blob: 5760c626c3092e9e7966582aef1c68c45f4da2d2 [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>
Dan Williams4b94ffd2016-01-15 16:56:22 -080046#include <linux/memremap.h>
Yasunori Goto68113782006-06-23 02:03:11 -070047#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070048#include <linux/sort.h>
49#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070050#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080051#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070052#include <linux/page-isolation.h>
Joonsoo Kimeefa8642014-12-12 16:55:46 -080053#include <linux/page_ext.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070054#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010055#include <linux/kmemleak.h>
Mel Gorman56de7262010-05-24 14:32:30 -070056#include <linux/compaction.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070057#include <trace/events/kmem.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070058#include <linux/prefetch.h>
Lisa Du6e543d52013-09-11 14:22:36 -070059#include <linux/mm_inline.h>
Michal Nazarewicz041d3a82011-12-29 13:09:50 +010060#include <linux/migrate.h>
Joonsoo Kime30825f2014-12-12 16:55:49 -080061#include <linux/page_ext.h>
David Rientjes949f7ec2013-04-29 15:07:48 -070062#include <linux/hugetlb.h>
Clark Williams8bd75c72013-02-07 09:47:07 -060063#include <linux/sched/rt.h>
Joonsoo Kim48c96a32014-12-12 16:56:01 -080064#include <linux/page_owner.h>
Mel Gorman0e1cc952015-06-30 14:57:27 -070065#include <linux/kthread.h>
Vladimir Davydov49491482016-07-26 15:24:24 -070066#include <linux/memcontrol.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
Jiang Liu7ee3d4e2013-07-03 15:03:41 -070068#include <asm/sections.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070069#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070070#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#include "internal.h"
72
Cody P Schaferc8e251f2013-07-03 15:01:29 -070073/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
74static DEFINE_MUTEX(pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -070075#define MIN_PERCPU_PAGELIST_FRACTION (8)
Cody P Schaferc8e251f2013-07-03 15:01:29 -070076
Lee Schermerhorn72812012010-05-26 14:44:56 -070077#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
78DEFINE_PER_CPU(int, numa_node);
79EXPORT_PER_CPU_SYMBOL(numa_node);
80#endif
81
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070082#ifdef CONFIG_HAVE_MEMORYLESS_NODES
83/*
84 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
85 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
86 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
87 * defined in <linux/topology.h>.
88 */
89DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
90EXPORT_PER_CPU_SYMBOL(_numa_mem_);
Joonsoo Kimad2c8142014-10-09 15:26:13 -070091int _node_numa_mem_[MAX_NUMNODES];
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070092#endif
93
Linus Torvalds1da177e2005-04-16 15:20:36 -070094/*
Christoph Lameter13808912007-10-16 01:25:27 -070095 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 */
Christoph Lameter13808912007-10-16 01:25:27 -070097nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
98 [N_POSSIBLE] = NODE_MASK_ALL,
99 [N_ONLINE] = { { [0] = 1UL } },
100#ifndef CONFIG_NUMA
101 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
102#ifdef CONFIG_HIGHMEM
103 [N_HIGH_MEMORY] = { { [0] = 1UL } },
104#endif
Lai Jiangshan20b2f522012-12-12 13:52:00 -0800105#ifdef CONFIG_MOVABLE_NODE
106 [N_MEMORY] = { { [0] = 1UL } },
107#endif
Christoph Lameter13808912007-10-16 01:25:27 -0700108 [N_CPU] = { { [0] = 1UL } },
109#endif /* NUMA */
110};
111EXPORT_SYMBOL(node_states);
112
Jiang Liuc3d5f5f2013-07-03 15:03:14 -0700113/* Protect totalram_pages and zone->managed_pages */
114static DEFINE_SPINLOCK(managed_page_count_lock);
115
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700116unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -0700117unsigned long totalreserve_pages __read_mostly;
Pintu Kumare48322a2014-12-18 16:17:15 -0800118unsigned long totalcma_pages __read_mostly;
Johannes Weinerab8fabd2012-01-10 15:07:42 -0800119
Hugh Dickins1b76b022012-05-11 01:00:07 -0700120int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000121gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -0700123/*
124 * A cached value of the page's pageblock's migratetype, used when the page is
125 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
126 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
127 * Also the migratetype set in the page does not necessarily match the pcplist
128 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
129 * other index - this ensures that it will be put on the correct CMA freelist.
130 */
131static inline int get_pcppage_migratetype(struct page *page)
132{
133 return page->index;
134}
135
136static inline void set_pcppage_migratetype(struct page *page, int migratetype)
137{
138 page->index = migratetype;
139}
140
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800141#ifdef CONFIG_PM_SLEEP
142/*
143 * The following functions are used by the suspend/hibernate code to temporarily
144 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
145 * while devices are suspended. To avoid races with the suspend/hibernate code,
146 * they should always be called with pm_mutex held (gfp_allowed_mask also should
147 * only be modified with pm_mutex held, unless the suspend/hibernate code is
148 * guaranteed not to run in parallel with that modification).
149 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100150
151static gfp_t saved_gfp_mask;
152
153void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800154{
155 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100156 if (saved_gfp_mask) {
157 gfp_allowed_mask = saved_gfp_mask;
158 saved_gfp_mask = 0;
159 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800160}
161
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100162void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800163{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800164 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100165 WARN_ON(saved_gfp_mask);
166 saved_gfp_mask = gfp_allowed_mask;
Mel Gormand0164ad2015-11-06 16:28:21 -0800167 gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800168}
Mel Gormanf90ac392012-01-10 15:07:15 -0800169
170bool pm_suspended_storage(void)
171{
Mel Gormand0164ad2015-11-06 16:28:21 -0800172 if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
Mel Gormanf90ac392012-01-10 15:07:15 -0800173 return false;
174 return true;
175}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800176#endif /* CONFIG_PM_SLEEP */
177
Mel Gormand9c23402007-10-16 01:26:01 -0700178#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -0800179unsigned int pageblock_order __read_mostly;
Mel Gormand9c23402007-10-16 01:26:01 -0700180#endif
181
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800182static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800183
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184/*
185 * results with 256, 32 in the lowmem_reserve sysctl:
186 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
187 * 1G machine -> (16M dma, 784M normal, 224M high)
188 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
189 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
Yaowei Bai84109e12015-02-12 15:00:22 -0800190 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100191 *
192 * TBD: should special case ZONE_DMA32 machines here - in those we normally
193 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700195int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800196#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700197 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800198#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700199#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700200 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700201#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700202#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700203 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700204#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700205 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700206};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
208EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
Helge Deller15ad7cd2006-12-06 20:40:36 -0800210static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800211#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700212 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800213#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700214#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700215 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700216#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700217 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700218#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700219 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700220#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700221 "Movable",
Dan Williams033fbae2015-08-09 15:29:06 -0400222#ifdef CONFIG_ZONE_DEVICE
223 "Device",
224#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700225};
226
Vlastimil Babka60f30352016-03-15 14:56:08 -0700227char * const migratetype_names[MIGRATE_TYPES] = {
228 "Unmovable",
229 "Movable",
230 "Reclaimable",
231 "HighAtomic",
232#ifdef CONFIG_CMA
233 "CMA",
234#endif
235#ifdef CONFIG_MEMORY_ISOLATION
236 "Isolate",
237#endif
238};
239
Kirill A. Shutemovf1e61552015-11-06 16:29:50 -0800240compound_page_dtor * const compound_page_dtors[] = {
241 NULL,
242 free_compound_page,
243#ifdef CONFIG_HUGETLB_PAGE
244 free_huge_page,
245#endif
Kirill A. Shutemov9a982252016-01-15 16:54:17 -0800246#ifdef CONFIG_TRANSPARENT_HUGEPAGE
247 free_transhuge_page,
248#endif
Kirill A. Shutemovf1e61552015-11-06 16:29:50 -0800249};
250
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251int min_free_kbytes = 1024;
Han Pingtian42aa83c2014-01-23 15:53:28 -0800252int user_min_free_kbytes = -1;
Johannes Weiner795ae7a2016-03-17 14:19:14 -0700253int watermark_scale_factor = 10;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
Jan Beulich2c85f512009-09-21 17:03:07 -0700255static unsigned long __meminitdata nr_kernel_pages;
256static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700257static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
Tejun Heo0ee332c2011-12-08 10:22:09 -0800259#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
260static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
261static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
262static unsigned long __initdata required_kernelcore;
263static unsigned long __initdata required_movablecore;
264static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Taku Izumi342332e2016-03-15 14:55:22 -0700265static bool mirrored_kernelcore;
Mel Gormanc7132162006-09-27 01:49:43 -0700266
Tejun Heo0ee332c2011-12-08 10:22:09 -0800267/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
268int movable_zone;
269EXPORT_SYMBOL(movable_zone);
270#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700271
Miklos Szeredi418508c2007-05-23 13:57:55 -0700272#if MAX_NUMNODES > 1
273int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700274int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700275EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700276EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700277#endif
278
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700279int page_group_by_mobility_disabled __read_mostly;
280
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700281#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
282static inline void reset_deferred_meminit(pg_data_t *pgdat)
283{
284 pgdat->first_deferred_pfn = ULONG_MAX;
285}
286
287/* Returns true if the struct page for the pfn is uninitialised */
Mel Gorman0e1cc952015-06-30 14:57:27 -0700288static inline bool __meminit early_page_uninitialised(unsigned long pfn)
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700289{
Mel Gormanef70b6f2016-07-14 12:07:23 -0700290 int nid = early_pfn_to_nid(pfn);
291
292 if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700293 return true;
294
295 return false;
296}
297
298/*
299 * Returns false when the remaining initialisation should be deferred until
300 * later in the boot cycle when it can be parallelised.
301 */
302static inline bool update_defer_init(pg_data_t *pgdat,
303 unsigned long pfn, unsigned long zone_end,
304 unsigned long *nr_initialised)
305{
Li Zhang987b3092016-03-17 14:20:16 -0700306 unsigned long max_initialise;
307
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700308 /* Always populate low zones for address-contrained allocations */
309 if (zone_end < pgdat_end_pfn(pgdat))
310 return true;
Li Zhang987b3092016-03-17 14:20:16 -0700311 /*
312 * Initialise at least 2G of a node but also take into account that
313 * two large system hashes that can take up 1GB for 0.25TB/node.
314 */
315 max_initialise = max(2UL << (30 - PAGE_SHIFT),
316 (pgdat->node_spanned_pages >> 8));
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700317
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700318 (*nr_initialised)++;
Li Zhang987b3092016-03-17 14:20:16 -0700319 if ((*nr_initialised > max_initialise) &&
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700320 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
321 pgdat->first_deferred_pfn = pfn;
322 return false;
323 }
324
325 return true;
326}
327#else
328static inline void reset_deferred_meminit(pg_data_t *pgdat)
329{
330}
331
332static inline bool early_page_uninitialised(unsigned long pfn)
333{
334 return false;
335}
336
337static inline bool update_defer_init(pg_data_t *pgdat,
338 unsigned long pfn, unsigned long zone_end,
339 unsigned long *nr_initialised)
340{
341 return true;
342}
343#endif
344
Mel Gorman0b423ca2016-05-19 17:14:27 -0700345/* Return a pointer to the bitmap storing bits affecting a block of pages */
346static inline unsigned long *get_pageblock_bitmap(struct page *page,
347 unsigned long pfn)
348{
349#ifdef CONFIG_SPARSEMEM
350 return __pfn_to_section(pfn)->pageblock_flags;
351#else
352 return page_zone(page)->pageblock_flags;
353#endif /* CONFIG_SPARSEMEM */
354}
355
356static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
357{
358#ifdef CONFIG_SPARSEMEM
359 pfn &= (PAGES_PER_SECTION-1);
360 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
361#else
362 pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
363 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
364#endif /* CONFIG_SPARSEMEM */
365}
366
367/**
368 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
369 * @page: The page within the block of interest
370 * @pfn: The target page frame number
371 * @end_bitidx: The last bit of interest to retrieve
372 * @mask: mask of bits that the caller is interested in
373 *
374 * Return: pageblock_bits flags
375 */
376static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page,
377 unsigned long pfn,
378 unsigned long end_bitidx,
379 unsigned long mask)
380{
381 unsigned long *bitmap;
382 unsigned long bitidx, word_bitidx;
383 unsigned long word;
384
385 bitmap = get_pageblock_bitmap(page, pfn);
386 bitidx = pfn_to_bitidx(page, pfn);
387 word_bitidx = bitidx / BITS_PER_LONG;
388 bitidx &= (BITS_PER_LONG-1);
389
390 word = bitmap[word_bitidx];
391 bitidx += end_bitidx;
392 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
393}
394
395unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
396 unsigned long end_bitidx,
397 unsigned long mask)
398{
399 return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask);
400}
401
402static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
403{
404 return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK);
405}
406
407/**
408 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
409 * @page: The page within the block of interest
410 * @flags: The flags to set
411 * @pfn: The target page frame number
412 * @end_bitidx: The last bit of interest
413 * @mask: mask of bits that the caller is interested in
414 */
415void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
416 unsigned long pfn,
417 unsigned long end_bitidx,
418 unsigned long mask)
419{
420 unsigned long *bitmap;
421 unsigned long bitidx, word_bitidx;
422 unsigned long old_word, word;
423
424 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
425
426 bitmap = get_pageblock_bitmap(page, pfn);
427 bitidx = pfn_to_bitidx(page, pfn);
428 word_bitidx = bitidx / BITS_PER_LONG;
429 bitidx &= (BITS_PER_LONG-1);
430
431 VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
432
433 bitidx += end_bitidx;
434 mask <<= (BITS_PER_LONG - bitidx - 1);
435 flags <<= (BITS_PER_LONG - bitidx - 1);
436
437 word = READ_ONCE(bitmap[word_bitidx]);
438 for (;;) {
439 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
440 if (word == old_word)
441 break;
442 word = old_word;
443 }
444}
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700445
Minchan Kimee6f5092012-07-31 16:43:50 -0700446void set_pageblock_migratetype(struct page *page, int migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700447{
KOSAKI Motohiro5d0f3f72013-11-12 15:08:18 -0800448 if (unlikely(page_group_by_mobility_disabled &&
449 migratetype < MIGRATE_PCPTYPES))
Mel Gorman49255c62009-06-16 15:31:58 -0700450 migratetype = MIGRATE_UNMOVABLE;
451
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700452 set_pageblock_flags_group(page, (unsigned long)migratetype,
453 PB_migrate, PB_migrate_end);
454}
455
Nick Piggin13e74442006-01-06 00:10:58 -0800456#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700457static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700459 int ret = 0;
460 unsigned seq;
461 unsigned long pfn = page_to_pfn(page);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800462 unsigned long sp, start_pfn;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700463
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700464 do {
465 seq = zone_span_seqbegin(zone);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800466 start_pfn = zone->zone_start_pfn;
467 sp = zone->spanned_pages;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800468 if (!zone_spans_pfn(zone, pfn))
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700469 ret = 1;
470 } while (zone_span_seqretry(zone, seq));
471
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800472 if (ret)
Dave Hansen613813e2014-06-04 16:07:27 -0700473 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
474 pfn, zone_to_nid(zone), zone->name,
475 start_pfn, start_pfn + sp);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800476
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700477 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700478}
479
480static int page_is_consistent(struct zone *zone, struct page *page)
481{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700482 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700483 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700485 return 0;
486
487 return 1;
488}
489/*
490 * Temporary debugging check for pages not lying within a given zone.
491 */
492static int bad_range(struct zone *zone, struct page *page)
493{
494 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700496 if (!page_is_consistent(zone, page))
497 return 1;
498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 return 0;
500}
Nick Piggin13e74442006-01-06 00:10:58 -0800501#else
502static inline int bad_range(struct zone *zone, struct page *page)
503{
504 return 0;
505}
506#endif
507
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700508static void bad_page(struct page *page, const char *reason,
509 unsigned long bad_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800511 static unsigned long resume;
512 static unsigned long nr_shown;
513 static unsigned long nr_unshown;
514
515 /*
516 * Allow a burst of 60 reports, then keep quiet for that minute;
517 * or allow a steady drip of one report per second.
518 */
519 if (nr_shown == 60) {
520 if (time_before(jiffies, resume)) {
521 nr_unshown++;
522 goto out;
523 }
524 if (nr_unshown) {
Vlastimil Babkaff8e8112016-03-15 14:56:24 -0700525 pr_alert(
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800526 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800527 nr_unshown);
528 nr_unshown = 0;
529 }
530 nr_shown = 0;
531 }
532 if (nr_shown++ == 0)
533 resume = jiffies + 60 * HZ;
534
Vlastimil Babkaff8e8112016-03-15 14:56:24 -0700535 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800536 current->comm, page_to_pfn(page));
Vlastimil Babkaff8e8112016-03-15 14:56:24 -0700537 __dump_page(page, reason);
538 bad_flags &= page->flags;
539 if (bad_flags)
540 pr_alert("bad because of flags: %#lx(%pGp)\n",
541 bad_flags, &bad_flags);
Vlastimil Babka4e462112016-03-15 14:56:21 -0700542 dump_page_owner(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700543
Dave Jones4f318882011-10-31 17:07:24 -0700544 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800546out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800547 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800548 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030549 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550}
551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552/*
553 * Higher-order pages are called "compound pages". They are structured thusly:
554 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800555 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800557 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
558 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800560 * The first tail page's ->compound_dtor holds the offset in array of compound
561 * page destructors. See compound_page_dtors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800563 * The first tail page's ->compound_order holds the order of allocation.
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800564 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800566
Kirill A. Shutemov9a982252016-01-15 16:54:17 -0800567void free_compound_page(struct page *page)
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800568{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700569 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800570}
571
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -0800572void prep_compound_page(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573{
574 int i;
575 int nr_pages = 1 << order;
576
Kirill A. Shutemovf1e61552015-11-06 16:29:50 -0800577 set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700578 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700579 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800580 for (i = 1; i < nr_pages; i++) {
581 struct page *p = page + i;
Youquan Song58a84aa2011-12-08 14:34:18 -0800582 set_page_count(p, 0);
Kirill A. Shutemov1c290f62016-01-15 16:52:07 -0800583 p->mapping = TAIL_MAPPING;
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800584 set_compound_head(p, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 }
Kirill A. Shutemov53f92632016-01-15 16:53:42 -0800586 atomic_set(compound_mapcount_ptr(page), -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587}
588
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800589#ifdef CONFIG_DEBUG_PAGEALLOC
590unsigned int _debug_guardpage_minorder;
Christian Borntraegerea6eabb2016-03-15 14:55:30 -0700591bool _debug_pagealloc_enabled __read_mostly
592 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
Joonsoo Kim505f6d22016-03-17 14:17:56 -0700593EXPORT_SYMBOL(_debug_pagealloc_enabled);
Joonsoo Kime30825f2014-12-12 16:55:49 -0800594bool _debug_guardpage_enabled __read_mostly;
595
Joonsoo Kim031bc572014-12-12 16:55:52 -0800596static int __init early_debug_pagealloc(char *buf)
597{
598 if (!buf)
599 return -EINVAL;
Minfei Huang2a138dc2016-05-20 16:58:13 -0700600 return kstrtobool(buf, &_debug_pagealloc_enabled);
Joonsoo Kim031bc572014-12-12 16:55:52 -0800601}
602early_param("debug_pagealloc", early_debug_pagealloc);
603
Joonsoo Kime30825f2014-12-12 16:55:49 -0800604static bool need_debug_guardpage(void)
605{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800606 /* If we don't use debug_pagealloc, we don't need guard page */
607 if (!debug_pagealloc_enabled())
608 return false;
609
Joonsoo Kime30825f2014-12-12 16:55:49 -0800610 return true;
611}
612
613static void init_debug_guardpage(void)
614{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800615 if (!debug_pagealloc_enabled())
616 return;
617
Joonsoo Kime30825f2014-12-12 16:55:49 -0800618 _debug_guardpage_enabled = true;
619}
620
621struct page_ext_operations debug_guardpage_ops = {
622 .need = need_debug_guardpage,
623 .init = init_debug_guardpage,
624};
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800625
626static int __init debug_guardpage_minorder_setup(char *buf)
627{
628 unsigned long res;
629
630 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
Joe Perches11705322016-03-17 14:19:50 -0700631 pr_err("Bad debug_guardpage_minorder value\n");
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800632 return 0;
633 }
634 _debug_guardpage_minorder = res;
Joe Perches11705322016-03-17 14:19:50 -0700635 pr_info("Setting debug_guardpage_minorder to %lu\n", res);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800636 return 0;
637}
638__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
639
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800640static inline void set_page_guard(struct zone *zone, struct page *page,
641 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800642{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800643 struct page_ext *page_ext;
644
645 if (!debug_guardpage_enabled())
646 return;
647
648 page_ext = lookup_page_ext(page);
Yang Shif86e4272016-06-03 14:55:38 -0700649 if (unlikely(!page_ext))
650 return;
651
Joonsoo Kime30825f2014-12-12 16:55:49 -0800652 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
653
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800654 INIT_LIST_HEAD(&page->lru);
655 set_page_private(page, order);
656 /* Guard pages are not available for any usage */
657 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800658}
659
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800660static inline void clear_page_guard(struct zone *zone, struct page *page,
661 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800662{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800663 struct page_ext *page_ext;
664
665 if (!debug_guardpage_enabled())
666 return;
667
668 page_ext = lookup_page_ext(page);
Yang Shif86e4272016-06-03 14:55:38 -0700669 if (unlikely(!page_ext))
670 return;
671
Joonsoo Kime30825f2014-12-12 16:55:49 -0800672 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
673
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800674 set_page_private(page, 0);
675 if (!is_migrate_isolate(migratetype))
676 __mod_zone_freepage_state(zone, (1 << order), migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800677}
678#else
Joonsoo Kime30825f2014-12-12 16:55:49 -0800679struct page_ext_operations debug_guardpage_ops = { NULL, };
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800680static inline void set_page_guard(struct zone *zone, struct page *page,
681 unsigned int order, int migratetype) {}
682static inline void clear_page_guard(struct zone *zone, struct page *page,
683 unsigned int order, int migratetype) {}
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800684#endif
685
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700686static inline void set_page_order(struct page *page, unsigned int order)
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700687{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700688 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000689 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690}
691
692static inline void rmv_page_order(struct page *page)
693{
Nick Piggin676165a2006-04-10 11:21:48 +1000694 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700695 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696}
697
698/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 * This function checks whether a page is free && is the buddy
700 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800701 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000702 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700703 * (c) a page and its buddy have the same order &&
704 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 *
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700706 * For recording whether a page is in the buddy system, we set ->_mapcount
707 * PAGE_BUDDY_MAPCOUNT_VALUE.
708 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
709 * serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000710 *
711 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700713static inline int page_is_buddy(struct page *page, struct page *buddy,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700714 unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700716 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800717 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800718
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800719 if (page_is_guard(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700720 if (page_zone_id(page) != page_zone_id(buddy))
721 return 0;
722
Weijie Yang4c5018c2015-02-10 14:11:39 -0800723 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
724
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800725 return 1;
726 }
727
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700728 if (PageBuddy(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700729 /*
730 * zone check is done late to avoid uselessly
731 * calculating zone/node ids for pages that could
732 * never merge.
733 */
734 if (page_zone_id(page) != page_zone_id(buddy))
735 return 0;
736
Weijie Yang4c5018c2015-02-10 14:11:39 -0800737 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
738
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700739 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000740 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700741 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742}
743
744/*
745 * Freeing function for a buddy system allocator.
746 *
747 * The concept of a buddy system is to maintain direct-mapped table
748 * (containing bit values) for memory blocks of various "orders".
749 * The bottom level table contains the map for the smallest allocatable
750 * units of memory (here, pages), and each level above it describes
751 * pairs of units from the levels below, hence, "buddies".
752 * At a high level, all that happens here is marking the table entry
753 * at the bottom level available, and propagating the changes upward
754 * as necessary, plus some accounting needed to play nicely with other
755 * parts of the VM system.
756 * At each level, we keep a list of pages, which are heads of continuous
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700757 * free pages of length of (1 << order) and marked with _mapcount
758 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
759 * field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100761 * other. That is, if we allocate a small block, and both were
762 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100764 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100766 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 */
768
Nick Piggin48db57f2006-01-08 01:00:42 -0800769static inline void __free_one_page(struct page *page,
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700770 unsigned long pfn,
Mel Gormaned0ae212009-06-16 15:32:07 -0700771 struct zone *zone, unsigned int order,
772 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773{
774 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700775 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800776 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700777 struct page *buddy;
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700778 unsigned int max_order;
779
780 max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
Cody P Schaferd29bb972013-02-22 16:35:25 -0800782 VM_BUG_ON(!zone_is_initialized(zone));
Kirill A. Shutemov6e9f0d52015-02-11 15:25:50 -0800783 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
Mel Gormaned0ae212009-06-16 15:32:07 -0700785 VM_BUG_ON(migratetype == -1);
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700786 if (likely(!is_migrate_isolate(migratetype)))
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800787 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Mel Gormaned0ae212009-06-16 15:32:07 -0700788
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700789 page_idx = pfn & ((1 << MAX_ORDER) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
Sasha Levin309381fea2014-01-23 15:52:54 -0800791 VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
792 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700794continue_merging:
Joonsoo Kim3c605092014-11-13 15:19:21 -0800795 while (order < max_order - 1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800796 buddy_idx = __find_buddy_index(page_idx, order);
797 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700798 if (!page_is_buddy(page, buddy, order))
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700799 goto done_merging;
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800800 /*
801 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
802 * merge with it and move up one order.
803 */
804 if (page_is_guard(buddy)) {
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800805 clear_page_guard(zone, buddy, order, migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800806 } else {
807 list_del(&buddy->lru);
808 zone->free_area[order].nr_free--;
809 rmv_page_order(buddy);
810 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800811 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 page = page + (combined_idx - page_idx);
813 page_idx = combined_idx;
814 order++;
815 }
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700816 if (max_order < MAX_ORDER) {
817 /* If we are here, it means order is >= pageblock_order.
818 * We want to prevent merge between freepages on isolate
819 * pageblock and normal pageblock. Without this, pageblock
820 * isolation could cause incorrect freepage or CMA accounting.
821 *
822 * We don't want to hit this code for the more frequent
823 * low-order merging.
824 */
825 if (unlikely(has_isolate_pageblock(zone))) {
826 int buddy_mt;
827
828 buddy_idx = __find_buddy_index(page_idx, order);
829 buddy = page + (buddy_idx - page_idx);
830 buddy_mt = get_pageblock_migratetype(buddy);
831
832 if (migratetype != buddy_mt
833 && (is_migrate_isolate(migratetype) ||
834 is_migrate_isolate(buddy_mt)))
835 goto done_merging;
836 }
837 max_order++;
838 goto continue_merging;
839 }
840
841done_merging:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700843
844 /*
845 * If this is not the largest possible page, check if the buddy
846 * of the next-highest order is free. If it is, it's possible
847 * that pages are being freed that will coalesce soon. In case,
848 * that is happening, add the free page to the tail of the list
849 * so it's less likely to be used soon and more likely to be merged
850 * as a higher order page
851 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700852 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700853 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800854 combined_idx = buddy_idx & page_idx;
855 higher_page = page + (combined_idx - page_idx);
856 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700857 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700858 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
859 list_add_tail(&page->lru,
860 &zone->free_area[order].free_list[migratetype]);
861 goto out;
862 }
863 }
864
865 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
866out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 zone->free_area[order].nr_free++;
868}
869
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700870/*
871 * A bad page could be due to a number of fields. Instead of multiple branches,
872 * try and check multiple fields with one check. The caller must do a detailed
873 * check if necessary.
874 */
875static inline bool page_expected_state(struct page *page,
876 unsigned long check_flags)
877{
878 if (unlikely(atomic_read(&page->_mapcount) != -1))
879 return false;
880
881 if (unlikely((unsigned long)page->mapping |
882 page_ref_count(page) |
883#ifdef CONFIG_MEMCG
884 (unsigned long)page->mem_cgroup |
885#endif
886 (page->flags & check_flags)))
887 return false;
888
889 return true;
890}
891
Mel Gormanbb552ac2016-05-19 17:14:18 -0700892static void free_pages_check_bad(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893{
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700894 const char *bad_reason;
895 unsigned long bad_flags;
896
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700897 bad_reason = NULL;
898 bad_flags = 0;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800899
Kirill A. Shutemov53f92632016-01-15 16:53:42 -0800900 if (unlikely(atomic_read(&page->_mapcount) != -1))
Dave Hansenf0b791a2014-01-23 15:52:49 -0800901 bad_reason = "nonzero mapcount";
902 if (unlikely(page->mapping != NULL))
903 bad_reason = "non-NULL mapping";
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700904 if (unlikely(page_ref_count(page) != 0))
Joonsoo Kim0139aa72016-05-19 17:10:49 -0700905 bad_reason = "nonzero _refcount";
Dave Hansenf0b791a2014-01-23 15:52:49 -0800906 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
907 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
908 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
909 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -0800910#ifdef CONFIG_MEMCG
911 if (unlikely(page->mem_cgroup))
912 bad_reason = "page still charged to cgroup";
913#endif
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700914 bad_page(page, bad_reason, bad_flags);
Mel Gormanbb552ac2016-05-19 17:14:18 -0700915}
916
917static inline int free_pages_check(struct page *page)
918{
Mel Gormanda838d42016-05-19 17:14:21 -0700919 if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
Mel Gormanbb552ac2016-05-19 17:14:18 -0700920 return 0;
Mel Gormanbb552ac2016-05-19 17:14:18 -0700921
922 /* Something has gone sideways, find it */
923 free_pages_check_bad(page);
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700924 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925}
926
Mel Gorman4db75482016-05-19 17:14:32 -0700927static int free_tail_pages_check(struct page *head_page, struct page *page)
928{
929 int ret = 1;
930
931 /*
932 * We rely page->lru.next never has bit 0 set, unless the page
933 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
934 */
935 BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
936
937 if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
938 ret = 0;
939 goto out;
940 }
941 switch (page - head_page) {
942 case 1:
943 /* the first tail page: ->mapping is compound_mapcount() */
944 if (unlikely(compound_mapcount(page))) {
945 bad_page(page, "nonzero compound_mapcount", 0);
946 goto out;
947 }
948 break;
949 case 2:
950 /*
951 * the second tail page: ->mapping is
952 * page_deferred_list().next -- ignore value.
953 */
954 break;
955 default:
956 if (page->mapping != TAIL_MAPPING) {
957 bad_page(page, "corrupted mapping in tail page", 0);
958 goto out;
959 }
960 break;
961 }
962 if (unlikely(!PageTail(page))) {
963 bad_page(page, "PageTail not set", 0);
964 goto out;
965 }
966 if (unlikely(compound_head(page) != head_page)) {
967 bad_page(page, "compound_head not consistent", 0);
968 goto out;
969 }
970 ret = 0;
971out:
972 page->mapping = NULL;
973 clear_compound_head(page);
974 return ret;
975}
976
Mel Gormane2769db2016-05-19 17:14:38 -0700977static __always_inline bool free_pages_prepare(struct page *page,
978 unsigned int order, bool check_free)
979{
980 int bad = 0;
981
982 VM_BUG_ON_PAGE(PageTail(page), page);
983
984 trace_mm_page_free(page, order);
985 kmemcheck_free_shadow(page, order);
Mel Gormane2769db2016-05-19 17:14:38 -0700986
987 /*
988 * Check tail pages before head page information is cleared to
989 * avoid checking PageCompound for order-0 pages.
990 */
991 if (unlikely(order)) {
992 bool compound = PageCompound(page);
993 int i;
994
995 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
996
Kirill A. Shutemov9a73f612016-07-26 15:25:53 -0700997 if (compound)
998 ClearPageDoubleMap(page);
Mel Gormane2769db2016-05-19 17:14:38 -0700999 for (i = 1; i < (1 << order); i++) {
1000 if (compound)
1001 bad += free_tail_pages_check(page, page + i);
1002 if (unlikely(free_pages_check(page + i))) {
1003 bad++;
1004 continue;
1005 }
1006 (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1007 }
1008 }
Minchan Kimbda807d2016-07-26 15:23:05 -07001009 if (PageMappingFlags(page))
Mel Gormane2769db2016-05-19 17:14:38 -07001010 page->mapping = NULL;
Vladimir Davydov49491482016-07-26 15:24:24 -07001011 if (memcg_kmem_enabled() && PageKmemcg(page)) {
1012 memcg_kmem_uncharge(page, order);
1013 __ClearPageKmemcg(page);
1014 }
Mel Gormane2769db2016-05-19 17:14:38 -07001015 if (check_free)
1016 bad += free_pages_check(page);
1017 if (bad)
1018 return false;
1019
1020 page_cpupid_reset_last(page);
1021 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1022 reset_page_owner(page, order);
1023
1024 if (!PageHighMem(page)) {
1025 debug_check_no_locks_freed(page_address(page),
1026 PAGE_SIZE << order);
1027 debug_check_no_obj_freed(page_address(page),
1028 PAGE_SIZE << order);
1029 }
1030 arch_free_page(page, order);
1031 kernel_poison_pages(page, 1 << order, 0);
1032 kernel_map_pages(page, 1 << order, 0);
seokhoon.yoon29b52de2016-05-20 16:58:47 -07001033 kasan_free_pages(page, order);
Mel Gormane2769db2016-05-19 17:14:38 -07001034
1035 return true;
1036}
Mel Gorman4db75482016-05-19 17:14:32 -07001037
1038#ifdef CONFIG_DEBUG_VM
1039static inline bool free_pcp_prepare(struct page *page)
1040{
Mel Gormane2769db2016-05-19 17:14:38 -07001041 return free_pages_prepare(page, 0, true);
Mel Gorman4db75482016-05-19 17:14:32 -07001042}
1043
1044static inline bool bulkfree_pcp_prepare(struct page *page)
1045{
1046 return false;
1047}
1048#else
1049static bool free_pcp_prepare(struct page *page)
1050{
Mel Gormane2769db2016-05-19 17:14:38 -07001051 return free_pages_prepare(page, 0, false);
Mel Gorman4db75482016-05-19 17:14:32 -07001052}
1053
1054static bool bulkfree_pcp_prepare(struct page *page)
1055{
1056 return free_pages_check(page);
1057}
1058#endif /* CONFIG_DEBUG_VM */
1059
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001061 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -07001063 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 *
1065 * If the zone was previously in an "all pages pinned" state then look to
1066 * see if this freeing clears that state.
