientjes8fe78042014-08-06 16:07:54 -07003136 /*
3137 * It can become very expensive to allocate transparent hugepages at
3138 * fault, so use asynchronous memory compaction for THP unless it is
3139 * khugepaged trying to collapse.
3140 */
3141 if ((gfp_mask & GFP_TRANSHUGE) != GFP_TRANSHUGE ||
3142 (current->flags & PF_KTHREAD))
3143 migration_mode = MIGRATE_SYNC_LIGHT;
3144
Mel Gorman11e33f62009-06-16 15:31:57 -07003145 /* Try direct reclaim and then allocating */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003146 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3147 &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07003148 if (page)
3149 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150
Johannes Weiner90839052015-06-24 16:57:21 -07003151 /* Do not loop if specifically requested */
3152 if (gfp_mask & __GFP_NORETRY)
3153 goto noretry;
3154
3155 /* Keep reclaiming pages as long as there is reasonable progress */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07003156 pages_reclaimed += did_some_progress;
Johannes Weiner90839052015-06-24 16:57:21 -07003157 if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) ||
3158 ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) {
Mel Gorman11e33f62009-06-16 15:31:57 -07003159 /* Wait for some write requests to complete then retry */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003160 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
Johannes Weiner9879de72015-01-26 12:58:32 -08003161 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162 }
3163
Johannes Weiner90839052015-06-24 16:57:21 -07003164 /* Reclaim has failed us, start killing things */
3165 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3166 if (page)
3167 goto got_pg;
3168
3169 /* Retry as long as the OOM killer is making progress */
3170 if (did_some_progress)
3171 goto retry;
3172
3173noretry:
3174 /*
3175 * High-order allocations do not necessarily loop after
3176 * direct reclaim and reclaim/compaction depends on compaction
3177 * being called after reclaim so call directly if necessary
3178 */
3179 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags,
3180 ac, migration_mode,
3181 &contended_compaction,
3182 &deferred_compaction);
3183 if (page)
3184 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07003186 warn_alloc_failed(gfp_mask, order, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187got_pg:
Mel Gorman072bb0a2012-07-31 16:43:58 -07003188 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003189}
Mel Gorman11e33f62009-06-16 15:31:57 -07003190
3191/*
3192 * This is the 'heart' of the zoned buddy allocator.
3193 */
3194struct page *
3195__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3196 struct zonelist *zonelist, nodemask_t *nodemask)
3197{
Mel Gormand8846372014-06-04 16:10:33 -07003198 struct zoneref *preferred_zoneref;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003199 struct page *page = NULL;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003200 unsigned int cpuset_mems_cookie;
Johannes Weiner3a025762014-04-07 15:37:48 -07003201 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
Andrew Morton91fbdc02015-02-11 15:25:04 -08003202 gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003203 struct alloc_context ac = {
3204 .high_zoneidx = gfp_zone(gfp_mask),
3205 .nodemask = nodemask,
3206 .migratetype = gfpflags_to_migratetype(gfp_mask),
3207 };
Mel Gorman11e33f62009-06-16 15:31:57 -07003208
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003209 gfp_mask &= gfp_allowed_mask;
3210
Mel Gorman11e33f62009-06-16 15:31:57 -07003211 lockdep_trace_alloc(gfp_mask);
3212
3213 might_sleep_if(gfp_mask & __GFP_WAIT);
3214
3215 if (should_fail_alloc_page(gfp_mask, order))
3216 return NULL;
3217
3218 /*
3219 * Check the zones suitable for the gfp_mask contain at least one
3220 * valid zone. It's possible to have an empty zonelist as a result
David Rientjes4167e9b2015-04-14 15:46:55 -07003221 * of __GFP_THISNODE and a memoryless node
Mel Gorman11e33f62009-06-16 15:31:57 -07003222 */
3223 if (unlikely(!zonelist->_zonerefs->zone))
3224 return NULL;
3225
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003226 if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
Vlastimil Babka21bb9bd2014-10-09 15:26:51 -07003227 alloc_flags |= ALLOC_CMA;
3228
Mel Gormancc9a6c82012-03-21 16:34:11 -07003229retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07003230 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003231
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003232 /* We set it here, as __alloc_pages_slowpath might have changed it */
3233 ac.zonelist = zonelist;
Mel Gorman5117f452009-06-16 15:31:59 -07003234 /* The preferred zone is used for statistics later */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003235 preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
3236 ac.nodemask ? : &cpuset_current_mems_allowed,
3237 &ac.preferred_zone);
3238 if (!ac.preferred_zone)
Mel Gormancc9a6c82012-03-21 16:34:11 -07003239 goto out;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003240 ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gorman5117f452009-06-16 15:31:59 -07003241
3242 /* First allocation attempt */
Andrew Morton91fbdc02015-02-11 15:25:04 -08003243 alloc_mask = gfp_mask|__GFP_HARDWALL;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003244 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08003245 if (unlikely(!page)) {
3246 /*
3247 * Runtime PM, block IO and its error handling path
3248 * can deadlock because I/O on the device might not
3249 * complete.
3250 */
Andrew Morton91fbdc02015-02-11 15:25:04 -08003251 alloc_mask = memalloc_noio_flags(gfp_mask);
3252
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003253 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08003254 }
Mel Gorman11e33f62009-06-16 15:31:57 -07003255
Xishi Qiu23f086f2015-02-11 15:25:07 -08003256 if (kmemcheck_enabled && page)
3257 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3258
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003259 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07003260
3261out:
3262 /*
3263 * When updating a task's mems_allowed, it is possible to race with
3264 * parallel threads in such a way that an allocation can fail while
3265 * the mask is being updated. If a page allocation is about to fail,
3266 * check if the cpuset changed during allocation and if so, retry.
3267 */
Mel Gormand26914d2014-04-03 14:47:24 -07003268 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07003269 goto retry_cpuset;
3270
Mel Gorman11e33f62009-06-16 15:31:57 -07003271 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003272}
Mel Gormand2391712009-06-16 15:31:52 -07003273EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274
3275/*
3276 * Common helper functions.
3277 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08003278unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279{
Akinobu Mita945a1112009-09-21 17:01:47 -07003280 struct page *page;
3281
3282 /*
3283 * __get_free_pages() returns a 32-bit address, which cannot represent
3284 * a highmem page
3285 */
3286 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3287
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288 page = alloc_pages(gfp_mask, order);
3289 if (!page)
3290 return 0;
3291 return (unsigned long) page_address(page);
3292}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003293EXPORT_SYMBOL(__get_free_pages);
3294
Harvey Harrison920c7a52008-02-04 22:29:26 -08003295unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296{
Akinobu Mita945a1112009-09-21 17:01:47 -07003297 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299EXPORT_SYMBOL(get_zeroed_page);
3300
Harvey Harrison920c7a52008-02-04 22:29:26 -08003301void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302{
Nick Pigginb5810032005-10-29 18:16:12 -07003303 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003304 if (order == 0)
Mel Gormanb745bc82014-06-04 16:10:22 -07003305 free_hot_cold_page(page, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306 else
3307 __free_pages_ok(page, order);
3308 }
3309}
3310
3311EXPORT_SYMBOL(__free_pages);
3312
Harvey Harrison920c7a52008-02-04 22:29:26 -08003313void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314{
3315 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07003316 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317 __free_pages(virt_to_page((void *)addr), order);
3318 }
3319}
3320
3321EXPORT_SYMBOL(free_pages);
3322
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003323/*
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003324 * Page Fragment:
3325 * An arbitrary-length arbitrary-offset area of memory which resides
3326 * within a 0 or higher order page. Multiple fragments within that page
3327 * are individually refcounted, in the page's reference counter.
3328 *
3329 * The page_frag functions below provide a simple allocation framework for
3330 * page fragments. This is used by the network stack and network device
3331 * drivers to provide a backing region of memory for use as either an
3332 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3333 */
3334static struct page *__page_frag_refill(struct page_frag_cache *nc,
3335 gfp_t gfp_mask)
3336{
3337 struct page *page = NULL;
3338 gfp_t gfp = gfp_mask;
3339
3340#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3341 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3342 __GFP_NOMEMALLOC;
3343 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3344 PAGE_FRAG_CACHE_MAX_ORDER);
3345 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3346#endif
3347 if (unlikely(!page))
3348 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3349
3350 nc->va = page ? page_address(page) : NULL;
3351
3352 return page;
3353}
3354
3355void *__alloc_page_frag(struct page_frag_cache *nc,
3356 unsigned int fragsz, gfp_t gfp_mask)
3357{
3358 unsigned int size = PAGE_SIZE;
3359 struct page *page;
3360 int offset;
3361
3362 if (unlikely(!nc->va)) {
3363refill:
3364 page = __page_frag_refill(nc, gfp_mask);
3365 if (!page)
3366 return NULL;
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 /* Even if we own the page, we do not use atomic_set().
3373 * This would break get_page_unless_zero() users.
3374 */
3375 atomic_add(size - 1, &page->_count);
3376
3377 /* reset page count bias and offset to start of new frag */
Michal Hocko2f064f32015-08-21 14:11:51 -07003378 nc->pfmemalloc = page_is_pfmemalloc(page);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003379 nc->pagecnt_bias = size;
3380 nc->offset = size;
3381 }
3382
3383 offset = nc->offset - fragsz;
3384 if (unlikely(offset < 0)) {
3385 page = virt_to_page(nc->va);
3386
3387 if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count))
3388 goto refill;
3389
3390#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3391 /* if size can vary use size else just use PAGE_SIZE */
3392 size = nc->size;
3393#endif
3394 /* OK, page count is 0, we can safely set it */
3395 atomic_set(&page->_count, size);
3396
3397 /* reset page count bias and offset to start of new frag */
3398 nc->pagecnt_bias = size;
3399 offset = size - fragsz;
3400 }
3401
3402 nc->pagecnt_bias--;
3403 nc->offset = offset;
3404
3405 return nc->va + offset;
3406}
3407EXPORT_SYMBOL(__alloc_page_frag);
3408
3409/*
3410 * Frees a page fragment allocated out of either a compound or order 0 page.
3411 */
3412void __free_page_frag(void *addr)
3413{
3414 struct page *page = virt_to_head_page(addr);
3415
3416 if (unlikely(put_page_testzero(page)))
3417 __free_pages_ok(page, compound_order(page));
3418}
3419EXPORT_SYMBOL(__free_page_frag);
3420
3421/*
Vladimir Davydov52383432014-06-04 16:06:39 -07003422 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3423 * of the current memory cgroup.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003424 *
Vladimir Davydov52383432014-06-04 16:06:39 -07003425 * It should be used when the caller would like to use kmalloc, but since the
3426 * allocation is large, it has to fall back to the page allocator.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003427 */
Vladimir Davydov52383432014-06-04 16:06:39 -07003428struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
3429{
3430 struct page *page;
3431 struct mem_cgroup *memcg = NULL;
3432
3433 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
3434 return NULL;
3435 page = alloc_pages(gfp_mask, order);
3436 memcg_kmem_commit_charge(page, memcg, order);
3437 return page;
3438}
3439
3440struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
3441{
3442 struct page *page;
3443 struct mem_cgroup *memcg = NULL;
3444
3445 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
3446 return NULL;
3447 page = alloc_pages_node(nid, gfp_mask, order);
3448 memcg_kmem_commit_charge(page, memcg, order);
3449 return page;
3450}
3451
3452/*
3453 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3454 * alloc_kmem_pages.
3455 */
3456void __free_kmem_pages(struct page *page, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003457{
3458 memcg_kmem_uncharge_pages(page, order);
3459 __free_pages(page, order);
3460}
3461
Vladimir Davydov52383432014-06-04 16:06:39 -07003462void free_kmem_pages(unsigned long addr, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003463{
3464 if (addr != 0) {
3465 VM_BUG_ON(!virt_addr_valid((void *)addr));
Vladimir Davydov52383432014-06-04 16:06:39 -07003466 __free_kmem_pages(virt_to_page((void *)addr), order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003467 }
3468}
3469
Andi Kleenee85c2e2011-05-11 15:13:34 -07003470static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
3471{
3472 if (addr) {
3473 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3474 unsigned long used = addr + PAGE_ALIGN(size);
3475
3476 split_page(virt_to_page((void *)addr), order);
3477 while (used < alloc_end) {
3478 free_page(used);
3479 used += PAGE_SIZE;
3480 }
3481 }
3482 return (void *)addr;
3483}
3484
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003485/**
3486 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3487 * @size: the number of bytes to allocate
3488 * @gfp_mask: GFP flags for the allocation
3489 *
3490 * This function is similar to alloc_pages(), except that it allocates the
3491 * minimum number of pages to satisfy the request. alloc_pages() can only
3492 * allocate memory in power-of-two pages.
3493 *
3494 * This function is also limited by MAX_ORDER.
3495 *
3496 * Memory allocated by this function must be released by free_pages_exact().
3497 */
3498void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3499{
3500 unsigned int order = get_order(size);
3501 unsigned long addr;
3502
3503 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07003504 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003505}
3506EXPORT_SYMBOL(alloc_pages_exact);
3507
3508/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07003509 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3510 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07003511 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07003512 * @size: the number of bytes to allocate
3513 * @gfp_mask: GFP flags for the allocation
3514 *
3515 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3516 * back.
Andi Kleenee85c2e2011-05-11 15:13:34 -07003517 */
Fabian Fredericke1931812014-08-06 16:04:59 -07003518void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
Andi Kleenee85c2e2011-05-11 15:13:34 -07003519{
3520 unsigned order = get_order(size);
3521 struct page *p = alloc_pages_node(nid, gfp_mask, order);
3522 if (!p)
3523 return NULL;
3524 return make_alloc_exact((unsigned long)page_address(p), order, size);
3525}
Andi Kleenee85c2e2011-05-11 15:13:34 -07003526
3527/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003528 * free_pages_exact - release memory allocated via alloc_pages_exact()
3529 * @virt: the value returned by alloc_pages_exact.
3530 * @size: size of allocation, same value as passed to alloc_pages_exact().
3531 *
3532 * Release the memory allocated by a previous call to alloc_pages_exact.
