blob: 3fbe73a6fe4b6869dcd44a45de43928d4c08fe0c [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 Gruszkac0a32fc52012-01-10 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 Gruszkac0a32fc52012-01-10 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 Gruszkac0a32fc52012-01-10 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 Gruszkac0a32fc52012-01-10 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 Gruszkac0a32fc52012-01-10 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 Gruszkac0a32fc52012-01-10 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 Gruszkac0a32fc52012-01-10 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 Gruszkac0a32fc52012-01-10 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 Gruszkac0a32fc52012-01-10 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 Gruszkac0a32fc52012-01-10 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 Gruszkac0a32fc52012-01-10 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 Gruszkac0a32fc52012-01-10 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 Gruszkac0a32fc52012-01-10 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 Davydovc4159a72016-08-08 23:03:12 +03001011 if (memcg_kmem_enabled() && PageKmemcg(page))
Vladimir Davydov49491482016-07-26 15:24:24 -07001012 memcg_kmem_uncharge(page, order);
Mel Gormane2769db2016-05-19 17:14:38 -07001013 if (check_free)
1014 bad += free_pages_check(page);
1015 if (bad)
1016 return false;
1017
1018 page_cpupid_reset_last(page);
1019 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1020 reset_page_owner(page, order);
1021
1022 if (!PageHighMem(page)) {
1023 debug_check_no_locks_freed(page_address(page),
1024 PAGE_SIZE << order);
1025 debug_check_no_obj_freed(page_address(page),
1026 PAGE_SIZE << order);
1027 }
1028 arch_free_page(page, order);
1029 kernel_poison_pages(page, 1 << order, 0);
1030 kernel_map_pages(page, 1 << order, 0);
seokhoon.yoon29b52de2016-05-20 16:58:47 -07001031 kasan_free_pages(page, order);
Mel Gormane2769db2016-05-19 17:14:38 -07001032
1033 return true;
1034}
Mel Gorman4db75482016-05-19 17:14:32 -07001035
1036#ifdef CONFIG_DEBUG_VM
1037static inline bool free_pcp_prepare(struct page *page)
1038{
Mel Gormane2769db2016-05-19 17:14:38 -07001039 return free_pages_prepare(page, 0, true);
Mel Gorman4db75482016-05-19 17:14:32 -07001040}
1041
1042static inline bool bulkfree_pcp_prepare(struct page *page)
1043{
1044 return false;
1045}
1046#else
1047static bool free_pcp_prepare(struct page *page)
1048{
Mel Gormane2769db2016-05-19 17:14:38 -07001049 return free_pages_prepare(page, 0, false);
Mel Gorman4db75482016-05-19 17:14:32 -07001050}
1051
1052static bool bulkfree_pcp_prepare(struct page *page)
1053{
1054 return free_pages_check(page);
1055}
1056#endif /* CONFIG_DEBUG_VM */
1057
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001059 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -07001061 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 *
1063 * If the zone was previously in an "all pages pinned" state then look to
1064 * see if this freeing clears that state.
1065 *
1066 * And clear the zone's pages_scanned counter, to hold off the "all pages are
1067 * pinned" detection logic.