1067 *
1068 * And clear the zone's pages_scanned counter, to hold off the "all pages are
1069 * pinned" detection logic.
1070 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001071static void free_pcppages_bulk(struct zone *zone, int count,
1072 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001074 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -07001075 int batch_free = 0;
Mel Gorman0d5d8232014-08-06 16:07:16 -07001076 unsigned long nr_scanned;
Mel Gorman37779992016-05-19 17:13:58 -07001077 bool isolated_pageblocks;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001078
Nick Pigginc54ad302006-01-06 00:10:56 -08001079 spin_lock(&zone->lock);
Mel Gorman37779992016-05-19 17:13:58 -07001080 isolated_pageblocks = has_isolate_pageblock(zone);
Mel Gorman0d5d8232014-08-06 16:07:16 -07001081 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
1082 if (nr_scanned)
1083 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -07001084
Mel Gormane5b31ac2016-05-19 17:14:24 -07001085 while (count) {
Nick Piggin48db57f2006-01-08 01:00:42 -08001086 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001087 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -08001088
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001089 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -07001090 * Remove pages from lists in a round-robin fashion. A
1091 * batch_free count is maintained that is incremented when an
1092 * empty list is encountered. This is so more pages are freed
1093 * off fuller lists instead of spinning excessively around empty
1094 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001095 */
1096 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -07001097 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001098 if (++migratetype == MIGRATE_PCPTYPES)
1099 migratetype = 0;
1100 list = &pcp->lists[migratetype];
1101 } while (list_empty(list));
1102
Namhyung Kim1d168712011-03-22 16:32:45 -07001103 /* This is the only non-empty list. Free them all. */
1104 if (batch_free == MIGRATE_PCPTYPES)
Mel Gormane5b31ac2016-05-19 17:14:24 -07001105 batch_free = count;
Namhyung Kim1d168712011-03-22 16:32:45 -07001106
Mel Gormana6f9edd2009-09-21 17:03:20 -07001107 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -07001108 int mt; /* migratetype of the to-be-freed page */
1109
Geliang Tanga16601c2016-01-14 15:20:30 -08001110 page = list_last_entry(list, struct page, lru);
Mel Gormana6f9edd2009-09-21 17:03:20 -07001111 /* must delete as __free_one_page list manipulates */
1112 list_del(&page->lru);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -07001113
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001114 mt = get_pcppage_migratetype(page);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -07001115 /* MIGRATE_ISOLATE page should not go to pcplists */
1116 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1117 /* Pageblock could have been isolated meanwhile */
Mel Gorman37779992016-05-19 17:13:58 -07001118 if (unlikely(isolated_pageblocks))
Joonsoo Kim51bb1a42014-11-13 15:19:14 -08001119 mt = get_pageblock_migratetype(page);
Joonsoo Kim51bb1a42014-11-13 15:19:14 -08001120
Mel Gorman4db75482016-05-19 17:14:32 -07001121 if (bulkfree_pcp_prepare(page))
1122 continue;
1123
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001124 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -07001125 trace_mm_page_pcpu_drain(page, 0, mt);
Mel Gormane5b31ac2016-05-19 17:14:24 -07001126 } while (--count && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
Nick Pigginc54ad302006-01-06 00:10:56 -08001128 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129}
1130
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001131static void free_one_page(struct zone *zone,
1132 struct page *page, unsigned long pfn,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001133 unsigned int order,
Mel Gormaned0ae212009-06-16 15:32:07 -07001134 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -08001135{
Mel Gorman0d5d8232014-08-06 16:07:16 -07001136 unsigned long nr_scanned;
Christoph Lameter006d22d2006-09-25 23:31:48 -07001137 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -07001138 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
1139 if (nr_scanned)
1140 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -07001141
Joonsoo Kimad53f922014-11-13 15:19:11 -08001142 if (unlikely(has_isolate_pageblock(zone) ||
1143 is_migrate_isolate(migratetype))) {
1144 migratetype = get_pfnblock_migratetype(page, pfn);
Joonsoo Kimad53f922014-11-13 15:19:11 -08001145 }
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001146 __free_one_page(page, pfn, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -07001147 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -08001148}
1149
Robin Holt1e8ce832015-06-30 14:56:45 -07001150static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1151 unsigned long zone, int nid)
1152{
Robin Holt1e8ce832015-06-30 14:56:45 -07001153 set_page_links(page, zone, nid, pfn);
Robin Holt1e8ce832015-06-30 14:56:45 -07001154 init_page_count(page);
1155 page_mapcount_reset(page);
1156 page_cpupid_reset_last(page);
Robin Holt1e8ce832015-06-30 14:56:45 -07001157
Robin Holt1e8ce832015-06-30 14:56:45 -07001158 INIT_LIST_HEAD(&page->lru);
1159#ifdef WANT_PAGE_VIRTUAL
1160 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1161 if (!is_highmem_idx(zone))
1162 set_page_address(page, __va(pfn << PAGE_SHIFT));
1163#endif
1164}
1165
1166static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
1167 int nid)
1168{
1169 return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
1170}
1171
Mel Gorman7e18adb2015-06-30 14:57:05 -07001172#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1173static void init_reserved_page(unsigned long pfn)
1174{
1175 pg_data_t *pgdat;
1176 int nid, zid;
1177
1178 if (!early_page_uninitialised(pfn))
1179 return;
1180
1181 nid = early_pfn_to_nid(pfn);
1182 pgdat = NODE_DATA(nid);
1183
1184 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1185 struct zone *zone = &pgdat->node_zones[zid];
1186
1187 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1188 break;
1189 }
1190 __init_single_pfn(pfn, zid, nid);
1191}
1192#else
1193static inline void init_reserved_page(unsigned long pfn)
1194{
1195}
1196#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1197
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001198/*
1199 * Initialised pages do not have PageReserved set. This function is
1200 * called for each range allocated by the bootmem allocator and
1201 * marks the pages PageReserved. The remaining valid pages are later
1202 * sent to the buddy page allocator.
1203 */
Stefan Bader4b50bcc2016-05-20 16:58:38 -07001204void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001205{
1206 unsigned long start_pfn = PFN_DOWN(start);
1207 unsigned long end_pfn = PFN_UP(end);
1208
Mel Gorman7e18adb2015-06-30 14:57:05 -07001209 for (; start_pfn < end_pfn; start_pfn++) {
1210 if (pfn_valid(start_pfn)) {
1211 struct page *page = pfn_to_page(start_pfn);
1212
1213 init_reserved_page(start_pfn);
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -08001214
1215 /* Avoid false-positive PageTail() */
1216 INIT_LIST_HEAD(&page->lru);
1217
Mel Gorman7e18adb2015-06-30 14:57:05 -07001218 SetPageReserved(page);
1219 }
1220 }
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001221}
1222
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001223static void __free_pages_ok(struct page *page, unsigned int order)
1224{
1225 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -07001226 int migratetype;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001227 unsigned long pfn = page_to_pfn(page);
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001228
Mel Gormane2769db2016-05-19 17:14:38 -07001229 if (!free_pages_prepare(page, order, true))
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001230 return;
1231
Mel Gormancfc47a22014-06-04 16:10:19 -07001232 migratetype = get_pfnblock_migratetype(page, pfn);
Nick Pigginc54ad302006-01-06 00:10:56 -08001233 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001234 __count_vm_events(PGFREE, 1 << order);
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001235 free_one_page(page_zone(page), page, pfn, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -08001236 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237}
1238
Li Zhang949698a2016-05-19 17:11:37 -07001239static void __init __free_pages_boot_core(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -08001240{
Johannes Weinerc3993072012-01-10 15:08:10 -08001241 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001242 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -08001243 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -08001244
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001245 prefetchw(p);
1246 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1247 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -08001248 __ClearPageReserved(p);
1249 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -08001250 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001251 __ClearPageReserved(p);
1252 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -08001253
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001254 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -08001255 set_page_refcounted(page);
1256 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -08001257}
1258
Mel Gorman75a592a2015-06-30 14:56:59 -07001259#if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1260 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
Mel Gorman7ace9912015-08-06 15:46:13 -07001261
Mel Gorman75a592a2015-06-30 14:56:59 -07001262static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1263
1264int __meminit early_pfn_to_nid(unsigned long pfn)
1265{
Mel Gorman7ace9912015-08-06 15:46:13 -07001266 static DEFINE_SPINLOCK(early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -07001267 int nid;
1268
Mel Gorman7ace9912015-08-06 15:46:13 -07001269 spin_lock(&early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -07001270 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
Mel Gorman7ace9912015-08-06 15:46:13 -07001271 if (nid < 0)
Mel Gormane4568d32016-07-14 12:07:20 -07001272 nid = first_online_node;
Mel Gorman7ace9912015-08-06 15:46:13 -07001273 spin_unlock(&early_pfn_lock);
1274
1275 return nid;
Mel Gorman75a592a2015-06-30 14:56:59 -07001276}
1277#endif
1278
1279#ifdef CONFIG_NODES_SPAN_OTHER_NODES
1280static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1281 struct mminit_pfnnid_cache *state)
1282{
1283 int nid;
1284
1285 nid = __early_pfn_to_nid(pfn, state);
1286 if (nid >= 0 && nid != node)
1287 return false;
1288 return true;
1289}
1290
1291/* Only safe to use early in boot when initialisation is single-threaded */
1292static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1293{
1294 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1295}
1296
1297#else
1298
1299static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1300{
1301 return true;
1302}
1303static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1304 struct mminit_pfnnid_cache *state)
1305{
1306 return true;
1307}
1308#endif
1309
1310
Mel Gorman0e1cc952015-06-30 14:57:27 -07001311void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001312 unsigned int order)
1313{
1314 if (early_page_uninitialised(pfn))
1315 return;
Li Zhang949698a2016-05-19 17:11:37 -07001316 return __free_pages_boot_core(page, order);
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001317}
1318
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001319/*
1320 * Check that the whole (or subset of) a pageblock given by the interval of
1321 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1322 * with the migration of free compaction scanner. The scanners then need to
1323 * use only pfn_valid_within() check for arches that allow holes within
1324 * pageblocks.
1325 *
1326 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1327 *
1328 * It's possible on some configurations to have a setup like node0 node1 node0
1329 * i.e. it's possible that all pages within a zones range of pages do not
1330 * belong to a single zone. We assume that a border between node0 and node1
1331 * can occur within a single pageblock, but not a node0 node1 node0
1332 * interleaving within a single pageblock. It is therefore sufficient to check
1333 * the first and last page of a pageblock and avoid checking each individual
1334 * page in a pageblock.
1335 */
1336struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1337 unsigned long end_pfn, struct zone *zone)
1338{
1339 struct page *start_page;
1340 struct page *end_page;
1341
1342 /* end_pfn is one past the range we are checking */
1343 end_pfn--;
1344
1345 if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1346 return NULL;
1347
1348 start_page = pfn_to_page(start_pfn);
1349
1350 if (page_zone(start_page) != zone)
1351 return NULL;
1352
1353 end_page = pfn_to_page(end_pfn);
1354
1355 /* This gives a shorter code than deriving page_zone(end_page) */
1356 if (page_zone_id(start_page) != page_zone_id(end_page))
1357 return NULL;
1358
1359 return start_page;
1360}
1361
1362void set_zone_contiguous(struct zone *zone)
1363{
1364 unsigned long block_start_pfn = zone->zone_start_pfn;
1365 unsigned long block_end_pfn;
1366
1367 block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1368 for (; block_start_pfn < zone_end_pfn(zone);
1369 block_start_pfn = block_end_pfn,
1370 block_end_pfn += pageblock_nr_pages) {
1371
1372 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1373
1374 if (!__pageblock_pfn_to_page(block_start_pfn,
1375 block_end_pfn, zone))
1376 return;
1377 }
1378
1379 /* We confirm that there is no hole */
1380 zone->contiguous = true;
1381}
1382
1383void clear_zone_contiguous(struct zone *zone)
1384{
1385 zone->contiguous = false;
1386}
1387
Mel Gorman7e18adb2015-06-30 14:57:05 -07001388#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
Mel Gorman0e1cc952015-06-30 14:57:27 -07001389static void __init deferred_free_range(struct page *page,
Mel Gormana4de83d2015-06-30 14:57:16 -07001390 unsigned long pfn, int nr_pages)
1391{
1392 int i;
1393
1394 if (!page)
1395 return;
1396
1397 /* Free a large naturally-aligned chunk if possible */
1398 if (nr_pages == MAX_ORDER_NR_PAGES &&
1399 (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
Mel Gormanac5d2532015-06-30 14:57:20 -07001400 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Li Zhang949698a2016-05-19 17:11:37 -07001401 __free_pages_boot_core(page, MAX_ORDER-1);
Mel Gormana4de83d2015-06-30 14:57:16 -07001402 return;
1403 }
1404
Li Zhang949698a2016-05-19 17:11:37 -07001405 for (i = 0; i < nr_pages; i++, page++)
1406 __free_pages_boot_core(page, 0);
Mel Gormana4de83d2015-06-30 14:57:16 -07001407}
1408
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001409/* Completion tracking for deferred_init_memmap() threads */
1410static atomic_t pgdat_init_n_undone __initdata;
1411static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1412
1413static inline void __init pgdat_init_report_one_done(void)
1414{
1415 if (atomic_dec_and_test(&pgdat_init_n_undone))
1416 complete(&pgdat_init_all_done_comp);
1417}
Mel Gorman0e1cc952015-06-30 14:57:27 -07001418
Mel Gorman7e18adb2015-06-30 14:57:05 -07001419/* Initialise remaining memory on a node */
Mel Gorman0e1cc952015-06-30 14:57:27 -07001420static int __init deferred_init_memmap(void *data)
Mel Gorman7e18adb2015-06-30 14:57:05 -07001421{
Mel Gorman0e1cc952015-06-30 14:57:27 -07001422 pg_data_t *pgdat = data;
1423 int nid = pgdat->node_id;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001424 struct mminit_pfnnid_cache nid_init_state = { };
1425 unsigned long start = jiffies;
1426 unsigned long nr_pages = 0;
1427 unsigned long walk_start, walk_end;
1428 int i, zid;
1429 struct zone *zone;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001430 unsigned long first_init_pfn = pgdat->first_deferred_pfn;
Mel Gorman0e1cc952015-06-30 14:57:27 -07001431 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001432
Mel Gorman0e1cc952015-06-30 14:57:27 -07001433 if (first_init_pfn == ULONG_MAX) {
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001434 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001435 return 0;
1436 }
1437
1438 /* Bind memory initialisation thread to a local node if possible */
1439 if (!cpumask_empty(cpumask))
1440 set_cpus_allowed_ptr(current, cpumask);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001441
1442 /* Sanity check boundaries */
1443 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1444 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1445 pgdat->first_deferred_pfn = ULONG_MAX;
1446
1447 /* Only the highest zone is deferred so find it */
1448 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1449 zone = pgdat->node_zones + zid;
1450 if (first_init_pfn < zone_end_pfn(zone))
1451 break;
1452 }
1453
1454 for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1455 unsigned long pfn, end_pfn;
Mel Gorman54608c32015-06-30 14:57:09 -07001456 struct page *page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001457 struct page *free_base_page = NULL;
1458 unsigned long free_base_pfn = 0;
1459 int nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001460
1461 end_pfn = min(walk_end, zone_end_pfn(zone));
1462 pfn = first_init_pfn;
1463 if (pfn < walk_start)
1464 pfn = walk_start;
1465 if (pfn < zone->zone_start_pfn)
1466 pfn = zone->zone_start_pfn;
1467
1468 for (; pfn < end_pfn; pfn++) {
Mel Gorman54608c32015-06-30 14:57:09 -07001469 if (!pfn_valid_within(pfn))
Mel Gormana4de83d2015-06-30 14:57:16 -07001470 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001471
Mel Gorman54608c32015-06-30 14:57:09 -07001472 /*
1473 * Ensure pfn_valid is checked every
1474 * MAX_ORDER_NR_PAGES for memory holes
1475 */
1476 if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1477 if (!pfn_valid(pfn)) {
1478 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001479 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001480 }
1481 }
1482
1483 if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1484 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001485 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001486 }
1487
1488 /* Minimise pfn page lookups and scheduler checks */
1489 if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1490 page++;
1491 } else {
Mel Gormana4de83d2015-06-30 14:57:16 -07001492 nr_pages += nr_to_free;
1493 deferred_free_range(free_base_page,
1494 free_base_pfn, nr_to_free);
1495 free_base_page = NULL;
1496 free_base_pfn = nr_to_free = 0;
1497
Mel Gorman54608c32015-06-30 14:57:09 -07001498 page = pfn_to_page(pfn);
1499 cond_resched();
1500 }
Mel Gorman7e18adb2015-06-30 14:57:05 -07001501
1502 if (page->flags) {
1503 VM_BUG_ON(page_zone(page) != zone);
Mel Gormana4de83d2015-06-30 14:57:16 -07001504 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001505 }
1506
1507 __init_single_page(page, pfn, zid, nid);
Mel Gormana4de83d2015-06-30 14:57:16 -07001508 if (!free_base_page) {
1509 free_base_page = page;
1510 free_base_pfn = pfn;
1511 nr_to_free = 0;
1512 }
1513 nr_to_free++;
1514
1515 /* Where possible, batch up pages for a single free */
1516 continue;
1517free_range:
1518 /* Free the current block of pages to allocator */
1519 nr_pages += nr_to_free;
1520 deferred_free_range(free_base_page, free_base_pfn,
1521 nr_to_free);
1522 free_base_page = NULL;
1523 free_base_pfn = nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001524 }
Mel Gormana4de83d2015-06-30 14:57:16 -07001525
Mel Gorman7e18adb2015-06-30 14:57:05 -07001526 first_init_pfn = max(end_pfn, first_init_pfn);
1527 }
1528
1529 /* Sanity check that the next zone really is unpopulated */
1530 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1531
Mel Gorman0e1cc952015-06-30 14:57:27 -07001532 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
Mel Gorman7e18adb2015-06-30 14:57:05 -07001533 jiffies_to_msecs(jiffies - start));
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001534
1535 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001536 return 0;
1537}
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001538#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
Mel Gorman0e1cc952015-06-30 14:57:27 -07001539
1540void __init page_alloc_init_late(void)
1541{
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001542 struct zone *zone;
1543
1544#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
Mel Gorman0e1cc952015-06-30 14:57:27 -07001545 int nid;
1546
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001547 /* There will be num_node_state(N_MEMORY) threads */
1548 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
Mel Gorman0e1cc952015-06-30 14:57:27 -07001549 for_each_node_state(nid, N_MEMORY) {
Mel Gorman0e1cc952015-06-30 14:57:27 -07001550 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1551 }
1552
1553 /* Block until all are initialised */
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001554 wait_for_completion(&pgdat_init_all_done_comp);
Mel Gorman4248b0d2015-08-06 15:46:20 -07001555
1556 /* Reinit limits that are based on free pages after the kernel is up */
1557 files_maxfiles_init();
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001558#endif
1559
1560 for_each_populated_zone(zone)
1561 set_zone_contiguous(zone);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001562}
Mel Gorman7e18adb2015-06-30 14:57:05 -07001563
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001564#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +08001565/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001566void __init init_cma_reserved_pageblock(struct page *page)
1567{
1568 unsigned i = pageblock_nr_pages;
1569 struct page *p = page;
1570
1571 do {
1572 __ClearPageReserved(p);
1573 set_page_count(p, 0);
1574 } while (++p, --i);
1575
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001576 set_pageblock_migratetype(page, MIGRATE_CMA);
Michal Nazarewiczdc783272014-07-02 15:22:35 -07001577
1578 if (pageblock_order >= MAX_ORDER) {
1579 i = pageblock_nr_pages;
1580 p = page;
1581 do {
1582 set_page_refcounted(p);
1583 __free_pages(p, MAX_ORDER - 1);
1584 p += MAX_ORDER_NR_PAGES;
1585 } while (i -= MAX_ORDER_NR_PAGES);
1586 } else {
1587 set_page_refcounted(page);
1588 __free_pages(page, pageblock_order);
1589 }
1590
Jiang Liu3dcc0572013-07-03 15:03:21 -07001591 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001592}
1593#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
1595/*
1596 * The order of subdivision here is critical for the IO subsystem.
1597 * Please do not alter this order without good reasons and regression
1598 * testing. Specifically, as large blocks of memory are subdivided,
1599 * the order in which smaller blocks are delivered depends on the order
1600 * they're subdivided in this function. This is the primary factor
1601 * influencing the order in which pages are delivered to the IO
1602 * subsystem according to empirical testing, and this is also justified
1603 * by considering the behavior of a buddy system containing a single
1604 * large block of memory acted on by a series of small allocations.
1605 * This behavior is a critical factor in sglist merging's success.
1606 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +01001607 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 */
Nick Piggin085cc7d2006-01-06 00:11:01 -08001609static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001610 int low, int high, struct free_area *area,
1611 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612{
1613 unsigned long size = 1 << high;
1614
1615 while (high > low) {
1616 area--;
1617 high--;
1618 size >>= 1;
Sasha Levin309381fea2014-01-23 15:52:54 -08001619 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08001620
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001621 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
Joonsoo Kime30825f2014-12-12 16:55:49 -08001622 debug_guardpage_enabled() &&
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001623 high < debug_guardpage_minorder()) {
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08001624 /*
1625 * Mark as guard pages (or page), that will allow to
1626 * merge back to allocator when buddy will be freed.
1627 * Corresponding page table entries will not be touched,
1628 * pages will stay not present in virtual address space
1629 */
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001630 set_page_guard(zone, &page[size], high, migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08001631 continue;
1632 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001633 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 area->nr_free++;
1635 set_page_order(&page[size], high);
1636 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637}
1638
Vlastimil Babka4e611802016-05-19 17:14:41 -07001639static void check_new_page_bad(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640{
Vlastimil Babka4e611802016-05-19 17:14:41 -07001641 const char *bad_reason = NULL;
1642 unsigned long bad_flags = 0;
Dave Hansenf0b791a2014-01-23 15:52:49 -08001643
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001644 if (unlikely(atomic_read(&page->_mapcount) != -1))
Dave Hansenf0b791a2014-01-23 15:52:49 -08001645 bad_reason = "nonzero mapcount";
1646 if (unlikely(page->mapping != NULL))
1647 bad_reason = "non-NULL mapping";
Joonsoo Kimfe896d12016-03-17 14:19:26 -07001648 if (unlikely(page_ref_count(page) != 0))
Dave Hansenf0b791a2014-01-23 15:52:49 -08001649 bad_reason = "nonzero _count";
Naoya Horiguchif4c18e62015-08-06 15:47:08 -07001650 if (unlikely(page->flags & __PG_HWPOISON)) {
1651 bad_reason = "HWPoisoned (hardware-corrupted)";
1652 bad_flags = __PG_HWPOISON;
Naoya Horiguchie570f562016-05-20 16:58:50 -07001653 /* Don't complain about hwpoisoned pages */
1654 page_mapcount_reset(page); /* remove PageBuddy */
1655 return;
Naoya Horiguchif4c18e62015-08-06 15:47:08 -07001656 }
Dave Hansenf0b791a2014-01-23 15:52:49 -08001657 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1658 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1659 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1660 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -08001661#ifdef CONFIG_MEMCG
1662 if (unlikely(page->mem_cgroup))
1663 bad_reason = "page still charged to cgroup";
1664#endif
Vlastimil Babka4e611802016-05-19 17:14:41 -07001665 bad_page(page, bad_reason, bad_flags);
1666}
1667
1668/*
1669 * This page is about to be returned from the page allocator
1670 */
1671static inline int check_new_page(struct page *page)
1672{
1673 if (likely(page_expected_state(page,
1674 PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
1675 return 0;
1676
1677 check_new_page_bad(page);
1678 return 1;
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001679}
1680
Laura Abbott1414c7f2016-03-15 14:56:30 -07001681static inline bool free_pages_prezeroed(bool poisoned)
1682{
1683 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1684 page_poisoning_enabled() && poisoned;
1685}
1686
Mel Gorman479f8542016-05-19 17:14:35 -07001687#ifdef CONFIG_DEBUG_VM
1688static bool check_pcp_refill(struct page *page)
1689{
1690 return false;
1691}
1692
1693static bool check_new_pcp(struct page *page)
1694{
1695 return check_new_page(page);
1696}
1697#else
1698static bool check_pcp_refill(struct page *page)
1699{
1700 return check_new_page(page);
1701}
1702static bool check_new_pcp(struct page *page)
1703{
1704 return false;
1705}
1706#endif /* CONFIG_DEBUG_VM */
1707
1708static bool check_new_pages(struct page *page, unsigned int order)
1709{
1710 int i;
1711 for (i = 0; i < (1 << order); i++) {
1712 struct page *p = page + i;
1713
1714 if (unlikely(check_new_page(p)))
1715 return true;
1716 }
1717
1718 return false;
1719}
1720
Joonsoo Kim46f24fd2016-07-26 15:23:58 -07001721inline void post_alloc_hook(struct page *page, unsigned int order,
1722 gfp_t gfp_flags)
1723{
1724 set_page_private(page, 0);
1725 set_page_refcounted(page);
1726
1727 arch_alloc_page(page, order);
1728 kernel_map_pages(page, 1 << order, 1);
1729 kernel_poison_pages(page, 1 << order, 1);
1730 kasan_alloc_pages(page, order);
1731 set_page_owner(page, order, gfp_flags);
1732}
1733
Mel Gorman479f8542016-05-19 17:14:35 -07001734static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
Mel Gormanc6038442016-05-19 17:13:38 -07001735 unsigned int alloc_flags)
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001736{
1737 int i;
Laura Abbott1414c7f2016-03-15 14:56:30 -07001738 bool poisoned = true;
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001739
1740 for (i = 0; i < (1 << order); i++) {
1741 struct page *p = page + i;
Laura Abbott1414c7f2016-03-15 14:56:30 -07001742 if (poisoned)
1743 poisoned &= page_is_poisoned(p);
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001744 }
Hugh Dickins689bceb2005-11-21 21:32:20 -08001745
Joonsoo Kim46f24fd2016-07-26 15:23:58 -07001746 post_alloc_hook(page, order, gfp_flags);
Nick Piggin17cf4402006-03-22 00:08:41 -08001747
Laura Abbott1414c7f2016-03-15 14:56:30 -07001748 if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO))
Anisse Astierf4d28972015-06-24 16:56:36 -07001749 for (i = 0; i < (1 << order); i++)
1750 clear_highpage(page + i);
Nick Piggin17cf4402006-03-22 00:08:41 -08001751
1752 if (order && (gfp_flags & __GFP_COMP))
1753 prep_compound_page(page, order);
1754
Vlastimil Babka75379192015-02-11 15:25:38 -08001755 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001756 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
Vlastimil Babka75379192015-02-11 15:25:38 -08001757 * allocate the page. The expectation is that the caller is taking
1758 * steps that will free more memory. The caller should avoid the page
1759 * being used for !PFMEMALLOC purposes.
1760 */
Michal Hocko2f064f32015-08-21 14:11:51 -07001761 if (alloc_flags & ALLOC_NO_WATERMARKS)
1762 set_page_pfmemalloc(page);
1763 else
1764 clear_page_pfmemalloc(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765}
1766
Mel Gorman56fd56b2007-10-16 01:25:58 -07001767/*
1768 * Go through the free lists for the given migratetype and remove
1769 * the smallest available page from the freelists
1770 */
Mel Gorman728ec982009-06-16 15:32:04 -07001771static inline
1772struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -07001773 int migratetype)
1774{
1775 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -07001776 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001777 struct page *page;
1778
1779 /* Find a page of the appropriate size in the preferred list */
1780 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1781 area = &(zone->free_area[current_order]);
Geliang Tanga16601c2016-01-14 15:20:30 -08001782 page = list_first_entry_or_null(&area->free_list[migratetype],
Mel Gorman56fd56b2007-10-16 01:25:58 -07001783 struct page, lru);
Geliang Tanga16601c2016-01-14 15:20:30 -08001784 if (!page)
1785 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001786 list_del(&page->lru);
1787 rmv_page_order(page);
1788 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001789 expand(zone, page, order, current_order, area, migratetype);
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001790 set_pcppage_migratetype(page, migratetype);
Mel Gorman56fd56b2007-10-16 01:25:58 -07001791 return page;
1792 }
1793
1794 return NULL;
1795}
1796
1797
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001798/*
1799 * This array describes the order lists are fallen back to when
1800 * the free lists for the desirable migrate type are depleted
1801 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001802static int fallbacks[MIGRATE_TYPES][4] = {
Mel Gorman974a7862015-11-06 16:28:34 -08001803 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1804 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1805 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
Joonsoo Kimdc676472015-04-14 15:45:15 -07001806#ifdef CONFIG_CMA
Mel Gorman974a7862015-11-06 16:28:34 -08001807 [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001808#endif
Minchan Kim194159f2013-02-22 16:33:58 -08001809#ifdef CONFIG_MEMORY_ISOLATION
Mel Gorman974a7862015-11-06 16:28:34 -08001810 [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001811#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001812};
1813
Joonsoo Kimdc676472015-04-14 15:45:15 -07001814#ifdef CONFIG_CMA
1815static struct page *__rmqueue_cma_fallback(struct zone *zone,
1816 unsigned int order)
1817{
1818 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1819}
1820#else
1821static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1822 unsigned int order) { return NULL; }
1823#endif
1824
Mel Gormanc361be52007-10-16 01:25:51 -07001825/*
1826 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -07001827 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -07001828 * boundary. If alignment is required, use move_freepages_block()
1829 */
Minchan Kim435b4052012-10-08 16:32:16 -07001830int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -07001831 struct page *start_page, struct page *end_page,
1832 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001833{
1834 struct page *page;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08001835 unsigned int order;
Mel Gormand1003132007-10-16 01:26:00 -07001836 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -07001837
1838#ifndef CONFIG_HOLES_IN_ZONE
1839 /*
1840 * page_zone is not safe to call in this context when
1841 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1842 * anyway as we check zone boundaries in move_freepages_block().
1843 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -07001844 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -07001845 */
Mel Gorman97ee4ba2014-10-09 15:28:28 -07001846 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
Mel Gormanc361be52007-10-16 01:25:51 -07001847#endif
1848
1849 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -07001850 /* Make sure we are not inadvertently changing nodes */
Sasha Levin309381fea2014-01-23 15:52:54 -08001851 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
Adam Litke344c7902008-09-02 14:35:38 -07001852
Mel Gormanc361be52007-10-16 01:25:51 -07001853 if (!pfn_valid_within(page_to_pfn(page))) {
1854 page++;
1855 continue;
1856 }
1857
1858 if (!PageBuddy(page)) {
1859 page++;
1860 continue;
1861 }
1862
1863 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001864 list_move(&page->lru,
1865 &zone->free_area[order].free_list[migratetype]);
Mel Gormanc361be52007-10-16 01:25:51 -07001866 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001867 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001868 }
1869
Mel Gormand1003132007-10-16 01:26:00 -07001870 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001871}
1872
Minchan Kimee6f5092012-07-31 16:43:50 -07001873int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001874 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001875{
1876 unsigned long start_pfn, end_pfn;
1877 struct page *start_page, *end_page;
1878
1879 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001880 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001881 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001882 end_page = start_page + pageblock_nr_pages - 1;
1883 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001884
1885 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001886 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001887 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001888 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001889 return 0;
1890
1891 return move_freepages(zone, start_page, end_page, migratetype);
1892}
1893
Mel Gorman2f66a682009-09-21 17:02:31 -07001894static void change_pageblock_range(struct page *pageblock_page,
1895 int start_order, int migratetype)
1896{
1897 int nr_pageblocks = 1 << (start_order - pageblock_order);
1898
1899 while (nr_pageblocks--) {
1900 set_pageblock_migratetype(pageblock_page, migratetype);
1901 pageblock_page += pageblock_nr_pages;
1902 }
1903}
1904
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001905/*
Vlastimil Babka9c0415e2015-02-11 15:28:21 -08001906 * When we are falling back to another migratetype during allocation, try to
1907 * steal extra free pages from the same pageblocks to satisfy further
1908 * allocations, instead of polluting multiple pageblocks.
1909 *
1910 * If we are stealing a relatively large buddy page, it is likely there will
1911 * be more free pages in the pageblock, so try to steal them all. For
1912 * reclaimable and unmovable allocations, we steal regardless of page size,
1913 * as fragmentation caused by those allocations polluting movable pageblocks
1914 * is worse than movable allocations stealing from unmovable and reclaimable
1915 * pageblocks.
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001916 */
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001917static bool can_steal_fallback(unsigned int order, int start_mt)
1918{
1919 /*
1920 * Leaving this order check is intended, although there is
1921 * relaxed order check in next check. The reason is that
1922 * we can actually steal whole pageblock if this condition met,
1923 * but, below check doesn't guarantee it and that is just heuristic
1924 * so could be changed anytime.