3533 */
3534void free_pages_exact(void *virt, size_t size)
3535{
3536 unsigned long addr = (unsigned long)virt;
3537 unsigned long end = addr + PAGE_ALIGN(size);
3538
3539 while (addr < end) {
3540 free_page(addr);
3541 addr += PAGE_SIZE;
3542 }
3543}
3544EXPORT_SYMBOL(free_pages_exact);
3545
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003546/**
3547 * nr_free_zone_pages - count number of pages beyond high watermark
3548 * @offset: The zone index of the highest zone
3549 *
3550 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3551 * high watermark within all zones at or below a given zone index. For each
3552 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07003553 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003554 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003555static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003556{
Mel Gormandd1a2392008-04-28 02:12:17 -07003557 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003558 struct zone *zone;
3559
Martin J. Blighe310fd42005-07-29 22:59:18 -07003560 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003561 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562
Mel Gorman0e884602008-04-28 02:12:14 -07003563 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003564
Mel Gorman54a6eb52008-04-28 02:12:16 -07003565 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08003566 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003567 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07003568 if (size > high)
3569 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570 }
3571
3572 return sum;
3573}
3574
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003575/**
3576 * nr_free_buffer_pages - count number of pages beyond high watermark
3577 *
3578 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3579 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003581unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582{
Al Viroaf4ca452005-10-21 02:55:38 -04003583 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584}
Meelap Shahc2f1a552007-07-17 04:04:39 -07003585EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003586
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003587/**
3588 * nr_free_pagecache_pages - count number of pages beyond high watermark
3589 *
3590 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3591 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003592 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003593unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594{
Mel Gorman2a1e2742007-07-17 04:03:12 -07003595 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003596}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07003597
3598static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003599{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08003600 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08003601 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003602}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003603
Linus Torvalds1da177e2005-04-16 15:20:36 -07003604void si_meminfo(struct sysinfo *val)
3605{
3606 val->totalram = totalram_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003607 val->sharedram = global_page_state(NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003608 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003609 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003610 val->totalhigh = totalhigh_pages;
3611 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612 val->mem_unit = PAGE_SIZE;
3613}
3614
3615EXPORT_SYMBOL(si_meminfo);
3616
3617#ifdef CONFIG_NUMA
3618void si_meminfo_node(struct sysinfo *val, int nid)
3619{
Jiang Liucdd91a72013-07-03 15:03:27 -07003620 int zone_type; /* needs to be signed */
3621 unsigned long managed_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003622 pg_data_t *pgdat = NODE_DATA(nid);
3623
Jiang Liucdd91a72013-07-03 15:03:27 -07003624 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3625 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3626 val->totalram = managed_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003627 val->sharedram = node_page_state(nid, NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003628 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003629#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08003630 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08003631 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3632 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003633#else
3634 val->totalhigh = 0;
3635 val->freehigh = 0;
3636#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003637 val->mem_unit = PAGE_SIZE;
3638}
3639#endif
3640
David Rientjesddd588b2011-03-22 16:30:46 -07003641/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07003642 * Determine whether the node should be displayed or not, depending on whether
3643 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07003644 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003645bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07003646{
3647 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003648 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07003649
3650 if (!(flags & SHOW_MEM_FILTER_NODES))
3651 goto out;
3652
Mel Gormancc9a6c82012-03-21 16:34:11 -07003653 do {
Mel Gormand26914d2014-04-03 14:47:24 -07003654 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003655 ret = !node_isset(nid, cpuset_current_mems_allowed);
Mel Gormand26914d2014-04-03 14:47:24 -07003656 } while (read_mems_allowed_retry(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07003657out:
3658 return ret;
3659}
3660
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661#define K(x) ((x) << (PAGE_SHIFT-10))
3662
Rabin Vincent377e4f12012-12-11 16:00:24 -08003663static void show_migration_types(unsigned char type)
3664{
3665 static const char types[MIGRATE_TYPES] = {
3666 [MIGRATE_UNMOVABLE] = 'U',
3667 [MIGRATE_RECLAIMABLE] = 'E',
3668 [MIGRATE_MOVABLE] = 'M',
3669 [MIGRATE_RESERVE] = 'R',
3670#ifdef CONFIG_CMA
3671 [MIGRATE_CMA] = 'C',
3672#endif
Minchan Kim194159f2013-02-22 16:33:58 -08003673#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08003674 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08003675#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08003676 };
3677 char tmp[MIGRATE_TYPES + 1];
3678 char *p = tmp;
3679 int i;
3680
3681 for (i = 0; i < MIGRATE_TYPES; i++) {
3682 if (type & (1 << i))
3683 *p++ = types[i];
3684 }
3685
3686 *p = '\0';
3687 printk("(%s) ", tmp);
3688}
3689
Linus Torvalds1da177e2005-04-16 15:20:36 -07003690/*
3691 * Show free area list (used inside shift_scroll-lock stuff)
3692 * We also calculate the percentage fragmentation. We do this by counting the
3693 * memory on each free list with the exception of the first item on the list.
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003694 *
3695 * Bits in @filter:
3696 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3697 * cpuset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003698 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003699void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003700{
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003701 unsigned long free_pcp = 0;
Jes Sorensenc7241912006-09-27 01:50:05 -07003702 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003703 struct zone *zone;
3704
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003705 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07003706 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003707 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003708
Konstantin Khlebnikov761b0672015-04-14 15:45:32 -07003709 for_each_online_cpu(cpu)
3710 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003711 }
3712
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003713 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3714 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003715 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3716 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003717 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003718 " free:%lu free_pcp:%lu free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003719 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003720 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003721 global_page_state(NR_ISOLATED_ANON),
3722 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003723 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003724 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003725 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003726 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003727 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003728 global_page_state(NR_UNSTABLE_NFS),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003729 global_page_state(NR_SLAB_RECLAIMABLE),
3730 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003731 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003732 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003733 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003734 global_page_state(NR_BOUNCE),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003735 global_page_state(NR_FREE_PAGES),
3736 free_pcp,
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003737 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003739 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003740 int i;
3741
David Rientjes7bf02ea2011-05-24 17:11:16 -07003742 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003743 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003744
3745 free_pcp = 0;
3746 for_each_online_cpu(cpu)
3747 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3748
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749 show_node(zone);
3750 printk("%s"
3751 " free:%lukB"
3752 " min:%lukB"
3753 " low:%lukB"
3754 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003755 " active_anon:%lukB"
3756 " inactive_anon:%lukB"
3757 " active_file:%lukB"
3758 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003759 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003760 " isolated(anon):%lukB"
3761 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003763 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003764 " mlocked:%lukB"
3765 " dirty:%lukB"
3766 " writeback:%lukB"
3767 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003768 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003769 " slab_reclaimable:%lukB"
3770 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003771 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003772 " pagetables:%lukB"
3773 " unstable:%lukB"
3774 " bounce:%lukB"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003775 " free_pcp:%lukB"
3776 " local_pcp:%ukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003777 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003778 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003779 " pages_scanned:%lu"
3780 " all_unreclaimable? %s"
3781 "\n",
3782 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003783 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003784 K(min_wmark_pages(zone)),
3785 K(low_wmark_pages(zone)),
3786 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003787 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3788 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3789 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3790 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003791 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003792 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3793 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003795 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003796 K(zone_page_state(zone, NR_MLOCK)),
3797 K(zone_page_state(zone, NR_FILE_DIRTY)),
3798 K(zone_page_state(zone, NR_WRITEBACK)),
3799 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003800 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003801 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3802 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003803 zone_page_state(zone, NR_KERNEL_STACK) *
3804 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003805 K(zone_page_state(zone, NR_PAGETABLE)),
3806 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3807 K(zone_page_state(zone, NR_BOUNCE)),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003808 K(free_pcp),
3809 K(this_cpu_read(zone->pageset->pcp.count)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003810 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003811 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Mel Gorman0d5d8232014-08-06 16:07:16 -07003812 K(zone_page_state(zone, NR_PAGES_SCANNED)),
Lisa Du6e543d52013-09-11 14:22:36 -07003813 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814 );
3815 printk("lowmem_reserve[]:");
3816 for (i = 0; i < MAX_NR_ZONES; i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07003817 printk(" %ld", zone->lowmem_reserve[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003818 printk("\n");
3819 }
3820
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003821 for_each_populated_zone(zone) {
Pintu Kumarb8af2942013-09-11 14:20:34 -07003822 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003823 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003824
David Rientjes7bf02ea2011-05-24 17:11:16 -07003825 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003826 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827 show_node(zone);
3828 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003829
3830 spin_lock_irqsave(&zone->lock, flags);
3831 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08003832 struct free_area *area = &zone->free_area[order];
3833 int type;
3834
3835 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003836 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003837
3838 types[order] = 0;
3839 for (type = 0; type < MIGRATE_TYPES; type++) {
3840 if (!list_empty(&area->free_list[type]))
3841 types[order] |= 1 << type;
3842 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003843 }
3844 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003845 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003846 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003847 if (nr[order])
3848 show_migration_types(types[order]);
3849 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003850 printk("= %lukB\n", K(total));
3851 }
3852
David Rientjes949f7ec2013-04-29 15:07:48 -07003853 hugetlb_show_meminfo();
3854
Larry Woodmane6f36022008-02-04 22:29:30 -08003855 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3856
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857 show_swap_cache_info();
3858}
3859
Mel Gorman19770b32008-04-28 02:12:18 -07003860static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3861{
3862 zoneref->zone = zone;
3863 zoneref->zone_idx = zone_idx(zone);
3864}
3865
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866/*
3867 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003868 *
3869 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003870 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003871static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003872 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003873{
Christoph Lameter1a932052006-01-06 00:11:16 -08003874 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003875 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003876
3877 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003878 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003879 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003880 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003881 zoneref_set_zone(zone,
3882 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003883 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003884 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003885 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003886
Christoph Lameter070f8032006-01-06 00:11:19 -08003887 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888}
3889
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003890
3891/*
3892 * zonelist_order:
3893 * 0 = automatic detection of better ordering.
3894 * 1 = order by ([node] distance, -zonetype)
3895 * 2 = order by (-zonetype, [node] distance)
3896 *
3897 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3898 * the same zonelist. So only NUMA can configure this param.
3899 */
3900#define ZONELIST_ORDER_DEFAULT 0
3901#define ZONELIST_ORDER_NODE 1
3902#define ZONELIST_ORDER_ZONE 2
3903
3904/* zonelist order in the kernel.
3905 * set_zonelist_order() will set this to NODE or ZONE.
3906 */
3907static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3908static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3909
3910
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003912/* The value user specified ....changed by config */
3913static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3914/* string for sysctl */
3915#define NUMA_ZONELIST_ORDER_LEN 16
3916char numa_zonelist_order[16] = "default";
3917
3918/*
3919 * interface for configure zonelist ordering.
3920 * command line option "numa_zonelist_order"
3921 * = "[dD]efault - default, automatic configuration.
3922 * = "[nN]ode - order by node locality, then by zone within node
3923 * = "[zZ]one - order by zone, then by locality within zone
3924 */
3925
3926static int __parse_numa_zonelist_order(char *s)
3927{
3928 if (*s == 'd' || *s == 'D') {
3929 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3930 } else if (*s == 'n' || *s == 'N') {
3931 user_zonelist_order = ZONELIST_ORDER_NODE;
3932 } else if (*s == 'z' || *s == 'Z') {
3933 user_zonelist_order = ZONELIST_ORDER_ZONE;
3934 } else {
3935 printk(KERN_WARNING
3936 "Ignoring invalid numa_zonelist_order value: "
3937 "%s\n", s);
3938 return -EINVAL;
3939 }
3940 return 0;
3941}
3942
3943static __init int setup_numa_zonelist_order(char *s)
3944{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003945 int ret;
3946
3947 if (!s)
3948 return 0;
3949
3950 ret = __parse_numa_zonelist_order(s);
3951 if (ret == 0)
3952 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3953
3954 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003955}
3956early_param("numa_zonelist_order", setup_numa_zonelist_order);
3957
3958/*
3959 * sysctl handler for numa_zonelist_order
3960 */
Joe Perchescccad5b2014-06-06 14:38:09 -07003961int numa_zonelist_order_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003962 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003963 loff_t *ppos)
3964{
3965 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3966 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003967 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003968
Andi Kleen443c6f12009-12-23 21:00:47 +01003969 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07003970 if (write) {
3971 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3972 ret = -EINVAL;
3973 goto out;
3974 }
3975 strcpy(saved_string, (char *)table->data);
3976 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003977 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003978 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003979 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003980 if (write) {
3981 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07003982
3983 ret = __parse_numa_zonelist_order((char *)table->data);
3984 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003985 /*
3986 * bogus value. restore saved string
3987 */
Chen Gangdacbde02013-07-03 15:02:35 -07003988 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003989 NUMA_ZONELIST_ORDER_LEN);
3990 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003991 } else if (oldval != user_zonelist_order) {
3992 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07003993 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003994 mutex_unlock(&zonelists_mutex);
3995 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003996 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003997out:
3998 mutex_unlock(&zl_order_mutex);
3999 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004000}
4001
4002
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004003#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004004static int node_load[MAX_NUMNODES];
4005
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07004007 * 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 -07004008 * @node: node whose fallback list we're appending
4009 * @used_node_mask: nodemask_t of already used nodes
4010 *
4011 * We use a number of factors to determine which is the next node that should
4012 * appear on a given node's fallback list. The node should not have appeared
4013 * already in @node's fallback list, and it should be the next closest node
4014 * according to the distance array (which contains arbitrary distance values
4015 * from each node to each node in the system), and should also prefer nodes
4016 * with no CPUs, since presumably they'll have very little allocation pressure
4017 * on them otherwise.
4018 * It returns -1 if no node is found.
4019 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004020static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004021{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01004022 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004023 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08004024 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10304025 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004026
Linus Torvalds4cf808e2006-02-17 20:38:21 +01004027 /* Use the local node if we haven't already */
4028 if (!node_isset(node, *used_node_mask)) {
4029 node_set(node, *used_node_mask);
4030 return node;
4031 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004032
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004033 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004034
4035 /* Don't want a node to appear more than once */
4036 if (node_isset(n, *used_node_mask))
4037 continue;
4038
Linus Torvalds1da177e2005-04-16 15:20:36 -07004039 /* Use the distance array to find the distance */
4040 val = node_distance(node, n);
4041
Linus Torvalds4cf808e2006-02-17 20:38:21 +01004042 /* Penalize nodes under us ("prefer the next node") */
4043 val += (n < node);
4044
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10304046 tmp = cpumask_of_node(n);
4047 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004048 val += PENALTY_FOR_NODE_WITH_CPUS;
4049
4050 /* Slight preference for less loaded node */
4051 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4052 val += node_load[n];
4053
4054 if (val < min_val) {
4055 min_val = val;
4056 best_node = n;
4057 }
4058 }
4059
4060 if (best_node >= 0)
4061 node_set(best_node, *used_node_mask);
4062
4063 return best_node;
4064}
4065
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004066
4067/*
4068 * Build zonelists ordered by node and zones within node.
4069 * This results in maximum locality--normal zone overflows into local
4070 * DMA zone, if any--but risks exhausting DMA zone.
4071 */
4072static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004073{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004074 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004076
Mel Gorman54a6eb52008-04-28 02:12:16 -07004077 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07004078 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07004079 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004080 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07004081 zonelist->_zonerefs[j].zone = NULL;
4082 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004083}
4084
4085/*
Christoph Lameter523b9452007-10-16 01:25:37 -07004086 * Build gfp_thisnode zonelists
4087 */
4088static void build_thisnode_zonelists(pg_data_t *pgdat)
4089{
Christoph Lameter523b9452007-10-16 01:25:37 -07004090 int j;
4091 struct zonelist *zonelist;
4092
Mel Gorman54a6eb52008-04-28 02:12:16 -07004093 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004094 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07004095 zonelist->_zonerefs[j].zone = NULL;
4096 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07004097}
4098
4099/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004100 * Build zonelists ordered by zone and nodes within zones.
4101 * This results in conserving DMA zone[s] until all Normal memory is
4102 * exhausted, but results in overflowing to remote node while memory
4103 * may still exist in local DMA zone.
4104 */
4105static int node_order[MAX_NUMNODES];
4106
4107static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4108{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004109 int pos, j, node;
4110 int zone_type; /* needs to be signed */
4111 struct zone *z;
4112 struct zonelist *zonelist;
4113
Mel Gorman54a6eb52008-04-28 02:12:16 -07004114 zonelist = &pgdat->node_zonelists[0];
4115 pos = 0;
4116 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4117 for (j = 0; j < nr_nodes; j++) {
4118 node = node_order[j];
4119 z = &NODE_DATA(node)->node_zones[zone_type];
4120 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07004121 zoneref_set_zone(z,
4122 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004123 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004124 }
4125 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004126 }
Mel Gormandd1a2392008-04-28 02:12:17 -07004127 zonelist->_zonerefs[pos].zone = NULL;
4128 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004129}
4130
Mel Gorman31939132014-10-09 15:28:30 -07004131#if defined(CONFIG_64BIT)
4132/*
4133 * Devices that require DMA32/DMA are relatively rare and do not justify a
4134 * penalty to every machine in case the specialised case applies. Default
4135 * to Node-ordering on 64-bit NUMA machines
4136 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004137static int default_zonelist_order(void)
4138{
Mel Gorman31939132014-10-09 15:28:30 -07004139 return ZONELIST_ORDER_NODE;
4140}
4141#else
4142/*
4143 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4144 * by the kernel. If processes running on node 0 deplete the low memory zone
4145 * then reclaim will occur more frequency increasing stalls and potentially
4146 * be easier to OOM if a large percentage of the zone is under writeback or
4147 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4148 * Hence, default to zone ordering on 32-bit.