1068 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001069static void free_pcppages_bulk(struct zone *zone, int count,
1070 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071{
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001072 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -07001073 int batch_free = 0;
Mel Gorman0d5d8232014-08-06 16:07:16 -07001074 unsigned long nr_scanned;
Mel Gorman37779992016-05-19 17:13:58 -07001075 bool isolated_pageblocks;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001076
Nick Pigginc54ad302006-01-06 00:10:56 -08001077 spin_lock(&zone->lock);
Mel Gorman37779992016-05-19 17:13:58 -07001078 isolated_pageblocks = has_isolate_pageblock(zone);
Mel Gorman599d0c92016-07-28 15:45:31 -07001079 nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED);
Mel Gorman0d5d8232014-08-06 16:07:16 -07001080 if (nr_scanned)
Mel Gorman599d0c92016-07-28 15:45:31 -07001081 __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -07001082
Mel Gormane5b31ac2016-05-19 17:14:24 -07001083 while (count) {
Nick Piggin48db57f2006-01-08 01:00:42 -08001084 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001085 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -08001086
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001087 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -07001088 * Remove pages from lists in a round-robin fashion. A
1089 * batch_free count is maintained that is incremented when an
1090 * empty list is encountered. This is so more pages are freed
1091 * off fuller lists instead of spinning excessively around empty
1092 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001093 */
1094 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -07001095 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001096 if (++migratetype == MIGRATE_PCPTYPES)
1097 migratetype = 0;
1098 list = &pcp->lists[migratetype];
1099 } while (list_empty(list));
1100
Namhyung Kim1d168712011-03-22 16:32:45 -07001101 /* This is the only non-empty list. Free them all. */
1102 if (batch_free == MIGRATE_PCPTYPES)
Mel Gormane5b31ac2016-05-19 17:14:24 -07001103 batch_free = count;
Namhyung Kim1d168712011-03-22 16:32:45 -07001104
Mel Gormana6f9edd2009-09-21 17:03:20 -07001105 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -07001106 int mt; /* migratetype of the to-be-freed page */
1107
Geliang Tanga16601c2016-01-14 15:20:30 -08001108 page = list_last_entry(list, struct page, lru);
Mel Gormana6f9edd2009-09-21 17:03:20 -07001109 /* must delete as __free_one_page list manipulates */
1110 list_del(&page->lru);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -07001111
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001112 mt = get_pcppage_migratetype(page);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -07001113 /* MIGRATE_ISOLATE page should not go to pcplists */
1114 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1115 /* Pageblock could have been isolated meanwhile */
Mel Gorman37779992016-05-19 17:13:58 -07001116 if (unlikely(isolated_pageblocks))
Joonsoo Kim51bb1a42014-11-13 15:19:14 -08001117 mt = get_pageblock_migratetype(page);
Joonsoo Kim51bb1a42014-11-13 15:19:14 -08001118
Mel Gorman4db75482016-05-19 17:14:32 -07001119 if (bulkfree_pcp_prepare(page))
1120 continue;
1121
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001122 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -07001123 trace_mm_page_pcpu_drain(page, 0, mt);
Mel Gormane5b31ac2016-05-19 17:14:24 -07001124 } while (--count && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 }
Nick Pigginc54ad302006-01-06 00:10:56 -08001126 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127}
1128
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001129static void free_one_page(struct zone *zone,
1130 struct page *page, unsigned long pfn,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001131 unsigned int order,
Mel Gormaned0ae212009-06-16 15:32:07 -07001132 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -08001133{
Mel Gorman0d5d8232014-08-06 16:07:16 -07001134 unsigned long nr_scanned;
Christoph Lameter006d22d2006-09-25 23:31:48 -07001135 spin_lock(&zone->lock);
Mel Gorman599d0c92016-07-28 15:45:31 -07001136 nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED);
Mel Gorman0d5d8232014-08-06 16:07:16 -07001137 if (nr_scanned)
Mel Gorman599d0c92016-07-28 15:45:31 -07001138 __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -07001139
Joonsoo Kimad53f922014-11-13 15:19:11 -08001140 if (unlikely(has_isolate_pageblock(zone) ||
1141 is_migrate_isolate(migratetype))) {