1925 */
1926 if (order >= pageblock_order)
1927 return true;
1928
1929 if (order >= pageblock_order / 2 ||
1930 start_mt == MIGRATE_RECLAIMABLE ||
1931 start_mt == MIGRATE_UNMOVABLE ||
1932 page_group_by_mobility_disabled)
1933 return true;
1934
1935 return false;
1936}
1937
1938/*
1939 * This function implements actual steal behaviour. If order is large enough,
1940 * we can steal whole pageblock. If not, we first move freepages in this
1941 * pageblock and check whether half of pages are moved or not. If half of
1942 * pages are moved, we can change migratetype of pageblock and permanently
1943 * use it's pages as requested migratetype in the future.
1944 */
1945static void steal_suitable_fallback(struct zone *zone, struct page *page,
1946 int start_type)
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001947{
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08001948 unsigned int current_order = page_order(page);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001949 int pages;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001950
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001951 /* Take ownership for orders >= pageblock_order */
1952 if (current_order >= pageblock_order) {
1953 change_pageblock_range(page, current_order, start_type);
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001954 return;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001955 }
1956
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001957 pages = move_freepages_block(zone, page, start_type);
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001958
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001959 /* Claim the whole block if over half of it is free */
1960 if (pages >= (1 << (pageblock_order-1)) ||
1961 page_group_by_mobility_disabled)
1962 set_pageblock_migratetype(page, start_type);
1963}
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001964
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001965/*
1966 * Check whether there is a suitable fallback freepage with requested order.
1967 * If only_stealable is true, this function returns fallback_mt only if
1968 * we can steal other freepages all together. This would help to reduce
1969 * fragmentation due to mixed migratetype pages in one pageblock.
1970 */
1971int find_suitable_fallback(struct free_area *area, unsigned int order,
1972 int migratetype, bool only_stealable, bool *can_steal)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001973{
1974 int i;
1975 int fallback_mt;
1976
1977 if (area->nr_free == 0)
1978 return -1;
1979
1980 *can_steal = false;
1981 for (i = 0;; i++) {
1982 fallback_mt = fallbacks[migratetype][i];
Mel Gorman974a7862015-11-06 16:28:34 -08001983 if (fallback_mt == MIGRATE_TYPES)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001984 break;
1985
1986 if (list_empty(&area->free_list[fallback_mt]))
1987 continue;
1988
1989 if (can_steal_fallback(order, migratetype))
1990 *can_steal = true;
1991
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001992 if (!only_stealable)
1993 return fallback_mt;
1994
1995 if (*can_steal)
1996 return fallback_mt;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001997 }
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001998
1999 return -1;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07002000}
2001
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002002/*
2003 * Reserve a pageblock for exclusive use of high-order atomic allocations if
2004 * there are no empty page blocks that contain a page with a suitable order
2005 */
2006static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2007 unsigned int alloc_order)
2008{
2009 int mt;
2010 unsigned long max_managed, flags;
2011
2012 /*
2013 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2014 * Check is race-prone but harmless.
2015 */
2016 max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
2017 if (zone->nr_reserved_highatomic >= max_managed)
2018 return;
2019
2020 spin_lock_irqsave(&zone->lock, flags);
2021
2022 /* Recheck the nr_reserved_highatomic limit under the lock */
2023 if (zone->nr_reserved_highatomic >= max_managed)
2024 goto out_unlock;
2025
2026 /* Yoink! */
2027 mt = get_pageblock_migratetype(page);
2028 if (mt != MIGRATE_HIGHATOMIC &&
2029 !is_migrate_isolate(mt) && !is_migrate_cma(mt)) {
2030 zone->nr_reserved_highatomic += pageblock_nr_pages;
2031 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2032 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC);
2033 }
2034
2035out_unlock:
2036 spin_unlock_irqrestore(&zone->lock, flags);
2037}
2038
2039/*
2040 * Used when an allocation is about to fail under memory pressure. This
2041 * potentially hurts the reliability of high-order allocations when under
2042 * intense memory pressure but failed atomic allocations should be easier
2043 * to recover from than an OOM.
2044 */
2045static void unreserve_highatomic_pageblock(const struct alloc_context *ac)
2046{
2047 struct zonelist *zonelist = ac->zonelist;
2048 unsigned long flags;
2049 struct zoneref *z;
2050 struct zone *zone;
2051 struct page *page;
2052 int order;
2053
2054 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2055 ac->nodemask) {
2056 /* Preserve at least one pageblock */
2057 if (zone->nr_reserved_highatomic <= pageblock_nr_pages)
2058 continue;
2059
2060 spin_lock_irqsave(&zone->lock, flags);
2061 for (order = 0; order < MAX_ORDER; order++) {
2062 struct free_area *area = &(zone->free_area[order]);
2063
Geliang Tanga16601c2016-01-14 15:20:30 -08002064 page = list_first_entry_or_null(
2065 &area->free_list[MIGRATE_HIGHATOMIC],
2066 struct page, lru);
2067 if (!page)
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002068 continue;
2069
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002070 /*
2071 * It should never happen but changes to locking could
2072 * inadvertently allow a per-cpu drain to add pages
2073 * to MIGRATE_HIGHATOMIC while unreserving so be safe
2074 * and watch for underflows.
2075 */
2076 zone->nr_reserved_highatomic -= min(pageblock_nr_pages,
2077 zone->nr_reserved_highatomic);
2078
2079 /*
2080 * Convert to ac->migratetype and avoid the normal
2081 * pageblock stealing heuristics. Minimally, the caller
2082 * is doing the work and needs the pages. More
2083 * importantly, if the block was always converted to
2084 * MIGRATE_UNMOVABLE or another type then the number
2085 * of pageblocks that cannot be completely freed
2086 * may increase.
2087 */
2088 set_pageblock_migratetype(page, ac->migratetype);
2089 move_freepages_block(zone, page, ac->migratetype);
2090 spin_unlock_irqrestore(&zone->lock, flags);
2091 return;
2092 }
2093 spin_unlock_irqrestore(&zone->lock, flags);
2094 }
2095}
2096
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002097/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07002098static inline struct page *
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002099__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002100{
Pintu Kumarb8af2942013-09-11 14:20:34 -07002101 struct free_area *area;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002102 unsigned int current_order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002103 struct page *page;
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002104 int fallback_mt;
2105 bool can_steal;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002106
2107 /* Find the largest possible block of pages in the other list */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002108 for (current_order = MAX_ORDER-1;
2109 current_order >= order && current_order <= MAX_ORDER-1;
2110 --current_order) {
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002111 area = &(zone->free_area[current_order]);
2112 fallback_mt = find_suitable_fallback(area, current_order,
Joonsoo Kim2149cda2015-04-14 15:45:21 -07002113 start_migratetype, false, &can_steal);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002114 if (fallback_mt == -1)
2115 continue;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002116
Geliang Tanga16601c2016-01-14 15:20:30 -08002117 page = list_first_entry(&area->free_list[fallback_mt],
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002118 struct page, lru);
2119 if (can_steal)
2120 steal_suitable_fallback(zone, page, start_migratetype);
Mel Gormane0104872007-10-16 01:25:53 -07002121
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002122 /* Remove the page from the freelists */
2123 area->nr_free--;
2124 list_del(&page->lru);
2125 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002126
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002127 expand(zone, page, order, current_order, area,
2128 start_migratetype);
2129 /*
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002130 * The pcppage_migratetype may differ from pageblock's
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002131 * migratetype depending on the decisions in
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002132 * find_suitable_fallback(). This is OK as long as it does not
2133 * differ for MIGRATE_CMA pageblocks. Those can be used as
2134 * fallback only via special __rmqueue_cma_fallback() function
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002135 */
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002136 set_pcppage_migratetype(page, start_migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002137
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002138 trace_mm_page_alloc_extfrag(page, order, current_order,
2139 start_migratetype, fallback_mt);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002140
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002141 return page;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002142 }
2143
Mel Gorman728ec982009-06-16 15:32:04 -07002144 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002145}
2146
Mel Gorman56fd56b2007-10-16 01:25:58 -07002147/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 * Do the hard work of removing an element from the buddy allocator.
2149 * Call me with the zone->lock already held.
2150 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002151static struct page *__rmqueue(struct zone *zone, unsigned int order,
Mel Gorman6ac02062016-01-14 15:20:28 -08002152 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 struct page *page;
2155
Mel Gorman56fd56b2007-10-16 01:25:58 -07002156 page = __rmqueue_smallest(zone, order, migratetype);
Mel Gorman974a7862015-11-06 16:28:34 -08002157 if (unlikely(!page)) {
Joonsoo Kimdc676472015-04-14 15:45:15 -07002158 if (migratetype == MIGRATE_MOVABLE)
2159 page = __rmqueue_cma_fallback(zone, order);
2160
2161 if (!page)
2162 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gorman728ec982009-06-16 15:32:04 -07002163 }
2164
Mel Gorman0d3d0622009-09-21 17:02:44 -07002165 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002166 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167}
2168
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01002169/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 * Obtain a specified number of elements from the buddy allocator, all under
2171 * a single hold of the lock, for efficiency. Add them to the supplied list.
2172 * Returns the number of new pages which were placed at *list.
2173 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01002174static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002175 unsigned long count, struct list_head *list,
Mel Gormanb745bc82014-06-04 16:10:22 -07002176 int migratetype, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177{
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07002178 int i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01002179
Nick Pigginc54ad302006-01-06 00:10:56 -08002180 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 for (i = 0; i < count; ++i) {
Mel Gorman6ac02062016-01-14 15:20:28 -08002182 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08002183 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08002185
Mel Gorman479f8542016-05-19 17:14:35 -07002186 if (unlikely(check_pcp_refill(page)))
2187 continue;
2188
Mel Gorman81eabcb2007-12-17 16:20:05 -08002189 /*
2190 * Split buddy pages returned by expand() are received here
2191 * in physical page order. The page is added to the callers and
2192 * list and the list head then moves forward. From the callers
2193 * perspective, the linked list is ordered by page number in
2194 * some conditions. This is useful for IO devices that can
2195 * merge IO requests if the physical pages are ordered
2196 * properly.
2197 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002198 if (likely(!cold))
Mel Gormane084b2d2009-07-29 15:02:04 -07002199 list_add(&page->lru, list);
2200 else
2201 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08002202 list = &page->lru;
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002203 if (is_migrate_cma(get_pcppage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07002204 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2205 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 }
Mel Gormanf2260e62009-06-16 15:32:13 -07002207 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08002208 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08002209 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210}
2211
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002212#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08002213/*
Christoph Lameter4037d452007-05-09 02:35:14 -07002214 * Called from the vmstat counter updater to drain pagesets of this
2215 * currently executing processor on remote nodes after they have
2216 * expired.
2217 *
Christoph Lameter879336c2006-03-22 00:09:08 -08002218 * Note that this function must be called with the thread pinned to
2219 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08002220 */
Christoph Lameter4037d452007-05-09 02:35:14 -07002221void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002222{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002223 unsigned long flags;
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07002224 int to_drain, batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002225
Christoph Lameter4037d452007-05-09 02:35:14 -07002226 local_irq_save(flags);
Jason Low4db0c3c2015-04-15 16:14:08 -07002227 batch = READ_ONCE(pcp->batch);
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07002228 to_drain = min(pcp->count, batch);
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07002229 if (to_drain > 0) {
2230 free_pcppages_bulk(zone, to_drain, pcp);
2231 pcp->count -= to_drain;
2232 }
Christoph Lameter4037d452007-05-09 02:35:14 -07002233 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002234}
2235#endif
2236
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002237/*
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002238 * Drain pcplists of the indicated processor and zone.
2239 *
2240 * The processor must either be the current processor and the
2241 * thread pinned to the current processor or a processor that
2242 * is not online.
2243 */
2244static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2245{
2246 unsigned long flags;
2247 struct per_cpu_pageset *pset;
2248 struct per_cpu_pages *pcp;
2249
2250 local_irq_save(flags);
2251 pset = per_cpu_ptr(zone->pageset, cpu);
2252
2253 pcp = &pset->pcp;
2254 if (pcp->count) {
2255 free_pcppages_bulk(zone, pcp->count, pcp);
2256 pcp->count = 0;
2257 }
2258 local_irq_restore(flags);
2259}
2260
2261/*
2262 * Drain pcplists of all zones on the indicated processor.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002263 *
2264 * The processor must either be the current processor and the
2265 * thread pinned to the current processor or a processor that
2266 * is not online.
2267 */
2268static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269{
2270 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002272 for_each_populated_zone(zone) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002273 drain_pages_zone(cpu, zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 }
2275}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002277/*
2278 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002279 *
2280 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2281 * the single zone's pages.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002282 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002283void drain_local_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002284{
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002285 int cpu = smp_processor_id();
2286
2287 if (zone)
2288 drain_pages_zone(cpu, zone);
2289 else
2290 drain_pages(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002291}
2292
2293/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002294 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2295 *
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002296 * When zone parameter is non-NULL, spill just the single zone's pages.
2297 *
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002298 * Note that this code is protected against sending an IPI to an offline
2299 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2300 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2301 * nothing keeps CPUs from showing up after we populated the cpumask and
2302 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002303 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002304void drain_all_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002305{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002306 int cpu;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002307
2308 /*
2309 * Allocate in the BSS so we wont require allocation in
2310 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2311 */
2312 static cpumask_t cpus_with_pcps;
2313
2314 /*
2315 * We don't care about racing with CPU hotplug event
2316 * as offline notification will cause the notified
2317 * cpu to drain that CPU pcps and on_each_cpu_mask
2318 * disables preemption as part of its processing
2319 */
2320 for_each_online_cpu(cpu) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002321 struct per_cpu_pageset *pcp;
2322 struct zone *z;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002323 bool has_pcps = false;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002324
2325 if (zone) {
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002326 pcp = per_cpu_ptr(zone->pageset, cpu);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002327 if (pcp->pcp.count)
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002328 has_pcps = true;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002329 } else {
2330 for_each_populated_zone(z) {
2331 pcp = per_cpu_ptr(z->pageset, cpu);
2332 if (pcp->pcp.count) {
2333 has_pcps = true;
2334 break;
2335 }
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002336 }
2337 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002338
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002339 if (has_pcps)
2340 cpumask_set_cpu(cpu, &cpus_with_pcps);
2341 else
2342 cpumask_clear_cpu(cpu, &cpus_with_pcps);
2343 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002344 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
2345 zone, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002346}
2347
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02002348#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349
2350void mark_free_pages(struct zone *zone)
2351{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002352 unsigned long pfn, max_zone_pfn;
2353 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002354 unsigned int order, t;
Geliang Tang86760a22016-01-14 15:20:33 -08002355 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356
Xishi Qiu8080fc02013-09-11 14:21:45 -07002357 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 return;
2359
2360 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002361
Cody P Schafer108bcc92013-02-22 16:35:23 -08002362 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002363 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2364 if (pfn_valid(pfn)) {
Geliang Tang86760a22016-01-14 15:20:33 -08002365 page = pfn_to_page(pfn);
Joonsoo Kimba6b0972016-05-19 17:12:16 -07002366
2367 if (page_zone(page) != zone)
2368 continue;
2369
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07002370 if (!swsusp_page_is_forbidden(page))
2371 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002374 for_each_migratetype_order(order, t) {
Geliang Tang86760a22016-01-14 15:20:33 -08002375 list_for_each_entry(page,
2376 &zone->free_area[order].free_list[t], lru) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002377 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378
Geliang Tang86760a22016-01-14 15:20:33 -08002379 pfn = page_to_pfn(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002380 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07002381 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002382 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002383 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 spin_unlock_irqrestore(&zone->lock, flags);
2385}
Mel Gormane2c55dc2007-10-16 01:25:50 -07002386#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387
2388/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 * Free a 0-order page
Mel Gormanb745bc82014-06-04 16:10:22 -07002390 * cold == true ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002392void free_hot_cold_page(struct page *page, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393{
2394 struct zone *zone = page_zone(page);
2395 struct per_cpu_pages *pcp;
2396 unsigned long flags;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002397 unsigned long pfn = page_to_pfn(page);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002398 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399
Mel Gorman4db75482016-05-19 17:14:32 -07002400 if (!free_pcp_prepare(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -08002401 return;
2402
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002403 migratetype = get_pfnblock_migratetype(page, pfn);
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002404 set_pcppage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07002406 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07002407
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002408 /*
2409 * We only track unmovable, reclaimable and movable on pcp lists.
2410 * Free ISOLATE pages back to the allocator because they are being
2411 * offlined but treat RESERVE as movable pages so we can get those
2412 * areas back if necessary. Otherwise, we may have to free
2413 * excessively into the page allocator
2414 */
2415 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08002416 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002417 free_one_page(zone, page, pfn, 0, migratetype);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002418 goto out;
2419 }
2420 migratetype = MIGRATE_MOVABLE;
2421 }
2422
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002423 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Mel Gormanb745bc82014-06-04 16:10:22 -07002424 if (!cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002425 list_add(&page->lru, &pcp->lists[migratetype]);
Mel Gormanb745bc82014-06-04 16:10:22 -07002426 else
2427 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08002429 if (pcp->count >= pcp->high) {
Jason Low4db0c3c2015-04-15 16:14:08 -07002430 unsigned long batch = READ_ONCE(pcp->batch);
Cody P Schafer998d39c2013-07-03 15:01:32 -07002431 free_pcppages_bulk(zone, batch, pcp);
2432 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08002433 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002434
2435out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437}
2438
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002439/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002440 * Free a list of 0-order pages
2441 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002442void free_hot_cold_page_list(struct list_head *list, bool cold)
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002443{
2444 struct page *page, *next;
2445
2446 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08002447 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002448 free_hot_cold_page(page, cold);
2449 }
2450}
2451
2452/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002453 * split_page takes a non-compound higher-order page, and splits it into
2454 * n (1<<order) sub-pages: page[0..n]
2455 * Each sub-page must be freed individually.
2456 *
2457 * Note: this is probably too low level an operation for use in drivers.
2458 * Please consult with lkml before using this in your driver.
2459 */
2460void split_page(struct page *page, unsigned int order)
2461{
2462 int i;
2463
Sasha Levin309381fea2014-01-23 15:52:54 -08002464 VM_BUG_ON_PAGE(PageCompound(page), page);
2465 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002466
2467#ifdef CONFIG_KMEMCHECK
2468 /*
2469 * Split shadow pages too, because free(page[0]) would
2470 * otherwise free the whole shadow.
2471 */
2472 if (kmemcheck_page_is_tracked(page))
2473 split_page(virt_to_page(page[0].shadow), order);
2474#endif
2475
Joonsoo Kima9627bc2016-07-26 15:23:49 -07002476 for (i = 1; i < (1 << order); i++)
Nick Piggin7835e982006-03-22 00:08:40 -08002477 set_page_refcounted(page + i);
Joonsoo Kima9627bc2016-07-26 15:23:49 -07002478 split_page_owner(page, order);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002479}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07002480EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002481
Joonsoo Kim3c605092014-11-13 15:19:21 -08002482int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07002483{
Mel Gorman748446b2010-05-24 14:32:27 -07002484 unsigned long watermark;
2485 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002486 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07002487
2488 BUG_ON(!PageBuddy(page));
2489
2490 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002491 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07002492
Minchan Kim194159f2013-02-22 16:33:58 -08002493 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002494 /* Obey watermarks as if the page was being allocated */
2495 watermark = low_wmark_pages(zone) + (1 << order);
2496 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2497 return 0;
2498
Mel Gorman8fb74b92013-01-11 14:32:16 -08002499 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002500 }
Mel Gorman748446b2010-05-24 14:32:27 -07002501
2502 /* Remove page from free list */
2503 list_del(&page->lru);
2504 zone->free_area[order].nr_free--;
2505 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002506
zhong jiang400bc7f2016-07-28 15:45:07 -07002507 /*
2508 * Set the pageblock if the isolated page is at least half of a
2509 * pageblock
2510 */
Mel Gorman748446b2010-05-24 14:32:27 -07002511 if (order >= pageblock_order - 1) {
2512 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002513 for (; page < endpage; page += pageblock_nr_pages) {
2514 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08002515 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002516 set_pageblock_migratetype(page,
2517 MIGRATE_MOVABLE);
2518 }
Mel Gorman748446b2010-05-24 14:32:27 -07002519 }
2520
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002521
Mel Gorman8fb74b92013-01-11 14:32:16 -08002522 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002523}
2524
2525/*
Mel Gorman060e7412016-05-19 17:13:27 -07002526 * Update NUMA hit/miss statistics
2527 *
2528 * Must be called with interrupts disabled.
2529 *
2530 * When __GFP_OTHER_NODE is set assume the node of the preferred
2531 * zone is the local node. This is useful for daemons who allocate
2532 * memory on behalf of other processes.
2533 */
2534static inline void zone_statistics(struct zone *preferred_zone, struct zone *z,
2535 gfp_t flags)
2536{
2537#ifdef CONFIG_NUMA
2538 int local_nid = numa_node_id();
2539 enum zone_stat_item local_stat = NUMA_LOCAL;
2540
2541 if (unlikely(flags & __GFP_OTHER_NODE)) {
2542 local_stat = NUMA_OTHER;
2543 local_nid = preferred_zone->node;
2544 }
2545
2546 if (z->node == local_nid) {
2547 __inc_zone_state(z, NUMA_HIT);
2548 __inc_zone_state(z, local_stat);
2549 } else {
2550 __inc_zone_state(z, NUMA_MISS);
2551 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
2552 }
2553#endif
2554}
2555
2556/*
Vlastimil Babka75379192015-02-11 15:25:38 -08002557 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07002559static inline
2560struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002561 struct zone *zone, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07002562 gfp_t gfp_flags, unsigned int alloc_flags,
2563 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564{
2565 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08002566 struct page *page;
Mel Gormanb745bc82014-06-04 16:10:22 -07002567 bool cold = ((gfp_flags & __GFP_COLD) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568
Nick Piggin48db57f2006-01-08 01:00:42 -08002569 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002571 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 local_irq_save(flags);
Mel Gorman479f8542016-05-19 17:14:35 -07002574 do {
2575 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2576 list = &pcp->lists[migratetype];
2577 if (list_empty(list)) {
2578 pcp->count += rmqueue_bulk(zone, 0,
2579 pcp->batch, list,
2580 migratetype, cold);
2581 if (unlikely(list_empty(list)))
2582 goto failed;
2583 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07002584
Mel Gorman479f8542016-05-19 17:14:35 -07002585 if (cold)
2586 page = list_last_entry(list, struct page, lru);
2587 else
2588 page = list_first_entry(list, struct page, lru);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002589
Vlastimil Babka83b93552016-06-03 14:55:52 -07002590 __dec_zone_state(zone, NR_ALLOC_BATCH);
2591 list_del(&page->lru);
2592 pcp->count--;
2593
2594 } while (check_new_pcp(page));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002595 } else {
Michal Hocko0f352e52016-03-17 14:19:32 -07002596 /*
2597 * We most definitely don't want callers attempting to
2598 * allocate greater than order-1 page units with __GFP_NOFAIL.
2599 */
2600 WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 spin_lock_irqsave(&zone->lock, flags);
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002602
Mel Gorman479f8542016-05-19 17:14:35 -07002603 do {
2604 page = NULL;
2605 if (alloc_flags & ALLOC_HARDER) {
2606 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
2607 if (page)
2608 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2609 }
2610 if (!page)
2611 page = __rmqueue(zone, order, migratetype);
2612 } while (page && check_new_pages(page, order));
Nick Piggina74609f2006-01-06 00:11:20 -08002613 spin_unlock(&zone->lock);
2614 if (!page)
2615 goto failed;
Mel Gorman754078e2016-05-19 17:13:33 -07002616 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07002617 __mod_zone_freepage_state(zone, -(1 << order),
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002618 get_pcppage_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 }
2620
Johannes Weinerabe5f972014-10-02 16:21:10 -07002621 if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
Johannes Weiner57054652014-10-09 15:28:17 -07002622 !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
2623 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Johannes Weiner27329362014-03-03 15:38:41 -08002624
Christoph Lameterf8891e52006-06-30 01:55:45 -07002625 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07002626 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002627 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628
Sasha Levin309381fea2014-01-23 15:52:54 -08002629 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08002631
2632failed:
2633 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002634 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635}
2636
Akinobu Mita933e3122006-12-08 02:39:45 -08002637#ifdef CONFIG_FAIL_PAGE_ALLOC
2638
Akinobu Mitab2588c42011-07-26 16:09:03 -07002639static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08002640 struct fault_attr attr;
2641
Viresh Kumar621a5f72015-09-26 15:04:07 -07002642 bool ignore_gfp_highmem;
Mel Gorman71baba42015-11-06 16:28:28 -08002643 bool ignore_gfp_reclaim;
Akinobu Mita54114992007-07-15 23:40:23 -07002644 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08002645} fail_page_alloc = {
2646 .attr = FAULT_ATTR_INITIALIZER,
Mel Gorman71baba42015-11-06 16:28:28 -08002647 .ignore_gfp_reclaim = true,
Viresh Kumar621a5f72015-09-26 15:04:07 -07002648 .ignore_gfp_highmem = true,
Akinobu Mita54114992007-07-15 23:40:23 -07002649 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08002650};
2651
2652static int __init setup_fail_page_alloc(char *str)
2653{
2654 return setup_fault_attr(&fail_page_alloc.attr, str);
2655}
2656__setup("fail_page_alloc=", setup_fail_page_alloc);
2657
Gavin Shandeaf3862012-07-31 16:41:51 -07002658static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002659{
Akinobu Mita54114992007-07-15 23:40:23 -07002660 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07002661 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002662 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07002663 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002664 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002665 return false;
Mel Gorman71baba42015-11-06 16:28:28 -08002666 if (fail_page_alloc.ignore_gfp_reclaim &&
2667 (gfp_mask & __GFP_DIRECT_RECLAIM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002668 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002669
2670 return should_fail(&fail_page_alloc.attr, 1 << order);
2671}
2672
2673#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2674
2675static int __init fail_page_alloc_debugfs(void)
2676{
Al Virof4ae40a2011-07-24 04:33:43 -04002677 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08002678 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08002679
Akinobu Mitadd48c082011-08-03 16:21:01 -07002680 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2681 &fail_page_alloc.attr);
2682 if (IS_ERR(dir))
2683 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002684
Akinobu Mitab2588c42011-07-26 16:09:03 -07002685 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
Mel Gorman71baba42015-11-06 16:28:28 -08002686 &fail_page_alloc.ignore_gfp_reclaim))
Akinobu Mitab2588c42011-07-26 16:09:03 -07002687 goto fail;
2688 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2689 &fail_page_alloc.ignore_gfp_highmem))
2690 goto fail;
2691 if (!debugfs_create_u32("min-order", mode, dir,
2692 &fail_page_alloc.min_order))
2693 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08002694
Akinobu Mitab2588c42011-07-26 16:09:03 -07002695 return 0;
2696fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07002697 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002698
Akinobu Mitab2588c42011-07-26 16:09:03 -07002699 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08002700}
2701
2702late_initcall(fail_page_alloc_debugfs);
2703
2704#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2705
2706#else /* CONFIG_FAIL_PAGE_ALLOC */
2707
Gavin Shandeaf3862012-07-31 16:41:51 -07002708static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002709{
Gavin Shandeaf3862012-07-31 16:41:51 -07002710 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002711}
2712
2713#endif /* CONFIG_FAIL_PAGE_ALLOC */
2714
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715/*
Mel Gorman97a16fc2015-11-06 16:28:40 -08002716 * Return true if free base pages are above 'mark'. For high-order checks it
2717 * will return true of the order-0 watermark is reached and there is at least
2718 * one free page of a suitable size. Checking now avoids taking the zone lock
2719 * to check in the allocation paths if no pages are free.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 */
Michal Hocko86a294a2016-05-20 16:57:12 -07002721bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2722 int classzone_idx, unsigned int alloc_flags,
2723 long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724{
Christoph Lameterd23ad422007-02-10 01:43:02 -08002725 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 int o;
Mel Gormanc6038442016-05-19 17:13:38 -07002727 const bool alloc_harder = (alloc_flags & ALLOC_HARDER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002729 /* free_pages may go negative - that's OK */
Michal Hockodf0a6da2012-01-10 15:08:02 -08002730 free_pages -= (1 << order) - 1;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002731
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002732 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 min -= min / 2;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002734
2735 /*
2736 * If the caller does not have rights to ALLOC_HARDER then subtract
2737 * the high-atomic reserves. This will over-estimate the size of the
2738 * atomic reserve but it avoids a search.
2739 */
Mel Gorman97a16fc2015-11-06 16:28:40 -08002740 if (likely(!alloc_harder))
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002741 free_pages -= z->nr_reserved_highatomic;
2742 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 min -= min / 4;
Mel Gormane2b19192015-11-06 16:28:09 -08002744
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002745#ifdef CONFIG_CMA
2746 /* If allocation can't use CMA areas don't use free CMA pages */
2747 if (!(alloc_flags & ALLOC_CMA))
Mel Gorman97a16fc2015-11-06 16:28:40 -08002748 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002749#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002750
Mel Gorman97a16fc2015-11-06 16:28:40 -08002751 /*
2752 * Check watermarks for an order-0 allocation request. If these
2753 * are not met, then a high-order request also cannot go ahead
2754 * even if a suitable page happened to be free.
2755 */
2756 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08002757 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758
Mel Gorman97a16fc2015-11-06 16:28:40 -08002759 /* If this is an order-0 request then the watermark is fine */
2760 if (!order)
2761 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762
Mel Gorman97a16fc2015-11-06 16:28:40 -08002763 /* For a high-order request, check at least one suitable page is free */
2764 for (o = order; o < MAX_ORDER; o++) {
2765 struct free_area *area = &z->free_area[o];
2766 int mt;
2767
2768 if (!area->nr_free)
2769 continue;
2770
2771 if (alloc_harder)
2772 return true;
2773
2774 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
2775 if (!list_empty(&area->free_list[mt]))
2776 return true;
2777 }
2778
2779#ifdef CONFIG_CMA
2780 if ((alloc_flags & ALLOC_CMA) &&
2781 !list_empty(&area->free_list[MIGRATE_CMA])) {
2782 return true;
2783 }
2784#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 }
Mel Gorman97a16fc2015-11-06 16:28:40 -08002786 return false;
Mel Gorman88f5acf2011-01-13 15:45:41 -08002787}
2788
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002789bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Mel Gormanc6038442016-05-19 17:13:38 -07002790 int classzone_idx, unsigned int alloc_flags)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002791{
2792 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2793 zone_page_state(z, NR_FREE_PAGES));
2794}
2795
Mel Gorman48ee5f32016-05-19 17:14:07 -07002796static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
2797 unsigned long mark, int classzone_idx, unsigned int alloc_flags)
2798{
2799 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2800 long cma_pages = 0;
2801
2802#ifdef CONFIG_CMA
2803 /* If allocation can't use CMA areas don't use free CMA pages */
2804 if (!(alloc_flags & ALLOC_CMA))
2805 cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES);
2806#endif
2807
2808 /*
2809 * Fast check for order-0 only. If this fails then the reserves
2810 * need to be calculated. There is a corner case where the check
2811 * passes but only the high-order atomic reserve are free. If
2812 * the caller is !atomic then it'll uselessly search the free
2813 * list. That corner case is then slower but it is harmless.
2814 */
2815 if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx])
2816 return true;
2817
2818 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2819 free_pages);
2820}
2821
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002822bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
Mel Gormane2b19192015-11-06 16:28:09 -08002823 unsigned long mark, int classzone_idx)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002824{
2825 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2826
2827 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2828 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2829
Mel Gormane2b19192015-11-06 16:28:09 -08002830 return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002831 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832}
2833
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002834#ifdef CONFIG_NUMA
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002835static bool zone_local(struct zone *local_zone, struct zone *zone)
2836{
Johannes Weinerfff40682013-12-20 14:54:12 +00002837 return local_zone->node == zone->node;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002838}
2839
David Rientjes957f8222012-10-08 16:33:24 -07002840static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2841{
Mel Gorman5f7a75a2014-06-04 16:07:15 -07002842 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2843 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07002844}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002845#else /* CONFIG_NUMA */
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002846static bool zone_local(struct zone *local_zone, struct zone *zone)
2847{
2848 return true;
2849}
2850
David Rientjes957f8222012-10-08 16:33:24 -07002851static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2852{
2853 return true;
2854}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002855#endif /* CONFIG_NUMA */
2856
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002857static void reset_alloc_batches(struct zone *preferred_zone)
2858{
2859 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2860
2861 do {
2862 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2863 high_wmark_pages(zone) - low_wmark_pages(zone) -
2864 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner57054652014-10-09 15:28:17 -07002865 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002866 } while (zone++ != preferred_zone);
2867}
2868
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002869/*
Paul Jackson0798e512006-12-06 20:31:38 -08002870 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002871 * a page.
2872 */
2873static struct page *
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002874get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2875 const struct alloc_context *ac)
Martin Hicks753ee722005-06-21 17:14:41 -07002876{
Mel Gormanc33d6c02016-05-19 17:14:10 -07002877 struct zoneref *z = ac->preferred_zoneref;
Mel Gorman5117f452009-06-16 15:31:59 -07002878 struct zone *zone;
Mel Gorman30534752016-05-19 17:14:04 -07002879 bool fair_skipped = false;
2880 bool apply_fair = (alloc_flags & ALLOC_FAIR);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002881
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002882zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002883 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002884 * Scan zonelist, looking for a zone with enough free.
Vladimir Davydov344736f2014-10-20 15:50:30 +04002885 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002886 */
Mel Gormanc33d6c02016-05-19 17:14:10 -07002887 for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002888 ac->nodemask) {
Mel Gormanbe06af02016-05-19 17:13:47 -07002889 struct page *page;
Johannes Weinere085dbc2013-09-11 14:20:46 -07002890 unsigned long mark;
2891
Mel Gorman664eedd2014-06-04 16:10:08 -07002892 if (cpusets_enabled() &&
2893 (alloc_flags & ALLOC_CPUSET) &&
Vlastimil Babka002f2902016-05-19 17:14:30 -07002894 !__cpuset_zone_allowed(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07002895 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08002896 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002897 * Distribute pages in proportion to the individual
2898 * zone size to ensure fair page aging. The zone a
2899 * page was allocated in should have no effect on the
2900 * time the page has in memory before being reclaimed.