4149 */
4150static int default_zonelist_order(void)
4151{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004152 return ZONELIST_ORDER_ZONE;
4153}
Mel Gorman31939132014-10-09 15:28:30 -07004154#endif /* CONFIG_64BIT */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004155
4156static void set_zonelist_order(void)
4157{
4158 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4159 current_zonelist_order = default_zonelist_order();
4160 else
4161 current_zonelist_order = user_zonelist_order;
4162}
4163
4164static void build_zonelists(pg_data_t *pgdat)
4165{
4166 int j, node, load;
4167 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004168 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004169 int local_node, prev_node;
4170 struct zonelist *zonelist;
4171 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004172
4173 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07004174 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004175 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07004176 zonelist->_zonerefs[0].zone = NULL;
4177 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004178 }
4179
4180 /* NUMA-aware ordering of nodes */
4181 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004182 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183 prev_node = local_node;
4184 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004185
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004186 memset(node_order, 0, sizeof(node_order));
4187 j = 0;
4188
Linus Torvalds1da177e2005-04-16 15:20:36 -07004189 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4190 /*
4191 * We don't want to pressure a particular node.
4192 * So adding penalty to the first node in same
4193 * distance group to make it round-robin.
4194 */
David Rientjes957f8222012-10-08 16:33:24 -07004195 if (node_distance(local_node, node) !=
4196 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004197 node_load[node] = load;
4198
Linus Torvalds1da177e2005-04-16 15:20:36 -07004199 prev_node = node;
4200 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004201 if (order == ZONELIST_ORDER_NODE)
4202 build_zonelists_in_node_order(pgdat, node);
4203 else
4204 node_order[j++] = node; /* remember order */
4205 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004206
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004207 if (order == ZONELIST_ORDER_ZONE) {
4208 /* calculate node order -- i.e., DMA last! */
4209 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004210 }
Christoph Lameter523b9452007-10-16 01:25:37 -07004211
4212 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004213}
4214
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004215/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004216static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004217{
Mel Gorman54a6eb52008-04-28 02:12:16 -07004218 struct zonelist *zonelist;
4219 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07004220 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004221
Mel Gorman54a6eb52008-04-28 02:12:16 -07004222 zonelist = &pgdat->node_zonelists[0];
4223 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
4224 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07004225 for (z = zonelist->_zonerefs; z->zone; z++)
4226 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004227}
4228
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004229#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4230/*
4231 * Return node id of node used for "local" allocations.
4232 * I.e., first node id of first zone in arg node's generic zonelist.
4233 * Used for initializing percpu 'numa_mem', which is used primarily
4234 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4235 */
4236int local_memory_node(int node)
4237{
4238 struct zone *zone;
4239
4240 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
4241 gfp_zone(GFP_KERNEL),
4242 NULL,
4243 &zone);
4244 return zone->node;
4245}
4246#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004247
Linus Torvalds1da177e2005-04-16 15:20:36 -07004248#else /* CONFIG_NUMA */
4249
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004250static void set_zonelist_order(void)
4251{
4252 current_zonelist_order = ZONELIST_ORDER_ZONE;
4253}
4254
4255static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004256{
Christoph Lameter19655d32006-09-25 23:31:19 -07004257 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07004258 enum zone_type j;
4259 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004260
4261 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004262
Mel Gorman54a6eb52008-04-28 02:12:16 -07004263 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004264 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004265
Mel Gorman54a6eb52008-04-28 02:12:16 -07004266 /*
4267 * Now we build the zonelist so that it contains the zones
4268 * of all the other nodes.
4269 * We don't want to pressure a particular node, so when
4270 * building the zones for node N, we make sure that the
4271 * zones coming right after the local ones are those from
4272 * node N+1 (modulo N)
4273 */
4274 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4275 if (!node_online(node))
4276 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004277 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004278 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07004279 for (node = 0; node < local_node; node++) {
4280 if (!node_online(node))
4281 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004282 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004283 }
4284
Mel Gormandd1a2392008-04-28 02:12:17 -07004285 zonelist->_zonerefs[j].zone = NULL;
4286 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004287}
4288
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004289/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004290static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004291{
Mel Gorman54a6eb52008-04-28 02:12:16 -07004292 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004293}
4294
Linus Torvalds1da177e2005-04-16 15:20:36 -07004295#endif /* CONFIG_NUMA */
4296
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004297/*
4298 * Boot pageset table. One per cpu which is going to be used for all
4299 * zones and all nodes. The parameters will be set in such a way
4300 * that an item put on a list will immediately be handed over to
4301 * the buddy list. This is safe since pageset manipulation is done
4302 * with interrupts disabled.
4303 *
4304 * The boot_pagesets must be kept even after bootup is complete for
4305 * unused processors and/or zones. They do play a role for bootstrapping
4306 * hotplugged processors.
4307 *
4308 * zoneinfo_show() and maybe other functions do
4309 * not check if the processor is online before following the pageset pointer.
4310 * Other parts of the kernel may not check if the zone is available.
4311 */
4312static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4313static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07004314static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004315
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004316/*
4317 * Global mutex to protect against size modification of zonelists
4318 * as well as to serialize pageset setup for the new populated zone.
4319 */
4320DEFINE_MUTEX(zonelists_mutex);
4321
Rusty Russell9b1a4d32008-07-28 12:16:30 -05004322/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07004323static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004324{
Yasunori Goto68113782006-06-23 02:03:11 -07004325 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004326 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07004327 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004328
Bo Liu7f9cfb32009-08-18 14:11:19 -07004329#ifdef CONFIG_NUMA
4330 memset(node_load, 0, sizeof(node_load));
4331#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07004332
4333 if (self && !node_online(self->node_id)) {
4334 build_zonelists(self);
4335 build_zonelist_cache(self);
4336 }
4337
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004338 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07004339 pg_data_t *pgdat = NODE_DATA(nid);
4340
4341 build_zonelists(pgdat);
4342 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004343 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004344
4345 /*
4346 * Initialize the boot_pagesets that are going to be used
4347 * for bootstrapping processors. The real pagesets for
4348 * each zone will be allocated later when the per cpu
4349 * allocator is available.
4350 *
4351 * boot_pagesets are used also for bootstrapping offline
4352 * cpus if the system is already booted because the pagesets
4353 * are needed to initialize allocators on a specific cpu too.
4354 * F.e. the percpu allocator needs the page allocator which
4355 * needs the percpu allocator in order to allocate its pagesets
4356 * (a chicken-egg dilemma).
4357 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004358 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004359 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4360
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004361#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4362 /*
4363 * We now know the "local memory node" for each node--
4364 * i.e., the node of the first zone in the generic zonelist.
4365 * Set up numa_mem percpu variable for on-line cpus. During
4366 * boot, only the boot cpu should be on-line; we'll init the
4367 * secondary cpus' numa_mem as they come on-line. During
4368 * node/memory hotplug, we'll fixup all on-line cpus.
4369 */
4370 if (cpu_online(cpu))
4371 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4372#endif
4373 }
4374
Yasunori Goto68113782006-06-23 02:03:11 -07004375 return 0;
4376}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004377
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004378static noinline void __init
4379build_all_zonelists_init(void)
4380{
4381 __build_all_zonelists(NULL);
4382 mminit_verify_zonelist();
4383 cpuset_init_current_mems_allowed();
4384}
4385
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004386/*
4387 * Called with zonelists_mutex held always
4388 * unless system_state == SYSTEM_BOOTING.
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004389 *
4390 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4391 * [we're only called with non-NULL zone through __meminit paths] and
4392 * (2) call of __init annotated helper build_all_zonelists_init
4393 * [protected by SYSTEM_BOOTING].
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004394 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07004395void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07004396{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004397 set_zonelist_order();
4398
Yasunori Goto68113782006-06-23 02:03:11 -07004399 if (system_state == SYSTEM_BOOTING) {
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004400 build_all_zonelists_init();
Yasunori Goto68113782006-06-23 02:03:11 -07004401 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08004402#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07004403 if (zone)
4404 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08004405#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07004406 /* we have to stop all cpus to guarantee there is no user
4407 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07004408 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07004409 /* cpuset refresh routine should be here */
4410 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07004411 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004412 /*
4413 * Disable grouping by mobility if the number of pages in the
4414 * system is too low to allow the mechanism to work. It would be
4415 * more accurate, but expensive to check per-zone. This check is
4416 * made on memory-hotadd so a system can start with mobility
4417 * disabled and enable it later
4418 */
Mel Gormand9c23402007-10-16 01:26:01 -07004419 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004420 page_group_by_mobility_disabled = 1;
4421 else
4422 page_group_by_mobility_disabled = 0;
4423
Anton Blanchardf88dfff2014-12-10 15:42:53 -08004424 pr_info("Built %i zonelists in %s order, mobility grouping %s. "
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004425 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004426 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004427 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004428 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004429 vm_total_pages);
4430#ifdef CONFIG_NUMA
Anton Blanchardf88dfff2014-12-10 15:42:53 -08004431 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004432#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004433}
4434
4435/*
4436 * Helper functions to size the waitqueue hash table.
4437 * Essentially these want to choose hash table sizes sufficiently
4438 * large so that collisions trying to wait on pages are rare.
4439 * But in fact, the number of active page waitqueues on typical
4440 * systems is ridiculously low, less than 200. So this is even
4441 * conservative, even though it seems large.
4442 *
4443 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4444 * waitqueues, i.e. the size of the waitq table given the number of pages.
4445 */
4446#define PAGES_PER_WAITQUEUE 256
4447
Yasunori Gotocca448f2006-06-23 02:03:10 -07004448#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07004449static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004450{
4451 unsigned long size = 1;
4452
4453 pages /= PAGES_PER_WAITQUEUE;
4454
4455 while (size < pages)
4456 size <<= 1;
4457
4458 /*
4459 * Once we have dozens or even hundreds of threads sleeping
4460 * on IO we've got bigger problems than wait queue collision.
4461 * Limit the size of the wait table to a reasonable size.
4462 */
4463 size = min(size, 4096UL);
4464
4465 return max(size, 4UL);
4466}
Yasunori Gotocca448f2006-06-23 02:03:10 -07004467#else
4468/*
4469 * A zone's size might be changed by hot-add, so it is not possible to determine
4470 * a suitable size for its wait_table. So we use the maximum size now.
4471 *
4472 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4473 *
4474 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4475 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4476 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4477 *
4478 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4479 * or more by the traditional way. (See above). It equals:
4480 *
4481 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4482 * ia64(16K page size) : = ( 8G + 4M)byte.
4483 * powerpc (64K page size) : = (32G +16M)byte.
4484 */
4485static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4486{
4487 return 4096UL;
4488}
4489#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004490
4491/*
4492 * This is an integer logarithm so that shifts can be used later
4493 * to extract the more random high bits from the multiplicative
4494 * hash function before the remainder is taken.
4495 */
4496static inline unsigned long wait_table_bits(unsigned long size)
4497{
4498 return ffz(~size);
4499}
4500
Mel Gorman56fd56b2007-10-16 01:25:58 -07004501/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07004502 * Check if a pageblock contains reserved pages
4503 */
4504static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
4505{
4506 unsigned long pfn;
4507
4508 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4509 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
4510 return 1;
4511 }
4512 return 0;
4513}
4514
4515/*
Mel Gormand9c23402007-10-16 01:26:01 -07004516 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07004517 * of blocks reserved is based on min_wmark_pages(zone). The memory within
4518 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07004519 * higher will lead to a bigger reserve which will get freed as contiguous
4520 * blocks as reclaim kicks in
4521 */
4522static void setup_zone_migrate_reserve(struct zone *zone)
4523{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07004524 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004525 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07004526 unsigned long block_migratetype;
4527 int reserve;
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004528 int old_reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004529
Michal Hockod02156382011-12-08 14:34:27 -08004530 /*
4531 * Get the start pfn, end pfn and the number of blocks to reserve
4532 * We have to be careful to be aligned to pageblock_nr_pages to
4533 * make sure that we always check pfn_valid for the first page in
4534 * the block.
4535 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07004536 start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08004537 end_pfn = zone_end_pfn(zone);
Michal Hockod02156382011-12-08 14:34:27 -08004538 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07004539 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07004540 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004541
Mel Gorman78986a62009-09-21 17:03:02 -07004542 /*
4543 * Reserve blocks are generally in place to help high-order atomic
4544 * allocations that are short-lived. A min_free_kbytes value that
4545 * would result in more than 2 reserve blocks for atomic allocations
4546 * is assumed to be in place to help anti-fragmentation for the
4547 * future allocation of hugepages at runtime.
4548 */
4549 reserve = min(2, reserve);
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004550 old_reserve = zone->nr_migrate_reserve_block;
4551
4552 /* When memory hot-add, we almost always need to do nothing */
4553 if (reserve == old_reserve)
4554 return;
4555 zone->nr_migrate_reserve_block = reserve;
Mel Gorman78986a62009-09-21 17:03:02 -07004556
Mel Gormand9c23402007-10-16 01:26:01 -07004557 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman7e18adb2015-06-30 14:57:05 -07004558 if (!early_page_nid_uninitialised(pfn, zone_to_nid(zone)))
4559 return;
4560
Mel Gorman56fd56b2007-10-16 01:25:58 -07004561 if (!pfn_valid(pfn))
4562 continue;
4563 page = pfn_to_page(pfn);
4564
Adam Litke344c7902008-09-02 14:35:38 -07004565 /* Watch out for overlapping nodes */
4566 if (page_to_nid(page) != zone_to_nid(zone))
4567 continue;
4568
Mel Gorman56fd56b2007-10-16 01:25:58 -07004569 block_migratetype = get_pageblock_migratetype(page);
4570
Mel Gorman938929f2012-01-10 15:07:14 -08004571 /* Only test what is necessary when the reserves are not met */
4572 if (reserve > 0) {
4573 /*
4574 * Blocks with reserved pages will never free, skip
4575 * them.
4576 */
4577 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4578 if (pageblock_is_reserved(pfn, block_end_pfn))
4579 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004580
Mel Gorman938929f2012-01-10 15:07:14 -08004581 /* If this block is reserved, account for it */
4582 if (block_migratetype == MIGRATE_RESERVE) {
4583 reserve--;
4584 continue;
4585 }
4586
4587 /* Suitable for reserving if this block is movable */
4588 if (block_migratetype == MIGRATE_MOVABLE) {
4589 set_pageblock_migratetype(page,
4590 MIGRATE_RESERVE);
4591 move_freepages_block(zone, page,
4592 MIGRATE_RESERVE);
4593 reserve--;
4594 continue;
4595 }
Yasuaki Ishimatsu943dca12014-01-21 15:49:06 -08004596 } else if (!old_reserve) {
4597 /*
4598 * At boot time we don't need to scan the whole zone
4599 * for turning off MIGRATE_RESERVE.