1142 migratetype = get_pfnblock_migratetype(page, pfn);
Joonsoo Kimad53f922014-11-13 15:19:11 -08001143 }
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001144 __free_one_page(page, pfn, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -07001145 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -08001146}
1147
Robin Holt1e8ce832015-06-30 14:56:45 -07001148static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1149 unsigned long zone, int nid)
1150{
Robin Holt1e8ce832015-06-30 14:56:45 -07001151 set_page_links(page, zone, nid, pfn);
Robin Holt1e8ce832015-06-30 14:56:45 -07001152 init_page_count(page);
1153 page_mapcount_reset(page);
1154 page_cpupid_reset_last(page);
Robin Holt1e8ce832015-06-30 14:56:45 -07001155
Robin Holt1e8ce832015-06-30 14:56:45 -07001156 INIT_LIST_HEAD(&page->lru);
1157#ifdef WANT_PAGE_VIRTUAL
1158 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1159 if (!is_highmem_idx(zone))
1160 set_page_address(page, __va(pfn << PAGE_SHIFT));
1161#endif
1162}
1163
1164static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
1165 int nid)
1166{
1167 return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
1168}
1169
Mel Gorman7e18adb2015-06-30 14:57:05 -07001170#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1171static void init_reserved_page(unsigned long pfn)
1172{
1173 pg_data_t *pgdat;
1174 int nid, zid;
1175
1176 if (!early_page_uninitialised(pfn))
1177 return;
1178
1179 nid = early_pfn_to_nid(pfn);
1180 pgdat = NODE_DATA(nid);
1181
1182 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1183 struct zone *zone = &pgdat->node_zones[zid];
1184
1185 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1186 break;
1187 }
1188 __init_single_pfn(pfn, zid, nid);
1189}
1190#else
1191static inline void init_reserved_page(unsigned long pfn)
1192{
1193}
1194#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1195
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001196/*
1197 * Initialised pages do not have PageReserved set. This function is
1198 * called for each range allocated by the bootmem allocator and
1199 * marks the pages PageReserved. The remaining valid pages are later
1200 * sent to the buddy page allocator.
1201 */
Stefan Bader4b50bcc2016-05-20 16:58:38 -07001202void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001203{
1204 unsigned long start_pfn = PFN_DOWN(start);
1205 unsigned long end_pfn = PFN_UP(end);
1206
Mel Gorman7e18adb2015-06-30 14:57:05 -07001207 for (; start_pfn < end_pfn; start_pfn++) {
1208 if (pfn_valid(start_pfn)) {
1209 struct page *page = pfn_to_page(start_pfn);
1210
1211 init_reserved_page(start_pfn);
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -08001212
1213 /* Avoid false-positive PageTail() */
1214 INIT_LIST_HEAD(&page->lru);
1215
Mel Gorman7e18adb2015-06-30 14:57:05 -07001216 SetPageReserved(page);
1217 }
1218 }
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001219}
1220
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001221static void __free_pages_ok(struct page *page, unsigned int order)
1222{
1223 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -07001224 int migratetype;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001225 unsigned long pfn = page_to_pfn(page);
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001226
Mel Gormane2769db2016-05-19 17:14:38 -07001227 if (!free_pages_prepare(page, order, true))
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001228 return;
1229
Mel Gormancfc47a22014-06-04 16:10:19 -07001230 migratetype = get_pfnblock_migratetype(page, pfn);
Nick Pigginc54ad302006-01-06 00:10:56 -08001231 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001232 __count_vm_events(PGFREE, 1 << order);
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001233 free_one_page(page_zone(page), page, pfn, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -08001234 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235}
1236
Li Zhang949698a2016-05-19 17:11:37 -07001237static void __init __free_pages_boot_core(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -08001238{
Johannes Weinerc3993072012-01-10 15:08:10 -08001239 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001240 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -08001241 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -08001242
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001243 prefetchw(p);
1244 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1245 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -08001246 __ClearPageReserved(p);