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002901 */
Mel Gorman30534752016-05-19 17:14:04 -07002902 if (apply_fair) {
Johannes Weiner57054652014-10-09 15:28:17 -07002903 if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
Mel Gormanfa379b92016-05-19 17:13:41 -07002904 fair_skipped = true;
Johannes Weiner3a025762014-04-07 15:37:48 -07002905 continue;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002906 }
Mel Gormanc33d6c02016-05-19 17:14:10 -07002907 if (!zone_local(ac->preferred_zoneref->zone, zone)) {
Mel Gorman30534752016-05-19 17:14:04 -07002908 if (fair_skipped)
2909 goto reset_fair;
2910 apply_fair = false;
2911 }
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002912 }
2913 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08002914 * When allocating a page cache page for writing, we
2915 * want to get it from a zone that is within its dirty
2916 * limit, such that no single zone holds more than its
2917 * proportional share of globally allowed dirty pages.
2918 * The dirty limits take into account the zone's
2919 * lowmem reserves and high watermark so that kswapd
2920 * should be able to balance it without having to
2921 * write pages from its LRU list.
2922 *
2923 * This may look like it could increase pressure on
2924 * lower zones by failing allocations in higher zones
2925 * before they are full. But the pages that do spill
2926 * over are limited as the lower zones are protected
2927 * by this very same mechanism. It should not become
2928 * a practical burden to them.
2929 *
2930 * XXX: For now, allow allocations to potentially
2931 * exceed the per-zone dirty limit in the slowpath
Mel Gormanc9ab0c42015-11-06 16:28:12 -08002932 * (spread_dirty_pages unset) before going into reclaim,
Johannes Weinera756cf52012-01-10 15:07:49 -08002933 * which is important when on a NUMA setup the allowed
2934 * zones are together not big enough to reach the
2935 * global limit. The proper fix for these situations
2936 * will require awareness of zones in the
2937 * dirty-throttling and the flusher threads.
2938 */
Mel Gormanc9ab0c42015-11-06 16:28:12 -08002939 if (ac->spread_dirty_pages && !zone_dirty_ok(zone))
Mel Gorman800a1e72014-06-04 16:10:06 -07002940 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002941
Johannes Weinere085dbc2013-09-11 14:20:46 -07002942 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gorman48ee5f32016-05-19 17:14:07 -07002943 if (!zone_watermark_fast(zone, order, mark,
Mel Gorman93ea9962016-05-19 17:14:13 -07002944 ac_classzone_idx(ac), alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07002945 int ret;
2946
Mel Gorman5dab2912014-06-04 16:10:14 -07002947 /* Checked here to keep the fast path fast */
2948 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2949 if (alloc_flags & ALLOC_NO_WATERMARKS)
2950 goto try_this_zone;
2951
David Rientjes957f8222012-10-08 16:33:24 -07002952 if (zone_reclaim_mode == 0 ||
Mel Gormanc33d6c02016-05-19 17:14:10 -07002953 !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
Mel Gormancd38b112011-07-25 17:12:29 -07002954 continue;
2955
Mel Gormanfa5e0842009-06-16 15:33:22 -07002956 ret = zone_reclaim(zone, gfp_mask, order);
2957 switch (ret) {
2958 case ZONE_RECLAIM_NOSCAN:
2959 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07002960 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002961 case ZONE_RECLAIM_FULL:
2962 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002963 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002964 default:
2965 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002966 if (zone_watermark_ok(zone, order, mark,
Mel Gorman93ea9962016-05-19 17:14:13 -07002967 ac_classzone_idx(ac), alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002968 goto try_this_zone;
2969
Mel Gormanfed27192013-04-29 15:07:57 -07002970 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002971 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002972 }
2973
Mel Gormanfa5e0842009-06-16 15:33:22 -07002974try_this_zone:
Mel Gormanc33d6c02016-05-19 17:14:10 -07002975 page = buffered_rmqueue(ac->preferred_zoneref->zone, zone, order,
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002976 gfp_mask, alloc_flags, ac->migratetype);
Vlastimil Babka75379192015-02-11 15:25:38 -08002977 if (page) {
Mel Gorman479f8542016-05-19 17:14:35 -07002978 prep_new_page(page, order, gfp_mask, alloc_flags);
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002979
2980 /*
2981 * If this is a high-order atomic allocation then check
2982 * if the pageblock should be reserved for the future
2983 */
2984 if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
2985 reserve_highatomic_pageblock(page, zone, order);
2986
Vlastimil Babka75379192015-02-11 15:25:38 -08002987 return page;
2988 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002989 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002990
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002991 /*
2992 * The first pass makes sure allocations are spread fairly within the
2993 * local node. However, the local node might have free pages left
2994 * after the fairness batches are exhausted, and remote zones haven't
2995 * even been considered yet. Try once more without fairness, and
2996 * include remote zones now, before entering the slowpath and waking
2997 * kswapd: prefer spilling to a remote zone over swapping locally.
2998 */
Mel Gorman30534752016-05-19 17:14:04 -07002999 if (fair_skipped) {
3000reset_fair:
3001 apply_fair = false;
3002 fair_skipped = false;
Mel Gormanc33d6c02016-05-19 17:14:10 -07003003 reset_alloc_batches(ac->preferred_zoneref->zone);
Mel Gorman0d0bd892016-06-03 14:55:58 -07003004 z = ac->preferred_zoneref;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07003005 goto zonelist_scan;
Mel Gorman30534752016-05-19 17:14:04 -07003006 }
Mel Gorman4ffeaf32014-08-06 16:07:22 -07003007
3008 return NULL;
Martin Hicks753ee722005-06-21 17:14:41 -07003009}
3010
David Rientjes29423e772011-03-22 16:30:47 -07003011/*
3012 * Large machines with many possible nodes should not always dump per-node
3013 * meminfo in irq context.
3014 */
3015static inline bool should_suppress_show_mem(void)
3016{
3017 bool ret = false;
3018
3019#if NODES_SHIFT > 8
3020 ret = in_interrupt();
3021#endif
3022 return ret;
3023}
3024
Dave Hansena238ab52011-05-24 17:12:16 -07003025static DEFINE_RATELIMIT_STATE(nopage_rs,
3026 DEFAULT_RATELIMIT_INTERVAL,
3027 DEFAULT_RATELIMIT_BURST);
3028
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08003029void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...)
Dave Hansena238ab52011-05-24 17:12:16 -07003030{
Dave Hansena238ab52011-05-24 17:12:16 -07003031 unsigned int filter = SHOW_MEM_FILTER_NODES;
3032
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08003033 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
3034 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07003035 return;
3036
3037 /*
3038 * This documents exceptions given to allocations in certain
3039 * contexts that are allowed to allocate outside current's set
3040 * of allowed nodes.
3041 */
3042 if (!(gfp_mask & __GFP_NOMEMALLOC))
3043 if (test_thread_flag(TIF_MEMDIE) ||
3044 (current->flags & (PF_MEMALLOC | PF_EXITING)))
3045 filter &= ~SHOW_MEM_FILTER_NODES;
Mel Gormand0164ad2015-11-06 16:28:21 -08003046 if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
Dave Hansena238ab52011-05-24 17:12:16 -07003047 filter &= ~SHOW_MEM_FILTER_NODES;
3048
3049 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07003050 struct va_format vaf;
3051 va_list args;
3052
Dave Hansena238ab52011-05-24 17:12:16 -07003053 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07003054
3055 vaf.fmt = fmt;
3056 vaf.va = &args;
3057
3058 pr_warn("%pV", &vaf);
3059
Dave Hansena238ab52011-05-24 17:12:16 -07003060 va_end(args);
3061 }
3062
Vlastimil Babkac5c990e2016-03-15 14:56:02 -07003063 pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n",
3064 current->comm, order, gfp_mask, &gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07003065 dump_stack();
3066 if (!should_suppress_show_mem())
3067 show_mem(filter);
3068}
3069
Mel Gorman11e33f62009-06-16 15:31:57 -07003070static inline struct page *
3071__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003072 const struct alloc_context *ac, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07003073{
David Rientjes6e0fc462015-09-08 15:00:36 -07003074 struct oom_control oc = {
3075 .zonelist = ac->zonelist,
3076 .nodemask = ac->nodemask,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07003077 .memcg = NULL,
David Rientjes6e0fc462015-09-08 15:00:36 -07003078 .gfp_mask = gfp_mask,
3079 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07003080 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082
Johannes Weiner9879de72015-01-26 12:58:32 -08003083 *did_some_progress = 0;
3084
Johannes Weiner9879de72015-01-26 12:58:32 -08003085 /*
Johannes Weinerdc564012015-06-24 16:57:19 -07003086 * Acquire the oom lock. If that fails, somebody else is
3087 * making progress for us.
Johannes Weiner9879de72015-01-26 12:58:32 -08003088 */
Johannes Weinerdc564012015-06-24 16:57:19 -07003089 if (!mutex_trylock(&oom_lock)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08003090 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07003091 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092 return NULL;
3093 }
Jens Axboe6b1de912005-11-17 21:35:02 +01003094
Mel Gorman11e33f62009-06-16 15:31:57 -07003095 /*
3096 * Go through the zonelist yet one more time, keep very high watermark
3097 * here, this is only to catch a parallel oom killing, we must fail if
3098 * we're still under heavy pressure.
3099 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003100 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
3101 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003102 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07003103 goto out;
3104
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08003105 if (!(gfp_mask & __GFP_NOFAIL)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08003106 /* Coredumps can quickly deplete all memory reserves */
3107 if (current->flags & PF_DUMPCORE)
3108 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08003109 /* The OOM killer will not help higher order allocs */
3110 if (order > PAGE_ALLOC_COSTLY_ORDER)
3111 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07003112 /* The OOM killer does not needlessly kill tasks for lowmem */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003113 if (ac->high_zoneidx < ZONE_NORMAL)
David Rientjes03668b32010-08-09 17:18:54 -07003114 goto out;
Johannes Weiner90839052015-06-24 16:57:21 -07003115 if (pm_suspended_storage())
3116 goto out;
Michal Hocko3da88fb2016-05-19 17:13:09 -07003117 /*
3118 * XXX: GFP_NOFS allocations should rather fail than rely on
3119 * other request to make a forward progress.
3120 * We are in an unfortunate situation where out_of_memory cannot
3121 * do much for this context but let's try it to at least get
3122 * access to memory reserved if the current task is killed (see
3123 * out_of_memory). Once filesystems are ready to handle allocation
3124 * failures more gracefully we should just bail out here.
3125 */
3126
David Rientjes4167e9b2015-04-14 15:46:55 -07003127 /* The OOM killer may not free memory on a specific node */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08003128 if (gfp_mask & __GFP_THISNODE)
3129 goto out;
3130 }
Mel Gorman11e33f62009-06-16 15:31:57 -07003131 /* Exhausted what can be done so it's blamo time */
Michal Hocko5020e282016-01-14 15:20:36 -08003132 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
Michal Hockoc32b3cb2015-02-11 15:26:24 -08003133 *did_some_progress = 1;
Michal Hocko5020e282016-01-14 15:20:36 -08003134
3135 if (gfp_mask & __GFP_NOFAIL) {
3136 page = get_page_from_freelist(gfp_mask, order,
3137 ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac);
3138 /*
3139 * fallback to ignore cpuset restriction if our nodes
3140 * are depleted
3141 */
3142 if (!page)
3143 page = get_page_from_freelist(gfp_mask, order,
3144 ALLOC_NO_WATERMARKS, ac);
3145 }
3146 }
Mel Gorman11e33f62009-06-16 15:31:57 -07003147out:
Johannes Weinerdc564012015-06-24 16:57:19 -07003148 mutex_unlock(&oom_lock);
Mel Gorman11e33f62009-06-16 15:31:57 -07003149 return page;
3150}
3151
Michal Hocko33c2d212016-05-20 16:57:06 -07003152
3153/*
3154 * Maximum number of compaction retries wit a progress before OOM
3155 * killer is consider as the only way to move forward.
3156 */
3157#define MAX_COMPACT_RETRIES 16
3158
Mel Gorman56de7262010-05-24 14:32:30 -07003159#ifdef CONFIG_COMPACTION
3160/* Try memory compaction for high-order allocations before reclaim */
3161static struct page *
3162__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07003163 unsigned int alloc_flags, const struct alloc_context *ac,
Michal Hockoc5d01d02016-05-20 16:56:53 -07003164 enum migrate_mode mode, enum compact_result *compact_result)
Mel Gorman56de7262010-05-24 14:32:30 -07003165{
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003166 struct page *page;
Michal Hockoc5d01d02016-05-20 16:56:53 -07003167 int contended_compaction;
Vlastimil Babka53853e22014-10-09 15:27:02 -07003168
Mel Gorman66199712012-01-12 17:19:41 -08003169 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07003170 return NULL;
3171
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003172 current->flags |= PF_MEMALLOC;
Michal Hockoc5d01d02016-05-20 16:56:53 -07003173 *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
3174 mode, &contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003175 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07003176
Michal Hockoc5d01d02016-05-20 16:56:53 -07003177 if (*compact_result <= COMPACT_INACTIVE)
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003178 return NULL;
Mel Gorman56de7262010-05-24 14:32:30 -07003179
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003180 /*
3181 * At least in one zone compaction wasn't deferred or skipped, so let's
3182 * count a compaction stall
3183 */
3184 count_vm_event(COMPACTSTALL);
3185
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003186 page = get_page_from_freelist(gfp_mask, order,
3187 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003188
3189 if (page) {
3190 struct zone *zone = page_zone(page);
3191
3192 zone->compact_blockskip_flush = false;
3193 compaction_defer_reset(zone, order, true);
3194 count_vm_event(COMPACTSUCCESS);
3195 return page;
3196 }
3197
3198 /*
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003199 * It's bad if compaction run occurs and fails. The most likely reason
3200 * is that pages exist, but not enough to satisfy watermarks.
3201 */
3202 count_vm_event(COMPACTFAIL);
3203
Michal Hockoc5d01d02016-05-20 16:56:53 -07003204 /*
3205 * In all zones where compaction was attempted (and not
3206 * deferred or skipped), lock contention has been detected.
3207 * For THP allocation we do not want to disrupt the others
3208 * so we fallback to base pages instead.
3209 */
3210 if (contended_compaction == COMPACT_CONTENDED_LOCK)
3211 *compact_result = COMPACT_CONTENDED;
3212
3213 /*
3214 * If compaction was aborted due to need_resched(), we do not
3215 * want to further increase allocation latency, unless it is
3216 * khugepaged trying to collapse.
3217 */
3218 if (contended_compaction == COMPACT_CONTENDED_SCHED
3219 && !(current->flags & PF_KTHREAD))
3220 *compact_result = COMPACT_CONTENDED;
3221
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003222 cond_resched();
3223
Mel Gorman56de7262010-05-24 14:32:30 -07003224 return NULL;
3225}
Michal Hocko33c2d212016-05-20 16:57:06 -07003226
3227static inline bool
Michal Hocko86a294a2016-05-20 16:57:12 -07003228should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
3229 enum compact_result compact_result, enum migrate_mode *migrate_mode,
Michal Hocko33c2d212016-05-20 16:57:06 -07003230 int compaction_retries)
3231{
Michal Hocko7854ea62016-05-20 16:57:09 -07003232 int max_retries = MAX_COMPACT_RETRIES;
3233
Michal Hocko33c2d212016-05-20 16:57:06 -07003234 if (!order)
3235 return false;
3236
3237 /*
3238 * compaction considers all the zone as desperately out of memory
3239 * so it doesn't really make much sense to retry except when the
3240 * failure could be caused by weak migration mode.
3241 */
3242 if (compaction_failed(compact_result)) {
3243 if (*migrate_mode == MIGRATE_ASYNC) {
3244 *migrate_mode = MIGRATE_SYNC_LIGHT;
3245 return true;
3246 }
3247 return false;
3248 }
3249
3250 /*
Michal Hocko7854ea62016-05-20 16:57:09 -07003251 * make sure the compaction wasn't deferred or didn't bail out early
3252 * due to locks contention before we declare that we should give up.
Michal Hocko86a294a2016-05-20 16:57:12 -07003253 * But do not retry if the given zonelist is not suitable for
3254 * compaction.
Michal Hocko33c2d212016-05-20 16:57:06 -07003255 */
Michal Hocko7854ea62016-05-20 16:57:09 -07003256 if (compaction_withdrawn(compact_result))
Michal Hocko86a294a2016-05-20 16:57:12 -07003257 return compaction_zonelist_suitable(ac, order, alloc_flags);
Michal Hocko7854ea62016-05-20 16:57:09 -07003258
3259 /*
3260 * !costly requests are much more important than __GFP_REPEAT
3261 * costly ones because they are de facto nofail and invoke OOM
3262 * killer to move on while costly can fail and users are ready
3263 * to cope with that. 1/4 retries is rather arbitrary but we
3264 * would need much more detailed feedback from compaction to
3265 * make a better decision.
3266 */
3267 if (order > PAGE_ALLOC_COSTLY_ORDER)
3268 max_retries /= 4;
3269 if (compaction_retries <= max_retries)
3270 return true;
Michal Hocko33c2d212016-05-20 16:57:06 -07003271
3272 return false;
3273}
Mel Gorman56de7262010-05-24 14:32:30 -07003274#else
3275static inline struct page *
3276__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07003277 unsigned int alloc_flags, const struct alloc_context *ac,
Michal Hockoc5d01d02016-05-20 16:56:53 -07003278 enum migrate_mode mode, enum compact_result *compact_result)
Mel Gorman56de7262010-05-24 14:32:30 -07003279{
Michal Hocko33c2d212016-05-20 16:57:06 -07003280 *compact_result = COMPACT_SKIPPED;
Mel Gorman56de7262010-05-24 14:32:30 -07003281 return NULL;
3282}
Michal Hocko33c2d212016-05-20 16:57:06 -07003283
3284static inline bool
Michal Hocko86a294a2016-05-20 16:57:12 -07003285should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
3286 enum compact_result compact_result,
Michal Hocko33c2d212016-05-20 16:57:06 -07003287 enum migrate_mode *migrate_mode,
3288 int compaction_retries)
3289{
Michal Hocko31e49bf2016-05-20 16:57:15 -07003290 struct zone *zone;
3291 struct zoneref *z;
3292
3293 if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
3294 return false;
3295
3296 /*
3297 * There are setups with compaction disabled which would prefer to loop
3298 * inside the allocator rather than hit the oom killer prematurely.
3299 * Let's give them a good hope and keep retrying while the order-0
3300 * watermarks are OK.
3301 */
3302 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3303 ac->nodemask) {
3304 if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
3305 ac_classzone_idx(ac), alloc_flags))
3306 return true;
3307 }
Michal Hocko33c2d212016-05-20 16:57:06 -07003308 return false;
3309}
Mel Gorman56de7262010-05-24 14:32:30 -07003310#endif /* CONFIG_COMPACTION */
3311
Marek Szyprowskibba90712012-01-25 12:09:52 +01003312/* Perform direct synchronous page reclaim */
3313static int
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003314__perform_reclaim(gfp_t gfp_mask, unsigned int order,
3315 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003316{
Mel Gorman11e33f62009-06-16 15:31:57 -07003317 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01003318 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07003319
3320 cond_resched();
3321
3322 /* We now go into synchronous reclaim */
3323 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003324 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07003325 lockdep_set_current_reclaim_state(gfp_mask);
3326 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003327 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07003328
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003329 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
3330 ac->nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07003331
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003332 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07003333 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003334 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07003335
3336 cond_resched();
3337
Marek Szyprowskibba90712012-01-25 12:09:52 +01003338 return progress;
3339}
3340
3341/* The really slow allocator path where we enter direct reclaim */
3342static inline struct page *
3343__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07003344 unsigned int alloc_flags, const struct alloc_context *ac,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003345 unsigned long *did_some_progress)
Marek Szyprowskibba90712012-01-25 12:09:52 +01003346{
3347 struct page *page = NULL;
3348 bool drained = false;
3349
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003350 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003351 if (unlikely(!(*did_some_progress)))
3352 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07003353
Mel Gorman9ee493c2010-09-09 16:38:18 -07003354retry:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003355 page = get_page_from_freelist(gfp_mask, order,
3356 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003357
3358 /*
3359 * If an allocation failed after direct reclaim, it could be because
Mel Gorman0aaa29a2015-11-06 16:28:37 -08003360 * pages are pinned on the per-cpu lists or in high alloc reserves.
3361 * Shrink them them and try again
Mel Gorman9ee493c2010-09-09 16:38:18 -07003362 */
3363 if (!page && !drained) {
Mel Gorman0aaa29a2015-11-06 16:28:37 -08003364 unreserve_highatomic_pageblock(ac);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08003365 drain_all_pages(NULL);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003366 drained = true;
3367 goto retry;
3368 }
3369
Mel Gorman11e33f62009-06-16 15:31:57 -07003370 return page;
3371}
3372
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003373static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003374{
3375 struct zoneref *z;
3376 struct zone *zone;
3377
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003378 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
3379 ac->high_zoneidx, ac->nodemask)
Mel Gorman93ea9962016-05-19 17:14:13 -07003380 wakeup_kswapd(zone, order, ac_classzone_idx(ac));
Mel Gorman11e33f62009-06-16 15:31:57 -07003381}
3382
Mel Gormanc6038442016-05-19 17:13:38 -07003383static inline unsigned int
Peter Zijlstra341ce062009-06-16 15:32:02 -07003384gfp_to_alloc_flags(gfp_t gfp_mask)
3385{
Mel Gormanc6038442016-05-19 17:13:38 -07003386 unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
Peter Zijlstra341ce062009-06-16 15:32:02 -07003387
Mel Gormana56f57f2009-06-16 15:32:02 -07003388 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07003389 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07003390
Peter Zijlstra341ce062009-06-16 15:32:02 -07003391 /*
3392 * The caller may dip into page reserves a bit more if the caller
3393 * cannot run direct reclaim, or if the caller has realtime scheduling
3394 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
Mel Gormand0164ad2015-11-06 16:28:21 -08003395 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
Peter Zijlstra341ce062009-06-16 15:32:02 -07003396 */
Namhyung Kime6223a32010-10-26 14:21:59 -07003397 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07003398
Mel Gormand0164ad2015-11-06 16:28:21 -08003399 if (gfp_mask & __GFP_ATOMIC) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003400 /*
David Rientjesb104a352014-07-30 16:08:24 -07003401 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3402 * if it can't schedule.
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003403 */
David Rientjesb104a352014-07-30 16:08:24 -07003404 if (!(gfp_mask & __GFP_NOMEMALLOC))
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003405 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07003406 /*
David Rientjesb104a352014-07-30 16:08:24 -07003407 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
Vladimir Davydov344736f2014-10-20 15:50:30 +04003408 * comment for __cpuset_node_allowed().
Peter Zijlstra341ce062009-06-16 15:32:02 -07003409 */
3410 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003411 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07003412 alloc_flags |= ALLOC_HARDER;
3413
Mel Gormanb37f1dd2012-07-31 16:44:03 -07003414 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
3415 if (gfp_mask & __GFP_MEMALLOC)
3416 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07003417 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
3418 alloc_flags |= ALLOC_NO_WATERMARKS;
3419 else if (!in_interrupt() &&
3420 ((current->flags & PF_MEMALLOC) ||
3421 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07003422 alloc_flags |= ALLOC_NO_WATERMARKS;
3423 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07003424#ifdef CONFIG_CMA
David Rientjes43e7a342014-10-09 15:27:25 -07003425 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07003426 alloc_flags |= ALLOC_CMA;
3427#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07003428 return alloc_flags;
3429}
3430
Mel Gorman072bb0a2012-07-31 16:43:58 -07003431bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
3432{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07003433 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07003434}
3435
Mel Gormand0164ad2015-11-06 16:28:21 -08003436static inline bool is_thp_gfp_mask(gfp_t gfp_mask)
3437{
3438 return (gfp_mask & (GFP_TRANSHUGE | __GFP_KSWAPD_RECLAIM)) == GFP_TRANSHUGE;
3439}
3440
Michal Hocko0a0337e2016-05-20 16:57:00 -07003441/*
3442 * Maximum number of reclaim retries without any progress before OOM killer
3443 * is consider as the only way to move forward.
3444 */
3445#define MAX_RECLAIM_RETRIES 16
3446
3447/*
3448 * Checks whether it makes sense to retry the reclaim to make a forward progress
3449 * for the given allocation request.
3450 * The reclaim feedback represented by did_some_progress (any progress during
Michal Hocko7854ea62016-05-20 16:57:09 -07003451 * the last reclaim round) and no_progress_loops (number of reclaim rounds without
3452 * any progress in a row) is considered as well as the reclaimable pages on the
3453 * applicable zone list (with a backoff mechanism which is a function of
3454 * no_progress_loops).
Michal Hocko0a0337e2016-05-20 16:57:00 -07003455 *
3456 * Returns true if a retry is viable or false to enter the oom path.
3457 */
3458static inline bool
3459should_reclaim_retry(gfp_t gfp_mask, unsigned order,
3460 struct alloc_context *ac, int alloc_flags,
Michal Hocko7854ea62016-05-20 16:57:09 -07003461 bool did_some_progress, int no_progress_loops)
Michal Hocko0a0337e2016-05-20 16:57:00 -07003462{
3463 struct zone *zone;
3464 struct zoneref *z;
3465
3466 /*
3467 * Make sure we converge to OOM if we cannot make any progress
3468 * several times in the row.
3469 */
3470 if (no_progress_loops > MAX_RECLAIM_RETRIES)
3471 return false;
3472
Michal Hocko0a0337e2016-05-20 16:57:00 -07003473 /*
3474 * Keep reclaiming pages while there is a chance this will lead somewhere.
3475 * If none of the target zones can satisfy our allocation request even
3476 * if all reclaimable pages are considered then we are screwed and have
3477 * to go OOM.
3478 */
3479 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3480 ac->nodemask) {
3481 unsigned long available;
Michal Hockoede37712016-05-20 16:57:03 -07003482 unsigned long reclaimable;
Michal Hocko0a0337e2016-05-20 16:57:00 -07003483
Michal Hockoede37712016-05-20 16:57:03 -07003484 available = reclaimable = zone_reclaimable_pages(zone);
Michal Hocko0a0337e2016-05-20 16:57:00 -07003485 available -= DIV_ROUND_UP(no_progress_loops * available,
3486 MAX_RECLAIM_RETRIES);
3487 available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
3488
3489 /*
3490 * Would the allocation succeed if we reclaimed the whole
3491 * available?
3492 */
3493 if (__zone_watermark_ok(zone, order, min_wmark_pages(zone),
Michal Hockoede37712016-05-20 16:57:03 -07003494 ac_classzone_idx(ac), alloc_flags, available)) {
3495 /*
3496 * If we didn't make any progress and have a lot of
3497 * dirty + writeback pages then we should wait for
3498 * an IO to complete to slow down the reclaim and
3499 * prevent from pre mature OOM
3500 */
3501 if (!did_some_progress) {
3502 unsigned long writeback;
3503 unsigned long dirty;
3504
3505 writeback = zone_page_state_snapshot(zone,
3506 NR_WRITEBACK);
3507 dirty = zone_page_state_snapshot(zone, NR_FILE_DIRTY);
3508
3509 if (2*(writeback + dirty) > reclaimable) {
3510 congestion_wait(BLK_RW_ASYNC, HZ/10);
3511 return true;
3512 }
3513 }
3514
3515 /*
3516 * Memory allocation/reclaim might be called from a WQ
3517 * context and the current implementation of the WQ
3518 * concurrency control doesn't recognize that
3519 * a particular WQ is congested if the worker thread is
3520 * looping without ever sleeping. Therefore we have to
3521 * do a short sleep here rather than calling
3522 * cond_resched().
3523 */
3524 if (current->flags & PF_WQ_WORKER)
3525 schedule_timeout_uninterruptible(1);
3526 else
3527 cond_resched();
3528
Michal Hocko0a0337e2016-05-20 16:57:00 -07003529 return true;
3530 }
3531 }
3532
3533 return false;
3534}
3535
Mel Gorman11e33f62009-06-16 15:31:57 -07003536static inline struct page *
3537__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003538 struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003539{
Mel Gormand0164ad2015-11-06 16:28:21 -08003540 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
Mel Gorman11e33f62009-06-16 15:31:57 -07003541 struct page *page = NULL;
Mel Gormanc6038442016-05-19 17:13:38 -07003542 unsigned int alloc_flags;
Mel Gorman11e33f62009-06-16 15:31:57 -07003543 unsigned long did_some_progress;
David Rientjese0b9dae2014-06-04 16:08:28 -07003544 enum migrate_mode migration_mode = MIGRATE_ASYNC;
Michal Hockoc5d01d02016-05-20 16:56:53 -07003545 enum compact_result compact_result;
Michal Hocko33c2d212016-05-20 16:57:06 -07003546 int compaction_retries = 0;
Michal Hocko0a0337e2016-05-20 16:57:00 -07003547 int no_progress_loops = 0;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003548
Christoph Lameter952f3b52006-12-06 20:33:26 -08003549 /*
Mel Gorman72807a72009-06-16 15:32:18 -07003550 * In the slowpath, we sanity check order to avoid ever trying to
3551 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3552 * be using allocators in order of preference for an area that is
3553 * too large.
3554 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07003555 if (order >= MAX_ORDER) {
3556 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07003557 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07003558 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003559
Christoph Lameter952f3b52006-12-06 20:33:26 -08003560 /*
Mel Gormand0164ad2015-11-06 16:28:21 -08003561 * We also sanity check to catch abuse of atomic reserves being used by
3562 * callers that are not in atomic context.
3563 */
3564 if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
3565 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
3566 gfp_mask &= ~__GFP_ATOMIC;
3567
Johannes Weiner9879de72015-01-26 12:58:32 -08003568retry:
Mel Gormand0164ad2015-11-06 16:28:21 -08003569 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003570 wake_all_kswapds(order, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003571
Paul Jackson9bf22292005-09-06 15:18:12 -07003572 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003573 * OK, we're below the kswapd watermark and have kicked background
3574 * reclaim. Now things get more complex, so set up alloc_flags according
3575 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07003576 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07003577 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578
Mel Gormane46e7b72016-06-03 14:56:01 -07003579 /*
3580 * Reset the zonelist iterators if memory policies can be ignored.
3581 * These allocations are high priority and system rather than user
3582 * orientated.
3583 */
3584 if ((alloc_flags & ALLOC_NO_WATERMARKS) || !(alloc_flags & ALLOC_CPUSET)) {
3585 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
3586 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
3587 ac->high_zoneidx, ac->nodemask);
3588 }
3589
Peter Zijlstra341ce062009-06-16 15:32:02 -07003590 /* This is the last chance, in general, before the goto nopage. */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003591 page = get_page_from_freelist(gfp_mask, order,
3592 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003593 if (page)
3594 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595
Mel Gorman11e33f62009-06-16 15:31:57 -07003596 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07003597 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Michal Hocko33d53102016-01-14 15:19:05 -08003598 page = get_page_from_freelist(gfp_mask, order,
3599 ALLOC_NO_WATERMARKS, ac);
3600 if (page)
3601 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003602 }
3603
Mel Gormand0164ad2015-11-06 16:28:21 -08003604 /* Caller is not willing to reclaim, we can't balance anything */
3605 if (!can_direct_reclaim) {
David Rientjesaed0a0e2014-01-21 15:51:12 -08003606 /*
Michal Hocko33d53102016-01-14 15:19:05 -08003607 * All existing users of the __GFP_NOFAIL are blockable, so warn
3608 * of any new users that actually allow this type of allocation
3609 * to fail.
David Rientjesaed0a0e2014-01-21 15:51:12 -08003610 */
3611 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612 goto nopage;
David Rientjesaed0a0e2014-01-21 15:51:12 -08003613 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614
Peter Zijlstra341ce062009-06-16 15:32:02 -07003615 /* Avoid recursion of direct reclaim */
Michal Hocko33d53102016-01-14 15:19:05 -08003616 if (current->flags & PF_MEMALLOC) {
3617 /*
3618 * __GFP_NOFAIL request from this context is rather bizarre
3619 * because we cannot reclaim anything and only can loop waiting
3620 * for somebody to do a work for us.
3621 */
3622 if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3623 cond_resched();
3624 goto retry;
3625 }
Peter Zijlstra341ce062009-06-16 15:32:02 -07003626 goto nopage;
Michal Hocko33d53102016-01-14 15:19:05 -08003627 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003628
David Rientjes6583bb62009-07-29 15:02:06 -07003629 /* Avoid allocations with no watermarks from looping endlessly */
3630 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3631 goto nopage;
3632
Mel Gorman77f1fe62011-01-13 15:45:57 -08003633 /*
3634 * Try direct compaction. The first pass is asynchronous. Subsequent
3635 * attempts after direct reclaim are synchronous
3636 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003637 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3638 migration_mode,
Michal Hockoc5d01d02016-05-20 16:56:53 -07003639 &compact_result);
Mel Gorman56de7262010-05-24 14:32:30 -07003640 if (page)
3641 goto got_pg;
David Rientjes75f30862014-06-04 16:08:30 -07003642
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003643 /* Checks for THP-specific high-order allocations */
Mel Gormand0164ad2015-11-06 16:28:21 -08003644 if (is_thp_gfp_mask(gfp_mask)) {
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003645 /*
3646 * If compaction is deferred for high-order allocations, it is
3647 * because sync compaction recently failed. If this is the case
3648 * and the caller requested a THP allocation, we do not want
3649 * to heavily disrupt the system, so we fail the allocation
3650 * instead of entering direct reclaim.