4600 */
4601 break;
Mel Gorman56fd56b2007-10-16 01:25:58 -07004602 }
4603
4604 /*
4605 * If the reserve is met and this is a previous reserved block,
4606 * take it back
4607 */
4608 if (block_migratetype == MIGRATE_RESERVE) {
4609 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4610 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4611 }
4612 }
4613}
Mel Gormanac0e5b72007-10-16 01:25:58 -07004614
Linus Torvalds1da177e2005-04-16 15:20:36 -07004615/*
4616 * Initially all pages are reserved - free ones are freed
4617 * up by free_all_bootmem() once the early boot process is
4618 * done. Non-atomic initialization, single-pass.
4619 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004620void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004621 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004622{
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004623 pg_data_t *pgdat = NODE_DATA(nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -07004624 unsigned long end_pfn = start_pfn + size;
4625 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004626 struct zone *z;
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004627 unsigned long nr_initialised = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004628
Hugh Dickins22b31ee2009-01-06 14:40:09 -08004629 if (highest_memmap_pfn < end_pfn - 1)
4630 highest_memmap_pfn = end_pfn - 1;
4631
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004632 z = &pgdat->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08004633 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08004634 /*
4635 * There can be holes in boot-time mem_map[]s
4636 * handed to this function. They do not
4637 * exist on hotplugged memory.
4638 */
4639 if (context == MEMMAP_EARLY) {
4640 if (!early_pfn_valid(pfn))
4641 continue;
4642 if (!early_pfn_in_nid(pfn, nid))
4643 continue;
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004644 if (!update_defer_init(pgdat, pfn, end_pfn,
4645 &nr_initialised))
4646 break;
Dave Hansena2f3aa022007-01-10 23:15:30 -08004647 }
Mel Gormanac5d2532015-06-30 14:57:20 -07004648
4649 /*
4650 * Mark the block movable so that blocks are reserved for
4651 * movable at startup. This will force kernel allocations
4652 * to reserve their blocks rather than leaking throughout
4653 * the address space during boot when many long-lived
4654 * kernel allocations are made. Later some blocks near
4655 * the start are marked MIGRATE_RESERVE by
4656 * setup_zone_migrate_reserve()
4657 *
4658 * bitmap is created for zone's valid pfn range. but memmap
4659 * can be created for invalid pages (for alignment)
4660 * check here not to call set_pageblock_migratetype() against
4661 * pfn out of zone.
4662 */
4663 if (!(pfn & (pageblock_nr_pages - 1))) {
4664 struct page *page = pfn_to_page(pfn);
4665
4666 __init_single_page(page, pfn, zone, nid);
4667 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4668 } else {
4669 __init_single_pfn(pfn, zone, nid);
4670 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004671 }
4672}
4673
Andi Kleen1e548de2008-02-04 22:29:26 -08004674static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004675{
Mel Gorman7aeb09f2014-06-04 16:10:21 -07004676 unsigned int order, t;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004677 for_each_migratetype_order(order, t) {
4678 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004679 zone->free_area[order].nr_free = 0;
4680 }
4681}
4682
4683#ifndef __HAVE_ARCH_MEMMAP_INIT
4684#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08004685 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004686#endif
4687
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004688static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004689{
David Howells3a6be872009-05-06 16:03:03 -07004690#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004691 int batch;
4692
4693 /*
4694 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07004695 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004696 *
4697 * OK, so we don't know how big the cache is. So guess.
4698 */
Jiang Liub40da042013-02-22 16:33:52 -08004699 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07004700 if (batch * PAGE_SIZE > 512 * 1024)
4701 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004702 batch /= 4; /* We effectively *= 4 below */
4703 if (batch < 1)
4704 batch = 1;
4705
4706 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004707 * Clamp the batch to a 2^n - 1 value. Having a power
4708 * of 2 value was found to be more likely to have
4709 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004710 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004711 * For example if 2 tasks are alternately allocating
4712 * batches of pages, one task can end up with a lot
4713 * of pages of one half of the possible page colors
4714 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004715 */
David Howells91552032009-05-06 16:03:02 -07004716 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004717
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004718 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004719
4720#else
4721 /* The deferral and batching of frees should be suppressed under NOMMU
4722 * conditions.
4723 *
4724 * The problem is that NOMMU needs to be able to allocate large chunks
4725 * of contiguous memory as there's no hardware page translation to
4726 * assemble apparent contiguous memory from discontiguous pages.
4727 *
4728 * Queueing large contiguous runs of pages for batching, however,
4729 * causes the pages to actually be freed in smaller chunks. As there
4730 * can be a significant delay between the individual batches being
4731 * recycled, this leads to the once large chunks of space being
4732 * fragmented and becoming unavailable for high-order allocations.
4733 */
4734 return 0;
4735#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004736}
4737
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004738/*
4739 * pcp->high and pcp->batch values are related and dependent on one another:
4740 * ->batch must never be higher then ->high.
4741 * The following function updates them in a safe manner without read side
4742 * locking.
4743 *
4744 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4745 * those fields changing asynchronously (acording the the above rule).
4746 *
4747 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4748 * outside of boot time (or some other assurance that no concurrent updaters
4749 * exist).
4750 */
4751static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4752 unsigned long batch)
4753{
4754 /* start with a fail safe value for batch */
4755 pcp->batch = 1;
4756 smp_wmb();
4757
4758 /* Update high, then batch, in order */
4759 pcp->high = high;
4760 smp_wmb();
4761
4762 pcp->batch = batch;
4763}
4764
Cody P Schafer36640332013-07-03 15:01:40 -07004765/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07004766static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4767{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004768 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07004769}
4770
Cody P Schafer88c90db2013-07-03 15:01:35 -07004771static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004772{
4773 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004774 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004775
Magnus Damm1c6fe942005-10-26 01:58:59 -07004776 memset(p, 0, sizeof(*p));
4777
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004778 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004779 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004780 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4781 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004782}
4783
Cody P Schafer88c90db2013-07-03 15:01:35 -07004784static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4785{
4786 pageset_init(p);
4787 pageset_set_batch(p, batch);
4788}
4789
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004790/*
Cody P Schafer36640332013-07-03 15:01:40 -07004791 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004792 * to the value high for the pageset p.
4793 */
Cody P Schafer36640332013-07-03 15:01:40 -07004794static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004795 unsigned long high)
4796{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004797 unsigned long batch = max(1UL, high / 4);
4798 if ((high / 4) > (PAGE_SHIFT * 8))
4799 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004800
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004801 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004802}
4803
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004804static void pageset_set_high_and_batch(struct zone *zone,
4805 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004806{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004807 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07004808 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004809 (zone->managed_pages /
4810 percpu_pagelist_fraction));
4811 else
4812 pageset_set_batch(pcp, zone_batchsize(zone));
4813}
4814
Cody P Schafer169f6c12013-07-03 15:01:41 -07004815static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4816{
4817 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4818
4819 pageset_init(pcp);
4820 pageset_set_high_and_batch(zone, pcp);
4821}
4822
Jiang Liu4ed7e022012-07-31 16:43:35 -07004823static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004824{
4825 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07004826 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004827 for_each_possible_cpu(cpu)
4828 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07004829}
4830
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004831/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004832 * Allocate per cpu pagesets and initialize them.
4833 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004834 */
Al Viro78d99552005-12-15 09:18:25 +00004835void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004836{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004837 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004838
Wu Fengguang319774e2010-05-24 14:32:49 -07004839 for_each_populated_zone(zone)
4840 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004841}
4842
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004843static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004844int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004845{
4846 int i;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004847 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004848
4849 /*
4850 * The per-page waitqueue mechanism uses hashed waitqueues
4851 * per zone.
4852 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004853 zone->wait_table_hash_nr_entries =
4854 wait_table_hash_nr_entries(zone_size_pages);
4855 zone->wait_table_bits =
4856 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004857 alloc_size = zone->wait_table_hash_nr_entries
4858 * sizeof(wait_queue_head_t);
4859
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004860 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004861 zone->wait_table = (wait_queue_head_t *)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004862 memblock_virt_alloc_node_nopanic(
4863 alloc_size, zone->zone_pgdat->node_id);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004864 } else {
4865 /*
4866 * This case means that a zone whose size was 0 gets new memory
4867 * via memory hot-add.
4868 * But it may be the case that a new node was hot-added. In
4869 * this case vmalloc() will not be able to use this new node's
4870 * memory - this wait_table must be initialized to use this new
4871 * node itself as well.
4872 * To use this new node's memory, further consideration will be
4873 * necessary.
4874 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004875 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004876 }
4877 if (!zone->wait_table)
4878 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004879
Pintu Kumarb8af2942013-09-11 14:20:34 -07004880 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004881 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004882
4883 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004884}
4885
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004886static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004887{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004888 /*
4889 * per cpu subsystem is not up at this point. The following code
4890 * relies on the ability of the linker to provide the
4891 * offset of a (static) per cpu variable into the per cpu area.
4892 */
4893 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004894
Xishi Qiub38a8722013-11-12 15:07:20 -08004895 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004896 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4897 zone->name, zone->present_pages,
4898 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004899}
4900
Jiang Liu4ed7e022012-07-31 16:43:35 -07004901int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004902 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004903 unsigned long size,
4904 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07004905{
4906 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004907 int ret;
4908 ret = zone_wait_table_init(zone, size);
4909 if (ret)
4910 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004911 pgdat->nr_zones = zone_idx(zone) + 1;
4912
Dave Hansened8ece22005-10-29 18:16:50 -07004913 zone->zone_start_pfn = zone_start_pfn;
4914
Mel Gorman708614e2008-07-23 21:26:51 -07004915 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4916 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4917 pgdat->node_id,
4918 (unsigned long)zone_idx(zone),
4919 zone_start_pfn, (zone_start_pfn + size));
4920
Andi Kleen1e548de2008-02-04 22:29:26 -08004921 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004922
4923 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004924}
4925
Tejun Heo0ee332c2011-12-08 10:22:09 -08004926#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004927#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
Mel Gorman8a942fd2015-06-30 14:56:55 -07004928
Mel Gormanc7132162006-09-27 01:49:43 -07004929/*
4930 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
Mel Gormanc7132162006-09-27 01:49:43 -07004931 */
Mel Gorman8a942fd2015-06-30 14:56:55 -07004932int __meminit __early_pfn_to_nid(unsigned long pfn,
4933 struct mminit_pfnnid_cache *state)
Mel Gormanc7132162006-09-27 01:49:43 -07004934{
Tejun Heoc13291a2011-07-12 10:46:30 +02004935 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004936 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004937
Mel Gorman8a942fd2015-06-30 14:56:55 -07004938 if (state->last_start <= pfn && pfn < state->last_end)
4939 return state->last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004940
Yinghai Lue76b63f2013-09-11 14:22:17 -07004941 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4942 if (nid != -1) {
Mel Gorman8a942fd2015-06-30 14:56:55 -07004943 state->last_start = start_pfn;
4944 state->last_end = end_pfn;
4945 state->last_nid = nid;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004946 }
4947
4948 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004949}
4950#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4951
Mel Gormanc7132162006-09-27 01:49:43 -07004952/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08004953 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004954 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08004955 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07004956 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004957 * If an architecture guarantees that all ranges registered contain no holes
4958 * and may be freed, this this function may be used instead of calling
4959 * memblock_free_early_nid() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004960 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004961void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004962{
Tejun Heoc13291a2011-07-12 10:46:30 +02004963 unsigned long start_pfn, end_pfn;
4964 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004965
Tejun Heoc13291a2011-07-12 10:46:30 +02004966 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4967 start_pfn = min(start_pfn, max_low_pfn);
4968 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004969
Tejun Heoc13291a2011-07-12 10:46:30 +02004970 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004971 memblock_free_early_nid(PFN_PHYS(start_pfn),
4972 (end_pfn - start_pfn) << PAGE_SHIFT,
4973 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07004974 }
4975}
4976
4977/**
4978 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004979 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004980 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004981 * If an architecture guarantees that all ranges registered contain no holes and may
4982 * be freed, this function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004983 */
4984void __init sparse_memory_present_with_active_regions(int nid)
4985{
Tejun Heoc13291a2011-07-12 10:46:30 +02004986 unsigned long start_pfn, end_pfn;
4987 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004988
Tejun Heoc13291a2011-07-12 10:46:30 +02004989 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4990 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004991}
4992
4993/**
4994 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004995 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4996 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4997 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004998 *
4999 * It returns the start and end page frame of a node based on information
Zhang Zhen7d018172014-06-04 16:10:53 -07005000 * provided by memblock_set_node(). If called for a node
Mel Gormanc7132162006-09-27 01:49:43 -07005001 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005002 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07005003 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005004void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005005 unsigned long *start_pfn, unsigned long *end_pfn)
5006{
Tejun Heoc13291a2011-07-12 10:46:30 +02005007 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005008 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02005009
Mel Gormanc7132162006-09-27 01:49:43 -07005010 *start_pfn = -1UL;
5011 *end_pfn = 0;
5012
Tejun Heoc13291a2011-07-12 10:46:30 +02005013 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5014 *start_pfn = min(*start_pfn, this_start_pfn);
5015 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005016 }
5017
Christoph Lameter633c0662007-10-16 01:25:37 -07005018 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07005019 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005020}
5021
5022/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07005023 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5024 * assumption is made that zones within a node are ordered in monotonic
5025 * increasing memory addresses so that the "highest" populated zone is used
5026 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005027static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005028{
5029 int zone_index;
5030 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5031 if (zone_index == ZONE_MOVABLE)
5032 continue;
5033
5034 if (arch_zone_highest_possible_pfn[zone_index] >
5035 arch_zone_lowest_possible_pfn[zone_index])
5036 break;
5037 }
5038
5039 VM_BUG_ON(zone_index == -1);
5040 movable_zone = zone_index;
5041}
5042
5043/*
5044 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03005045 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005046 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5047 * in each node depending on the size of each node and how evenly kernelcore
5048 * is distributed. This helper function adjusts the zone ranges
5049 * provided by the architecture for a given node by using the end of the
5050 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5051 * zones within a node are in order of monotonic increases memory addresses
5052 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005053static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005054 unsigned long zone_type,
5055 unsigned long node_start_pfn,
5056 unsigned long node_end_pfn,
5057 unsigned long *zone_start_pfn,
5058 unsigned long *zone_end_pfn)
5059{
5060 /* Only adjust if ZONE_MOVABLE is on this node */
5061 if (zone_movable_pfn[nid]) {
5062 /* Size ZONE_MOVABLE */
5063 if (zone_type == ZONE_MOVABLE) {
5064 *zone_start_pfn = zone_movable_pfn[nid];
5065 *zone_end_pfn = min(node_end_pfn,
5066 arch_zone_highest_possible_pfn[movable_zone]);
5067
5068 /* Adjust for ZONE_MOVABLE starting within this range */
5069 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
5070 *zone_end_pfn > zone_movable_pfn[nid]) {
5071 *zone_end_pfn = zone_movable_pfn[nid];
5072
5073 /* Check if this whole range is within ZONE_MOVABLE */
5074 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5075 *zone_start_pfn = *zone_end_pfn;
5076 }
5077}
5078
5079/*
Mel Gormanc7132162006-09-27 01:49:43 -07005080 * Return the number of pages a zone spans in a node, including holes