1247 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -08001248 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001249 __ClearPageReserved(p);
1250 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -08001251
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001252 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -08001253 set_page_refcounted(page);
1254 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -08001255}
1256
Mel Gorman75a592a2015-06-30 14:56:59 -07001257#if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1258 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
Mel Gorman7ace9912015-08-06 15:46:13 -07001259
Mel Gorman75a592a2015-06-30 14:56:59 -07001260static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1261
1262int __meminit early_pfn_to_nid(unsigned long pfn)
1263{
Mel Gorman7ace9912015-08-06 15:46:13 -07001264 static DEFINE_SPINLOCK(early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -07001265 int nid;
1266
Mel Gorman7ace9912015-08-06 15:46:13 -07001267 spin_lock(&early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -07001268 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
Mel Gorman7ace9912015-08-06 15:46:13 -07001269 if (nid < 0)
Mel Gormane4568d32016-07-14 12:07:20 -07001270 nid = first_online_node;
Mel Gorman7ace9912015-08-06 15:46:13 -07001271 spin_unlock(&early_pfn_lock);
1272
1273 return nid;
Mel Gorman75a592a2015-06-30 14:56:59 -07001274}
1275#endif
1276
1277#ifdef CONFIG_NODES_SPAN_OTHER_NODES
1278static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1279 struct mminit_pfnnid_cache *state)
1280{
1281 int nid;
1282
1283 nid = __early_pfn_to_nid(pfn, state);
1284 if (nid >= 0 && nid != node)
1285 return false;
1286 return true;
1287}
1288
1289/* Only safe to use early in boot when initialisation is single-threaded */
1290static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1291{
1292 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1293}
1294
1295#else
1296
1297static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1298{
1299 return true;
1300}
1301static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1302 struct mminit_pfnnid_cache *state)
1303{
1304 return true;
1305}
1306#endif
1307
1308
Mel Gorman0e1cc952015-06-30 14:57:27 -07001309void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001310 unsigned int order)
1311{
1312 if (early_page_uninitialised(pfn))
1313 return;
Li Zhang949698a2016-05-19 17:11:37 -07001314 return __free_pages_boot_core(page, order);
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001315}
1316
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001317/*
1318 * Check that the whole (or subset of) a pageblock given by the interval of
1319 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1320 * with the migration of free compaction scanner. The scanners then need to
1321 * use only pfn_valid_within() check for arches that allow holes within
1322 * pageblocks.
1323 *
1324 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1325 *
1326 * It's possible on some configurations to have a setup like node0 node1 node0
1327 * i.e. it's possible that all pages within a zones range of pages do not
1328 * belong to a single zone. We assume that a border between node0 and node1
1329 * can occur within a single pageblock, but not a node0 node1 node0
1330 * interleaving within a single pageblock. It is therefore sufficient to check
1331 * the first and last page of a pageblock and avoid checking each individual
1332 * page in a pageblock.
1333 */
1334struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1335 unsigned long end_pfn, struct zone *zone)
1336{
1337 struct page *start_page;
1338 struct page *end_page;
1339
1340 /* end_pfn is one past the range we are checking */
1341 end_pfn--;
1342
1343 if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1344 return NULL;
1345
1346 start_page = pfn_to_page(start_pfn);
1347
1348 if (page_zone(start_page) != zone)
1349 return NULL;
1350
1351 end_page = pfn_to_page(end_pfn);
1352
1353 /* This gives a shorter code than deriving page_zone(end_page) */
1354 if (page_zone_id(start_page) != page_zone_id(end_page))
1355 return NULL;
1356
1357 return start_page;
1358}
1359
1360void set_zone_contiguous(struct zone *zone)
1361{
1362 unsigned long block_start_pfn = zone->zone_start_pfn;
1363 unsigned long block_end_pfn;
1364
1365 block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1366 for (; block_start_pfn < zone_end_pfn(zone);
1367 block_start_pfn = block_end_pfn,
1368 block_end_pfn += pageblock_nr_pages) {