3651 */
Michal Hockoc5d01d02016-05-20 16:56:53 -07003652 if (compact_result == COMPACT_DEFERRED)
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003653 goto nopage;
3654
3655 /*
Michal Hockoc5d01d02016-05-20 16:56:53 -07003656 * Compaction is contended so rather back off than cause
3657 * excessive stalls.
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003658 */
Michal Hockoc5d01d02016-05-20 16:56:53 -07003659 if(compact_result == COMPACT_CONTENDED)
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003660 goto nopage;
3661 }
Mel Gorman66199712012-01-12 17:19:41 -08003662
Michal Hocko33c2d212016-05-20 16:57:06 -07003663 if (order && compaction_made_progress(compact_result))
3664 compaction_retries++;
David Rientjes8fe78042014-08-06 16:07:54 -07003665
Mel Gorman11e33f62009-06-16 15:31:57 -07003666 /* Try direct reclaim and then allocating */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003667 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3668 &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07003669 if (page)
3670 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003671
Johannes Weiner90839052015-06-24 16:57:21 -07003672 /* Do not loop if specifically requested */
3673 if (gfp_mask & __GFP_NORETRY)
3674 goto noretry;
3675
Michal Hocko0a0337e2016-05-20 16:57:00 -07003676 /*
3677 * Do not retry costly high order allocations unless they are
3678 * __GFP_REPEAT
3679 */
3680 if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_REPEAT))
3681 goto noretry;
3682
Michal Hocko7854ea62016-05-20 16:57:09 -07003683 /*
3684 * Costly allocations might have made a progress but this doesn't mean
3685 * their order will become available due to high fragmentation so
3686 * always increment the no progress counter for them
3687 */
3688 if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
Michal Hocko0a0337e2016-05-20 16:57:00 -07003689 no_progress_loops = 0;
Michal Hocko7854ea62016-05-20 16:57:09 -07003690 else
Michal Hocko0a0337e2016-05-20 16:57:00 -07003691 no_progress_loops++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692
Michal Hocko0a0337e2016-05-20 16:57:00 -07003693 if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
Michal Hocko7854ea62016-05-20 16:57:09 -07003694 did_some_progress > 0, no_progress_loops))
Michal Hocko0a0337e2016-05-20 16:57:00 -07003695 goto retry;
3696
Michal Hocko33c2d212016-05-20 16:57:06 -07003697 /*
3698 * It doesn't make any sense to retry for the compaction if the order-0
3699 * reclaim is not able to make any progress because the current
3700 * implementation of the compaction depends on the sufficient amount
3701 * of free memory (see __compaction_suitable)
3702 */
3703 if (did_some_progress > 0 &&
Michal Hocko86a294a2016-05-20 16:57:12 -07003704 should_compact_retry(ac, order, alloc_flags,
3705 compact_result, &migration_mode,
3706 compaction_retries))
Michal Hocko33c2d212016-05-20 16:57:06 -07003707 goto retry;
3708
Johannes Weiner90839052015-06-24 16:57:21 -07003709 /* Reclaim has failed us, start killing things */
3710 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3711 if (page)
3712 goto got_pg;
3713
3714 /* Retry as long as the OOM killer is making progress */
Michal Hocko0a0337e2016-05-20 16:57:00 -07003715 if (did_some_progress) {
3716 no_progress_loops = 0;
Johannes Weiner90839052015-06-24 16:57:21 -07003717 goto retry;
Michal Hocko0a0337e2016-05-20 16:57:00 -07003718 }
Johannes Weiner90839052015-06-24 16:57:21 -07003719
3720noretry:
3721 /*
Michal Hocko33c2d212016-05-20 16:57:06 -07003722 * High-order allocations do not necessarily loop after direct reclaim
3723 * and reclaim/compaction depends on compaction being called after
3724 * reclaim so call directly if necessary.
3725 * It can become very expensive to allocate transparent hugepages at
3726 * fault, so use asynchronous memory compaction for THP unless it is
3727 * khugepaged trying to collapse. All other requests should tolerate
3728 * at least light sync migration.
Johannes Weiner90839052015-06-24 16:57:21 -07003729 */
Michal Hocko33c2d212016-05-20 16:57:06 -07003730 if (is_thp_gfp_mask(gfp_mask) && !(current->flags & PF_KTHREAD))
3731 migration_mode = MIGRATE_ASYNC;
3732 else
3733 migration_mode = MIGRATE_SYNC_LIGHT;
Johannes Weiner90839052015-06-24 16:57:21 -07003734 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags,
3735 ac, migration_mode,
Michal Hockoc5d01d02016-05-20 16:56:53 -07003736 &compact_result);
Johannes Weiner90839052015-06-24 16:57:21 -07003737 if (page)
3738 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003739nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07003740 warn_alloc_failed(gfp_mask, order, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741got_pg:
Mel Gorman072bb0a2012-07-31 16:43:58 -07003742 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743}
Mel Gorman11e33f62009-06-16 15:31:57 -07003744
3745/*
3746 * This is the 'heart' of the zoned buddy allocator.
3747 */
3748struct page *
3749__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3750 struct zonelist *zonelist, nodemask_t *nodemask)
3751{
Mel Gorman5bb1b162016-05-19 17:13:50 -07003752 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003753 unsigned int cpuset_mems_cookie;
Mel Gormanc6038442016-05-19 17:13:38 -07003754 unsigned int alloc_flags = ALLOC_WMARK_LOW|ALLOC_FAIR;
Mel Gorman83d4ca82016-05-19 17:13:56 -07003755 gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003756 struct alloc_context ac = {
3757 .high_zoneidx = gfp_zone(gfp_mask),
Mel Gorman682a3382016-05-19 17:13:30 -07003758 .zonelist = zonelist,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003759 .nodemask = nodemask,
3760 .migratetype = gfpflags_to_migratetype(gfp_mask),
3761 };
Mel Gorman11e33f62009-06-16 15:31:57 -07003762
Mel Gorman682a3382016-05-19 17:13:30 -07003763 if (cpusets_enabled()) {
Mel Gorman83d4ca82016-05-19 17:13:56 -07003764 alloc_mask |= __GFP_HARDWALL;
Mel Gorman682a3382016-05-19 17:13:30 -07003765 alloc_flags |= ALLOC_CPUSET;
3766 if (!ac.nodemask)
3767 ac.nodemask = &cpuset_current_mems_allowed;
3768 }
3769
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003770 gfp_mask &= gfp_allowed_mask;
3771
Mel Gorman11e33f62009-06-16 15:31:57 -07003772 lockdep_trace_alloc(gfp_mask);
3773
Mel Gormand0164ad2015-11-06 16:28:21 -08003774 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
Mel Gorman11e33f62009-06-16 15:31:57 -07003775
3776 if (should_fail_alloc_page(gfp_mask, order))
3777 return NULL;
3778
3779 /*
3780 * Check the zones suitable for the gfp_mask contain at least one
3781 * valid zone. It's possible to have an empty zonelist as a result
David Rientjes4167e9b2015-04-14 15:46:55 -07003782 * of __GFP_THISNODE and a memoryless node
Mel Gorman11e33f62009-06-16 15:31:57 -07003783 */
3784 if (unlikely(!zonelist->_zonerefs->zone))
3785 return NULL;
3786
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003787 if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
Vlastimil Babka21bb9bd2014-10-09 15:26:51 -07003788 alloc_flags |= ALLOC_CMA;
3789
Mel Gormancc9a6c82012-03-21 16:34:11 -07003790retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07003791 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003792
Mel Gormanc9ab0c42015-11-06 16:28:12 -08003793 /* Dirty zone balancing only done in the fast path */
3794 ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE);
3795
Mel Gormane46e7b72016-06-03 14:56:01 -07003796 /*
3797 * The preferred zone is used for statistics but crucially it is
3798 * also used as the starting point for the zonelist iterator. It
3799 * may get reset for allocations that ignore memory policies.
3800 */
Mel Gormanc33d6c02016-05-19 17:14:10 -07003801 ac.preferred_zoneref = first_zones_zonelist(ac.zonelist,
3802 ac.high_zoneidx, ac.nodemask);
3803 if (!ac.preferred_zoneref) {
Mel Gorman5bb1b162016-05-19 17:13:50 -07003804 page = NULL;
Mel Gorman4fcb0972016-05-19 17:14:01 -07003805 goto no_zone;
Mel Gorman5bb1b162016-05-19 17:13:50 -07003806 }
3807
Mel Gorman5117f452009-06-16 15:31:59 -07003808 /* First allocation attempt */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003809 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
Mel Gorman4fcb0972016-05-19 17:14:01 -07003810 if (likely(page))
3811 goto out;
Andrew Morton91fbdc02015-02-11 15:25:04 -08003812
Mel Gorman4fcb0972016-05-19 17:14:01 -07003813 /*
3814 * Runtime PM, block IO and its error handling path can deadlock
3815 * because I/O on the device might not complete.
3816 */
3817 alloc_mask = memalloc_noio_flags(gfp_mask);
3818 ac.spread_dirty_pages = false;
Mel Gorman11e33f62009-06-16 15:31:57 -07003819
Mel Gorman47415262016-05-19 17:14:44 -07003820 /*
3821 * Restore the original nodemask if it was potentially replaced with
3822 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
3823 */
3824 if (cpusets_enabled())
3825 ac.nodemask = nodemask;
Mel Gorman4fcb0972016-05-19 17:14:01 -07003826 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
Xishi Qiu23f086f2015-02-11 15:25:07 -08003827
Mel Gorman4fcb0972016-05-19 17:14:01 -07003828no_zone:
Mel Gormancc9a6c82012-03-21 16:34:11 -07003829 /*
3830 * When updating a task's mems_allowed, it is possible to race with
3831 * parallel threads in such a way that an allocation can fail while
3832 * the mask is being updated. If a page allocation is about to fail,
3833 * check if the cpuset changed during allocation and if so, retry.
3834 */
Mel Gorman83d4ca82016-05-19 17:13:56 -07003835 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) {
3836 alloc_mask = gfp_mask;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003837 goto retry_cpuset;
Mel Gorman83d4ca82016-05-19 17:13:56 -07003838 }
Mel Gormancc9a6c82012-03-21 16:34:11 -07003839
Mel Gorman4fcb0972016-05-19 17:14:01 -07003840out:
Vladimir Davydov49491482016-07-26 15:24:24 -07003841 if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page) {
3842 if (unlikely(memcg_kmem_charge(page, gfp_mask, order))) {
3843 __free_pages(page, order);
3844 page = NULL;
3845 } else
3846 __SetPageKmemcg(page);
3847 }
3848
Mel Gorman4fcb0972016-05-19 17:14:01 -07003849 if (kmemcheck_enabled && page)
3850 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3851
3852 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
3853
Mel Gorman11e33f62009-06-16 15:31:57 -07003854 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003855}
Mel Gormand2391712009-06-16 15:31:52 -07003856EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857
3858/*
3859 * Common helper functions.
3860 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08003861unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003862{
Akinobu Mita945a1112009-09-21 17:01:47 -07003863 struct page *page;
3864
3865 /*
3866 * __get_free_pages() returns a 32-bit address, which cannot represent
3867 * a highmem page
3868 */
3869 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3870
Linus Torvalds1da177e2005-04-16 15:20:36 -07003871 page = alloc_pages(gfp_mask, order);
3872 if (!page)
3873 return 0;
3874 return (unsigned long) page_address(page);
3875}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003876EXPORT_SYMBOL(__get_free_pages);
3877
Harvey Harrison920c7a52008-02-04 22:29:26 -08003878unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003879{
Akinobu Mita945a1112009-09-21 17:01:47 -07003880 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003881}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882EXPORT_SYMBOL(get_zeroed_page);
3883
Harvey Harrison920c7a52008-02-04 22:29:26 -08003884void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003885{
Nick Pigginb5810032005-10-29 18:16:12 -07003886 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003887 if (order == 0)
Mel Gormanb745bc82014-06-04 16:10:22 -07003888 free_hot_cold_page(page, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003889 else
3890 __free_pages_ok(page, order);
3891 }
3892}
3893
3894EXPORT_SYMBOL(__free_pages);
3895
Harvey Harrison920c7a52008-02-04 22:29:26 -08003896void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003897{
3898 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07003899 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003900 __free_pages(virt_to_page((void *)addr), order);
3901 }
3902}
3903
3904EXPORT_SYMBOL(free_pages);
3905
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003906/*
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003907 * Page Fragment:
3908 * An arbitrary-length arbitrary-offset area of memory which resides
3909 * within a 0 or higher order page. Multiple fragments within that page
3910 * are individually refcounted, in the page's reference counter.
3911 *
3912 * The page_frag functions below provide a simple allocation framework for
3913 * page fragments. This is used by the network stack and network device
3914 * drivers to provide a backing region of memory for use as either an
3915 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3916 */
3917static struct page *__page_frag_refill(struct page_frag_cache *nc,
3918 gfp_t gfp_mask)
3919{
3920 struct page *page = NULL;
3921 gfp_t gfp = gfp_mask;
3922
3923#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3924 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3925 __GFP_NOMEMALLOC;
3926 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3927 PAGE_FRAG_CACHE_MAX_ORDER);
3928 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3929#endif
3930 if (unlikely(!page))
3931 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3932
3933 nc->va = page ? page_address(page) : NULL;
3934
3935 return page;
3936}
3937
3938void *__alloc_page_frag(struct page_frag_cache *nc,
3939 unsigned int fragsz, gfp_t gfp_mask)
3940{
3941 unsigned int size = PAGE_SIZE;
3942 struct page *page;
3943 int offset;
3944
3945 if (unlikely(!nc->va)) {
3946refill:
3947 page = __page_frag_refill(nc, gfp_mask);
3948 if (!page)
3949 return NULL;
3950
3951#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3952 /* if size can vary use size else just use PAGE_SIZE */
3953 size = nc->size;
3954#endif
3955 /* Even if we own the page, we do not use atomic_set().
3956 * This would break get_page_unless_zero() users.
3957 */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07003958 page_ref_add(page, size - 1);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003959
3960 /* reset page count bias and offset to start of new frag */
Michal Hocko2f064f32015-08-21 14:11:51 -07003961 nc->pfmemalloc = page_is_pfmemalloc(page);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003962 nc->pagecnt_bias = size;
3963 nc->offset = size;
3964 }
3965
3966 offset = nc->offset - fragsz;
3967 if (unlikely(offset < 0)) {
3968 page = virt_to_page(nc->va);
3969
Joonsoo Kimfe896d12016-03-17 14:19:26 -07003970 if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003971 goto refill;
3972
3973#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3974 /* if size can vary use size else just use PAGE_SIZE */
3975 size = nc->size;
3976#endif
3977 /* OK, page count is 0, we can safely set it */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07003978 set_page_count(page, size);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003979
3980 /* reset page count bias and offset to start of new frag */
3981 nc->pagecnt_bias = size;
3982 offset = size - fragsz;
3983 }
3984
3985 nc->pagecnt_bias--;
3986 nc->offset = offset;
3987
3988 return nc->va + offset;
3989}
3990EXPORT_SYMBOL(__alloc_page_frag);
3991
3992/*
3993 * Frees a page fragment allocated out of either a compound or order 0 page.
3994 */
3995void __free_page_frag(void *addr)
3996{
3997 struct page *page = virt_to_head_page(addr);
3998
3999 if (unlikely(put_page_testzero(page)))
4000 __free_pages_ok(page, compound_order(page));
4001}
4002EXPORT_SYMBOL(__free_page_frag);
4003
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004004static void *make_alloc_exact(unsigned long addr, unsigned int order,
4005 size_t size)
Andi Kleenee85c2e2011-05-11 15:13:34 -07004006{
4007 if (addr) {
4008 unsigned long alloc_end = addr + (PAGE_SIZE << order);
4009 unsigned long used = addr + PAGE_ALIGN(size);
4010
4011 split_page(virt_to_page((void *)addr), order);
4012 while (used < alloc_end) {
4013 free_page(used);
4014 used += PAGE_SIZE;
4015 }
4016 }
4017 return (void *)addr;
4018}
4019
Timur Tabi2be0ffe2008-07-23 21:28:11 -07004020/**
4021 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4022 * @size: the number of bytes to allocate
4023 * @gfp_mask: GFP flags for the allocation
4024 *
4025 * This function is similar to alloc_pages(), except that it allocates the
4026 * minimum number of pages to satisfy the request. alloc_pages() can only
4027 * allocate memory in power-of-two pages.
4028 *
4029 * This function is also limited by MAX_ORDER.
4030 *
4031 * Memory allocated by this function must be released by free_pages_exact().
4032 */
4033void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
4034{
4035 unsigned int order = get_order(size);
4036 unsigned long addr;
4037
4038 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07004039 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07004040}
4041EXPORT_SYMBOL(alloc_pages_exact);
4042
4043/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07004044 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4045 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07004046 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07004047 * @size: the number of bytes to allocate
4048 * @gfp_mask: GFP flags for the allocation
4049 *
4050 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4051 * back.
Andi Kleenee85c2e2011-05-11 15:13:34 -07004052 */
Fabian Fredericke1931812014-08-06 16:04:59 -07004053void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
Andi Kleenee85c2e2011-05-11 15:13:34 -07004054{
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004055 unsigned int order = get_order(size);
Andi Kleenee85c2e2011-05-11 15:13:34 -07004056 struct page *p = alloc_pages_node(nid, gfp_mask, order);
4057 if (!p)
4058 return NULL;
4059 return make_alloc_exact((unsigned long)page_address(p), order, size);
4060}
Andi Kleenee85c2e2011-05-11 15:13:34 -07004061
4062/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07004063 * free_pages_exact - release memory allocated via alloc_pages_exact()
4064 * @virt: the value returned by alloc_pages_exact.
4065 * @size: size of allocation, same value as passed to alloc_pages_exact().
4066 *
4067 * Release the memory allocated by a previous call to alloc_pages_exact.
4068 */
4069void free_pages_exact(void *virt, size_t size)
4070{
4071 unsigned long addr = (unsigned long)virt;
4072 unsigned long end = addr + PAGE_ALIGN(size);
4073
4074 while (addr < end) {
4075 free_page(addr);
4076 addr += PAGE_SIZE;
4077 }
4078}
4079EXPORT_SYMBOL(free_pages_exact);
4080
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004081/**
4082 * nr_free_zone_pages - count number of pages beyond high watermark
4083 * @offset: The zone index of the highest zone
4084 *
4085 * nr_free_zone_pages() counts the number of counts pages which are beyond the
4086 * high watermark within all zones at or below a given zone index. For each
4087 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07004088 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004089 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004090static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091{
Mel Gormandd1a2392008-04-28 02:12:17 -07004092 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07004093 struct zone *zone;
4094
Martin J. Blighe310fd42005-07-29 22:59:18 -07004095 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004096 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004097
Mel Gorman0e884602008-04-28 02:12:14 -07004098 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004099
Mel Gorman54a6eb52008-04-28 02:12:16 -07004100 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08004101 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07004102 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07004103 if (size > high)
4104 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004105 }
4106
4107 return sum;
4108}
4109
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004110/**
4111 * nr_free_buffer_pages - count number of pages beyond high watermark
4112 *
4113 * nr_free_buffer_pages() counts the number of pages which are beyond the high
4114 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004115 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004116unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004117{
Al Viroaf4ca452005-10-21 02:55:38 -04004118 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004119}
Meelap Shahc2f1a552007-07-17 04:04:39 -07004120EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004121
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004122/**
4123 * nr_free_pagecache_pages - count number of pages beyond high watermark
4124 *
4125 * nr_free_pagecache_pages() counts the number of pages which are beyond the
4126 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004127 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004128unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004129{
Mel Gorman2a1e2742007-07-17 04:03:12 -07004130 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004131}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07004132
4133static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004134{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08004135 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08004136 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004137}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004138
Igor Redkod02bd272016-03-17 14:19:05 -07004139long si_mem_available(void)
4140{
4141 long available;
4142 unsigned long pagecache;
4143 unsigned long wmark_low = 0;
4144 unsigned long pages[NR_LRU_LISTS];
4145 struct zone *zone;
4146 int lru;
4147
4148 for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
4149 pages[lru] = global_page_state(NR_LRU_BASE + lru);
4150
4151 for_each_zone(zone)
4152 wmark_low += zone->watermark[WMARK_LOW];
4153
4154 /*
4155 * Estimate the amount of memory available for userspace allocations,
4156 * without causing swapping.
4157 */
4158 available = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
4159
4160 /*
4161 * Not all the page cache can be freed, otherwise the system will
4162 * start swapping. Assume at least half of the page cache, or the
4163 * low watermark worth of cache, needs to stay.
4164 */
4165 pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
4166 pagecache -= min(pagecache / 2, wmark_low);
4167 available += pagecache;
4168
4169 /*
4170 * Part of the reclaimable slab consists of items that are in use,
4171 * and cannot be freed. Cap this estimate at the low watermark.
4172 */
4173 available += global_page_state(NR_SLAB_RECLAIMABLE) -
4174 min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low);
4175
4176 if (available < 0)
4177 available = 0;
4178 return available;
4179}
4180EXPORT_SYMBOL_GPL(si_mem_available);
4181
Linus Torvalds1da177e2005-04-16 15:20:36 -07004182void si_meminfo(struct sysinfo *val)
4183{
4184 val->totalram = totalram_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07004185 val->sharedram = global_page_state(NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08004186 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004187 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188 val->totalhigh = totalhigh_pages;
4189 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004190 val->mem_unit = PAGE_SIZE;
4191}
4192
4193EXPORT_SYMBOL(si_meminfo);
4194
4195#ifdef CONFIG_NUMA
4196void si_meminfo_node(struct sysinfo *val, int nid)
4197{
Jiang Liucdd91a72013-07-03 15:03:27 -07004198 int zone_type; /* needs to be signed */
4199 unsigned long managed_pages = 0;
Joonsoo Kimfc2bd792016-05-19 17:12:23 -07004200 unsigned long managed_highpages = 0;
4201 unsigned long free_highpages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004202 pg_data_t *pgdat = NODE_DATA(nid);
4203
Jiang Liucdd91a72013-07-03 15:03:27 -07004204 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
4205 managed_pages += pgdat->node_zones[zone_type].managed_pages;
4206 val->totalram = managed_pages;
Mel Gorman75ef7182016-07-28 15:45:24 -07004207 val->sharedram = sum_zone_node_page_state(nid, NR_SHMEM);
4208 val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004209#ifdef CONFIG_HIGHMEM
Joonsoo Kimfc2bd792016-05-19 17:12:23 -07004210 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
4211 struct zone *zone = &pgdat->node_zones[zone_type];
4212
4213 if (is_highmem(zone)) {
4214 managed_highpages += zone->managed_pages;
4215 free_highpages += zone_page_state(zone, NR_FREE_PAGES);
4216 }
4217 }
4218 val->totalhigh = managed_highpages;
4219 val->freehigh = free_highpages;
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004220#else
Joonsoo Kimfc2bd792016-05-19 17:12:23 -07004221 val->totalhigh = managed_highpages;
4222 val->freehigh = free_highpages;
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004223#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004224 val->mem_unit = PAGE_SIZE;
4225}
4226#endif
4227
David Rientjesddd588b2011-03-22 16:30:46 -07004228/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07004229 * Determine whether the node should be displayed or not, depending on whether
4230 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07004231 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07004232bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07004233{
4234 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07004235 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07004236
4237 if (!(flags & SHOW_MEM_FILTER_NODES))
4238 goto out;
4239
Mel Gormancc9a6c82012-03-21 16:34:11 -07004240 do {
Mel Gormand26914d2014-04-03 14:47:24 -07004241 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07004242 ret = !node_isset(nid, cpuset_current_mems_allowed);
Mel Gormand26914d2014-04-03 14:47:24 -07004243 } while (read_mems_allowed_retry(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07004244out:
4245 return ret;
4246}
4247
Linus Torvalds1da177e2005-04-16 15:20:36 -07004248#define K(x) ((x) << (PAGE_SHIFT-10))
4249
Rabin Vincent377e4f12012-12-11 16:00:24 -08004250static void show_migration_types(unsigned char type)
4251{
4252 static const char types[MIGRATE_TYPES] = {
4253 [MIGRATE_UNMOVABLE] = 'U',
Rabin Vincent377e4f12012-12-11 16:00:24 -08004254 [MIGRATE_MOVABLE] = 'M',
Vlastimil Babka475a2f92015-12-11 13:40:29 -08004255 [MIGRATE_RECLAIMABLE] = 'E',
4256 [MIGRATE_HIGHATOMIC] = 'H',
Rabin Vincent377e4f12012-12-11 16:00:24 -08004257#ifdef CONFIG_CMA
4258 [MIGRATE_CMA] = 'C',
4259#endif
Minchan Kim194159f2013-02-22 16:33:58 -08004260#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08004261 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08004262#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08004263 };
4264 char tmp[MIGRATE_TYPES + 1];
4265 char *p = tmp;
4266 int i;
4267
4268 for (i = 0; i < MIGRATE_TYPES; i++) {
4269 if (type & (1 << i))
4270 *p++ = types[i];
4271 }
4272
4273 *p = '\0';
4274 printk("(%s) ", tmp);
4275}
4276
Linus Torvalds1da177e2005-04-16 15:20:36 -07004277/*
4278 * Show free area list (used inside shift_scroll-lock stuff)
4279 * We also calculate the percentage fragmentation. We do this by counting the
4280 * memory on each free list with the exception of the first item on the list.
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004281 *
4282 * Bits in @filter:
4283 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4284 * cpuset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004285 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07004286void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004287{
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004288 unsigned long free_pcp = 0;
Jes Sorensenc7241912006-09-27 01:50:05 -07004289 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004290 struct zone *zone;
4291
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07004292 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07004293 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07004294 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004295
Konstantin Khlebnikov761b0672015-04-14 15:45:32 -07004296 for_each_online_cpu(cpu)
4297 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004298 }
4299
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07004300 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4301 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004302 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4303 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004304 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
Kirill A. Shutemov65c45372016-07-26 15:26:10 -07004305#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4306 " anon_thp: %lu shmem_thp: %lu shmem_pmdmapped: %lu\n"
4307#endif
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004308 " free:%lu free_pcp:%lu free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07004309 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07004310 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07004311 global_page_state(NR_ISOLATED_ANON),
4312 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07004313 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07004314 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07004315 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07004316 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07004317 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07004318 global_page_state(NR_UNSTABLE_NFS),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07004319 global_page_state(NR_SLAB_RECLAIMABLE),
4320 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07004321 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07004322 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08004323 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004324 global_page_state(NR_BOUNCE),
Kirill A. Shutemov65c45372016-07-26 15:26:10 -07004325#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4326 global_page_state(NR_ANON_THPS) * HPAGE_PMD_NR,
4327 global_page_state(NR_SHMEM_THPS) * HPAGE_PMD_NR,
4328 global_page_state(NR_SHMEM_PMDMAPPED) * HPAGE_PMD_NR,
4329#endif
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004330 global_page_state(NR_FREE_PAGES),
4331 free_pcp,
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004332 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004333
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07004334 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004335 int i;
4336
David Rientjes7bf02ea2011-05-24 17:11:16 -07004337 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07004338 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004339
4340 free_pcp = 0;
4341 for_each_online_cpu(cpu)
4342 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
4343
Linus Torvalds1da177e2005-04-16 15:20:36 -07004344 show_node(zone);
4345 printk("%s"
4346 " free:%lukB"
4347 " min:%lukB"
4348 " low:%lukB"
4349 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07004350 " active_anon:%lukB"
4351 " inactive_anon:%lukB"
4352 " active_file:%lukB"
4353 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07004354 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07004355 " isolated(anon):%lukB"
4356 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07004357 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08004358 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004359 " mlocked:%lukB"
4360 " dirty:%lukB"
4361 " writeback:%lukB"
4362 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07004363 " shmem:%lukB"
Kirill A. Shutemov65c45372016-07-26 15:26:10 -07004364#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4365 " shmem_thp: %lukB"
4366 " shmem_pmdmapped: %lukB"
4367 " anon_thp: %lukB"
4368#endif
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004369 " slab_reclaimable:%lukB"
4370 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07004371 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004372 " pagetables:%lukB"
4373 " unstable:%lukB"
4374 " bounce:%lukB"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004375 " free_pcp:%lukB"
4376 " local_pcp:%ukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004377 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004378 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07004379 " pages_scanned:%lu"
4380 " all_unreclaimable? %s"
4381 "\n",
4382 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08004383 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07004384 K(min_wmark_pages(zone)),
4385 K(low_wmark_pages(zone)),
4386 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07004387 K(zone_page_state(zone, NR_ACTIVE_ANON)),
4388 K(zone_page_state(zone, NR_INACTIVE_ANON)),
4389 K(zone_page_state(zone, NR_ACTIVE_FILE)),
4390 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07004391 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07004392 K(zone_page_state(zone, NR_ISOLATED_ANON)),
4393 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07004394 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08004395 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004396 K(zone_page_state(zone, NR_MLOCK)),
4397 K(zone_page_state(zone, NR_FILE_DIRTY)),
4398 K(zone_page_state(zone, NR_WRITEBACK)),
4399 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07004400 K(zone_page_state(zone, NR_SHMEM)),
Kirill A. Shutemov65c45372016-07-26 15:26:10 -07004401#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4402 K(zone_page_state(zone, NR_SHMEM_THPS) * HPAGE_PMD_NR),
4403 K(zone_page_state(zone, NR_SHMEM_PMDMAPPED)
4404 * HPAGE_PMD_NR),
4405 K(zone_page_state(zone, NR_ANON_THPS) * HPAGE_PMD_NR),
4406#endif
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004407 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
4408 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07004409 zone_page_state(zone, NR_KERNEL_STACK) *
4410 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004411 K(zone_page_state(zone, NR_PAGETABLE)),
4412 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
4413 K(zone_page_state(zone, NR_BOUNCE)),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004414 K(free_pcp),
4415 K(this_cpu_read(zone->pageset->pcp.count)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004416 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004417 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Mel Gorman0d5d8232014-08-06 16:07:16 -07004418 K(zone_page_state(zone, NR_PAGES_SCANNED)),
Lisa Du6e543d52013-09-11 14:22:36 -07004419 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07004420 );
4421 printk("lowmem_reserve[]:");
4422 for (i = 0; i < MAX_NR_ZONES; i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07004423 printk(" %ld", zone->lowmem_reserve[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004424 printk("\n");
4425 }
4426
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07004427 for_each_populated_zone(zone) {
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004428 unsigned int order;
4429 unsigned long nr[MAX_ORDER], flags, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08004430 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004431
David Rientjes7bf02ea2011-05-24 17:11:16 -07004432 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07004433 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004434 show_node(zone);
4435 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004436
4437 spin_lock_irqsave(&zone->lock, flags);
4438 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08004439 struct free_area *area = &zone->free_area[order];
4440 int type;
4441
4442 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07004443 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08004444
4445 types[order] = 0;
4446 for (type = 0; type < MIGRATE_TYPES; type++) {
4447 if (!list_empty(&area->free_list[type]))
4448 types[order] |= 1 << type;
4449 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004450 }
4451 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08004452 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07004453 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08004454 if (nr[order])
4455 show_migration_types(types[order]);
4456 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004457 printk("= %lukB\n", K(total));
4458 }
4459
David Rientjes949f7ec2013-04-29 15:07:48 -07004460 hugetlb_show_meminfo();
4461
Larry Woodmane6f36022008-02-04 22:29:30 -08004462 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
4463
Linus Torvalds1da177e2005-04-16 15:20:36 -07004464 show_swap_cache_info();
4465}
4466
Mel Gorman19770b32008-04-28 02:12:18 -07004467static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
4468{
4469 zoneref->zone = zone;
4470 zoneref->zone_idx = zone_idx(zone);
4471}
4472
Linus Torvalds1da177e2005-04-16 15:20:36 -07004473/*
4474 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08004475 *
4476 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004477 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004478static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004479 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004480{
Christoph Lameter1a932052006-01-06 00:11:16 -08004481 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004482 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08004483
4484 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004485 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08004486 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08004487 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07004488 zoneref_set_zone(zone,
4489 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08004490 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004491 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004492 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004493
Christoph Lameter070f8032006-01-06 00:11:19 -08004494 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004495}
4496
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004497
4498/*
4499 * zonelist_order:
4500 * 0 = automatic detection of better ordering.
4501 * 1 = order by ([node] distance, -zonetype)
4502 * 2 = order by (-zonetype, [node] distance)
4503 *
4504 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4505 * the same zonelist. So only NUMA can configure this param.
4506 */
4507#define ZONELIST_ORDER_DEFAULT 0
4508#define ZONELIST_ORDER_NODE 1
4509#define ZONELIST_ORDER_ZONE 2
4510
4511/* zonelist order in the kernel.
4512 * set_zonelist_order() will set this to NODE or ZONE.
4513 */
4514static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
4515static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
4516
4517
Linus Torvalds1da177e2005-04-16 15:20:36 -07004518#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004519/* The value user specified ....changed by config */
4520static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4521/* string for sysctl */
4522#define NUMA_ZONELIST_ORDER_LEN 16
4523char numa_zonelist_order[16] = "default";
4524
4525/*
4526 * interface for configure zonelist ordering.
4527 * command line option "numa_zonelist_order"
4528 * = "[dD]efault - default, automatic configuration.