5081 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5082 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005083static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005084 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005085 unsigned long node_start_pfn,
5086 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005087 unsigned long *ignored)
5088{
Mel Gormanc7132162006-09-27 01:49:43 -07005089 unsigned long zone_start_pfn, zone_end_pfn;
5090
Xishi Qiub5685e92015-09-08 15:04:16 -07005091 /* When hotadd a new node from cpu_up(), the node should be empty */
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005092 if (!node_start_pfn && !node_end_pfn)
5093 return 0;
5094
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005095 /* Get the start and end of the zone */
Mel Gormanc7132162006-09-27 01:49:43 -07005096 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5097 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07005098 adjust_zone_range_for_zone_movable(nid, zone_type,
5099 node_start_pfn, node_end_pfn,
5100 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005101
5102 /* Check that this node has pages within the zone's required range */
5103 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
5104 return 0;
5105
5106 /* Move the zone boundaries inside the node if necessary */
5107 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
5108 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
5109
5110 /* Return the spanned pages */
5111 return zone_end_pfn - zone_start_pfn;
5112}
5113
5114/*
5115 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005116 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07005117 */
Yinghai Lu32996252009-12-15 17:59:02 -08005118unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005119 unsigned long range_start_pfn,
5120 unsigned long range_end_pfn)
5121{
Tejun Heo96e907d2011-07-12 10:46:29 +02005122 unsigned long nr_absent = range_end_pfn - range_start_pfn;
5123 unsigned long start_pfn, end_pfn;
5124 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07005125
Tejun Heo96e907d2011-07-12 10:46:29 +02005126 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5127 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5128 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5129 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005130 }
Tejun Heo96e907d2011-07-12 10:46:29 +02005131 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07005132}
5133
5134/**
5135 * absent_pages_in_range - Return number of page frames in holes within a range
5136 * @start_pfn: The start PFN to start searching for holes
5137 * @end_pfn: The end PFN to stop searching for holes
5138 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005139 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07005140 */
5141unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5142 unsigned long end_pfn)
5143{
5144 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5145}
5146
5147/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005148static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005149 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005150 unsigned long node_start_pfn,
5151 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005152 unsigned long *ignored)
5153{
Tejun Heo96e907d2011-07-12 10:46:29 +02005154 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5155 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07005156 unsigned long zone_start_pfn, zone_end_pfn;
5157
Xishi Qiub5685e92015-09-08 15:04:16 -07005158 /* When hotadd a new node from cpu_up(), the node should be empty */
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005159 if (!node_start_pfn && !node_end_pfn)
5160 return 0;
5161
Tejun Heo96e907d2011-07-12 10:46:29 +02005162 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5163 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07005164
Mel Gorman2a1e2742007-07-17 04:03:12 -07005165 adjust_zone_range_for_zone_movable(nid, zone_type,
5166 node_start_pfn, node_end_pfn,
5167 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07005168 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005169}
Mel Gorman0e0b8642006-09-27 01:49:56 -07005170
Tejun Heo0ee332c2011-12-08 10:22:09 -08005171#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005172static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005173 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005174 unsigned long node_start_pfn,
5175 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005176 unsigned long *zones_size)
5177{
5178 return zones_size[zone_type];
5179}
5180
Paul Mundt6ea6e682007-07-15 23:38:20 -07005181static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005182 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005183 unsigned long node_start_pfn,
5184 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005185 unsigned long *zholes_size)
5186{
5187 if (!zholes_size)
5188 return 0;
5189
5190 return zholes_size[zone_type];
5191}
Yinghai Lu20e69262013-03-01 14:51:27 -08005192
Tejun Heo0ee332c2011-12-08 10:22:09 -08005193#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005194
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005195static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005196 unsigned long node_start_pfn,
5197 unsigned long node_end_pfn,
5198 unsigned long *zones_size,
5199 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07005200{
Gu Zhengfebd5942015-06-24 16:57:02 -07005201 unsigned long realtotalpages = 0, totalpages = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005202 enum zone_type i;
5203
Gu Zhengfebd5942015-06-24 16:57:02 -07005204 for (i = 0; i < MAX_NR_ZONES; i++) {
5205 struct zone *zone = pgdat->node_zones + i;
5206 unsigned long size, real_size;
Mel Gormanc7132162006-09-27 01:49:43 -07005207
Gu Zhengfebd5942015-06-24 16:57:02 -07005208 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5209 node_start_pfn,
5210 node_end_pfn,
5211 zones_size);
5212 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005213 node_start_pfn, node_end_pfn,
5214 zholes_size);
Gu Zhengfebd5942015-06-24 16:57:02 -07005215 zone->spanned_pages = size;
5216 zone->present_pages = real_size;
5217
5218 totalpages += size;
5219 realtotalpages += real_size;
5220 }
5221
5222 pgdat->node_spanned_pages = totalpages;
Mel Gormanc7132162006-09-27 01:49:43 -07005223 pgdat->node_present_pages = realtotalpages;
5224 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5225 realtotalpages);
5226}
5227
Mel Gorman835c1342007-10-16 01:25:47 -07005228#ifndef CONFIG_SPARSEMEM
5229/*
5230 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07005231 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5232 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07005233 * round what is now in bits to nearest long in bits, then return it in
5234 * bytes.
5235 */
Linus Torvalds7c455122013-02-18 09:58:02 -08005236static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005237{
5238 unsigned long usemapsize;
5239
Linus Torvalds7c455122013-02-18 09:58:02 -08005240 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005241 usemapsize = roundup(zonesize, pageblock_nr_pages);
5242 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07005243 usemapsize *= NR_PAGEBLOCK_BITS;
5244 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5245
5246 return usemapsize / 8;
5247}
5248
5249static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08005250 struct zone *zone,
5251 unsigned long zone_start_pfn,
5252 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005253{
Linus Torvalds7c455122013-02-18 09:58:02 -08005254 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07005255 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08005256 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005257 zone->pageblock_flags =
5258 memblock_virt_alloc_node_nopanic(usemapsize,
5259 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07005260}
5261#else
Linus Torvalds7c455122013-02-18 09:58:02 -08005262static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5263 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07005264#endif /* CONFIG_SPARSEMEM */
5265
Mel Gormand9c23402007-10-16 01:26:01 -07005266#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08005267
Mel Gormand9c23402007-10-16 01:26:01 -07005268/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07005269void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07005270{
Andrew Morton955c1cd2012-05-29 15:06:31 -07005271 unsigned int order;
5272
Mel Gormand9c23402007-10-16 01:26:01 -07005273 /* Check that pageblock_nr_pages has not already been setup */
5274 if (pageblock_order)
5275 return;
5276
Andrew Morton955c1cd2012-05-29 15:06:31 -07005277 if (HPAGE_SHIFT > PAGE_SHIFT)
5278 order = HUGETLB_PAGE_ORDER;
5279 else
5280 order = MAX_ORDER - 1;
5281
Mel Gormand9c23402007-10-16 01:26:01 -07005282 /*
5283 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07005284 * This value may be variable depending on boot parameters on IA64 and
5285 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07005286 */
5287 pageblock_order = order;
5288}
5289#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5290
Mel Gormanba72cb82007-11-28 16:21:13 -08005291/*
5292 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07005293 * is unused as pageblock_order is set at compile-time. See
5294 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5295 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08005296 */
Chen Gang15ca2202013-09-11 14:20:27 -07005297void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08005298{
Mel Gormanba72cb82007-11-28 16:21:13 -08005299}
Mel Gormand9c23402007-10-16 01:26:01 -07005300
5301#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5302
Jiang Liu01cefae2012-12-12 13:52:19 -08005303static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5304 unsigned long present_pages)
5305{
5306 unsigned long pages = spanned_pages;
5307
5308 /*
5309 * Provide a more accurate estimation if there are holes within
5310 * the zone and SPARSEMEM is in use. If there are holes within the
5311 * zone, each populated memory region may cost us one or two extra
5312 * memmap pages due to alignment because memmap pages for each
5313 * populated regions may not naturally algined on page boundary.
5314 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5315 */
5316 if (spanned_pages > present_pages + (present_pages >> 4) &&
5317 IS_ENABLED(CONFIG_SPARSEMEM))
5318 pages = present_pages;
5319
5320 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5321}
5322
Linus Torvalds1da177e2005-04-16 15:20:36 -07005323/*
5324 * Set up the zone data structures:
5325 * - mark all pages reserved
5326 * - mark all memory queues empty
5327 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07005328 *
5329 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005330 */
Wei Yang7f3eb552015-09-08 14:59:50 -07005331static void __paginginit free_area_init_core(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005332{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07005333 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07005334 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005335 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07005336 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005337
Dave Hansen208d54e2005-10-29 18:16:52 -07005338 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01005339#ifdef CONFIG_NUMA_BALANCING
5340 spin_lock_init(&pgdat->numabalancing_migrate_lock);
5341 pgdat->numabalancing_migrate_nr_pages = 0;
5342 pgdat->numabalancing_migrate_next_window = jiffies;
5343#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005344 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07005345 init_waitqueue_head(&pgdat->pfmemalloc_wait);
Joonsoo Kimeefa8642014-12-12 16:55:46 -08005346 pgdat_page_ext_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01005347
Linus Torvalds1da177e2005-04-16 15:20:36 -07005348 for (j = 0; j < MAX_NR_ZONES; j++) {
5349 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08005350 unsigned long size, realsize, freesize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005351
Gu Zhengfebd5942015-06-24 16:57:02 -07005352 size = zone->spanned_pages;
5353 realsize = freesize = zone->present_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005354
Mel Gorman0e0b8642006-09-27 01:49:56 -07005355 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08005356 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07005357 * is used by this zone for memmap. This affects the watermark
5358 * and per-cpu initialisations
5359 */
Jiang Liu01cefae2012-12-12 13:52:19 -08005360 memmap_pages = calc_memmap_size(size, realsize);
Zhong Hongboba914f42014-12-12 16:56:21 -08005361 if (!is_highmem_idx(j)) {
5362 if (freesize >= memmap_pages) {
5363 freesize -= memmap_pages;
5364 if (memmap_pages)
5365 printk(KERN_DEBUG
5366 " %s zone: %lu pages used for memmap\n",
5367 zone_names[j], memmap_pages);
5368 } else
5369 printk(KERN_WARNING
5370 " %s zone: %lu pages exceeds freesize %lu\n",
5371 zone_names[j], memmap_pages, freesize);
5372 }
Mel Gorman0e0b8642006-09-27 01:49:56 -07005373
Christoph Lameter62672762007-02-10 01:43:07 -08005374 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08005375 if (j == 0 && freesize > dma_reserve) {
5376 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07005377 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08005378 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07005379 }
5380
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07005381 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08005382 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08005383 /* Charge for highmem memmap if there are enough kernel pages */
5384 else if (nr_kernel_pages > memmap_pages * 2)
5385 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08005386 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005387
Jiang Liu9feedc92012-12-12 13:52:12 -08005388 /*
5389 * Set an approximate value for lowmem here, it will be adjusted
5390 * when the bootmem allocator frees pages into the buddy system.
5391 * And all highmem pages will be managed by the buddy system.
5392 */
5393 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07005394#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07005395 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08005396 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07005397 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08005398 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07005399#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005400 zone->name = zone_names[j];
5401 spin_lock_init(&zone->lock);
5402 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07005403 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005404 zone->zone_pgdat = pgdat;
Dave Hansened8ece22005-10-29 18:16:50 -07005405 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005406
5407 /* For bootup, initialized properly in watermark setup */
5408 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
5409
Hugh Dickinsbea8c152012-11-16 14:14:54 -08005410 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005411 if (!size)
5412 continue;
5413
Andrew Morton955c1cd2012-05-29 15:06:31 -07005414 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08005415 setup_usemap(pgdat, zone, zone_start_pfn, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08005416 ret = init_currently_empty_zone(zone, zone_start_pfn,
5417 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07005418 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07005419 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005420 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005421 }
5422}
5423
Sam Ravnborg577a32f2007-05-17 23:29:25 +02005424static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005425{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005426 /* Skip empty nodes */
5427 if (!pgdat->node_spanned_pages)
5428 return;
5429
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005430#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07005431 /* ia64 gets its own node_mem_map, before this, without bootmem */
5432 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07005433 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005434 struct page *map;
5435
Bob Piccoe984bb42006-05-20 15:00:31 -07005436 /*
5437 * The zone's endpoints aren't required to be MAX_ORDER
5438 * aligned but the node_mem_map endpoints must be in order
5439 * for the buddy allocator to function correctly.
5440 */
5441 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
Cody P Schafer108bcc92013-02-22 16:35:23 -08005442 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07005443 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5444 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07005445 map = alloc_remap(pgdat->node_id, size);
5446 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005447 map = memblock_virt_alloc_node_nopanic(size,
5448 pgdat->node_id);
Bob Piccoe984bb42006-05-20 15:00:31 -07005449 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005450 }
Roman Zippel12d810c2007-05-31 00:40:54 -07005451#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07005452 /*
5453 * With no DISCONTIG, the global mem_map is just set as node 0's
5454 */
Mel Gormanc7132162006-09-27 01:49:43 -07005455 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005456 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08005457#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07005458 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08005459 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08005460#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005461 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005462#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005463#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005464}
5465
Johannes Weiner9109fb72008-07-23 21:27:20 -07005466void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5467 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005468{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005469 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005470 unsigned long start_pfn = 0;
5471 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07005472
Minchan Kim88fdf752012-07-31 16:46:14 -07005473 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07005474 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07005475
Mel Gorman3a80a7f2015-06-30 14:57:02 -07005476 reset_deferred_meminit(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005477 pgdat->node_id = nid;
5478 pgdat->node_start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005479#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5480 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
Juergen Gross8d29e182015-02-11 15:26:01 -08005481 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
Zhen Lei4ada0c52015-09-08 15:04:19 -07005482 (u64)start_pfn << PAGE_SHIFT,
5483 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005484#endif
5485 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5486 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005487
5488 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07005489#ifdef CONFIG_FLAT_NODE_MEM_MAP
5490 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5491 nid, (unsigned long)pgdat,
5492 (unsigned long)pgdat->node_mem_map);
5493#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005494
Wei Yang7f3eb552015-09-08 14:59:50 -07005495 free_area_init_core(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005496}
5497
Tejun Heo0ee332c2011-12-08 10:22:09 -08005498#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07005499
5500#if MAX_NUMNODES > 1
5501/*
5502 * Figure out the number of possible node ids.
5503 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07005504void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07005505{
Wei Yang904a9552015-09-08 14:59:48 -07005506 unsigned int highest;
Miklos Szeredi418508c2007-05-23 13:57:55 -07005507
Wei Yang904a9552015-09-08 14:59:48 -07005508 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
Miklos Szeredi418508c2007-05-23 13:57:55 -07005509 nr_node_ids = highest + 1;
5510}
Miklos Szeredi418508c2007-05-23 13:57:55 -07005511#endif
5512
Mel Gormanc7132162006-09-27 01:49:43 -07005513/**
Tejun Heo1e019792011-07-12 09:45:34 +02005514 * node_map_pfn_alignment - determine the maximum internode alignment
5515 *
5516 * This function should be called after node map is populated and sorted.
5517 * It calculates the maximum power of two alignment which can distinguish
5518 * all the nodes.
5519 *
5520 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5521 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5522 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5523 * shifted, 1GiB is enough and this function will indicate so.
5524 *
5525 * This is used to test whether pfn -> nid mapping of the chosen memory
5526 * model has fine enough granularity to avoid incorrect mapping for the
5527 * populated node map.
5528 *
5529 * Returns the determined alignment in pfn's. 0 if there is no alignment
5530 * requirement (single node).