1369
1370 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1371
1372 if (!__pageblock_pfn_to_page(block_start_pfn,
1373 block_end_pfn, zone))
1374 return;
1375 }
1376
1377 /* We confirm that there is no hole */
1378 zone->contiguous = true;
1379}
1380
1381void clear_zone_contiguous(struct zone *zone)
1382{
1383 zone->contiguous = false;
1384}
1385
Mel Gorman7e18adb2015-06-30 14:57:05 -07001386#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
Mel Gorman0e1cc952015-06-30 14:57:27 -07001387static void __init deferred_free_range(struct page *page,
Mel Gormana4de83d2015-06-30 14:57:16 -07001388 unsigned long pfn, int nr_pages)
1389{
1390 int i;
1391
1392 if (!page)
1393 return;
1394
1395 /* Free a large naturally-aligned chunk if possible */
1396 if (nr_pages == MAX_ORDER_NR_PAGES &&
1397 (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
Mel Gormanac5d2532015-06-30 14:57:20 -07001398 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Li Zhang949698a2016-05-19 17:11:37 -07001399 __free_pages_boot_core(page, MAX_ORDER-1);
Mel Gormana4de83d2015-06-30 14:57:16 -07001400 return;
1401 }
1402
Li Zhang949698a2016-05-19 17:11:37 -07001403 for (i = 0; i < nr_pages; i++, page++)
1404 __free_pages_boot_core(page, 0);
Mel Gormana4de83d2015-06-30 14:57:16 -07001405}
1406
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001407/* Completion tracking for deferred_init_memmap() threads */
1408static atomic_t pgdat_init_n_undone __initdata;
1409static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1410
1411static inline void __init pgdat_init_report_one_done(void)
1412{
1413 if (atomic_dec_and_test(&pgdat_init_n_undone))
1414 complete(&pgdat_init_all_done_comp);
1415}
Mel Gorman0e1cc952015-06-30 14:57:27 -07001416
Mel Gorman7e18adb2015-06-30 14:57:05 -07001417/* Initialise remaining memory on a node */
Mel Gorman0e1cc952015-06-30 14:57:27 -07001418static int __init deferred_init_memmap(void *data)
Mel Gorman7e18adb2015-06-30 14:57:05 -07001419{
Mel Gorman0e1cc952015-06-30 14:57:27 -07001420 pg_data_t *pgdat = data;
1421 int nid = pgdat->node_id;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001422 struct mminit_pfnnid_cache nid_init_state = { };
1423 unsigned long start = jiffies;
1424 unsigned long nr_pages = 0;
1425 unsigned long walk_start, walk_end;
1426 int i, zid;
1427 struct zone *zone;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001428 unsigned long first_init_pfn = pgdat->first_deferred_pfn;
Mel Gorman0e1cc952015-06-30 14:57:27 -07001429 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001430
Mel Gorman0e1cc952015-06-30 14:57:27 -07001431 if (first_init_pfn == ULONG_MAX) {
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001432 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001433 return 0;
1434 }
1435
1436 /* Bind memory initialisation thread to a local node if possible */
1437 if (!cpumask_empty(cpumask))
1438 set_cpus_allowed_ptr(current, cpumask);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001439
1440 /* Sanity check boundaries */
1441 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1442 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1443 pgdat->first_deferred_pfn = ULONG_MAX;
1444
1445 /* Only the highest zone is deferred so find it */
1446 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1447 zone = pgdat->node_zones + zid;
1448 if (first_init_pfn < zone_end_pfn(zone))
1449 break;
1450 }
1451
1452 for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1453 unsigned long pfn, end_pfn;
Mel Gorman54608c32015-06-30 14:57:09 -07001454 struct page *page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001455 struct page *free_base_page = NULL;
1456 unsigned long free_base_pfn = 0;
1457 int nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001458
1459 end_pfn = min(walk_end, zone_end_pfn(zone));
1460 pfn = first_init_pfn;
1461 if (pfn < walk_start)
1462 pfn = walk_start;
1463 if (pfn < zone->zone_start_pfn)
1464 pfn = zone->zone_start_pfn;
1465
1466 for (; pfn < end_pfn; pfn++) {
Mel Gorman54608c32015-06-30 14:57:09 -07001467 if (!pfn_valid_within(pfn))
Mel Gormana4de83d2015-06-30 14:57:16 -07001468 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001469
Mel Gorman54608c32015-06-30 14:57:09 -07001470 /*
1471 * Ensure pfn_valid is checked every
1472 * MAX_ORDER_NR_PAGES for memory holes
1473 */
1474 if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1475 if (!pfn_valid(pfn)) {
1476 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001477 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001478 }