4529 * = "[nN]ode - order by node locality, then by zone within node
4530 * = "[zZ]one - order by zone, then by locality within zone
4531 */
4532
4533static int __parse_numa_zonelist_order(char *s)
4534{
4535 if (*s == 'd' || *s == 'D') {
4536 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4537 } else if (*s == 'n' || *s == 'N') {
4538 user_zonelist_order = ZONELIST_ORDER_NODE;
4539 } else if (*s == 'z' || *s == 'Z') {
4540 user_zonelist_order = ZONELIST_ORDER_ZONE;
4541 } else {
Joe Perches11705322016-03-17 14:19:50 -07004542 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004543 return -EINVAL;
4544 }
4545 return 0;
4546}
4547
4548static __init int setup_numa_zonelist_order(char *s)
4549{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08004550 int ret;
4551
4552 if (!s)
4553 return 0;
4554
4555 ret = __parse_numa_zonelist_order(s);
4556 if (ret == 0)
4557 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
4558
4559 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004560}
4561early_param("numa_zonelist_order", setup_numa_zonelist_order);
4562
4563/*
4564 * sysctl handler for numa_zonelist_order
4565 */
Joe Perchescccad5b2014-06-06 14:38:09 -07004566int numa_zonelist_order_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004567 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004568 loff_t *ppos)
4569{
4570 char saved_string[NUMA_ZONELIST_ORDER_LEN];
4571 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01004572 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004573
Andi Kleen443c6f12009-12-23 21:00:47 +01004574 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07004575 if (write) {
4576 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
4577 ret = -EINVAL;
4578 goto out;
4579 }
4580 strcpy(saved_string, (char *)table->data);
4581 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004582 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004583 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01004584 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004585 if (write) {
4586 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07004587
4588 ret = __parse_numa_zonelist_order((char *)table->data);
4589 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004590 /*
4591 * bogus value. restore saved string
4592 */
Chen Gangdacbde02013-07-03 15:02:35 -07004593 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004594 NUMA_ZONELIST_ORDER_LEN);
4595 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004596 } else if (oldval != user_zonelist_order) {
4597 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07004598 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004599 mutex_unlock(&zonelists_mutex);
4600 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004601 }
Andi Kleen443c6f12009-12-23 21:00:47 +01004602out:
4603 mutex_unlock(&zl_order_mutex);
4604 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004605}
4606
4607
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004608#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004609static int node_load[MAX_NUMNODES];
4610
Linus Torvalds1da177e2005-04-16 15:20:36 -07004611/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07004612 * 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 -07004613 * @node: node whose fallback list we're appending
4614 * @used_node_mask: nodemask_t of already used nodes
4615 *
4616 * We use a number of factors to determine which is the next node that should
4617 * appear on a given node's fallback list. The node should not have appeared
4618 * already in @node's fallback list, and it should be the next closest node
4619 * according to the distance array (which contains arbitrary distance values
4620 * from each node to each node in the system), and should also prefer nodes
4621 * with no CPUs, since presumably they'll have very little allocation pressure
4622 * on them otherwise.
4623 * It returns -1 if no node is found.
4624 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004625static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004626{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01004627 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004628 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08004629 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10304630 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004631
Linus Torvalds4cf808e2006-02-17 20:38:21 +01004632 /* Use the local node if we haven't already */
4633 if (!node_isset(node, *used_node_mask)) {
4634 node_set(node, *used_node_mask);
4635 return node;
4636 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004637
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004638 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004639
4640 /* Don't want a node to appear more than once */
4641 if (node_isset(n, *used_node_mask))
4642 continue;
4643
Linus Torvalds1da177e2005-04-16 15:20:36 -07004644 /* Use the distance array to find the distance */
4645 val = node_distance(node, n);
4646
Linus Torvalds4cf808e2006-02-17 20:38:21 +01004647 /* Penalize nodes under us ("prefer the next node") */
4648 val += (n < node);
4649
Linus Torvalds1da177e2005-04-16 15:20:36 -07004650 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10304651 tmp = cpumask_of_node(n);
4652 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004653 val += PENALTY_FOR_NODE_WITH_CPUS;
4654
4655 /* Slight preference for less loaded node */
4656 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4657 val += node_load[n];
4658
4659 if (val < min_val) {
4660 min_val = val;
4661 best_node = n;
4662 }
4663 }
4664
4665 if (best_node >= 0)
4666 node_set(best_node, *used_node_mask);
4667
4668 return best_node;
4669}
4670
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004671
4672/*
4673 * Build zonelists ordered by node and zones within node.
4674 * This results in maximum locality--normal zone overflows into local
4675 * DMA zone, if any--but risks exhausting DMA zone.
4676 */
4677static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004678{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004679 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004680 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004681
Mel Gorman54a6eb52008-04-28 02:12:16 -07004682 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07004683 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07004684 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004685 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07004686 zonelist->_zonerefs[j].zone = NULL;
4687 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004688}
4689
4690/*
Christoph Lameter523b9452007-10-16 01:25:37 -07004691 * Build gfp_thisnode zonelists
4692 */
4693static void build_thisnode_zonelists(pg_data_t *pgdat)
4694{
Christoph Lameter523b9452007-10-16 01:25:37 -07004695 int j;
4696 struct zonelist *zonelist;
4697
Mel Gorman54a6eb52008-04-28 02:12:16 -07004698 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004699 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07004700 zonelist->_zonerefs[j].zone = NULL;
4701 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07004702}
4703
4704/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004705 * Build zonelists ordered by zone and nodes within zones.
4706 * This results in conserving DMA zone[s] until all Normal memory is
4707 * exhausted, but results in overflowing to remote node while memory
4708 * may still exist in local DMA zone.
4709 */
4710static int node_order[MAX_NUMNODES];
4711
4712static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4713{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004714 int pos, j, node;
4715 int zone_type; /* needs to be signed */
4716 struct zone *z;
4717 struct zonelist *zonelist;
4718
Mel Gorman54a6eb52008-04-28 02:12:16 -07004719 zonelist = &pgdat->node_zonelists[0];
4720 pos = 0;
4721 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4722 for (j = 0; j < nr_nodes; j++) {
4723 node = node_order[j];
4724 z = &NODE_DATA(node)->node_zones[zone_type];
4725 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07004726 zoneref_set_zone(z,
4727 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004728 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004729 }
4730 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004731 }
Mel Gormandd1a2392008-04-28 02:12:17 -07004732 zonelist->_zonerefs[pos].zone = NULL;
4733 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004734}
4735
Mel Gorman31939132014-10-09 15:28:30 -07004736#if defined(CONFIG_64BIT)
4737/*
4738 * Devices that require DMA32/DMA are relatively rare and do not justify a
4739 * penalty to every machine in case the specialised case applies. Default
4740 * to Node-ordering on 64-bit NUMA machines
4741 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004742static int default_zonelist_order(void)
4743{
Mel Gorman31939132014-10-09 15:28:30 -07004744 return ZONELIST_ORDER_NODE;
4745}
4746#else
4747/*
4748 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4749 * by the kernel. If processes running on node 0 deplete the low memory zone
4750 * then reclaim will occur more frequency increasing stalls and potentially
4751 * be easier to OOM if a large percentage of the zone is under writeback or
4752 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4753 * Hence, default to zone ordering on 32-bit.
4754 */
4755static int default_zonelist_order(void)
4756{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004757 return ZONELIST_ORDER_ZONE;
4758}
Mel Gorman31939132014-10-09 15:28:30 -07004759#endif /* CONFIG_64BIT */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004760
4761static void set_zonelist_order(void)
4762{
4763 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4764 current_zonelist_order = default_zonelist_order();
4765 else
4766 current_zonelist_order = user_zonelist_order;
4767}
4768
4769static void build_zonelists(pg_data_t *pgdat)
4770{
Yaowei Baic00eb152016-01-14 15:19:00 -08004771 int i, node, load;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004772 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004773 int local_node, prev_node;
4774 struct zonelist *zonelist;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004775 unsigned int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004776
4777 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07004778 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004779 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07004780 zonelist->_zonerefs[0].zone = NULL;
4781 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004782 }
4783
4784 /* NUMA-aware ordering of nodes */
4785 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004786 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004787 prev_node = local_node;
4788 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004789
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004790 memset(node_order, 0, sizeof(node_order));
Yaowei Baic00eb152016-01-14 15:19:00 -08004791 i = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004792
Linus Torvalds1da177e2005-04-16 15:20:36 -07004793 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4794 /*
4795 * We don't want to pressure a particular node.
4796 * So adding penalty to the first node in same
4797 * distance group to make it round-robin.
4798 */
David Rientjes957f8222012-10-08 16:33:24 -07004799 if (node_distance(local_node, node) !=
4800 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004801 node_load[node] = load;
4802
Linus Torvalds1da177e2005-04-16 15:20:36 -07004803 prev_node = node;
4804 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004805 if (order == ZONELIST_ORDER_NODE)
4806 build_zonelists_in_node_order(pgdat, node);
4807 else
Yaowei Baic00eb152016-01-14 15:19:00 -08004808 node_order[i++] = node; /* remember order */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004809 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004810
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004811 if (order == ZONELIST_ORDER_ZONE) {
4812 /* calculate node order -- i.e., DMA last! */
Yaowei Baic00eb152016-01-14 15:19:00 -08004813 build_zonelists_in_zone_order(pgdat, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004814 }
Christoph Lameter523b9452007-10-16 01:25:37 -07004815
4816 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004817}
4818
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004819#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4820/*
4821 * Return node id of node used for "local" allocations.
4822 * I.e., first node id of first zone in arg node's generic zonelist.
4823 * Used for initializing percpu 'numa_mem', which is used primarily
4824 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4825 */
4826int local_memory_node(int node)
4827{
Mel Gormanc33d6c02016-05-19 17:14:10 -07004828 struct zoneref *z;
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004829
Mel Gormanc33d6c02016-05-19 17:14:10 -07004830 z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004831 gfp_zone(GFP_KERNEL),
Mel Gormanc33d6c02016-05-19 17:14:10 -07004832 NULL);
4833 return z->zone->node;
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004834}
4835#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004836
Linus Torvalds1da177e2005-04-16 15:20:36 -07004837#else /* CONFIG_NUMA */
4838
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004839static void set_zonelist_order(void)
4840{
4841 current_zonelist_order = ZONELIST_ORDER_ZONE;
4842}
4843
4844static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004845{
Christoph Lameter19655d32006-09-25 23:31:19 -07004846 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07004847 enum zone_type j;
4848 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004849
4850 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004851
Mel Gorman54a6eb52008-04-28 02:12:16 -07004852 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004853 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004854
Mel Gorman54a6eb52008-04-28 02:12:16 -07004855 /*
4856 * Now we build the zonelist so that it contains the zones
4857 * of all the other nodes.
4858 * We don't want to pressure a particular node, so when
4859 * building the zones for node N, we make sure that the
4860 * zones coming right after the local ones are those from
4861 * node N+1 (modulo N)
4862 */
4863 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4864 if (!node_online(node))
4865 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004866 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004867 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07004868 for (node = 0; node < local_node; node++) {
4869 if (!node_online(node))
4870 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004871 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004872 }
4873
Mel Gormandd1a2392008-04-28 02:12:17 -07004874 zonelist->_zonerefs[j].zone = NULL;
4875 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004876}
4877
4878#endif /* CONFIG_NUMA */
4879
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004880/*
4881 * Boot pageset table. One per cpu which is going to be used for all
4882 * zones and all nodes. The parameters will be set in such a way
4883 * that an item put on a list will immediately be handed over to
4884 * the buddy list. This is safe since pageset manipulation is done
4885 * with interrupts disabled.
4886 *
4887 * The boot_pagesets must be kept even after bootup is complete for
4888 * unused processors and/or zones. They do play a role for bootstrapping
4889 * hotplugged processors.
4890 *
4891 * zoneinfo_show() and maybe other functions do
4892 * not check if the processor is online before following the pageset pointer.
4893 * Other parts of the kernel may not check if the zone is available.
4894 */
4895static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4896static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07004897static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004898
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004899/*
4900 * Global mutex to protect against size modification of zonelists
4901 * as well as to serialize pageset setup for the new populated zone.
4902 */
4903DEFINE_MUTEX(zonelists_mutex);
4904
Rusty Russell9b1a4d32008-07-28 12:16:30 -05004905/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07004906static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004907{
Yasunori Goto68113782006-06-23 02:03:11 -07004908 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004909 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07004910 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004911
Bo Liu7f9cfb32009-08-18 14:11:19 -07004912#ifdef CONFIG_NUMA
4913 memset(node_load, 0, sizeof(node_load));
4914#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07004915
4916 if (self && !node_online(self->node_id)) {
4917 build_zonelists(self);
Jiang Liu9adb62a2012-07-31 16:43:28 -07004918 }
4919
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004920 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07004921 pg_data_t *pgdat = NODE_DATA(nid);
4922
4923 build_zonelists(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004924 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004925
4926 /*
4927 * Initialize the boot_pagesets that are going to be used
4928 * for bootstrapping processors. The real pagesets for
4929 * each zone will be allocated later when the per cpu
4930 * allocator is available.
4931 *
4932 * boot_pagesets are used also for bootstrapping offline
4933 * cpus if the system is already booted because the pagesets
4934 * are needed to initialize allocators on a specific cpu too.
4935 * F.e. the percpu allocator needs the page allocator which
4936 * needs the percpu allocator in order to allocate its pagesets
4937 * (a chicken-egg dilemma).
4938 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004939 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004940 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4941
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004942#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4943 /*
4944 * We now know the "local memory node" for each node--
4945 * i.e., the node of the first zone in the generic zonelist.
4946 * Set up numa_mem percpu variable for on-line cpus. During
4947 * boot, only the boot cpu should be on-line; we'll init the
4948 * secondary cpus' numa_mem as they come on-line. During
4949 * node/memory hotplug, we'll fixup all on-line cpus.
4950 */
4951 if (cpu_online(cpu))
4952 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4953#endif
4954 }
4955
Yasunori Goto68113782006-06-23 02:03:11 -07004956 return 0;
4957}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004958
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004959static noinline void __init
4960build_all_zonelists_init(void)
4961{
4962 __build_all_zonelists(NULL);
4963 mminit_verify_zonelist();
4964 cpuset_init_current_mems_allowed();
4965}
4966
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004967/*
4968 * Called with zonelists_mutex held always
4969 * unless system_state == SYSTEM_BOOTING.
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004970 *
4971 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4972 * [we're only called with non-NULL zone through __meminit paths] and
4973 * (2) call of __init annotated helper build_all_zonelists_init
4974 * [protected by SYSTEM_BOOTING].
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004975 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07004976void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07004977{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004978 set_zonelist_order();
4979
Yasunori Goto68113782006-06-23 02:03:11 -07004980 if (system_state == SYSTEM_BOOTING) {
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004981 build_all_zonelists_init();
Yasunori Goto68113782006-06-23 02:03:11 -07004982 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08004983#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07004984 if (zone)
4985 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08004986#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07004987 /* we have to stop all cpus to guarantee there is no user
4988 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07004989 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07004990 /* cpuset refresh routine should be here */
4991 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07004992 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004993 /*
4994 * Disable grouping by mobility if the number of pages in the
4995 * system is too low to allow the mechanism to work. It would be
4996 * more accurate, but expensive to check per-zone. This check is
4997 * made on memory-hotadd so a system can start with mobility
4998 * disabled and enable it later
4999 */
Mel Gormand9c23402007-10-16 01:26:01 -07005000 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07005001 page_group_by_mobility_disabled = 1;
5002 else
5003 page_group_by_mobility_disabled = 0;
5004
Joe Perches756a0252016-03-17 14:19:47 -07005005 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
5006 nr_online_nodes,
5007 zonelist_order_name[current_zonelist_order],
5008 page_group_by_mobility_disabled ? "off" : "on",
5009 vm_total_pages);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07005010#ifdef CONFIG_NUMA
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005011 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07005012#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005013}
5014
5015/*
5016 * Helper functions to size the waitqueue hash table.
5017 * Essentially these want to choose hash table sizes sufficiently
5018 * large so that collisions trying to wait on pages are rare.
5019 * But in fact, the number of active page waitqueues on typical
5020 * systems is ridiculously low, less than 200. So this is even
5021 * conservative, even though it seems large.
5022 *
5023 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
5024 * waitqueues, i.e. the size of the waitq table given the number of pages.
5025 */
5026#define PAGES_PER_WAITQUEUE 256
5027
Yasunori Gotocca448f2006-06-23 02:03:10 -07005028#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07005029static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005030{
5031 unsigned long size = 1;
5032
5033 pages /= PAGES_PER_WAITQUEUE;
5034
5035 while (size < pages)
5036 size <<= 1;
5037
5038 /*
5039 * Once we have dozens or even hundreds of threads sleeping
5040 * on IO we've got bigger problems than wait queue collision.
5041 * Limit the size of the wait table to a reasonable size.
5042 */
5043 size = min(size, 4096UL);
5044
5045 return max(size, 4UL);
5046}
Yasunori Gotocca448f2006-06-23 02:03:10 -07005047#else
5048/*
5049 * A zone's size might be changed by hot-add, so it is not possible to determine
5050 * a suitable size for its wait_table. So we use the maximum size now.
5051 *
5052 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
5053 *
5054 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
5055 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
5056 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
5057 *
5058 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
5059 * or more by the traditional way. (See above). It equals:
5060 *
5061 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
5062 * ia64(16K page size) : = ( 8G + 4M)byte.
5063 * powerpc (64K page size) : = (32G +16M)byte.
5064 */
5065static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
5066{
5067 return 4096UL;
5068}
5069#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005070
5071/*
5072 * This is an integer logarithm so that shifts can be used later
5073 * to extract the more random high bits from the multiplicative
5074 * hash function before the remainder is taken.
5075 */
5076static inline unsigned long wait_table_bits(unsigned long size)
5077{
5078 return ffz(~size);
5079}
5080
Mel Gorman56fd56b2007-10-16 01:25:58 -07005081/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005082 * Initially all pages are reserved - free ones are freed
5083 * up by free_all_bootmem() once the early boot process is
5084 * done. Non-atomic initialization, single-pass.
5085 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01005086void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08005087 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005088{
Dan Williams4b94ffd2016-01-15 16:56:22 -08005089 struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn));
Andy Whitcroft29751f62005-06-23 00:08:00 -07005090 unsigned long end_pfn = start_pfn + size;
Dan Williams4b94ffd2016-01-15 16:56:22 -08005091 pg_data_t *pgdat = NODE_DATA(nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -07005092 unsigned long pfn;
Mel Gorman3a80a7f2015-06-30 14:57:02 -07005093 unsigned long nr_initialised = 0;
Taku Izumi342332e2016-03-15 14:55:22 -07005094#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5095 struct memblock_region *r = NULL, *tmp;
5096#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005097
Hugh Dickins22b31ee2009-01-06 14:40:09 -08005098 if (highest_memmap_pfn < end_pfn - 1)
5099 highest_memmap_pfn = end_pfn - 1;
5100
Dan Williams4b94ffd2016-01-15 16:56:22 -08005101 /*
5102 * Honor reservation requested by the driver for this ZONE_DEVICE
5103 * memory
5104 */
5105 if (altmap && start_pfn == altmap->base_pfn)
5106 start_pfn += altmap->reserve;
5107
Greg Ungerercbe8dd42006-01-12 01:05:24 -08005108 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08005109 /*
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005110 * There can be holes in boot-time mem_map[]s handed to this
5111 * function. They do not exist on hotplugged memory.
Dave Hansena2f3aa022007-01-10 23:15:30 -08005112 */
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005113 if (context != MEMMAP_EARLY)
5114 goto not_early;
5115
5116 if (!early_pfn_valid(pfn))
5117 continue;
5118 if (!early_pfn_in_nid(pfn, nid))
5119 continue;
5120 if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised))
5121 break;
Taku Izumi342332e2016-03-15 14:55:22 -07005122
5123#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005124 /*
5125 * If not mirrored_kernelcore and ZONE_MOVABLE exists, range
5126 * from zone_movable_pfn[nid] to end of each node should be
5127 * ZONE_MOVABLE not ZONE_NORMAL. skip it.
5128 */
5129 if (!mirrored_kernelcore && zone_movable_pfn[nid])
5130 if (zone == ZONE_NORMAL && pfn >= zone_movable_pfn[nid])
5131 continue;
Taku Izumi342332e2016-03-15 14:55:22 -07005132
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005133 /*
5134 * Check given memblock attribute by firmware which can affect
5135 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
5136 * mirrored, it's an overlapped memmap init. skip it.
5137 */
5138 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
5139 if (!r || pfn >= memblock_region_memory_end_pfn(r)) {
5140 for_each_memblock(memory, tmp)
5141 if (pfn < memblock_region_memory_end_pfn(tmp))
5142 break;
5143 r = tmp;
Taku Izumi342332e2016-03-15 14:55:22 -07005144 }
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005145 if (pfn >= memblock_region_memory_base_pfn(r) &&
5146 memblock_is_mirror(r)) {
5147 /* already initialized as NORMAL */
5148 pfn = memblock_region_memory_end_pfn(r);
5149 continue;
5150 }
Dave Hansena2f3aa022007-01-10 23:15:30 -08005151 }
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005152#endif
Mel Gormanac5d2532015-06-30 14:57:20 -07005153
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005154not_early:
Mel Gormanac5d2532015-06-30 14:57:20 -07005155 /*
5156 * Mark the block movable so that blocks are reserved for
5157 * movable at startup. This will force kernel allocations
5158 * to reserve their blocks rather than leaking throughout
5159 * the address space during boot when many long-lived
Mel Gorman974a7862015-11-06 16:28:34 -08005160 * kernel allocations are made.
Mel Gormanac5d2532015-06-30 14:57:20 -07005161 *
5162 * bitmap is created for zone's valid pfn range. but memmap
5163 * can be created for invalid pages (for alignment)
5164 * check here not to call set_pageblock_migratetype() against
5165 * pfn out of zone.
5166 */
5167 if (!(pfn & (pageblock_nr_pages - 1))) {
5168 struct page *page = pfn_to_page(pfn);
5169
5170 __init_single_page(page, pfn, zone, nid);
5171 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5172 } else {
5173 __init_single_pfn(pfn, zone, nid);
5174 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005175 }
5176}
5177
Andi Kleen1e548de2008-02-04 22:29:26 -08005178static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005179{
Mel Gorman7aeb09f2014-06-04 16:10:21 -07005180 unsigned int order, t;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07005181 for_each_migratetype_order(order, t) {
5182 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005183 zone->free_area[order].nr_free = 0;
5184 }
5185}
5186
5187#ifndef __HAVE_ARCH_MEMMAP_INIT
5188#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08005189 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005190#endif
5191
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005192static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005193{
David Howells3a6be872009-05-06 16:03:03 -07005194#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005195 int batch;
5196
5197 /*
5198 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07005199 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005200 *
5201 * OK, so we don't know how big the cache is. So guess.
5202 */
Jiang Liub40da042013-02-22 16:33:52 -08005203 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07005204 if (batch * PAGE_SIZE > 512 * 1024)
5205 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005206 batch /= 4; /* We effectively *= 4 below */
5207 if (batch < 1)
5208 batch = 1;
5209
5210 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11005211 * Clamp the batch to a 2^n - 1 value. Having a power
5212 * of 2 value was found to be more likely to have
5213 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005214 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11005215 * For example if 2 tasks are alternately allocating
5216 * batches of pages, one task can end up with a lot
5217 * of pages of one half of the possible page colors
5218 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005219 */
David Howells91552032009-05-06 16:03:02 -07005220 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07005221
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005222 return batch;
David Howells3a6be872009-05-06 16:03:03 -07005223
5224#else
5225 /* The deferral and batching of frees should be suppressed under NOMMU
5226 * conditions.
5227 *
5228 * The problem is that NOMMU needs to be able to allocate large chunks
5229 * of contiguous memory as there's no hardware page translation to
5230 * assemble apparent contiguous memory from discontiguous pages.
5231 *
5232 * Queueing large contiguous runs of pages for batching, however,
5233 * causes the pages to actually be freed in smaller chunks. As there
5234 * can be a significant delay between the individual batches being
5235 * recycled, this leads to the once large chunks of space being
5236 * fragmented and becoming unavailable for high-order allocations.
5237 */
5238 return 0;
5239#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005240}
5241
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005242/*
5243 * pcp->high and pcp->batch values are related and dependent on one another:
5244 * ->batch must never be higher then ->high.
5245 * The following function updates them in a safe manner without read side
5246 * locking.
5247 *
5248 * Any new users of pcp->batch and pcp->high should ensure they can cope with
5249 * those fields changing asynchronously (acording the the above rule).
5250 *
5251 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5252 * outside of boot time (or some other assurance that no concurrent updaters
5253 * exist).
5254 */
5255static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
5256 unsigned long batch)
5257{
5258 /* start with a fail safe value for batch */
5259 pcp->batch = 1;
5260 smp_wmb();
5261
5262 /* Update high, then batch, in order */
5263 pcp->high = high;
5264 smp_wmb();
5265
5266 pcp->batch = batch;
5267}
5268
Cody P Schafer36640332013-07-03 15:01:40 -07005269/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07005270static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
5271{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005272 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07005273}
5274
Cody P Schafer88c90db2013-07-03 15:01:35 -07005275static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07005276{
5277 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07005278 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07005279
Magnus Damm1c6fe942005-10-26 01:58:59 -07005280 memset(p, 0, sizeof(*p));
5281
Christoph Lameter3dfa5722008-02-04 22:29:19 -08005282 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07005283 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07005284 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
5285 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07005286}
5287
Cody P Schafer88c90db2013-07-03 15:01:35 -07005288static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
5289{
5290 pageset_init(p);
5291 pageset_set_batch(p, batch);
5292}
5293
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005294/*
Cody P Schafer36640332013-07-03 15:01:40 -07005295 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005296 * to the value high for the pageset p.
5297 */
Cody P Schafer36640332013-07-03 15:01:40 -07005298static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005299 unsigned long high)
5300{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005301 unsigned long batch = max(1UL, high / 4);
5302 if ((high / 4) > (PAGE_SHIFT * 8))
5303 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005304
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005305 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005306}
5307
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005308static void pageset_set_high_and_batch(struct zone *zone,
5309 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005310{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005311 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07005312 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005313 (zone->managed_pages /
5314 percpu_pagelist_fraction));
5315 else
5316 pageset_set_batch(pcp, zone_batchsize(zone));
5317}
5318
Cody P Schafer169f6c12013-07-03 15:01:41 -07005319static void __meminit zone_pageset_init(struct zone *zone, int cpu)
5320{
5321 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
5322
5323 pageset_init(pcp);
5324 pageset_set_high_and_batch(zone, pcp);
5325}
5326
Jiang Liu4ed7e022012-07-31 16:43:35 -07005327static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07005328{
5329 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07005330 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005331 for_each_possible_cpu(cpu)
5332 zone_pageset_init(zone, cpu);
Mel Gorman75ef7182016-07-28 15:45:24 -07005333
5334 if (!zone->zone_pgdat->per_cpu_nodestats) {
5335 zone->zone_pgdat->per_cpu_nodestats =
5336 alloc_percpu(struct per_cpu_nodestat);
5337 }
Wu Fengguang319774e2010-05-24 14:32:49 -07005338}
5339
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005340/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005341 * Allocate per cpu pagesets and initialize them.
5342 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07005343 */
Al Viro78d99552005-12-15 09:18:25 +00005344void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005345{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005346 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005347
Wu Fengguang319774e2010-05-24 14:32:49 -07005348 for_each_populated_zone(zone)
5349 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005350}
5351
Sam Ravnborg577a32f2007-05-17 23:29:25 +02005352static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07005353int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07005354{
5355 int i;
Yasunori Gotocca448f2006-06-23 02:03:10 -07005356 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07005357
5358 /*
5359 * The per-page waitqueue mechanism uses hashed waitqueues
5360 * per zone.
5361 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07005362 zone->wait_table_hash_nr_entries =
5363 wait_table_hash_nr_entries(zone_size_pages);
5364 zone->wait_table_bits =
5365 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07005366 alloc_size = zone->wait_table_hash_nr_entries
5367 * sizeof(wait_queue_head_t);
5368
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07005369 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07005370 zone->wait_table = (wait_queue_head_t *)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005371 memblock_virt_alloc_node_nopanic(
5372 alloc_size, zone->zone_pgdat->node_id);
Yasunori Gotocca448f2006-06-23 02:03:10 -07005373 } else {
5374 /*
5375 * This case means that a zone whose size was 0 gets new memory
5376 * via memory hot-add.
5377 * But it may be the case that a new node was hot-added. In
5378 * this case vmalloc() will not be able to use this new node's
5379 * memory - this wait_table must be initialized to use this new
5380 * node itself as well.
5381 * To use this new node's memory, further consideration will be
5382 * necessary.
5383 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07005384 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07005385 }
5386 if (!zone->wait_table)
5387 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07005388
Pintu Kumarb8af2942013-09-11 14:20:34 -07005389 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07005390 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07005391
5392 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07005393}
5394
Matt Tolentinoc09b4242006-01-17 07:03:44 +01005395static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07005396{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005397 /*
5398 * per cpu subsystem is not up at this point. The following code
5399 * relies on the ability of the linker to provide the
5400 * offset of a (static) per cpu variable into the per cpu area.
5401 */
5402 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07005403
Xishi Qiub38a8722013-11-12 15:07:20 -08005404 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005405 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
5406 zone->name, zone->present_pages,
5407 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07005408}
5409
Jiang Liu4ed7e022012-07-31 16:43:35 -07005410int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07005411 unsigned long zone_start_pfn,
Yaowei Baib171e402015-11-05 18:47:06 -08005412 unsigned long size)
Dave Hansened8ece22005-10-29 18:16:50 -07005413{
5414 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07005415 int ret;
5416 ret = zone_wait_table_init(zone, size);
5417 if (ret)
5418 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07005419 pgdat->nr_zones = zone_idx(zone) + 1;
5420
Dave Hansened8ece22005-10-29 18:16:50 -07005421 zone->zone_start_pfn = zone_start_pfn;
5422
Mel Gorman708614e2008-07-23 21:26:51 -07005423 mminit_dprintk(MMINIT_TRACE, "memmap_init",
5424 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5425 pgdat->node_id,
5426 (unsigned long)zone_idx(zone),
5427 zone_start_pfn, (zone_start_pfn + size));
5428
Andi Kleen1e548de2008-02-04 22:29:26 -08005429 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07005430
5431 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07005432}
5433
Tejun Heo0ee332c2011-12-08 10:22:09 -08005434#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07005435#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
Mel Gorman8a942fd2015-06-30 14:56:55 -07005436
Mel Gormanc7132162006-09-27 01:49:43 -07005437/*
5438 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
Mel Gormanc7132162006-09-27 01:49:43 -07005439 */
Mel Gorman8a942fd2015-06-30 14:56:55 -07005440int __meminit __early_pfn_to_nid(unsigned long pfn,
5441 struct mminit_pfnnid_cache *state)
Mel Gormanc7132162006-09-27 01:49:43 -07005442{
Tejun Heoc13291a2011-07-12 10:46:30 +02005443 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07005444 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07005445
Mel Gorman8a942fd2015-06-30 14:56:55 -07005446 if (state->last_start <= pfn && pfn < state->last_end)
5447 return state->last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005448
Yinghai Lue76b63f2013-09-11 14:22:17 -07005449 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
5450 if (nid != -1) {
Mel Gorman8a942fd2015-06-30 14:56:55 -07005451 state->last_start = start_pfn;
5452 state->last_end = end_pfn;
5453 state->last_nid = nid;
Yinghai Lue76b63f2013-09-11 14:22:17 -07005454 }
5455
5456 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005457}
5458#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5459
Mel Gormanc7132162006-09-27 01:49:43 -07005460/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08005461 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005462 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08005463 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07005464 *
Zhang Zhen7d018172014-06-04 16:10:53 -07005465 * If an architecture guarantees that all ranges registered contain no holes
5466 * and may be freed, this this function may be used instead of calling
5467 * memblock_free_early_nid() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07005468 */
Tejun Heoc13291a2011-07-12 10:46:30 +02005469void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07005470{
Tejun Heoc13291a2011-07-12 10:46:30 +02005471 unsigned long start_pfn, end_pfn;
5472 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005473
Tejun Heoc13291a2011-07-12 10:46:30 +02005474 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
5475 start_pfn = min(start_pfn, max_low_pfn);
5476 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005477
Tejun Heoc13291a2011-07-12 10:46:30 +02005478 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005479 memblock_free_early_nid(PFN_PHYS(start_pfn),
5480 (end_pfn - start_pfn) << PAGE_SHIFT,
5481 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005482 }
5483}
5484
5485/**
5486 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005487 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07005488 *
Zhang Zhen7d018172014-06-04 16:10:53 -07005489 * If an architecture guarantees that all ranges registered contain no holes and may
5490 * be freed, this function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07005491 */
5492void __init sparse_memory_present_with_active_regions(int nid)
5493{
Tejun Heoc13291a2011-07-12 10:46:30 +02005494 unsigned long start_pfn, end_pfn;
5495 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005496
Tejun Heoc13291a2011-07-12 10:46:30 +02005497 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
5498 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005499}
5500
5501/**
5502 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005503 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5504 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5505 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07005506 *
5507 * It returns the start and end page frame of a node based on information
Zhang Zhen7d018172014-06-04 16:10:53 -07005508 * provided by memblock_set_node(). If called for a node
Mel Gormanc7132162006-09-27 01:49:43 -07005509 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005510 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07005511 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005512void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005513 unsigned long *start_pfn, unsigned long *end_pfn)
5514{
Tejun Heoc13291a2011-07-12 10:46:30 +02005515 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005516 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02005517
Mel Gormanc7132162006-09-27 01:49:43 -07005518 *start_pfn = -1UL;
5519 *end_pfn = 0;
5520
Tejun Heoc13291a2011-07-12 10:46:30 +02005521 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5522 *start_pfn = min(*start_pfn, this_start_pfn);
5523 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005524 }
5525
Christoph Lameter633c0662007-10-16 01:25:37 -07005526 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07005527 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005528}
5529
5530/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07005531 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5532 * assumption is made that zones within a node are ordered in monotonic
5533 * increasing memory addresses so that the "highest" populated zone is used
5534 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005535static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005536{
5537 int zone_index;
5538 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5539 if (zone_index == ZONE_MOVABLE)
5540 continue;
5541
5542 if (arch_zone_highest_possible_pfn[zone_index] >
5543 arch_zone_lowest_possible_pfn[zone_index])
5544 break;
5545 }
5546
5547 VM_BUG_ON(zone_index == -1);
5548 movable_zone = zone_index;
5549}
5550
5551/*
5552 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03005553 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005554 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5555 * in each node depending on the size of each node and how evenly kernelcore
5556 * is distributed. This helper function adjusts the zone ranges
5557 * provided by the architecture for a given node by using the end of the
5558 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5559 * zones within a node are in order of monotonic increases memory addresses
5560 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005561static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005562 unsigned long zone_type,
5563 unsigned long node_start_pfn,
5564 unsigned long node_end_pfn,
5565 unsigned long *zone_start_pfn,
5566 unsigned long *zone_end_pfn)
5567{
5568 /* Only adjust if ZONE_MOVABLE is on this node */
5569 if (zone_movable_pfn[nid]) {
5570 /* Size ZONE_MOVABLE */
5571 if (zone_type == ZONE_MOVABLE) {
5572 *zone_start_pfn = zone_movable_pfn[nid];
5573 *zone_end_pfn = min(node_end_pfn,
5574 arch_zone_highest_possible_pfn[movable_zone]);
5575
Mel Gorman2a1e2742007-07-17 04:03:12 -07005576 /* Check if this whole range is within ZONE_MOVABLE */
5577 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5578 *zone_start_pfn = *zone_end_pfn;
5579 }
5580}
5581
5582/*
Mel Gormanc7132162006-09-27 01:49:43 -07005583 * Return the number of pages a zone spans in a node, including holes
5584 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5585 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005586static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005587 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005588 unsigned long node_start_pfn,
5589 unsigned long node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005590 unsigned long *zone_start_pfn,
5591 unsigned long *zone_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005592 unsigned long *ignored)
5593{
Xishi Qiub5685e92015-09-08 15:04:16 -07005594 /* When hotadd a new node from cpu_up(), the node should be empty */
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005595 if (!node_start_pfn && !node_end_pfn)
5596 return 0;
5597
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005598 /* Get the start and end of the zone */
Taku Izumid91749c2016-03-15 14:55:18 -07005599 *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5600 *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07005601 adjust_zone_range_for_zone_movable(nid, zone_type,
5602 node_start_pfn, node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005603 zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005604
5605 /* Check that this node has pages within the zone's required range */
Taku Izumid91749c2016-03-15 14:55:18 -07005606 if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07005607 return 0;
5608
5609 /* Move the zone boundaries inside the node if necessary */
Taku Izumid91749c2016-03-15 14:55:18 -07005610 *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
5611 *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005612
5613 /* Return the spanned pages */
Taku Izumid91749c2016-03-15 14:55:18 -07005614 return *zone_end_pfn - *zone_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005615}
5616
5617/*
5618 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005619 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07005620 */
Yinghai Lu32996252009-12-15 17:59:02 -08005621unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005622 unsigned long range_start_pfn,
5623 unsigned long range_end_pfn)
5624{
Tejun Heo96e907d2011-07-12 10:46:29 +02005625 unsigned long nr_absent = range_end_pfn - range_start_pfn;
5626 unsigned long start_pfn, end_pfn;
5627 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07005628
Tejun Heo96e907d2011-07-12 10:46:29 +02005629 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5630 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5631 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5632 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005633 }
Tejun Heo96e907d2011-07-12 10:46:29 +02005634 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07005635}
5636
5637/**
5638 * absent_pages_in_range - Return number of page frames in holes within a range
5639 * @start_pfn: The start PFN to start searching for holes
5640 * @end_pfn: The end PFN to stop searching for holes
5641 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005642 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07005643 */
5644unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5645 unsigned long end_pfn)
5646{
5647 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5648}
5649
5650/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005651static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005652 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005653 unsigned long node_start_pfn,
5654 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005655 unsigned long *ignored)
5656{
Tejun Heo96e907d2011-07-12 10:46:29 +02005657 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5658 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07005659 unsigned long zone_start_pfn, zone_end_pfn;
Taku Izumi342332e2016-03-15 14:55:22 -07005660 unsigned long nr_absent;
Mel Gorman9c7cd682006-09-27 01:49:58 -07005661
Xishi Qiub5685e92015-09-08 15:04:16 -07005662 /* When hotadd a new node from cpu_up(), the node should be empty */
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005663 if (!node_start_pfn && !node_end_pfn)
5664 return 0;
5665
Tejun Heo96e907d2011-07-12 10:46:29 +02005666 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5667 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07005668
Mel Gorman2a1e2742007-07-17 04:03:12 -07005669 adjust_zone_range_for_zone_movable(nid, zone_type,
5670 node_start_pfn, node_end_pfn,
5671 &zone_start_pfn, &zone_end_pfn);
Taku Izumi342332e2016-03-15 14:55:22 -07005672 nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5673
5674 /*
5675 * ZONE_MOVABLE handling.