5531 */
5532unsigned long __init node_map_pfn_alignment(void)
5533{
5534 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005535 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02005536 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02005537 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02005538
Tejun Heoc13291a2011-07-12 10:46:30 +02005539 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02005540 if (!start || last_nid < 0 || last_nid == nid) {
5541 last_nid = nid;
5542 last_end = end;
5543 continue;
5544 }
5545
5546 /*
5547 * Start with a mask granular enough to pin-point to the
5548 * start pfn and tick off bits one-by-one until it becomes
5549 * too coarse to separate the current node from the last.
5550 */
5551 mask = ~((1 << __ffs(start)) - 1);
5552 while (mask && last_end <= (start & (mask << 1)))
5553 mask <<= 1;
5554
5555 /* accumulate all internode masks */
5556 accl_mask |= mask;
5557 }
5558
5559 /* convert mask to number of pages */
5560 return ~accl_mask + 1;
5561}
5562
Mel Gormana6af2bc2007-02-10 01:42:57 -08005563/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005564static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07005565{
Mel Gormana6af2bc2007-02-10 01:42:57 -08005566 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02005567 unsigned long start_pfn;
5568 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00005569
Tejun Heoc13291a2011-07-12 10:46:30 +02005570 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5571 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005572
Mel Gormana6af2bc2007-02-10 01:42:57 -08005573 if (min_pfn == ULONG_MAX) {
5574 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07005575 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08005576 return 0;
5577 }
5578
5579 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005580}
5581
5582/**
5583 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5584 *
5585 * It returns the minimum PFN based on information provided via
Zhang Zhen7d018172014-06-04 16:10:53 -07005586 * memblock_set_node().
Mel Gormanc7132162006-09-27 01:49:43 -07005587 */
5588unsigned long __init find_min_pfn_with_active_regions(void)
5589{
5590 return find_min_pfn_for_node(MAX_NUMNODES);
5591}
5592
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005593/*
5594 * early_calculate_totalpages()
5595 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005596 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005597 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07005598static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07005599{
Mel Gorman7e63efe2007-07-17 04:03:15 -07005600 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005601 unsigned long start_pfn, end_pfn;
5602 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005603
Tejun Heoc13291a2011-07-12 10:46:30 +02005604 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5605 unsigned long pages = end_pfn - start_pfn;
5606
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005607 totalpages += pages;
5608 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005609 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005610 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07005611 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005612}
5613
Mel Gorman2a1e2742007-07-17 04:03:12 -07005614/*
5615 * Find the PFN the Movable zone begins in each node. Kernel memory
5616 * is spread evenly between nodes as long as the nodes have enough
5617 * memory. When they don't, some nodes will have more kernelcore than
5618 * others
5619 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07005620static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005621{
5622 int i, nid;
5623 unsigned long usable_startpfn;
5624 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07005625 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005626 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005627 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005628 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07005629 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005630
5631 /* Need to find movable_zone earlier when movable_node is specified. */
5632 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07005633
Mel Gorman7e63efe2007-07-17 04:03:15 -07005634 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005635 * If movable_node is specified, ignore kernelcore and movablecore
5636 * options.
5637 */
5638 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07005639 for_each_memblock(memory, r) {
5640 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005641 continue;
5642
Emil Medve136199f2014-04-07 15:37:52 -07005643 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005644
Emil Medve136199f2014-04-07 15:37:52 -07005645 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005646 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5647 min(usable_startpfn, zone_movable_pfn[nid]) :
5648 usable_startpfn;
5649 }
5650
5651 goto out2;
5652 }
5653
5654 /*
5655 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07005656 * kernelcore that corresponds so that memory usable for
5657 * any allocation type is evenly spread. If both kernelcore
5658 * and movablecore are specified, then the value of kernelcore
5659 * will be used for required_kernelcore if it's greater than
5660 * what movablecore would have allowed.
5661 */
5662 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07005663 unsigned long corepages;
5664
5665 /*
5666 * Round-up so that ZONE_MOVABLE is at least as large as what
5667 * was requested by the user
5668 */
5669 required_movablecore =
5670 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5671 corepages = totalpages - required_movablecore;
5672
5673 required_kernelcore = max(required_kernelcore, corepages);
5674 }
5675
Yinghai Lu20e69262013-03-01 14:51:27 -08005676 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5677 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07005678 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005679
5680 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005681 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5682
5683restart:
5684 /* Spread kernelcore memory as evenly as possible throughout nodes */
5685 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005686 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02005687 unsigned long start_pfn, end_pfn;
5688
Mel Gorman2a1e2742007-07-17 04:03:12 -07005689 /*
5690 * Recalculate kernelcore_node if the division per node
5691 * now exceeds what is necessary to satisfy the requested
5692 * amount of memory for the kernel
5693 */
5694 if (required_kernelcore < kernelcore_node)
5695 kernelcore_node = required_kernelcore / usable_nodes;
5696
5697 /*
5698 * As the map is walked, we track how much memory is usable
5699 * by the kernel using kernelcore_remaining. When it is
5700 * 0, the rest of the node is usable by ZONE_MOVABLE
5701 */
5702 kernelcore_remaining = kernelcore_node;
5703
5704 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02005705 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005706 unsigned long size_pages;
5707
Tejun Heoc13291a2011-07-12 10:46:30 +02005708 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005709 if (start_pfn >= end_pfn)
5710 continue;
5711
5712 /* Account for what is only usable for kernelcore */
5713 if (start_pfn < usable_startpfn) {
5714 unsigned long kernel_pages;
5715 kernel_pages = min(end_pfn, usable_startpfn)
5716 - start_pfn;
5717
5718 kernelcore_remaining -= min(kernel_pages,
5719 kernelcore_remaining);
5720 required_kernelcore -= min(kernel_pages,
5721 required_kernelcore);
5722
5723 /* Continue if range is now fully accounted */
5724 if (end_pfn <= usable_startpfn) {
5725
5726 /*
5727 * Push zone_movable_pfn to the end so
5728 * that if we have to rebalance
5729 * kernelcore across nodes, we will
5730 * not double account here
5731 */
5732 zone_movable_pfn[nid] = end_pfn;
5733 continue;
5734 }
5735 start_pfn = usable_startpfn;
5736 }
5737
5738 /*
5739 * The usable PFN range for ZONE_MOVABLE is from
5740 * start_pfn->end_pfn. Calculate size_pages as the
5741 * number of pages used as kernelcore
5742 */
5743 size_pages = end_pfn - start_pfn;
5744 if (size_pages > kernelcore_remaining)
5745 size_pages = kernelcore_remaining;
5746 zone_movable_pfn[nid] = start_pfn + size_pages;
5747
5748 /*
5749 * Some kernelcore has been met, update counts and
5750 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07005751 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005752 */
5753 required_kernelcore -= min(required_kernelcore,
5754 size_pages);
5755 kernelcore_remaining -= size_pages;
5756 if (!kernelcore_remaining)
5757 break;
5758 }
5759 }
5760
5761 /*
5762 * If there is still required_kernelcore, we do another pass with one
5763 * less node in the count. This will push zone_movable_pfn[nid] further
5764 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07005765 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005766 */
5767 usable_nodes--;
5768 if (usable_nodes && required_kernelcore > usable_nodes)
5769 goto restart;
5770
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005771out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07005772 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5773 for (nid = 0; nid < MAX_NUMNODES; nid++)
5774 zone_movable_pfn[nid] =
5775 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005776
Yinghai Lu20e69262013-03-01 14:51:27 -08005777out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005778 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005779 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005780}
5781
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005782/* Any regular or high memory on that node ? */
5783static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005784{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005785 enum zone_type zone_type;
5786
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005787 if (N_MEMORY == N_NORMAL_MEMORY)
5788 return;
5789
5790 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005791 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08005792 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005793 node_set_state(nid, N_HIGH_MEMORY);
5794 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5795 zone_type <= ZONE_NORMAL)
5796 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005797 break;
5798 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005799 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005800}
5801
Mel Gormanc7132162006-09-27 01:49:43 -07005802/**
5803 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005804 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005805 *
5806 * This will call free_area_init_node() for each active node in the system.
Zhang Zhen7d018172014-06-04 16:10:53 -07005807 * Using the page ranges provided by memblock_set_node(), the size of each
Mel Gormanc7132162006-09-27 01:49:43 -07005808 * zone in each node and their holes is calculated. If the maximum PFN
5809 * between two adjacent zones match, it is assumed that the zone is empty.
5810 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5811 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5812 * starts where the previous one ended. For example, ZONE_DMA32 starts
5813 * at arch_max_dma_pfn.
5814 */
5815void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5816{
Tejun Heoc13291a2011-07-12 10:46:30 +02005817 unsigned long start_pfn, end_pfn;
5818 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005819
Mel Gormanc7132162006-09-27 01:49:43 -07005820 /* Record where the zone boundaries are */
5821 memset(arch_zone_lowest_possible_pfn, 0,
5822 sizeof(arch_zone_lowest_possible_pfn));
5823 memset(arch_zone_highest_possible_pfn, 0,
5824 sizeof(arch_zone_highest_possible_pfn));
5825 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5826 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5827 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005828 if (i == ZONE_MOVABLE)
5829 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005830 arch_zone_lowest_possible_pfn[i] =
5831 arch_zone_highest_possible_pfn[i-1];
5832 arch_zone_highest_possible_pfn[i] =
5833 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5834 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005835 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5836 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5837
5838 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5839 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005840 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005841
Mel Gormanc7132162006-09-27 01:49:43 -07005842 /* Print out the zone ranges */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005843 pr_info("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005844 for (i = 0; i < MAX_NR_ZONES; i++) {
5845 if (i == ZONE_MOVABLE)
5846 continue;
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005847 pr_info(" %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08005848 if (arch_zone_lowest_possible_pfn[i] ==
5849 arch_zone_highest_possible_pfn[i])
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005850 pr_cont("empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08005851 else
Juergen Gross8d29e182015-02-11 15:26:01 -08005852 pr_cont("[mem %#018Lx-%#018Lx]\n",
5853 (u64)arch_zone_lowest_possible_pfn[i]
5854 << PAGE_SHIFT,
5855 ((u64)arch_zone_highest_possible_pfn[i]
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005856 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005857 }
5858
5859 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005860 pr_info("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005861 for (i = 0; i < MAX_NUMNODES; i++) {
5862 if (zone_movable_pfn[i])
Juergen Gross8d29e182015-02-11 15:26:01 -08005863 pr_info(" Node %d: %#018Lx\n", i,
5864 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005865 }
Mel Gormanc7132162006-09-27 01:49:43 -07005866
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07005867 /* Print out the early node map */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005868 pr_info("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02005869 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Juergen Gross8d29e182015-02-11 15:26:01 -08005870 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
5871 (u64)start_pfn << PAGE_SHIFT,
5872 ((u64)end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07005873
5874 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005875 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005876 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005877 for_each_online_node(nid) {
5878 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005879 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005880 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005881
5882 /* Any memory on that node */
5883 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005884 node_set_state(nid, N_MEMORY);
5885 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005886 }
5887}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005888
Mel Gorman7e63efe2007-07-17 04:03:15 -07005889static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005890{
5891 unsigned long long coremem;
5892 if (!p)
5893 return -EINVAL;
5894
5895 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005896 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005897
Mel Gorman7e63efe2007-07-17 04:03:15 -07005898 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005899 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5900
5901 return 0;
5902}
Mel Gormaned7ed362007-07-17 04:03:14 -07005903
Mel Gorman7e63efe2007-07-17 04:03:15 -07005904/*
5905 * kernelcore=size sets the amount of memory for use for allocations that
5906 * cannot be reclaimed or migrated.
5907 */
5908static int __init cmdline_parse_kernelcore(char *p)
5909{
5910 return cmdline_parse_core(p, &required_kernelcore);
5911}
5912
5913/*
5914 * movablecore=size sets the amount of memory for use for allocations that
5915 * can be reclaimed or migrated.
5916 */
5917static int __init cmdline_parse_movablecore(char *p)
5918{
5919 return cmdline_parse_core(p, &required_movablecore);
5920}
5921
Mel Gormaned7ed362007-07-17 04:03:14 -07005922early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005923early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005924
Tejun Heo0ee332c2011-12-08 10:22:09 -08005925#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005926
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005927void adjust_managed_page_count(struct page *page, long count)
5928{
5929 spin_lock(&managed_page_count_lock);
5930 page_zone(page)->managed_pages += count;
5931 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07005932#ifdef CONFIG_HIGHMEM
5933 if (PageHighMem(page))
5934 totalhigh_pages += count;
5935#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005936 spin_unlock(&managed_page_count_lock);
5937}
Jiang Liu3dcc0572013-07-03 15:03:21 -07005938EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005939
Jiang Liu11199692013-07-03 15:02:48 -07005940unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07005941{
Jiang Liu11199692013-07-03 15:02:48 -07005942 void *pos;
5943 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07005944
Jiang Liu11199692013-07-03 15:02:48 -07005945 start = (void *)PAGE_ALIGN((unsigned long)start);
5946 end = (void *)((unsigned long)end & PAGE_MASK);
5947 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07005948 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07005949 memset(pos, poison, PAGE_SIZE);
5950 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07005951 }
5952
5953 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07005954 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07005955 s, pages << (PAGE_SHIFT - 10), start, end);
5956
5957 return pages;
5958}
Jiang Liu11199692013-07-03 15:02:48 -07005959EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07005960
Jiang Liucfa11e02013-04-29 15:07:00 -07005961#ifdef CONFIG_HIGHMEM
5962void free_highmem_page(struct page *page)
5963{
5964 __free_reserved_page(page);
5965 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07005966 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07005967 totalhigh_pages++;
5968}
5969#endif
5970
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005971
5972void __init mem_init_print_info(const char *str)
5973{
5974 unsigned long physpages, codesize, datasize, rosize, bss_size;
5975 unsigned long init_code_size, init_data_size;
5976
5977 physpages = get_num_physpages();
5978 codesize = _etext - _stext;
5979 datasize = _edata - _sdata;
5980 rosize = __end_rodata - __start_rodata;
5981 bss_size = __bss_stop - __bss_start;
5982 init_data_size = __init_end - __init_begin;
5983 init_code_size = _einittext - _sinittext;
5984
5985 /*
5986 * Detect special cases and adjust section sizes accordingly:
5987 * 1) .init.* may be embedded into .data sections
5988 * 2) .init.text.* may be out of [__init_begin, __init_end],
5989 * please refer to arch/tile/kernel/vmlinux.lds.S.
5990 * 3) .rodata.* may be embedded into .text or .data sections.
5991 */
5992#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07005993 do { \
5994 if (start <= pos && pos < end && size > adj) \
5995 size -= adj; \
5996 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005997
5998 adj_init_size(__init_begin, __init_end, init_data_size,
5999 _sinittext, init_code_size);
6000 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
6001 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
6002 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
6003 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
6004
6005#undef adj_init_size
6006
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006007 pr_info("Memory: %luK/%luK available "
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006008 "(%luK kernel code, %luK rwdata, %luK rodata, "
Pintu Kumare48322a2014-12-18 16:17:15 -08006009 "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006010#ifdef CONFIG_HIGHMEM
6011 ", %luK highmem"
6012#endif
6013 "%s%s)\n",
6014 nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
6015 codesize >> 10, datasize >> 10, rosize >> 10,
6016 (init_data_size + init_code_size) >> 10, bss_size >> 10,
Pintu Kumare48322a2014-12-18 16:17:15 -08006017 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10),
6018 totalcma_pages << (PAGE_SHIFT-10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006019#ifdef CONFIG_HIGHMEM
6020 totalhigh_pages << (PAGE_SHIFT-10),
6021#endif
6022 str ? ", " : "", str ? str : "");
6023}
6024
Mel Gorman0e0b8642006-09-27 01:49:56 -07006025/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006026 * set_dma_reserve - set the specified number of pages reserved in the first zone
6027 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07006028 *
Yaowei Bai013110a2015-09-08 15:04:10 -07006029 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
Mel Gorman0e0b8642006-09-27 01:49:56 -07006030 * In the DMA zone, a significant percentage may be consumed by kernel image
6031 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006032 * function may optionally be used to account for unfreeable pages in the
6033 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6034 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07006035 */
6036void __init set_dma_reserve(unsigned long new_dma_reserve)
6037{
6038 dma_reserve = new_dma_reserve;
6039}
6040
Linus Torvalds1da177e2005-04-16 15:20:36 -07006041void __init free_area_init(unsigned long *zones_size)
6042{
Johannes Weiner9109fb72008-07-23 21:27:20 -07006043 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006044 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
6045}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006046
Linus Torvalds1da177e2005-04-16 15:20:36 -07006047static int page_alloc_cpu_notify(struct notifier_block *self,
6048 unsigned long action, void *hcpu)
6049{
6050 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006051
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07006052 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07006053 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006054 drain_pages(cpu);
6055
6056 /*
6057 * Spill the event counters of the dead processor
6058 * into the current processors event counters.