1479 }
1480
1481 if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1482 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001483 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001484 }
1485
1486 /* Minimise pfn page lookups and scheduler checks */
1487 if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1488 page++;
1489 } else {
Mel Gormana4de83d2015-06-30 14:57:16 -07001490 nr_pages += nr_to_free;
1491 deferred_free_range(free_base_page,
1492 free_base_pfn, nr_to_free);
1493 free_base_page = NULL;
1494 free_base_pfn = nr_to_free = 0;
1495
Mel Gorman54608c32015-06-30 14:57:09 -07001496 page = pfn_to_page(pfn);
1497 cond_resched();
1498 }
Mel Gorman7e18adb2015-06-30 14:57:05 -07001499
1500 if (page->flags) {
1501 VM_BUG_ON(page_zone(page) != zone);
Mel Gormana4de83d2015-06-30 14:57:16 -07001502 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001503 }
1504
1505 __init_single_page(page, pfn, zid, nid);
Mel Gormana4de83d2015-06-30 14:57:16 -07001506 if (!free_base_page) {
1507 free_base_page = page;
1508 free_base_pfn = pfn;
1509 nr_to_free = 0;
1510 }
1511 nr_to_free++;
1512
1513 /* Where possible, batch up pages for a single free */
1514 continue;
1515free_range:
1516 /* Free the current block of pages to allocator */
1517 nr_pages += nr_to_free;
1518 deferred_free_range(free_base_page, free_base_pfn,
1519 nr_to_free);
1520 free_base_page = NULL;
1521 free_base_pfn = nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001522 }
Mel Gormana4de83d2015-06-30 14:57:16 -07001523
Mel Gorman7e18adb2015-06-30 14:57:05 -07001524 first_init_pfn = max(end_pfn, first_init_pfn);
1525 }
1526
1527 /* Sanity check that the next zone really is unpopulated */
1528 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1529
Mel Gorman0e1cc952015-06-30 14:57:27 -07001530 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
Mel Gorman7e18adb2015-06-30 14:57:05 -07001531 jiffies_to_msecs(jiffies - start));
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001532
1533 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001534 return 0;
1535}
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001536#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
Mel Gorman0e1cc952015-06-30 14:57:27 -07001537
1538void __init page_alloc_init_late(void)
1539{
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001540 struct zone *zone;
1541
1542#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
Mel Gorman0e1cc952015-06-30 14:57:27 -07001543 int nid;
1544
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001545 /* There will be num_node_state(N_MEMORY) threads */
1546 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
Mel Gorman0e1cc952015-06-30 14:57:27 -07001547 for_each_node_state(nid, N_MEMORY) {
Mel Gorman0e1cc952015-06-30 14:57:27 -07001548 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1549 }
1550
1551 /* Block until all are initialised */
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001552 wait_for_completion(&pgdat_init_all_done_comp);
Mel Gorman4248b0d2015-08-06 15:46:20 -07001553
1554 /* Reinit limits that are based on free pages after the kernel is up */
1555 files_maxfiles_init();
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001556#endif
1557
1558 for_each_populated_zone(zone)
1559 set_zone_contiguous(zone);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001560}
Mel Gorman7e18adb2015-06-30 14:57:05 -07001561
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001562#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +08001563/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001564void __init init_cma_reserved_pageblock(struct page *page)
1565{
1566 unsigned i = pageblock_nr_pages;
1567 struct page *p = page;
1568
1569 do {
1570 __ClearPageReserved(p);
1571 set_page_count(p, 0);
1572 } while (++p, --i);
1573
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001574 set_pageblock_migratetype(page, MIGRATE_CMA);
Michal Nazarewiczdc783272014-07-02 15:22:35 -07001575
1576 if (pageblock_order >= MAX_ORDER) {
1577 i = pageblock_nr_pages;
1578 p = page;
1579 do {
1580 set_page_refcounted(p);
1581 __free_pages(p, MAX_ORDER - 1);
1582 p += MAX_ORDER_NR_PAGES;
1583 } while (i -= MAX_ORDER_NR_PAGES);
1584 } else {
1585 set_page_refcounted(page);
1586 __free_pages(page, pageblock_order);
1587 }
1588
Jiang Liu3dcc0572013-07-03 15:03:21 -07001589 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001590}
1591#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592
1593/*
1594 * The order of subdivision here is critical for the IO subsystem.