5676 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5677 * and vice versa.
5678 */
5679 if (zone_movable_pfn[nid]) {
5680 if (mirrored_kernelcore) {
5681 unsigned long start_pfn, end_pfn;
5682 struct memblock_region *r;
5683
5684 for_each_memblock(memory, r) {
5685 start_pfn = clamp(memblock_region_memory_base_pfn(r),
5686 zone_start_pfn, zone_end_pfn);
5687 end_pfn = clamp(memblock_region_memory_end_pfn(r),
5688 zone_start_pfn, zone_end_pfn);
5689
5690 if (zone_type == ZONE_MOVABLE &&
5691 memblock_is_mirror(r))
5692 nr_absent += end_pfn - start_pfn;
5693
5694 if (zone_type == ZONE_NORMAL &&
5695 !memblock_is_mirror(r))
5696 nr_absent += end_pfn - start_pfn;
5697 }
5698 } else {
5699 if (zone_type == ZONE_NORMAL)
5700 nr_absent += node_end_pfn - zone_movable_pfn[nid];
5701 }
5702 }
5703
5704 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07005705}
Mel Gorman0e0b8642006-09-27 01:49:56 -07005706
Tejun Heo0ee332c2011-12-08 10:22:09 -08005707#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005708static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005709 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005710 unsigned long node_start_pfn,
5711 unsigned long node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005712 unsigned long *zone_start_pfn,
5713 unsigned long *zone_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005714 unsigned long *zones_size)
5715{
Taku Izumid91749c2016-03-15 14:55:18 -07005716 unsigned int zone;
5717
5718 *zone_start_pfn = node_start_pfn;
5719 for (zone = 0; zone < zone_type; zone++)
5720 *zone_start_pfn += zones_size[zone];
5721
5722 *zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
5723
Mel Gormanc7132162006-09-27 01:49:43 -07005724 return zones_size[zone_type];
5725}
5726
Paul Mundt6ea6e682007-07-15 23:38:20 -07005727static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005728 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005729 unsigned long node_start_pfn,
5730 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005731 unsigned long *zholes_size)
5732{
5733 if (!zholes_size)
5734 return 0;
5735
5736 return zholes_size[zone_type];
5737}
Yinghai Lu20e69262013-03-01 14:51:27 -08005738
Tejun Heo0ee332c2011-12-08 10:22:09 -08005739#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005740
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005741static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005742 unsigned long node_start_pfn,
5743 unsigned long node_end_pfn,
5744 unsigned long *zones_size,
5745 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07005746{
Gu Zhengfebd5942015-06-24 16:57:02 -07005747 unsigned long realtotalpages = 0, totalpages = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005748 enum zone_type i;
5749
Gu Zhengfebd5942015-06-24 16:57:02 -07005750 for (i = 0; i < MAX_NR_ZONES; i++) {
5751 struct zone *zone = pgdat->node_zones + i;
Taku Izumid91749c2016-03-15 14:55:18 -07005752 unsigned long zone_start_pfn, zone_end_pfn;
Gu Zhengfebd5942015-06-24 16:57:02 -07005753 unsigned long size, real_size;
Mel Gormanc7132162006-09-27 01:49:43 -07005754
Gu Zhengfebd5942015-06-24 16:57:02 -07005755 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5756 node_start_pfn,
5757 node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005758 &zone_start_pfn,
5759 &zone_end_pfn,
Gu Zhengfebd5942015-06-24 16:57:02 -07005760 zones_size);
5761 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005762 node_start_pfn, node_end_pfn,
5763 zholes_size);
Taku Izumid91749c2016-03-15 14:55:18 -07005764 if (size)
5765 zone->zone_start_pfn = zone_start_pfn;
5766 else
5767 zone->zone_start_pfn = 0;
Gu Zhengfebd5942015-06-24 16:57:02 -07005768 zone->spanned_pages = size;
5769 zone->present_pages = real_size;
5770
5771 totalpages += size;
5772 realtotalpages += real_size;
5773 }
5774
5775 pgdat->node_spanned_pages = totalpages;
Mel Gormanc7132162006-09-27 01:49:43 -07005776 pgdat->node_present_pages = realtotalpages;
5777 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5778 realtotalpages);
5779}
5780
Mel Gorman835c1342007-10-16 01:25:47 -07005781#ifndef CONFIG_SPARSEMEM
5782/*
5783 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07005784 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5785 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07005786 * round what is now in bits to nearest long in bits, then return it in
5787 * bytes.
5788 */
Linus Torvalds7c455122013-02-18 09:58:02 -08005789static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005790{
5791 unsigned long usemapsize;
5792
Linus Torvalds7c455122013-02-18 09:58:02 -08005793 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005794 usemapsize = roundup(zonesize, pageblock_nr_pages);
5795 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07005796 usemapsize *= NR_PAGEBLOCK_BITS;
5797 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5798
5799 return usemapsize / 8;
5800}
5801
5802static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08005803 struct zone *zone,
5804 unsigned long zone_start_pfn,
5805 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005806{
Linus Torvalds7c455122013-02-18 09:58:02 -08005807 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07005808 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08005809 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005810 zone->pageblock_flags =
5811 memblock_virt_alloc_node_nopanic(usemapsize,
5812 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07005813}
5814#else
Linus Torvalds7c455122013-02-18 09:58:02 -08005815static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5816 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07005817#endif /* CONFIG_SPARSEMEM */
5818
Mel Gormand9c23402007-10-16 01:26:01 -07005819#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08005820
Mel Gormand9c23402007-10-16 01:26:01 -07005821/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07005822void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07005823{
Andrew Morton955c1cd2012-05-29 15:06:31 -07005824 unsigned int order;
5825
Mel Gormand9c23402007-10-16 01:26:01 -07005826 /* Check that pageblock_nr_pages has not already been setup */
5827 if (pageblock_order)
5828 return;
5829
Andrew Morton955c1cd2012-05-29 15:06:31 -07005830 if (HPAGE_SHIFT > PAGE_SHIFT)
5831 order = HUGETLB_PAGE_ORDER;
5832 else
5833 order = MAX_ORDER - 1;
5834
Mel Gormand9c23402007-10-16 01:26:01 -07005835 /*
5836 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07005837 * This value may be variable depending on boot parameters on IA64 and
5838 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07005839 */
5840 pageblock_order = order;
5841}
5842#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5843
Mel Gormanba72cb82007-11-28 16:21:13 -08005844/*
5845 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07005846 * is unused as pageblock_order is set at compile-time. See
5847 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5848 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08005849 */
Chen Gang15ca2202013-09-11 14:20:27 -07005850void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08005851{
Mel Gormanba72cb82007-11-28 16:21:13 -08005852}
Mel Gormand9c23402007-10-16 01:26:01 -07005853
5854#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5855
Jiang Liu01cefae2012-12-12 13:52:19 -08005856static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5857 unsigned long present_pages)
5858{
5859 unsigned long pages = spanned_pages;
5860
5861 /*
5862 * Provide a more accurate estimation if there are holes within
5863 * the zone and SPARSEMEM is in use. If there are holes within the
5864 * zone, each populated memory region may cost us one or two extra
5865 * memmap pages due to alignment because memmap pages for each
5866 * populated regions may not naturally algined on page boundary.
5867 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5868 */
5869 if (spanned_pages > present_pages + (present_pages >> 4) &&
5870 IS_ENABLED(CONFIG_SPARSEMEM))
5871 pages = present_pages;
5872
5873 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5874}
5875
Linus Torvalds1da177e2005-04-16 15:20:36 -07005876/*
5877 * Set up the zone data structures:
5878 * - mark all pages reserved
5879 * - mark all memory queues empty
5880 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07005881 *
5882 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005883 */
Wei Yang7f3eb552015-09-08 14:59:50 -07005884static void __paginginit free_area_init_core(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005885{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07005886 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07005887 int nid = pgdat->node_id;
Yasunori Goto718127c2006-06-23 02:03:10 -07005888 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005889
Dave Hansen208d54e2005-10-29 18:16:52 -07005890 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01005891#ifdef CONFIG_NUMA_BALANCING
5892 spin_lock_init(&pgdat->numabalancing_migrate_lock);
5893 pgdat->numabalancing_migrate_nr_pages = 0;
5894 pgdat->numabalancing_migrate_next_window = jiffies;
5895#endif
Kirill A. Shutemova3d0a9182016-02-02 16:57:08 -08005896#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5897 spin_lock_init(&pgdat->split_queue_lock);
5898 INIT_LIST_HEAD(&pgdat->split_queue);
5899 pgdat->split_queue_len = 0;
5900#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005901 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07005902 init_waitqueue_head(&pgdat->pfmemalloc_wait);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07005903#ifdef CONFIG_COMPACTION
5904 init_waitqueue_head(&pgdat->kcompactd_wait);
5905#endif
Joonsoo Kimeefa8642014-12-12 16:55:46 -08005906 pgdat_page_ext_init(pgdat);
Mel Gormana52633d2016-07-28 15:45:28 -07005907 spin_lock_init(&pgdat->lru_lock);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01005908
Linus Torvalds1da177e2005-04-16 15:20:36 -07005909 for (j = 0; j < MAX_NR_ZONES; j++) {
5910 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08005911 unsigned long size, realsize, freesize, memmap_pages;
Taku Izumid91749c2016-03-15 14:55:18 -07005912 unsigned long zone_start_pfn = zone->zone_start_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005913
Gu Zhengfebd5942015-06-24 16:57:02 -07005914 size = zone->spanned_pages;
5915 realsize = freesize = zone->present_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005916
Mel Gorman0e0b8642006-09-27 01:49:56 -07005917 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08005918 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07005919 * is used by this zone for memmap. This affects the watermark
5920 * and per-cpu initialisations
5921 */
Jiang Liu01cefae2012-12-12 13:52:19 -08005922 memmap_pages = calc_memmap_size(size, realsize);
Zhong Hongboba914f42014-12-12 16:56:21 -08005923 if (!is_highmem_idx(j)) {
5924 if (freesize >= memmap_pages) {
5925 freesize -= memmap_pages;
5926 if (memmap_pages)
5927 printk(KERN_DEBUG
5928 " %s zone: %lu pages used for memmap\n",
5929 zone_names[j], memmap_pages);
5930 } else
Joe Perches11705322016-03-17 14:19:50 -07005931 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
Zhong Hongboba914f42014-12-12 16:56:21 -08005932 zone_names[j], memmap_pages, freesize);
5933 }
Mel Gorman0e0b8642006-09-27 01:49:56 -07005934
Christoph Lameter62672762007-02-10 01:43:07 -08005935 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08005936 if (j == 0 && freesize > dma_reserve) {
5937 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07005938 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08005939 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07005940 }
5941
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07005942 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08005943 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08005944 /* Charge for highmem memmap if there are enough kernel pages */
5945 else if (nr_kernel_pages > memmap_pages * 2)
5946 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08005947 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005948
Jiang Liu9feedc92012-12-12 13:52:12 -08005949 /*
5950 * Set an approximate value for lowmem here, it will be adjusted
5951 * when the bootmem allocator frees pages into the buddy system.
5952 * And all highmem pages will be managed by the buddy system.
5953 */
5954 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07005955#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07005956 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08005957 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07005958 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08005959 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07005960#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005961 zone->name = zone_names[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -07005962 zone->zone_pgdat = pgdat;
Mel Gormana52633d2016-07-28 15:45:28 -07005963 spin_lock_init(&zone->lock);
5964 zone_seqlock_init(zone);
Dave Hansened8ece22005-10-29 18:16:50 -07005965 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005966
5967 /* For bootup, initialized properly in watermark setup */
5968 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
5969
Hugh Dickinsbea8c152012-11-16 14:14:54 -08005970 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005971 if (!size)
5972 continue;
5973
Andrew Morton955c1cd2012-05-29 15:06:31 -07005974 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08005975 setup_usemap(pgdat, zone, zone_start_pfn, size);
Yaowei Baib171e402015-11-05 18:47:06 -08005976 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
Yasunori Goto718127c2006-06-23 02:03:10 -07005977 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07005978 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005979 }
5980}
5981
Sam Ravnborg577a32f2007-05-17 23:29:25 +02005982static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005983{
Tony Luckb0aeba72015-11-10 10:09:47 -08005984 unsigned long __maybe_unused start = 0;
Laura Abbotta1c34a32015-11-05 18:48:46 -08005985 unsigned long __maybe_unused offset = 0;
5986
Linus Torvalds1da177e2005-04-16 15:20:36 -07005987 /* Skip empty nodes */
5988 if (!pgdat->node_spanned_pages)
5989 return;
5990
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005991#ifdef CONFIG_FLAT_NODE_MEM_MAP
Tony Luckb0aeba72015-11-10 10:09:47 -08005992 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
5993 offset = pgdat->node_start_pfn - start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005994 /* ia64 gets its own node_mem_map, before this, without bootmem */
5995 if (!pgdat->node_mem_map) {
Tony Luckb0aeba72015-11-10 10:09:47 -08005996 unsigned long size, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005997 struct page *map;
5998
Bob Piccoe984bb42006-05-20 15:00:31 -07005999 /*
6000 * The zone's endpoints aren't required to be MAX_ORDER
6001 * aligned but the node_mem_map endpoints must be in order
6002 * for the buddy allocator to function correctly.
6003 */
Cody P Schafer108bcc92013-02-22 16:35:23 -08006004 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07006005 end = ALIGN(end, MAX_ORDER_NR_PAGES);
6006 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07006007 map = alloc_remap(pgdat->node_id, size);
6008 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08006009 map = memblock_virt_alloc_node_nopanic(size,
6010 pgdat->node_id);
Laura Abbotta1c34a32015-11-05 18:48:46 -08006011 pgdat->node_mem_map = map + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006012 }
Roman Zippel12d810c2007-05-31 00:40:54 -07006013#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07006014 /*
6015 * With no DISCONTIG, the global mem_map is just set as node 0's
6016 */
Mel Gormanc7132162006-09-27 01:49:43 -07006017 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006018 mem_map = NODE_DATA(0)->node_mem_map;
Laura Abbotta1c34a32015-11-05 18:48:46 -08006019#if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
Mel Gormanc7132162006-09-27 01:49:43 -07006020 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Laura Abbotta1c34a32015-11-05 18:48:46 -08006021 mem_map -= offset;
Tejun Heo0ee332c2011-12-08 10:22:09 -08006022#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07006023 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006024#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07006025#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006026}
6027
Johannes Weiner9109fb72008-07-23 21:27:20 -07006028void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
6029 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006030{
Johannes Weiner9109fb72008-07-23 21:27:20 -07006031 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07006032 unsigned long start_pfn = 0;
6033 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07006034
Minchan Kim88fdf752012-07-31 16:46:14 -07006035 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07006036 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07006037
Mel Gorman3a80a7f2015-06-30 14:57:02 -07006038 reset_deferred_meminit(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006039 pgdat->node_id = nid;
6040 pgdat->node_start_pfn = node_start_pfn;
Mel Gorman75ef7182016-07-28 15:45:24 -07006041 pgdat->per_cpu_nodestats = NULL;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07006042#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6043 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
Juergen Gross8d29e182015-02-11 15:26:01 -08006044 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
Zhen Lei4ada0c52015-09-08 15:04:19 -07006045 (u64)start_pfn << PAGE_SHIFT,
6046 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
Taku Izumid91749c2016-03-15 14:55:18 -07006047#else
6048 start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07006049#endif
6050 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
6051 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006052
6053 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07006054#ifdef CONFIG_FLAT_NODE_MEM_MAP
6055 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
6056 nid, (unsigned long)pgdat,
6057 (unsigned long)pgdat->node_mem_map);
6058#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07006059
Wei Yang7f3eb552015-09-08 14:59:50 -07006060 free_area_init_core(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006061}
6062
Tejun Heo0ee332c2011-12-08 10:22:09 -08006063#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07006064
6065#if MAX_NUMNODES > 1
6066/*
6067 * Figure out the number of possible node ids.
6068 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07006069void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07006070{
Wei Yang904a9552015-09-08 14:59:48 -07006071 unsigned int highest;
Miklos Szeredi418508c2007-05-23 13:57:55 -07006072
Wei Yang904a9552015-09-08 14:59:48 -07006073 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
Miklos Szeredi418508c2007-05-23 13:57:55 -07006074 nr_node_ids = highest + 1;
6075}
Miklos Szeredi418508c2007-05-23 13:57:55 -07006076#endif
6077
Mel Gormanc7132162006-09-27 01:49:43 -07006078/**
Tejun Heo1e019792011-07-12 09:45:34 +02006079 * node_map_pfn_alignment - determine the maximum internode alignment
6080 *
6081 * This function should be called after node map is populated and sorted.
6082 * It calculates the maximum power of two alignment which can distinguish
6083 * all the nodes.
6084 *
6085 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6086 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
6087 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
6088 * shifted, 1GiB is enough and this function will indicate so.
6089 *
6090 * This is used to test whether pfn -> nid mapping of the chosen memory
6091 * model has fine enough granularity to avoid incorrect mapping for the
6092 * populated node map.
6093 *
6094 * Returns the determined alignment in pfn's. 0 if there is no alignment
6095 * requirement (single node).
6096 */
6097unsigned long __init node_map_pfn_alignment(void)
6098{
6099 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02006100 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02006101 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02006102 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02006103
Tejun Heoc13291a2011-07-12 10:46:30 +02006104 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02006105 if (!start || last_nid < 0 || last_nid == nid) {
6106 last_nid = nid;
6107 last_end = end;
6108 continue;
6109 }
6110
6111 /*
6112 * Start with a mask granular enough to pin-point to the
6113 * start pfn and tick off bits one-by-one until it becomes
6114 * too coarse to separate the current node from the last.
6115 */
6116 mask = ~((1 << __ffs(start)) - 1);
6117 while (mask && last_end <= (start & (mask << 1)))
6118 mask <<= 1;
6119
6120 /* accumulate all internode masks */
6121 accl_mask |= mask;
6122 }
6123
6124 /* convert mask to number of pages */
6125 return ~accl_mask + 1;
6126}
6127
Mel Gormana6af2bc2007-02-10 01:42:57 -08006128/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07006129static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07006130{
Mel Gormana6af2bc2007-02-10 01:42:57 -08006131 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02006132 unsigned long start_pfn;
6133 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00006134
Tejun Heoc13291a2011-07-12 10:46:30 +02006135 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
6136 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07006137
Mel Gormana6af2bc2007-02-10 01:42:57 -08006138 if (min_pfn == ULONG_MAX) {
Joe Perches11705322016-03-17 14:19:50 -07006139 pr_warn("Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08006140 return 0;
6141 }
6142
6143 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07006144}
6145
6146/**
6147 * find_min_pfn_with_active_regions - Find the minimum PFN registered
6148 *
6149 * It returns the minimum PFN based on information provided via
Zhang Zhen7d018172014-06-04 16:10:53 -07006150 * memblock_set_node().
Mel Gormanc7132162006-09-27 01:49:43 -07006151 */
6152unsigned long __init find_min_pfn_with_active_regions(void)
6153{
6154 return find_min_pfn_for_node(MAX_NUMNODES);
6155}
6156
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006157/*
6158 * early_calculate_totalpages()
6159 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006160 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006161 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07006162static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07006163{
Mel Gorman7e63efe2007-07-17 04:03:15 -07006164 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02006165 unsigned long start_pfn, end_pfn;
6166 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07006167
Tejun Heoc13291a2011-07-12 10:46:30 +02006168 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
6169 unsigned long pages = end_pfn - start_pfn;
6170
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006171 totalpages += pages;
6172 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006173 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006174 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07006175 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07006176}
6177
Mel Gorman2a1e2742007-07-17 04:03:12 -07006178/*
6179 * Find the PFN the Movable zone begins in each node. Kernel memory
6180 * is spread evenly between nodes as long as the nodes have enough
6181 * memory. When they don't, some nodes will have more kernelcore than
6182 * others
6183 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07006184static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07006185{
6186 int i, nid;
6187 unsigned long usable_startpfn;
6188 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07006189 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006190 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006191 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006192 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07006193 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006194
6195 /* Need to find movable_zone earlier when movable_node is specified. */
6196 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07006197
Mel Gorman7e63efe2007-07-17 04:03:15 -07006198 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006199 * If movable_node is specified, ignore kernelcore and movablecore
6200 * options.
6201 */
6202 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07006203 for_each_memblock(memory, r) {
6204 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006205 continue;
6206
Emil Medve136199f2014-04-07 15:37:52 -07006207 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006208
Emil Medve136199f2014-04-07 15:37:52 -07006209 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006210 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6211 min(usable_startpfn, zone_movable_pfn[nid]) :
6212 usable_startpfn;
6213 }
6214
6215 goto out2;
6216 }
6217
6218 /*
Taku Izumi342332e2016-03-15 14:55:22 -07006219 * If kernelcore=mirror is specified, ignore movablecore option
6220 */
6221 if (mirrored_kernelcore) {
6222 bool mem_below_4gb_not_mirrored = false;
6223
6224 for_each_memblock(memory, r) {
6225 if (memblock_is_mirror(r))
6226 continue;
6227
6228 nid = r->nid;
6229
6230 usable_startpfn = memblock_region_memory_base_pfn(r);
6231
6232 if (usable_startpfn < 0x100000) {
6233 mem_below_4gb_not_mirrored = true;
6234 continue;
6235 }
6236
6237 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6238 min(usable_startpfn, zone_movable_pfn[nid]) :
6239 usable_startpfn;
6240 }
6241
6242 if (mem_below_4gb_not_mirrored)
6243 pr_warn("This configuration results in unmirrored kernel memory.");
6244
6245 goto out2;
6246 }
6247
6248 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006249 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07006250 * kernelcore that corresponds so that memory usable for
6251 * any allocation type is evenly spread. If both kernelcore
6252 * and movablecore are specified, then the value of kernelcore
6253 * will be used for required_kernelcore if it's greater than
6254 * what movablecore would have allowed.
6255 */
6256 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07006257 unsigned long corepages;
6258
6259 /*
6260 * Round-up so that ZONE_MOVABLE is at least as large as what
6261 * was requested by the user
6262 */
6263 required_movablecore =
6264 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
Xishi Qiu9fd745d2015-11-05 18:48:11 -08006265 required_movablecore = min(totalpages, required_movablecore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07006266 corepages = totalpages - required_movablecore;
6267
6268 required_kernelcore = max(required_kernelcore, corepages);
6269 }
6270
Xishi Qiubde304b2015-11-05 18:48:56 -08006271 /*
6272 * If kernelcore was not specified or kernelcore size is larger
6273 * than totalpages, there is no ZONE_MOVABLE.
6274 */
6275 if (!required_kernelcore || required_kernelcore >= totalpages)
Yinghai Lu66918dc2009-06-30 11:41:37 -07006276 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07006277
6278 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07006279 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
6280
6281restart:
6282 /* Spread kernelcore memory as evenly as possible throughout nodes */
6283 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006284 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02006285 unsigned long start_pfn, end_pfn;
6286
Mel Gorman2a1e2742007-07-17 04:03:12 -07006287 /*
6288 * Recalculate kernelcore_node if the division per node
6289 * now exceeds what is necessary to satisfy the requested
6290 * amount of memory for the kernel
6291 */
6292 if (required_kernelcore < kernelcore_node)
6293 kernelcore_node = required_kernelcore / usable_nodes;
6294
6295 /*
6296 * As the map is walked, we track how much memory is usable
6297 * by the kernel using kernelcore_remaining. When it is
6298 * 0, the rest of the node is usable by ZONE_MOVABLE
6299 */
6300 kernelcore_remaining = kernelcore_node;
6301
6302 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02006303 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07006304 unsigned long size_pages;
6305
Tejun Heoc13291a2011-07-12 10:46:30 +02006306 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07006307 if (start_pfn >= end_pfn)
6308 continue;
6309
6310 /* Account for what is only usable for kernelcore */
6311 if (start_pfn < usable_startpfn) {
6312 unsigned long kernel_pages;
6313 kernel_pages = min(end_pfn, usable_startpfn)
6314 - start_pfn;
6315
6316 kernelcore_remaining -= min(kernel_pages,
6317 kernelcore_remaining);
6318 required_kernelcore -= min(kernel_pages,
6319 required_kernelcore);
6320
6321 /* Continue if range is now fully accounted */
6322 if (end_pfn <= usable_startpfn) {
6323
6324 /*
6325 * Push zone_movable_pfn to the end so
6326 * that if we have to rebalance
6327 * kernelcore across nodes, we will
6328 * not double account here
6329 */
6330 zone_movable_pfn[nid] = end_pfn;
6331 continue;
6332 }
6333 start_pfn = usable_startpfn;
6334 }
6335
6336 /*
6337 * The usable PFN range for ZONE_MOVABLE is from
6338 * start_pfn->end_pfn. Calculate size_pages as the
6339 * number of pages used as kernelcore
6340 */
6341 size_pages = end_pfn - start_pfn;
6342 if (size_pages > kernelcore_remaining)
6343 size_pages = kernelcore_remaining;
6344 zone_movable_pfn[nid] = start_pfn + size_pages;
6345
6346 /*
6347 * Some kernelcore has been met, update counts and
6348 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07006349 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07006350 */
6351 required_kernelcore -= min(required_kernelcore,
6352 size_pages);
6353 kernelcore_remaining -= size_pages;
6354 if (!kernelcore_remaining)
6355 break;
6356 }
6357 }
6358
6359 /*
6360 * If there is still required_kernelcore, we do another pass with one
6361 * less node in the count. This will push zone_movable_pfn[nid] further
6362 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07006363 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07006364 */
6365 usable_nodes--;
6366 if (usable_nodes && required_kernelcore > usable_nodes)
6367 goto restart;
6368
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006369out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07006370 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6371 for (nid = 0; nid < MAX_NUMNODES; nid++)
6372 zone_movable_pfn[nid] =
6373 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07006374
Yinghai Lu20e69262013-03-01 14:51:27 -08006375out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07006376 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006377 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07006378}
6379
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006380/* Any regular or high memory on that node ? */
6381static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006382{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006383 enum zone_type zone_type;
6384
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006385 if (N_MEMORY == N_NORMAL_MEMORY)
6386 return;
6387
6388 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006389 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08006390 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006391 node_set_state(nid, N_HIGH_MEMORY);
6392 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
6393 zone_type <= ZONE_NORMAL)
6394 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08006395 break;
6396 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006397 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006398}
6399
Mel Gormanc7132162006-09-27 01:49:43 -07006400/**
6401 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006402 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07006403 *
6404 * This will call free_area_init_node() for each active node in the system.
Zhang Zhen7d018172014-06-04 16:10:53 -07006405 * Using the page ranges provided by memblock_set_node(), the size of each
Mel Gormanc7132162006-09-27 01:49:43 -07006406 * zone in each node and their holes is calculated. If the maximum PFN
6407 * between two adjacent zones match, it is assumed that the zone is empty.
6408 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6409 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6410 * starts where the previous one ended. For example, ZONE_DMA32 starts
6411 * at arch_max_dma_pfn.
6412 */
6413void __init free_area_init_nodes(unsigned long *max_zone_pfn)
6414{
Tejun Heoc13291a2011-07-12 10:46:30 +02006415 unsigned long start_pfn, end_pfn;
6416 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08006417
Mel Gormanc7132162006-09-27 01:49:43 -07006418 /* Record where the zone boundaries are */
6419 memset(arch_zone_lowest_possible_pfn, 0,
6420 sizeof(arch_zone_lowest_possible_pfn));
6421 memset(arch_zone_highest_possible_pfn, 0,
6422 sizeof(arch_zone_highest_possible_pfn));
Oliver O'Halloran90cae1f2016-07-26 15:22:17 -07006423
6424 start_pfn = find_min_pfn_with_active_regions();
6425
6426 for (i = 0; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07006427 if (i == ZONE_MOVABLE)
6428 continue;
Oliver O'Halloran90cae1f2016-07-26 15:22:17 -07006429
6430 end_pfn = max(max_zone_pfn[i], start_pfn);
6431 arch_zone_lowest_possible_pfn[i] = start_pfn;
6432 arch_zone_highest_possible_pfn[i] = end_pfn;
6433
6434 start_pfn = end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07006435 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07006436 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
6437 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
6438
6439 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6440 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07006441 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07006442
Mel Gormanc7132162006-09-27 01:49:43 -07006443 /* Print out the zone ranges */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006444 pr_info("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07006445 for (i = 0; i < MAX_NR_ZONES; i++) {
6446 if (i == ZONE_MOVABLE)
6447 continue;
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006448 pr_info(" %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08006449 if (arch_zone_lowest_possible_pfn[i] ==
6450 arch_zone_highest_possible_pfn[i])
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006451 pr_cont("empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08006452 else
Juergen Gross8d29e182015-02-11 15:26:01 -08006453 pr_cont("[mem %#018Lx-%#018Lx]\n",
6454 (u64)arch_zone_lowest_possible_pfn[i]
6455 << PAGE_SHIFT,
6456 ((u64)arch_zone_highest_possible_pfn[i]
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07006457 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07006458 }
6459
6460 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006461 pr_info("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07006462 for (i = 0; i < MAX_NUMNODES; i++) {
6463 if (zone_movable_pfn[i])
Juergen Gross8d29e182015-02-11 15:26:01 -08006464 pr_info(" Node %d: %#018Lx\n", i,
6465 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07006466 }
Mel Gormanc7132162006-09-27 01:49:43 -07006467
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07006468 /* Print out the early node map */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006469 pr_info("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02006470 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Juergen Gross8d29e182015-02-11 15:26:01 -08006471 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
6472 (u64)start_pfn << PAGE_SHIFT,
6473 ((u64)end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07006474
6475 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07006476 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08006477 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07006478 for_each_online_node(nid) {
6479 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07006480 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07006481 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006482
6483 /* Any memory on that node */
6484 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006485 node_set_state(nid, N_MEMORY);
6486 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07006487 }
6488}
Mel Gorman2a1e2742007-07-17 04:03:12 -07006489
Mel Gorman7e63efe2007-07-17 04:03:15 -07006490static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07006491{
6492 unsigned long long coremem;
6493 if (!p)
6494 return -EINVAL;
6495
6496 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07006497 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07006498
Mel Gorman7e63efe2007-07-17 04:03:15 -07006499 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07006500 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
6501
6502 return 0;
6503}
Mel Gormaned7ed362007-07-17 04:03:14 -07006504
Mel Gorman7e63efe2007-07-17 04:03:15 -07006505/*
6506 * kernelcore=size sets the amount of memory for use for allocations that
6507 * cannot be reclaimed or migrated.