6059 * This artificially elevates the count of the current
6060 * processor.
6061 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07006062 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006063
6064 /*
6065 * Zero the differential counters of the dead processor
6066 * so that the vm statistics are consistent.
6067 *
6068 * This is only okay since the processor is dead and cannot
6069 * race with what we are doing.
6070 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07006071 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006072 }
6073 return NOTIFY_OK;
6074}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006075
6076void __init page_alloc_init(void)
6077{
6078 hotcpu_notifier(page_alloc_cpu_notify, 0);
6079}
6080
6081/*
Yaowei Bai34b10062015-09-08 15:04:13 -07006082 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006083 * or min_free_kbytes changes.
6084 */
6085static void calculate_totalreserve_pages(void)
6086{
6087 struct pglist_data *pgdat;
6088 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006089 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006090
6091 for_each_online_pgdat(pgdat) {
6092 for (i = 0; i < MAX_NR_ZONES; i++) {
6093 struct zone *zone = pgdat->node_zones + i;
Mel Gorman3484b2d2014-08-06 16:07:14 -07006094 long max = 0;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006095
6096 /* Find valid and maximum lowmem_reserve in the zone */
6097 for (j = i; j < MAX_NR_ZONES; j++) {
6098 if (zone->lowmem_reserve[j] > max)
6099 max = zone->lowmem_reserve[j];
6100 }
6101
Mel Gorman41858962009-06-16 15:32:12 -07006102 /* we treat the high watermark as reserved pages. */
6103 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006104
Jiang Liub40da042013-02-22 16:33:52 -08006105 if (max > zone->managed_pages)
6106 max = zone->managed_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006107 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08006108 /*
6109 * Lowmem reserves are not available to
6110 * GFP_HIGHUSER page cache allocations and
6111 * kswapd tries to balance zones to their high
6112 * watermark. As a result, neither should be
6113 * regarded as dirtyable memory, to prevent a
6114 * situation where reclaim has to clean pages
6115 * in order to balance the zones.
6116 */
6117 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006118 }
6119 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08006120 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006121 totalreserve_pages = reserve_pages;
6122}
6123
6124/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07006125 * setup_per_zone_lowmem_reserve - called whenever
Yaowei Bai34b10062015-09-08 15:04:13 -07006126 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
Linus Torvalds1da177e2005-04-16 15:20:36 -07006127 * has a correct pages reserved value, so an adequate number of
6128 * pages are left in the zone after a successful __alloc_pages().
6129 */
6130static void setup_per_zone_lowmem_reserve(void)
6131{
6132 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006133 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006134
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08006135 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006136 for (j = 0; j < MAX_NR_ZONES; j++) {
6137 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08006138 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006139
6140 zone->lowmem_reserve[j] = 0;
6141
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006142 idx = j;
6143 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006144 struct zone *lower_zone;
6145
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006146 idx--;
6147
Linus Torvalds1da177e2005-04-16 15:20:36 -07006148 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6149 sysctl_lowmem_reserve_ratio[idx] = 1;
6150
6151 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08006152 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07006153 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08006154 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006155 }
6156 }
6157 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006158
6159 /* update totalreserve_pages */
6160 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006161}
6162
Mel Gormancfd3da12011-04-25 21:36:42 +00006163static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006164{
6165 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6166 unsigned long lowmem_pages = 0;
6167 struct zone *zone;
6168 unsigned long flags;
6169
6170 /* Calculate total number of !ZONE_HIGHMEM pages */
6171 for_each_zone(zone) {
6172 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08006173 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006174 }
6175
6176 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07006177 u64 tmp;
6178
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006179 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08006180 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07006181 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006182 if (is_highmem(zone)) {
6183 /*
Nick Piggin669ed172005-11-13 16:06:45 -08006184 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6185 * need highmem pages, so cap pages_min to a small
6186 * value here.
6187 *
Mel Gorman41858962009-06-16 15:32:12 -07006188 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Yaowei Bai42ff2702015-04-14 15:47:14 -07006189 * deltas control asynch page reclaim, and so should
Nick Piggin669ed172005-11-13 16:06:45 -08006190 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006191 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08006192 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006193
Jiang Liub40da042013-02-22 16:33:52 -08006194 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08006195 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07006196 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006197 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08006198 /*
6199 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07006200 * proportionate to the zone's size.
6201 */
Mel Gorman41858962009-06-16 15:32:12 -07006202 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006203 }
6204
Mel Gorman41858962009-06-16 15:32:12 -07006205 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
6206 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006207
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006208 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
Johannes Weinerabe5f972014-10-02 16:21:10 -07006209 high_wmark_pages(zone) - low_wmark_pages(zone) -
6210 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006211
Mel Gorman56fd56b2007-10-16 01:25:58 -07006212 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006213 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006214 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006215
6216 /* update totalreserve_pages */
6217 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006218}
6219
Mel Gormancfd3da12011-04-25 21:36:42 +00006220/**
6221 * setup_per_zone_wmarks - called when min_free_kbytes changes
6222 * or when memory is hot-{added|removed}
6223 *
6224 * Ensures that the watermark[min,low,high] values for each zone are set
6225 * correctly with respect to min_free_kbytes.
6226 */
6227void setup_per_zone_wmarks(void)
6228{
6229 mutex_lock(&zonelists_mutex);
6230 __setup_per_zone_wmarks();
6231 mutex_unlock(&zonelists_mutex);
6232}
6233
Randy Dunlap55a44622009-09-21 17:01:20 -07006234/*
Rik van Riel556adec2008-10-18 20:26:34 -07006235 * The inactive anon list should be small enough that the VM never has to
6236 * do too much work, but large enough that each inactive page has a chance
6237 * to be referenced again before it is swapped out.
6238 *
6239 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6240 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6241 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6242 * the anonymous pages are kept on the inactive list.
6243 *
6244 * total target max
6245 * memory ratio inactive anon
6246 * -------------------------------------
6247 * 10MB 1 5MB
6248 * 100MB 1 50MB
6249 * 1GB 3 250MB
6250 * 10GB 10 0.9GB
6251 * 100GB 31 3GB
6252 * 1TB 101 10GB
6253 * 10TB 320 32GB
6254 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07006255static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07006256{
6257 unsigned int gb, ratio;
6258
6259 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08006260 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07006261 if (gb)
6262 ratio = int_sqrt(10 * gb);
6263 else
6264 ratio = 1;
6265
6266 zone->inactive_ratio = ratio;
6267}
6268
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07006269static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07006270{
6271 struct zone *zone;
6272
Minchan Kim96cb4df2009-06-16 15:32:49 -07006273 for_each_zone(zone)
6274 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07006275}
6276
Linus Torvalds1da177e2005-04-16 15:20:36 -07006277/*
6278 * Initialise min_free_kbytes.
6279 *
6280 * For small machines we want it small (128k min). For large machines
6281 * we want it large (64MB max). But it is not linear, because network
6282 * bandwidth does not increase linearly with machine size. We use
6283 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006284 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07006285 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6286 *
6287 * which yields
6288 *
6289 * 16MB: 512k
6290 * 32MB: 724k
6291 * 64MB: 1024k
6292 * 128MB: 1448k
6293 * 256MB: 2048k
6294 * 512MB: 2896k
6295 * 1024MB: 4096k
6296 * 2048MB: 5792k
6297 * 4096MB: 8192k
6298 * 8192MB: 11584k
6299 * 16384MB: 16384k
6300 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07006301int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006302{
6303 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07006304 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006305
6306 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07006307 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006308
Michal Hocko5f127332013-07-08 16:00:40 -07006309 if (new_min_free_kbytes > user_min_free_kbytes) {
6310 min_free_kbytes = new_min_free_kbytes;
6311 if (min_free_kbytes < 128)
6312 min_free_kbytes = 128;
6313 if (min_free_kbytes > 65536)
6314 min_free_kbytes = 65536;
6315 } else {
6316 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6317 new_min_free_kbytes, user_min_free_kbytes);
6318 }
Minchan Kimbc75d332009-06-16 15:32:48 -07006319 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07006320 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006321 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07006322 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006323 return 0;
6324}
Minchan Kimbc75d332009-06-16 15:32:48 -07006325module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006326
6327/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07006328 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07006329 * that we can call two helper functions whenever min_free_kbytes
6330 * changes.
6331 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006332int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006333 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006334{
Han Pingtianda8c7572014-01-23 15:53:17 -08006335 int rc;
6336
6337 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6338 if (rc)
6339 return rc;
6340
Michal Hocko5f127332013-07-08 16:00:40 -07006341 if (write) {
6342 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07006343 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07006344 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006345 return 0;
6346}
6347
Christoph Lameter96146342006-07-03 00:24:13 -07006348#ifdef CONFIG_NUMA
Joe Perchescccad5b2014-06-06 14:38:09 -07006349int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006350 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07006351{
6352 struct zone *zone;
6353 int rc;
6354
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006355 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07006356 if (rc)
6357 return rc;
6358
6359 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08006360 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07006361 sysctl_min_unmapped_ratio) / 100;
6362 return 0;
6363}
Christoph Lameter0ff38492006-09-25 23:31:52 -07006364
Joe Perchescccad5b2014-06-06 14:38:09 -07006365int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006366 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07006367{
6368 struct zone *zone;
6369 int rc;
6370
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006371 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07006372 if (rc)
6373 return rc;
6374
6375 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08006376 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07006377 sysctl_min_slab_ratio) / 100;
6378 return 0;
6379}
Christoph Lameter96146342006-07-03 00:24:13 -07006380#endif
6381
Linus Torvalds1da177e2005-04-16 15:20:36 -07006382/*
6383 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6384 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6385 * whenever sysctl_lowmem_reserve_ratio changes.
6386 *
6387 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07006388 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07006389 * if in function of the boot time zone sizes.
6390 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006391int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006392 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006393{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006394 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006395 setup_per_zone_lowmem_reserve();
6396 return 0;
6397}
6398
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006399/*
6400 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07006401 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6402 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006403 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006404int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006405 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006406{
6407 struct zone *zone;
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006408 int old_percpu_pagelist_fraction;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006409 int ret;
6410
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006411 mutex_lock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006412 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6413
6414 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
6415 if (!write || ret < 0)
6416 goto out;
6417
6418 /* Sanity checking to avoid pcp imbalance */
6419 if (percpu_pagelist_fraction &&
6420 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6421 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6422 ret = -EINVAL;
6423 goto out;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006424 }
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006425
6426 /* No change? */
6427 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6428 goto out;
6429
6430 for_each_populated_zone(zone) {
6431 unsigned int cpu;
6432
6433 for_each_possible_cpu(cpu)
6434 pageset_set_high_and_batch(zone,
6435 per_cpu_ptr(zone->pageset, cpu));
6436 }
6437out:
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006438 mutex_unlock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006439 return ret;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006440}
6441
Rasmus Villemoesa9919c72015-06-24 16:56:28 -07006442#ifdef CONFIG_NUMA
David S. Millerf034b5d2006-08-24 03:08:07 -07006443int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006444
Linus Torvalds1da177e2005-04-16 15:20:36 -07006445static int __init set_hashdist(char *str)
6446{
6447 if (!str)
6448 return 0;
6449 hashdist = simple_strtoul(str, &str, 0);
6450 return 1;
6451}
6452__setup("hashdist=", set_hashdist);
6453#endif
6454
6455/*
6456 * allocate a large system hash table from bootmem
6457 * - it is assumed that the hash table must contain an exact power-of-2
6458 * quantity of entries
6459 * - limit is the number of hash buckets, not the total allocation size
6460 */
6461void *__init alloc_large_system_hash(const char *tablename,
6462 unsigned long bucketsize,
6463 unsigned long numentries,
6464 int scale,
6465 int flags,
6466 unsigned int *_hash_shift,
6467 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00006468 unsigned long low_limit,
6469 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006470{
Tim Bird31fe62b2012-05-23 13:33:35 +00006471 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006472 unsigned long log2qty, size;
6473 void *table = NULL;
6474
6475 /* allow the kernel cmdline to have a say */
6476 if (!numentries) {
6477 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08006478 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07006479
6480 /* It isn't necessary when PAGE_SIZE >= 1MB */
6481 if (PAGE_SHIFT < 20)
6482 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006483
6484 /* limit to 1 bucket per 2^scale bytes of low memory */
6485 if (scale > PAGE_SHIFT)
6486 numentries >>= (scale - PAGE_SHIFT);
6487 else
6488 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08006489
6490 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07006491 if (unlikely(flags & HASH_SMALL)) {
6492 /* Makes no sense without HASH_EARLY */
6493 WARN_ON(!(flags & HASH_EARLY));
6494 if (!(numentries >> *_hash_shift)) {
6495 numentries = 1UL << *_hash_shift;
6496 BUG_ON(!numentries);
6497 }
6498 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08006499 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006500 }
John Hawkes6e692ed2006-03-25 03:08:02 -08006501 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006502
6503 /* limit allocation size to 1/16 total memory by default */
6504 if (max == 0) {
6505 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6506 do_div(max, bucketsize);
6507 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08006508 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006509
Tim Bird31fe62b2012-05-23 13:33:35 +00006510 if (numentries < low_limit)
6511 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006512 if (numentries > max)
6513 numentries = max;
6514
David Howellsf0d1b0b2006-12-08 02:37:49 -08006515 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006516
6517 do {
6518 size = bucketsize << log2qty;
6519 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08006520 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006521 else if (hashdist)
6522 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6523 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07006524 /*
6525 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07006526 * some pages at the end of hash table which
6527 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07006528 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006529 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07006530 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006531 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6532 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006533 }
6534 } while (!table && size > PAGE_SIZE && --log2qty);
6535
6536 if (!table)
6537 panic("Failed to allocate %s hash table\n", tablename);
6538
Robin Holtf241e6602010-10-07 12:59:26 -07006539 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07006540 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07006541 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08006542 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006543 size);
6544
6545 if (_hash_shift)
6546 *_hash_shift = log2qty;
6547 if (_hash_mask)
6548 *_hash_mask = (1 << log2qty) - 1;
6549
6550 return table;
6551}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08006552
Mel Gorman835c1342007-10-16 01:25:47 -07006553/* Return a pointer to the bitmap storing bits affecting a block of pages */
6554static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6555 unsigned long pfn)
6556{
6557#ifdef CONFIG_SPARSEMEM
6558 return __pfn_to_section(pfn)->pageblock_flags;
6559#else
6560 return zone->pageblock_flags;
6561#endif /* CONFIG_SPARSEMEM */
6562}
Andrew Morton6220ec72006-10-19 23:29:05 -07006563
Mel Gorman835c1342007-10-16 01:25:47 -07006564static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6565{
6566#ifdef CONFIG_SPARSEMEM
6567 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07006568 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006569#else
Laura Abbottc060f942013-01-11 14:31:51 -08006570 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07006571 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006572#endif /* CONFIG_SPARSEMEM */
6573}
6574
6575/**
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006576 * 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 -07006577 * @page: The page within the block of interest
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006578 * @pfn: The target page frame number
6579 * @end_bitidx: The last bit of interest to retrieve
6580 * @mask: mask of bits that the caller is interested in
6581 *
6582 * Return: pageblock_bits flags
Mel Gorman835c1342007-10-16 01:25:47 -07006583 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006584unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006585 unsigned long end_bitidx,
6586 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006587{
6588 struct zone *zone;
6589 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006590 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006591 unsigned long word;
Mel Gorman835c1342007-10-16 01:25:47 -07006592
6593 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006594 bitmap = get_pageblock_bitmap(zone, pfn);
6595 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006596 word_bitidx = bitidx / BITS_PER_LONG;
6597 bitidx &= (BITS_PER_LONG-1);
Mel Gorman835c1342007-10-16 01:25:47 -07006598
Mel Gormane58469b2014-06-04 16:10:16 -07006599 word = bitmap[word_bitidx];
6600 bitidx += end_bitidx;
6601 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
Mel Gorman835c1342007-10-16 01:25:47 -07006602}
6603
6604/**
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006605 * 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 -07006606 * @page: The page within the block of interest
Mel Gorman835c1342007-10-16 01:25:47 -07006607 * @flags: The flags to set
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006608 * @pfn: The target page frame number
6609 * @end_bitidx: The last bit of interest
6610 * @mask: mask of bits that the caller is interested in
Mel Gorman835c1342007-10-16 01:25:47 -07006611 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006612void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6613 unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006614 unsigned long end_bitidx,
6615 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006616{
6617 struct zone *zone;
6618 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006619 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006620 unsigned long old_word, word;
6621
6622 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
Mel Gorman835c1342007-10-16 01:25:47 -07006623
6624 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006625 bitmap = get_pageblock_bitmap(zone, pfn);
6626 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006627 word_bitidx = bitidx / BITS_PER_LONG;
6628 bitidx &= (BITS_PER_LONG-1);
6629
Sasha Levin309381fea2014-01-23 15:52:54 -08006630 VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
Mel Gorman835c1342007-10-16 01:25:47 -07006631
Mel Gormane58469b2014-06-04 16:10:16 -07006632 bitidx += end_bitidx;
6633 mask <<= (BITS_PER_LONG - bitidx - 1);
6634 flags <<= (BITS_PER_LONG - bitidx - 1);
6635
Jason Low4db0c3c2015-04-15 16:14:08 -07006636 word = READ_ONCE(bitmap[word_bitidx]);
Mel Gormane58469b2014-06-04 16:10:16 -07006637 for (;;) {
6638 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6639 if (word == old_word)
6640 break;
6641 word = old_word;
6642 }
Mel Gorman835c1342007-10-16 01:25:47 -07006643}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006644
6645/*
Minchan Kim80934512012-07-31 16:43:01 -07006646 * This function checks whether pageblock includes unmovable pages or not.
6647 * If @count is not zero, it is okay to include less @count unmovable pages
6648 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006649 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07006650 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6651 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006652 */
Wen Congyangb023f462012-12-11 16:00:45 -08006653bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6654 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006655{
6656 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006657 int mt;
6658
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006659 /*
6660 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07006661 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006662 */
6663 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07006664 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006665 mt = get_pageblock_migratetype(page);
6666 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07006667 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006668
6669 pfn = page_to_pfn(page);
6670 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6671 unsigned long check = pfn + iter;
6672
Namhyung Kim29723fc2011-02-25 14:44:25 -08006673 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006674 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08006675
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006676 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07006677
6678 /*
6679 * Hugepages are not in LRU lists, but they're movable.
6680 * We need not scan over tail pages bacause we don't
6681 * handle each tail page individually in migration.
6682 */
6683 if (PageHuge(page)) {
6684 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6685 continue;
6686 }
6687
Minchan Kim97d255c2012-07-31 16:42:59 -07006688 /*
6689 * We can't use page_count without pin a page
6690 * because another CPU can free compound page.
6691 * This check already skips compound tails of THP
6692 * because their page->_count is zero at all time.
6693 */
6694 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006695 if (PageBuddy(page))
6696 iter += (1 << page_order(page)) - 1;
6697 continue;
6698 }
Minchan Kim97d255c2012-07-31 16:42:59 -07006699
Wen Congyangb023f462012-12-11 16:00:45 -08006700 /*
6701 * The HWPoisoned page may be not in buddy system, and
6702 * page_count() is not 0.
6703 */
6704 if (skip_hwpoisoned_pages && PageHWPoison(page))
6705 continue;
6706
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006707 if (!PageLRU(page))
6708 found++;
6709 /*
Johannes Weiner6b4f7792014-12-12 16:56:13 -08006710 * If there are RECLAIMABLE pages, we need to check
6711 * it. But now, memory offline itself doesn't call
6712 * shrink_node_slabs() and it still to be fixed.
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006713 */
6714 /*
6715 * If the page is not RAM, page_count()should be 0.
6716 * we don't need more check. This is an _used_ not-movable page.
6717 *
6718 * The problematic thing here is PG_reserved pages. PG_reserved
6719 * is set to both of a memory hole page and a _used_ kernel
6720 * page at boot.
6721 */
6722 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07006723 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006724 }
Minchan Kim80934512012-07-31 16:43:01 -07006725 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006726}
6727
6728bool is_pageblock_removable_nolock(struct page *page)
6729{
Michal Hocko656a0702012-01-20 14:33:58 -08006730 struct zone *zone;
6731 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08006732
6733 /*
6734 * We have to be careful here because we are iterating over memory
6735 * sections which are not zone aware so we might end up outside of
6736 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08006737 * We have to take care about the node as well. If the node is offline
6738 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08006739 */
Michal Hocko656a0702012-01-20 14:33:58 -08006740 if (!node_online(page_to_nid(page)))
6741 return false;
6742
6743 zone = page_zone(page);
6744 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08006745 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08006746 return false;
6747
Wen Congyangb023f462012-12-11 16:00:45 -08006748 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006749}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006750
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006751#ifdef CONFIG_CMA
6752
6753static unsigned long pfn_max_align_down(unsigned long pfn)
6754{
6755 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6756 pageblock_nr_pages) - 1);
6757}
6758
6759static unsigned long pfn_max_align_up(unsigned long pfn)
6760{
6761 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6762 pageblock_nr_pages));
6763}
6764
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006765/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006766static int __alloc_contig_migrate_range(struct compact_control *cc,
6767 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006768{
6769 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006770 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006771 unsigned long pfn = start;
6772 unsigned int tries = 0;
6773 int ret = 0;
6774
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08006775 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006776
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006777 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006778 if (fatal_signal_pending(current)) {
6779 ret = -EINTR;
6780 break;
6781 }
6782
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006783 if (list_empty(&cc->migratepages)) {
6784 cc->nr_migratepages = 0;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07006785 pfn = isolate_migratepages_range(cc, pfn, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006786 if (!pfn) {
6787 ret = -EINTR;
6788 break;
6789 }
6790 tries = 0;
6791 } else if (++tries == 5) {
6792 ret = ret < 0 ? ret : -EBUSY;
6793 break;
6794 }
6795
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006796 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6797 &cc->migratepages);
6798 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07006799
Hugh Dickins9c620e22013-02-22 16:35:14 -08006800 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07006801 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006802 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08006803 if (ret < 0) {
6804 putback_movable_pages(&cc->migratepages);
6805 return ret;
6806 }
6807 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006808}
6809
6810/**
6811 * alloc_contig_range() -- tries to allocate given range of pages
6812 * @start: start PFN to allocate
6813 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006814 * @migratetype: migratetype of the underlaying pageblocks (either
6815 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6816 * in range must have the same migratetype and it must
6817 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006818 *
6819 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6820 * aligned, however it's the caller's responsibility to guarantee that
6821 * we are the only thread that changes migrate type of pageblocks the
6822 * pages fall in.
6823 *
6824 * The PFN range must belong to a single zone.
6825 *
6826 * Returns zero on success or negative error code. On success all
6827 * pages which PFN is in [start, end) are allocated for the caller and
6828 * need to be freed with free_contig_range().
6829 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006830int alloc_contig_range(unsigned long start, unsigned long end,
6831 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006832{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006833 unsigned long outer_start, outer_end;
6834 int ret = 0, order;
6835
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006836 struct compact_control cc = {
6837 .nr_migratepages = 0,
6838 .order = -1,
6839 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07006840 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006841 .ignore_skip_hint = true,
6842 };
6843 INIT_LIST_HEAD(&cc.migratepages);
6844
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006845 /*
6846 * What we do here is we mark all pageblocks in range as
6847 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6848 * have different sizes, and due to the way page allocator
6849 * work, we align the range to biggest of the two pages so
6850 * that page allocator won't try to merge buddies from
6851 * different pageblocks and change MIGRATE_ISOLATE to some
6852 * other migration type.
6853 *
6854 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6855 * migrate the pages from an unaligned range (ie. pages that
6856 * we are interested in). This will put all the pages in
6857 * range back to page allocator as MIGRATE_ISOLATE.
6858 *
6859 * When this is done, we take the pages in range from page
6860 * allocator removing them from the buddy system. This way
6861 * page allocator will never consider using them.
6862 *
6863 * This lets us mark the pageblocks back as
6864 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6865 * aligned range but not in the unaligned, original range are
6866 * put back to page allocator so that buddy can use them.
6867 */
6868
6869 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08006870 pfn_max_align_up(end), migratetype,
6871 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006872 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07006873 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006874
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006875 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006876 if (ret)
6877 goto done;
6878
6879 /*
6880 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6881 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6882 * more, all pages in [start, end) are free in page allocator.
6883 * What we are going to do is to allocate all pages from
6884 * [start, end) (that is remove them from page allocator).
6885 *
6886 * The only problem is that pages at the beginning and at the
6887 * end of interesting range may be not aligned with pages that
6888 * page allocator holds, ie. they can be part of higher order
6889 * pages. Because of this, we reserve the bigger range and
6890 * once this is done free the pages we are not interested in.
6891 *
6892 * We don't have to hold zone->lock here because the pages are
6893 * isolated thus they won't get removed from buddy.
6894 */
6895
6896 lru_add_drain_all();
Vlastimil Babka510f5502014-12-10 15:43:07 -08006897 drain_all_pages(cc.zone);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006898
6899 order = 0;
6900 outer_start = start;
6901 while (!PageBuddy(pfn_to_page(outer_start))) {
6902 if (++order >= MAX_ORDER) {
6903 ret = -EBUSY;
6904 goto done;
6905 }
6906 outer_start &= ~0UL << order;
6907 }
6908
6909 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08006910 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewiczdae803e2014-11-13 15:19:27 -08006911 pr_info("%s: [%lx, %lx) PFNs busy\n",
6912 __func__, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006913 ret = -EBUSY;
6914 goto done;
6915 }
6916
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006917 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006918 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006919 if (!outer_end) {
6920 ret = -EBUSY;
6921 goto done;
6922 }
6923
6924 /* Free head and tail (if any) */
6925 if (start != outer_start)
6926 free_contig_range(outer_start, start - outer_start);
6927 if (end != outer_end)
6928 free_contig_range(end, outer_end - end);
6929
6930done:
6931 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006932 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006933 return ret;
6934}
6935
6936void free_contig_range(unsigned long pfn, unsigned nr_pages)
6937{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08006938 unsigned int count = 0;
6939
6940 for (; nr_pages--; pfn++) {
6941 struct page *page = pfn_to_page(pfn);
6942
6943 count += page_count(page) != 1;
6944 __free_page(page);
6945 }
6946 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006947}
6948#endif
6949
Jiang Liu4ed7e022012-07-31 16:43:35 -07006950#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07006951/*
6952 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6953 * page high values need to be recalulated.
6954 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07006955void __meminit zone_pcp_update(struct zone *zone)
6956{
Cody P Schafer0a647f32013-07-03 15:01:33 -07006957 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006958 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07006959 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07006960 pageset_set_high_and_batch(zone,
6961 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006962 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07006963}
6964#endif
6965
Jiang Liu340175b2012-07-31 16:43:32 -07006966void zone_pcp_reset(struct zone *zone)
6967{
6968 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07006969 int cpu;
6970 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07006971
6972 /* avoid races with drain_pages() */
6973 local_irq_save(flags);
6974 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07006975 for_each_online_cpu(cpu) {
6976 pset = per_cpu_ptr(zone->pageset, cpu);
6977 drain_zonestat(zone, pset);
6978 }
Jiang Liu340175b2012-07-31 16:43:32 -07006979 free_percpu(zone->pageset);
6980 zone->pageset = &boot_pageset;
6981 }
6982 local_irq_restore(flags);
6983}
6984
Wen Congyang6dcd73d2012-12-11 16:01:01 -08006985#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006986/*
6987 * All pages in the range must be isolated before calling this.
6988 */
6989void
6990__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6991{
6992 struct page *page;
6993 struct zone *zone;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006994 unsigned int order, i;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006995 unsigned long pfn;
6996 unsigned long flags;
6997 /* find the first valid pfn */
6998 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6999 if (pfn_valid(pfn))
7000 break;
7001 if (pfn == end_pfn)
7002 return;
7003 zone = page_zone(pfn_to_page(pfn));
7004 spin_lock_irqsave(&zone->lock, flags);
7005 pfn = start_pfn;
7006 while (pfn < end_pfn) {
7007 if (!pfn_valid(pfn)) {
7008 pfn++;
7009 continue;
7010 }
7011 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08007012 /*
7013 * The HWPoisoned page may be not in buddy system, and
7014 * page_count() is not 0.
7015 */
7016 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
7017 pfn++;
7018 SetPageReserved(page);
7019 continue;
7020 }
7021
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007022 BUG_ON(page_count(page));
7023 BUG_ON(!PageBuddy(page));
7024 order = page_order(page);
7025#ifdef CONFIG_DEBUG_VM
7026 printk(KERN_INFO "remove from free list %lx %d %lx\n",
7027 pfn, 1 << order, end_pfn);
7028#endif
7029 list_del(&page->lru);
7030 rmv_page_order(page);
7031 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007032 for (i = 0; i < (1 << order); i++)
7033 SetPageReserved((page+i));
7034 pfn += (1 << order);
7035 }
7036 spin_unlock_irqrestore(&zone->lock, flags);
7037}
7038#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007039
7040#ifdef CONFIG_MEMORY_FAILURE
7041bool is_free_buddy_page(struct page *page)
7042{
7043 struct zone *zone = page_zone(page);
7044 unsigned long pfn = page_to_pfn(page);
7045 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07007046 unsigned int order;
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007047
7048 spin_lock_irqsave(&zone->lock, flags);
7049 for (order = 0; order < MAX_ORDER; order++) {
7050 struct page *page_head = page - (pfn & ((1 << order) - 1));
7051
7052 if (PageBuddy(page_head) && page_order(page_head) >= order)
7053 break;
7054 }
7055 spin_unlock_irqrestore(&zone->lock, flags);
7056
7057 return order < MAX_ORDER;
7058}
7059#endif