6508 */
6509static int __init cmdline_parse_kernelcore(char *p)
6510{
Taku Izumi342332e2016-03-15 14:55:22 -07006511 /* parse kernelcore=mirror */
6512 if (parse_option_str(p, "mirror")) {
6513 mirrored_kernelcore = true;
6514 return 0;
6515 }
6516
Mel Gorman7e63efe2007-07-17 04:03:15 -07006517 return cmdline_parse_core(p, &required_kernelcore);
6518}
6519
6520/*
6521 * movablecore=size sets the amount of memory for use for allocations that
6522 * can be reclaimed or migrated.
6523 */
6524static int __init cmdline_parse_movablecore(char *p)
6525{
6526 return cmdline_parse_core(p, &required_movablecore);
6527}
6528
Mel Gormaned7ed362007-07-17 04:03:14 -07006529early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07006530early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07006531
Tejun Heo0ee332c2011-12-08 10:22:09 -08006532#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07006533
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006534void adjust_managed_page_count(struct page *page, long count)
6535{
6536 spin_lock(&managed_page_count_lock);
6537 page_zone(page)->managed_pages += count;
6538 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07006539#ifdef CONFIG_HIGHMEM
6540 if (PageHighMem(page))
6541 totalhigh_pages += count;
6542#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006543 spin_unlock(&managed_page_count_lock);
6544}
Jiang Liu3dcc0572013-07-03 15:03:21 -07006545EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006546
Jiang Liu11199692013-07-03 15:02:48 -07006547unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07006548{
Jiang Liu11199692013-07-03 15:02:48 -07006549 void *pos;
6550 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07006551
Jiang Liu11199692013-07-03 15:02:48 -07006552 start = (void *)PAGE_ALIGN((unsigned long)start);
6553 end = (void *)((unsigned long)end & PAGE_MASK);
6554 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07006555 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07006556 memset(pos, poison, PAGE_SIZE);
6557 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07006558 }
6559
6560 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07006561 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07006562 s, pages << (PAGE_SHIFT - 10), start, end);
6563
6564 return pages;
6565}
Jiang Liu11199692013-07-03 15:02:48 -07006566EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07006567
Jiang Liucfa11e02013-04-29 15:07:00 -07006568#ifdef CONFIG_HIGHMEM
6569void free_highmem_page(struct page *page)
6570{
6571 __free_reserved_page(page);
6572 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07006573 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07006574 totalhigh_pages++;
6575}
6576#endif
6577
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006578
6579void __init mem_init_print_info(const char *str)
6580{
6581 unsigned long physpages, codesize, datasize, rosize, bss_size;
6582 unsigned long init_code_size, init_data_size;
6583
6584 physpages = get_num_physpages();
6585 codesize = _etext - _stext;
6586 datasize = _edata - _sdata;
6587 rosize = __end_rodata - __start_rodata;
6588 bss_size = __bss_stop - __bss_start;
6589 init_data_size = __init_end - __init_begin;
6590 init_code_size = _einittext - _sinittext;
6591
6592 /*
6593 * Detect special cases and adjust section sizes accordingly:
6594 * 1) .init.* may be embedded into .data sections
6595 * 2) .init.text.* may be out of [__init_begin, __init_end],
6596 * please refer to arch/tile/kernel/vmlinux.lds.S.
6597 * 3) .rodata.* may be embedded into .text or .data sections.
6598 */
6599#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07006600 do { \
6601 if (start <= pos && pos < end && size > adj) \
6602 size -= adj; \
6603 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006604
6605 adj_init_size(__init_begin, __init_end, init_data_size,
6606 _sinittext, init_code_size);
6607 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
6608 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
6609 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
6610 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
6611
6612#undef adj_init_size
6613
Joe Perches756a0252016-03-17 14:19:47 -07006614 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006615#ifdef CONFIG_HIGHMEM
Joe Perches756a0252016-03-17 14:19:47 -07006616 ", %luK highmem"
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006617#endif
Joe Perches756a0252016-03-17 14:19:47 -07006618 "%s%s)\n",
6619 nr_free_pages() << (PAGE_SHIFT - 10),
6620 physpages << (PAGE_SHIFT - 10),
6621 codesize >> 10, datasize >> 10, rosize >> 10,
6622 (init_data_size + init_code_size) >> 10, bss_size >> 10,
6623 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10),
6624 totalcma_pages << (PAGE_SHIFT - 10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006625#ifdef CONFIG_HIGHMEM
Joe Perches756a0252016-03-17 14:19:47 -07006626 totalhigh_pages << (PAGE_SHIFT - 10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006627#endif
Joe Perches756a0252016-03-17 14:19:47 -07006628 str ? ", " : "", str ? str : "");
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006629}
6630
Mel Gorman0e0b8642006-09-27 01:49:56 -07006631/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006632 * set_dma_reserve - set the specified number of pages reserved in the first zone
6633 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07006634 *
Yaowei Bai013110a2015-09-08 15:04:10 -07006635 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
Mel Gorman0e0b8642006-09-27 01:49:56 -07006636 * In the DMA zone, a significant percentage may be consumed by kernel image
6637 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006638 * function may optionally be used to account for unfreeable pages in the
6639 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6640 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07006641 */
6642void __init set_dma_reserve(unsigned long new_dma_reserve)
6643{
6644 dma_reserve = new_dma_reserve;
6645}
6646
Linus Torvalds1da177e2005-04-16 15:20:36 -07006647void __init free_area_init(unsigned long *zones_size)
6648{
Johannes Weiner9109fb72008-07-23 21:27:20 -07006649 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006650 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
6651}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006652
Linus Torvalds1da177e2005-04-16 15:20:36 -07006653static int page_alloc_cpu_notify(struct notifier_block *self,
6654 unsigned long action, void *hcpu)
6655{
6656 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006657
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07006658 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07006659 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006660 drain_pages(cpu);
6661
6662 /*
6663 * Spill the event counters of the dead processor
6664 * into the current processors event counters.
6665 * This artificially elevates the count of the current
6666 * processor.
6667 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07006668 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006669
6670 /*
6671 * Zero the differential counters of the dead processor
6672 * so that the vm statistics are consistent.
6673 *
6674 * This is only okay since the processor is dead and cannot
6675 * race with what we are doing.
6676 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07006677 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006678 }
6679 return NOTIFY_OK;
6680}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006681
6682void __init page_alloc_init(void)
6683{
6684 hotcpu_notifier(page_alloc_cpu_notify, 0);
6685}
6686
6687/*
Yaowei Bai34b10062015-09-08 15:04:13 -07006688 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006689 * or min_free_kbytes changes.
6690 */
6691static void calculate_totalreserve_pages(void)
6692{
6693 struct pglist_data *pgdat;
6694 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006695 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006696
6697 for_each_online_pgdat(pgdat) {
6698 for (i = 0; i < MAX_NR_ZONES; i++) {
6699 struct zone *zone = pgdat->node_zones + i;
Mel Gorman3484b2d2014-08-06 16:07:14 -07006700 long max = 0;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006701
6702 /* Find valid and maximum lowmem_reserve in the zone */
6703 for (j = i; j < MAX_NR_ZONES; j++) {
6704 if (zone->lowmem_reserve[j] > max)
6705 max = zone->lowmem_reserve[j];
6706 }
6707
Mel Gorman41858962009-06-16 15:32:12 -07006708 /* we treat the high watermark as reserved pages. */
6709 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006710
Jiang Liub40da042013-02-22 16:33:52 -08006711 if (max > zone->managed_pages)
6712 max = zone->managed_pages;
Johannes Weinera8d01432016-01-14 15:20:15 -08006713
6714 zone->totalreserve_pages = max;
6715
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006716 reserve_pages += max;
6717 }
6718 }
6719 totalreserve_pages = reserve_pages;
6720}
6721
6722/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07006723 * setup_per_zone_lowmem_reserve - called whenever
Yaowei Bai34b10062015-09-08 15:04:13 -07006724 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
Linus Torvalds1da177e2005-04-16 15:20:36 -07006725 * has a correct pages reserved value, so an adequate number of
6726 * pages are left in the zone after a successful __alloc_pages().
6727 */
6728static void setup_per_zone_lowmem_reserve(void)
6729{
6730 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006731 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006732
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08006733 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006734 for (j = 0; j < MAX_NR_ZONES; j++) {
6735 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08006736 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006737
6738 zone->lowmem_reserve[j] = 0;
6739
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006740 idx = j;
6741 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006742 struct zone *lower_zone;
6743
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006744 idx--;
6745
Linus Torvalds1da177e2005-04-16 15:20:36 -07006746 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6747 sysctl_lowmem_reserve_ratio[idx] = 1;
6748
6749 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08006750 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07006751 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08006752 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006753 }
6754 }
6755 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006756
6757 /* update totalreserve_pages */
6758 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006759}
6760
Mel Gormancfd3da12011-04-25 21:36:42 +00006761static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006762{
6763 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6764 unsigned long lowmem_pages = 0;
6765 struct zone *zone;
6766 unsigned long flags;
6767
6768 /* Calculate total number of !ZONE_HIGHMEM pages */
6769 for_each_zone(zone) {
6770 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08006771 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006772 }
6773
6774 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07006775 u64 tmp;
6776
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006777 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08006778 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07006779 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006780 if (is_highmem(zone)) {
6781 /*
Nick Piggin669ed172005-11-13 16:06:45 -08006782 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6783 * need highmem pages, so cap pages_min to a small
6784 * value here.
6785 *
Mel Gorman41858962009-06-16 15:32:12 -07006786 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Yaowei Bai42ff2702015-04-14 15:47:14 -07006787 * deltas control asynch page reclaim, and so should
Nick Piggin669ed172005-11-13 16:06:45 -08006788 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006789 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08006790 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006791
Jiang Liub40da042013-02-22 16:33:52 -08006792 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08006793 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07006794 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006795 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08006796 /*
6797 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07006798 * proportionate to the zone's size.
6799 */
Mel Gorman41858962009-06-16 15:32:12 -07006800 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006801 }
6802
Johannes Weiner795ae7a2016-03-17 14:19:14 -07006803 /*
6804 * Set the kswapd watermarks distance according to the
6805 * scale factor in proportion to available memory, but
6806 * ensure a minimum size on small systems.
6807 */
6808 tmp = max_t(u64, tmp >> 2,
6809 mult_frac(zone->managed_pages,
6810 watermark_scale_factor, 10000));
6811
6812 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp;
6813 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006814
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006815 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
Johannes Weinerabe5f972014-10-02 16:21:10 -07006816 high_wmark_pages(zone) - low_wmark_pages(zone) -
6817 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006818
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006819 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006820 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006821
6822 /* update totalreserve_pages */
6823 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006824}
6825
Mel Gormancfd3da12011-04-25 21:36:42 +00006826/**
6827 * setup_per_zone_wmarks - called when min_free_kbytes changes
6828 * or when memory is hot-{added|removed}
6829 *
6830 * Ensures that the watermark[min,low,high] values for each zone are set
6831 * correctly with respect to min_free_kbytes.
6832 */
6833void setup_per_zone_wmarks(void)
6834{
6835 mutex_lock(&zonelists_mutex);
6836 __setup_per_zone_wmarks();
6837 mutex_unlock(&zonelists_mutex);
6838}
6839
Randy Dunlap55a44622009-09-21 17:01:20 -07006840/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07006841 * Initialise min_free_kbytes.
6842 *
6843 * For small machines we want it small (128k min). For large machines
6844 * we want it large (64MB max). But it is not linear, because network
6845 * bandwidth does not increase linearly with machine size. We use
6846 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006847 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07006848 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6849 *
6850 * which yields
6851 *
6852 * 16MB: 512k
6853 * 32MB: 724k
6854 * 64MB: 1024k
6855 * 128MB: 1448k
6856 * 256MB: 2048k
6857 * 512MB: 2896k
6858 * 1024MB: 4096k
6859 * 2048MB: 5792k
6860 * 4096MB: 8192k
6861 * 8192MB: 11584k
6862 * 16384MB: 16384k
6863 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07006864int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006865{
6866 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07006867 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006868
6869 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07006870 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006871
Michal Hocko5f127332013-07-08 16:00:40 -07006872 if (new_min_free_kbytes > user_min_free_kbytes) {
6873 min_free_kbytes = new_min_free_kbytes;
6874 if (min_free_kbytes < 128)
6875 min_free_kbytes = 128;
6876 if (min_free_kbytes > 65536)
6877 min_free_kbytes = 65536;
6878 } else {
6879 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6880 new_min_free_kbytes, user_min_free_kbytes);
6881 }
Minchan Kimbc75d332009-06-16 15:32:48 -07006882 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07006883 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006884 setup_per_zone_lowmem_reserve();
6885 return 0;
6886}
Jason Baronbc22af72016-05-05 16:22:12 -07006887core_initcall(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006888
6889/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07006890 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07006891 * that we can call two helper functions whenever min_free_kbytes
6892 * changes.
6893 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006894int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006895 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006896{
Han Pingtianda8c7572014-01-23 15:53:17 -08006897 int rc;
6898
6899 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6900 if (rc)
6901 return rc;
6902
Michal Hocko5f127332013-07-08 16:00:40 -07006903 if (write) {
6904 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07006905 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07006906 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006907 return 0;
6908}
6909
Johannes Weiner795ae7a2016-03-17 14:19:14 -07006910int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
6911 void __user *buffer, size_t *length, loff_t *ppos)
6912{
6913 int rc;
6914
6915 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6916 if (rc)
6917 return rc;
6918
6919 if (write)
6920 setup_per_zone_wmarks();
6921
6922 return 0;
6923}
6924
Christoph Lameter96146342006-07-03 00:24:13 -07006925#ifdef CONFIG_NUMA
Joe Perchescccad5b2014-06-06 14:38:09 -07006926int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006927 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07006928{
6929 struct zone *zone;
6930 int rc;
6931
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006932 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07006933 if (rc)
6934 return rc;
6935
6936 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08006937 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07006938 sysctl_min_unmapped_ratio) / 100;
6939 return 0;
6940}
Christoph Lameter0ff38492006-09-25 23:31:52 -07006941
Joe Perchescccad5b2014-06-06 14:38:09 -07006942int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006943 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07006944{
6945 struct zone *zone;
6946 int rc;
6947
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006948 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07006949 if (rc)
6950 return rc;
6951
6952 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08006953 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07006954 sysctl_min_slab_ratio) / 100;
6955 return 0;
6956}
Christoph Lameter96146342006-07-03 00:24:13 -07006957#endif
6958
Linus Torvalds1da177e2005-04-16 15:20:36 -07006959/*
6960 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6961 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6962 * whenever sysctl_lowmem_reserve_ratio changes.
6963 *
6964 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07006965 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07006966 * if in function of the boot time zone sizes.
6967 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006968int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006969 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006970{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006971 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006972 setup_per_zone_lowmem_reserve();
6973 return 0;
6974}
6975
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006976/*
6977 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07006978 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6979 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006980 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006981int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006982 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006983{
6984 struct zone *zone;
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006985 int old_percpu_pagelist_fraction;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006986 int ret;
6987
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006988 mutex_lock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006989 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6990
6991 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
6992 if (!write || ret < 0)
6993 goto out;
6994
6995 /* Sanity checking to avoid pcp imbalance */
6996 if (percpu_pagelist_fraction &&
6997 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6998 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6999 ret = -EINVAL;
7000 goto out;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08007001 }
David Rientjes7cd2b0a2014-06-23 13:22:04 -07007002
7003 /* No change? */
7004 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
7005 goto out;
7006
7007 for_each_populated_zone(zone) {
7008 unsigned int cpu;
7009
7010 for_each_possible_cpu(cpu)
7011 pageset_set_high_and_batch(zone,
7012 per_cpu_ptr(zone->pageset, cpu));
7013 }
7014out:
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007015 mutex_unlock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07007016 return ret;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08007017}
7018
Rasmus Villemoesa9919c72015-06-24 16:56:28 -07007019#ifdef CONFIG_NUMA
David S. Millerf034b5d2006-08-24 03:08:07 -07007020int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007021
Linus Torvalds1da177e2005-04-16 15:20:36 -07007022static int __init set_hashdist(char *str)
7023{
7024 if (!str)
7025 return 0;
7026 hashdist = simple_strtoul(str, &str, 0);
7027 return 1;
7028}
7029__setup("hashdist=", set_hashdist);
7030#endif
7031
7032/*
7033 * allocate a large system hash table from bootmem
7034 * - it is assumed that the hash table must contain an exact power-of-2
7035 * quantity of entries
7036 * - limit is the number of hash buckets, not the total allocation size
7037 */
7038void *__init alloc_large_system_hash(const char *tablename,
7039 unsigned long bucketsize,
7040 unsigned long numentries,
7041 int scale,
7042 int flags,
7043 unsigned int *_hash_shift,
7044 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00007045 unsigned long low_limit,
7046 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007047{
Tim Bird31fe62b2012-05-23 13:33:35 +00007048 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007049 unsigned long log2qty, size;
7050 void *table = NULL;
7051
7052 /* allow the kernel cmdline to have a say */
7053 if (!numentries) {
7054 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08007055 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07007056
7057 /* It isn't necessary when PAGE_SIZE >= 1MB */
7058 if (PAGE_SHIFT < 20)
7059 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007060
7061 /* limit to 1 bucket per 2^scale bytes of low memory */
7062 if (scale > PAGE_SHIFT)
7063 numentries >>= (scale - PAGE_SHIFT);
7064 else
7065 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08007066
7067 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07007068 if (unlikely(flags & HASH_SMALL)) {
7069 /* Makes no sense without HASH_EARLY */
7070 WARN_ON(!(flags & HASH_EARLY));
7071 if (!(numentries >> *_hash_shift)) {
7072 numentries = 1UL << *_hash_shift;
7073 BUG_ON(!numentries);
7074 }
7075 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08007076 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007077 }
John Hawkes6e692ed2006-03-25 03:08:02 -08007078 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007079
7080 /* limit allocation size to 1/16 total memory by default */
7081 if (max == 0) {
7082 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
7083 do_div(max, bucketsize);
7084 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08007085 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007086
Tim Bird31fe62b2012-05-23 13:33:35 +00007087 if (numentries < low_limit)
7088 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007089 if (numentries > max)
7090 numentries = max;
7091
David Howellsf0d1b0b2006-12-08 02:37:49 -08007092 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007093
7094 do {
7095 size = bucketsize << log2qty;
7096 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08007097 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007098 else if (hashdist)
7099 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
7100 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07007101 /*
7102 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07007103 * some pages at the end of hash table which
7104 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07007105 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01007106 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07007107 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01007108 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
7109 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007110 }
7111 } while (!table && size > PAGE_SIZE && --log2qty);
7112
7113 if (!table)
7114 panic("Failed to allocate %s hash table\n", tablename);
7115
Joe Perches11705322016-03-17 14:19:50 -07007116 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7117 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007118
7119 if (_hash_shift)
7120 *_hash_shift = log2qty;
7121 if (_hash_mask)
7122 *_hash_mask = (1 << log2qty) - 1;
7123
7124 return table;
7125}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08007126
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07007127/*
Minchan Kim80934512012-07-31 16:43:01 -07007128 * This function checks whether pageblock includes unmovable pages or not.
7129 * If @count is not zero, it is okay to include less @count unmovable pages
7130 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07007131 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07007132 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
7133 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07007134 */
Wen Congyangb023f462012-12-11 16:00:45 -08007135bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
7136 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007137{
7138 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01007139 int mt;
7140
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007141 /*
7142 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07007143 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007144 */
7145 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07007146 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01007147 mt = get_pageblock_migratetype(page);
7148 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07007149 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007150
7151 pfn = page_to_pfn(page);
7152 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
7153 unsigned long check = pfn + iter;
7154
Namhyung Kim29723fc2011-02-25 14:44:25 -08007155 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007156 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08007157
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007158 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07007159
7160 /*
7161 * Hugepages are not in LRU lists, but they're movable.
7162 * We need not scan over tail pages bacause we don't
7163 * handle each tail page individually in migration.
7164 */
7165 if (PageHuge(page)) {
7166 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
7167 continue;
7168 }
7169
Minchan Kim97d255c2012-07-31 16:42:59 -07007170 /*
7171 * We can't use page_count without pin a page
7172 * because another CPU can free compound page.
7173 * This check already skips compound tails of THP
Joonsoo Kim0139aa72016-05-19 17:10:49 -07007174 * because their page->_refcount is zero at all time.
Minchan Kim97d255c2012-07-31 16:42:59 -07007175 */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07007176 if (!page_ref_count(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007177 if (PageBuddy(page))
7178 iter += (1 << page_order(page)) - 1;
7179 continue;
7180 }
Minchan Kim97d255c2012-07-31 16:42:59 -07007181
Wen Congyangb023f462012-12-11 16:00:45 -08007182 /*
7183 * The HWPoisoned page may be not in buddy system, and
7184 * page_count() is not 0.
7185 */
7186 if (skip_hwpoisoned_pages && PageHWPoison(page))
7187 continue;
7188
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007189 if (!PageLRU(page))
7190 found++;
7191 /*
Johannes Weiner6b4f7792014-12-12 16:56:13 -08007192 * If there are RECLAIMABLE pages, we need to check
7193 * it. But now, memory offline itself doesn't call
7194 * shrink_node_slabs() and it still to be fixed.
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007195 */
7196 /*
7197 * If the page is not RAM, page_count()should be 0.
7198 * we don't need more check. This is an _used_ not-movable page.
7199 *
7200 * The problematic thing here is PG_reserved pages. PG_reserved
7201 * is set to both of a memory hole page and a _used_ kernel
7202 * page at boot.
7203 */
7204 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07007205 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007206 }
Minchan Kim80934512012-07-31 16:43:01 -07007207 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007208}
7209
7210bool is_pageblock_removable_nolock(struct page *page)
7211{
Michal Hocko656a0702012-01-20 14:33:58 -08007212 struct zone *zone;
7213 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08007214
7215 /*
7216 * We have to be careful here because we are iterating over memory
7217 * sections which are not zone aware so we might end up outside of
7218 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08007219 * We have to take care about the node as well. If the node is offline
7220 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08007221 */
Michal Hocko656a0702012-01-20 14:33:58 -08007222 if (!node_online(page_to_nid(page)))
7223 return false;
7224
7225 zone = page_zone(page);
7226 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08007227 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08007228 return false;
7229
Wen Congyangb023f462012-12-11 16:00:45 -08007230 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07007231}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007232
Vlastimil Babka080fe202016-02-05 15:36:41 -08007233#if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007234
7235static unsigned long pfn_max_align_down(unsigned long pfn)
7236{
7237 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
7238 pageblock_nr_pages) - 1);
7239}
7240
7241static unsigned long pfn_max_align_up(unsigned long pfn)
7242{
7243 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
7244 pageblock_nr_pages));
7245}
7246
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007247/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007248static int __alloc_contig_migrate_range(struct compact_control *cc,
7249 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007250{
7251 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07007252 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007253 unsigned long pfn = start;
7254 unsigned int tries = 0;
7255 int ret = 0;
7256
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08007257 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007258
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007259 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007260 if (fatal_signal_pending(current)) {
7261 ret = -EINTR;
7262 break;
7263 }
7264
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007265 if (list_empty(&cc->migratepages)) {
7266 cc->nr_migratepages = 0;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07007267 pfn = isolate_migratepages_range(cc, pfn, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007268 if (!pfn) {
7269 ret = -EINTR;
7270 break;
7271 }
7272 tries = 0;
7273 } else if (++tries == 5) {
7274 ret = ret < 0 ? ret : -EBUSY;
7275 break;
7276 }
7277
Minchan Kimbeb51ea2012-10-08 16:33:51 -07007278 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
7279 &cc->migratepages);
7280 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07007281
Hugh Dickins9c620e22013-02-22 16:35:14 -08007282 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07007283 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007284 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08007285 if (ret < 0) {
7286 putback_movable_pages(&cc->migratepages);
7287 return ret;
7288 }
7289 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007290}
7291
7292/**
7293 * alloc_contig_range() -- tries to allocate given range of pages
7294 * @start: start PFN to allocate
7295 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02007296 * @migratetype: migratetype of the underlaying pageblocks (either
7297 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
7298 * in range must have the same migratetype and it must
7299 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007300 *
7301 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7302 * aligned, however it's the caller's responsibility to guarantee that
7303 * we are the only thread that changes migrate type of pageblocks the
7304 * pages fall in.
7305 *
7306 * The PFN range must belong to a single zone.
7307 *
7308 * Returns zero on success or negative error code. On success all
7309 * pages which PFN is in [start, end) are allocated for the caller and
7310 * need to be freed with free_contig_range().
7311 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02007312int alloc_contig_range(unsigned long start, unsigned long end,
7313 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007314{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007315 unsigned long outer_start, outer_end;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08007316 unsigned int order;
7317 int ret = 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007318
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007319 struct compact_control cc = {
7320 .nr_migratepages = 0,
7321 .order = -1,
7322 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07007323 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007324 .ignore_skip_hint = true,
7325 };
7326 INIT_LIST_HEAD(&cc.migratepages);
7327
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007328 /*
7329 * What we do here is we mark all pageblocks in range as
7330 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7331 * have different sizes, and due to the way page allocator
7332 * work, we align the range to biggest of the two pages so
7333 * that page allocator won't try to merge buddies from
7334 * different pageblocks and change MIGRATE_ISOLATE to some
7335 * other migration type.
7336 *
7337 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7338 * migrate the pages from an unaligned range (ie. pages that
7339 * we are interested in). This will put all the pages in
7340 * range back to page allocator as MIGRATE_ISOLATE.
7341 *
7342 * When this is done, we take the pages in range from page
7343 * allocator removing them from the buddy system. This way
7344 * page allocator will never consider using them.
7345 *
7346 * This lets us mark the pageblocks back as
7347 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7348 * aligned range but not in the unaligned, original range are
7349 * put back to page allocator so that buddy can use them.
7350 */
7351
7352 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08007353 pfn_max_align_up(end), migratetype,
7354 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007355 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07007356 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007357
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007358 /*
7359 * In case of -EBUSY, we'd like to know which page causes problem.
7360 * So, just fall through. We will check it in test_pages_isolated().
7361 */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007362 ret = __alloc_contig_migrate_range(&cc, start, end);
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007363 if (ret && ret != -EBUSY)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007364 goto done;
7365
7366 /*
7367 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7368 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7369 * more, all pages in [start, end) are free in page allocator.
7370 * What we are going to do is to allocate all pages from
7371 * [start, end) (that is remove them from page allocator).
7372 *
7373 * The only problem is that pages at the beginning and at the
7374 * end of interesting range may be not aligned with pages that
7375 * page allocator holds, ie. they can be part of higher order
7376 * pages. Because of this, we reserve the bigger range and
7377 * once this is done free the pages we are not interested in.
7378 *
7379 * We don't have to hold zone->lock here because the pages are
7380 * isolated thus they won't get removed from buddy.
7381 */
7382
7383 lru_add_drain_all();
Vlastimil Babka510f5502014-12-10 15:43:07 -08007384 drain_all_pages(cc.zone);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007385
7386 order = 0;
7387 outer_start = start;
7388 while (!PageBuddy(pfn_to_page(outer_start))) {
7389 if (++order >= MAX_ORDER) {
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007390 outer_start = start;
7391 break;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007392 }
7393 outer_start &= ~0UL << order;
7394 }
7395
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007396 if (outer_start != start) {
7397 order = page_order(pfn_to_page(outer_start));
7398
7399 /*
7400 * outer_start page could be small order buddy page and
7401 * it doesn't include start page. Adjust outer_start
7402 * in this case to report failed page properly
7403 * on tracepoint in test_pages_isolated()
7404 */
7405 if (outer_start + (1UL << order) <= start)
7406 outer_start = start;
7407 }
7408
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007409 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08007410 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewiczdae803e2014-11-13 15:19:27 -08007411 pr_info("%s: [%lx, %lx) PFNs busy\n",
7412 __func__, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007413 ret = -EBUSY;
7414 goto done;
7415 }
7416
Marek Szyprowski49f223a2012-01-25 12:49:24 +01007417 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007418 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007419 if (!outer_end) {
7420 ret = -EBUSY;
7421 goto done;
7422 }
7423
7424 /* Free head and tail (if any) */
7425 if (start != outer_start)
7426 free_contig_range(outer_start, start - outer_start);
7427 if (end != outer_end)
7428 free_contig_range(end, outer_end - end);
7429
7430done:
7431 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02007432 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007433 return ret;
7434}
7435
7436void free_contig_range(unsigned long pfn, unsigned nr_pages)
7437{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08007438 unsigned int count = 0;
7439
7440 for (; nr_pages--; pfn++) {
7441 struct page *page = pfn_to_page(pfn);
7442
7443 count += page_count(page) != 1;
7444 __free_page(page);
7445 }
7446 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007447}
7448#endif
7449
Jiang Liu4ed7e022012-07-31 16:43:35 -07007450#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07007451/*
7452 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7453 * page high values need to be recalulated.
7454 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07007455void __meminit zone_pcp_update(struct zone *zone)
7456{
Cody P Schafer0a647f32013-07-03 15:01:33 -07007457 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007458 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07007459 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07007460 pageset_set_high_and_batch(zone,
7461 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007462 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07007463}
7464#endif
7465
Jiang Liu340175b2012-07-31 16:43:32 -07007466void zone_pcp_reset(struct zone *zone)
7467{
7468 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07007469 int cpu;
7470 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07007471
7472 /* avoid races with drain_pages() */
7473 local_irq_save(flags);
7474 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07007475 for_each_online_cpu(cpu) {
7476 pset = per_cpu_ptr(zone->pageset, cpu);
7477 drain_zonestat(zone, pset);
7478 }
Jiang Liu340175b2012-07-31 16:43:32 -07007479 free_percpu(zone->pageset);
7480 zone->pageset = &boot_pageset;
7481 }
7482 local_irq_restore(flags);
7483}
7484
Wen Congyang6dcd73d2012-12-11 16:01:01 -08007485#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007486/*
Joonsoo Kimb9eb6312016-05-19 17:12:06 -07007487 * All pages in the range must be in a single zone and isolated
7488 * before calling this.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007489 */
7490void
7491__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
7492{
7493 struct page *page;
7494 struct zone *zone;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07007495 unsigned int order, i;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007496 unsigned long pfn;
7497 unsigned long flags;
7498 /* find the first valid pfn */
7499 for (pfn = start_pfn; pfn < end_pfn; pfn++)
7500 if (pfn_valid(pfn))
7501 break;
7502 if (pfn == end_pfn)
7503 return;
7504 zone = page_zone(pfn_to_page(pfn));
7505 spin_lock_irqsave(&zone->lock, flags);
7506 pfn = start_pfn;
7507 while (pfn < end_pfn) {
7508 if (!pfn_valid(pfn)) {
7509 pfn++;
7510 continue;
7511 }
7512 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08007513 /*
7514 * The HWPoisoned page may be not in buddy system, and
7515 * page_count() is not 0.
7516 */
7517 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
7518 pfn++;
7519 SetPageReserved(page);
7520 continue;
7521 }
7522
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007523 BUG_ON(page_count(page));
7524 BUG_ON(!PageBuddy(page));
7525 order = page_order(page);
7526#ifdef CONFIG_DEBUG_VM
Joe Perches11705322016-03-17 14:19:50 -07007527 pr_info("remove from free list %lx %d %lx\n",
7528 pfn, 1 << order, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007529#endif
7530 list_del(&page->lru);
7531 rmv_page_order(page);
7532 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007533 for (i = 0; i < (1 << order); i++)
7534 SetPageReserved((page+i));
7535 pfn += (1 << order);
7536 }
7537 spin_unlock_irqrestore(&zone->lock, flags);
7538}
7539#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007540
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007541bool is_free_buddy_page(struct page *page)
7542{
7543 struct zone *zone = page_zone(page);
7544 unsigned long pfn = page_to_pfn(page);
7545 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07007546 unsigned int order;
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007547
7548 spin_lock_irqsave(&zone->lock, flags);
7549 for (order = 0; order < MAX_ORDER; order++) {
7550 struct page *page_head = page - (pfn & ((1 << order) - 1));
7551
7552 if (PageBuddy(page_head) && page_order(page_head) >= order)
7553 break;
7554 }
7555 spin_unlock_irqrestore(&zone->lock, flags);
7556
7557 return order < MAX_ORDER;
7558}