1595 * Please do not alter this order without good reasons and regression
1596 * testing. Specifically, as large blocks of memory are subdivided,
1597 * the order in which smaller blocks are delivered depends on the order
1598 * they're subdivided in this function. This is the primary factor
1599 * influencing the order in which pages are delivered to the IO
1600 * subsystem according to empirical testing, and this is also justified
1601 * by considering the behavior of a buddy system containing a single
1602 * large block of memory acted on by a series of small allocations.
1603 * This behavior is a critical factor in sglist merging's success.
1604 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +01001605 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 */
Nick Piggin085cc7d2006-01-06 00:11:01 -08001607static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001608 int low, int high, struct free_area *area,
1609 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610{
1611 unsigned long size = 1 << high;
1612
1613 while (high > low) {
1614 area--;
1615 high--;
1616 size >>= 1;
Sasha Levin309381fea2014-01-23 15:52:54 -08001617 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08001618
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001619 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
Joonsoo Kime30825f2014-12-12 16:55:49 -08001620 debug_guardpage_enabled() &&
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001621 high < debug_guardpage_minorder()) {
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08001622 /*
1623 * Mark as guard pages (or page), that will allow to
1624 * merge back to allocator when buddy will be freed.
1625 * Corresponding page table entries will not be touched,
1626 * pages will stay not present in virtual address space
1627 */
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001628 set_page_guard(zone, &page[size], high, migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08001629 continue;
1630 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001631 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 area->nr_free++;
1633 set_page_order(&page[size], high);
1634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635}
1636
Vlastimil Babka4e611802016-05-19 17:14:41 -07001637static void check_new_page_bad(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638{
Vlastimil Babka4e611802016-05-19 17:14:41 -07001639 const char *bad_reason = NULL;
1640 unsigned long bad_flags = 0;
Dave Hansenf0b791a2014-01-23 15:52:49 -08001641
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001642 if (unlikely(atomic_read(&page->_mapcount) != -1))
Dave Hansenf0b791a2014-01-23 15:52:49 -08001643 bad_reason = "nonzero mapcount";
1644 if (unlikely(page->mapping != NULL))
1645 bad_reason = "non-NULL mapping";
Joonsoo Kimfe896d12016-03-17 14:19:26 -07001646 if (unlikely(page_ref_count(page) != 0))
Dave Hansenf0b791a2014-01-23 15:52:49 -08001647 bad_reason = "nonzero _count";
Naoya Horiguchif4c18e62015-08-06 15:47:08 -07001648 if (unlikely(page->flags & __PG_HWPOISON)) {
1649 bad_reason = "HWPoisoned (hardware-corrupted)";
1650 bad_flags = __PG_HWPOISON;
Naoya Horiguchie570f562016-05-20 16:58:50 -07001651 /* Don't complain about hwpoisoned pages */
1652 page_mapcount_reset(page); /* remove PageBuddy */
1653 return;
Naoya Horiguchif4c18e62015-08-06 15:47:08 -07001654 }
Dave Hansenf0b791a2014-01-23 15:52:49 -08001655 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1656 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1657 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1658 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -08001659#ifdef CONFIG_MEMCG
1660 if (unlikely(page->mem_cgroup))
1661 bad_reason = "page still charged to cgroup";
1662#endif
Vlastimil Babka4e611802016-05-19 17:14:41 -07001663 bad_page(page, bad_reason, bad_flags);
1664}
1665
1666/*
1667 * This page is about to be returned from the page allocator
1668 */
1669static inline int check_new_page(struct page *page)
1670{
1671 if (likely(page_expected_state(page,
1672 PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
1673 return 0;
1674
1675 check_new_page_bad(page);
1676 return 1;
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001677}
1678
Laura Abbott1414c7f2016-03-15 14:56:30 -07001679static inline bool free_pages_prezeroed(bool poisoned)
1680{
1681 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1682 page_poisoning_enabled() && poisoned;
1683}
1684
Mel Gorman479f8542016-05-19 17:14:35 -07001685#ifdef CONFIG_DEBUG_VM
1686static bool check_pcp_refill(struct page *page)
1687{
1688 return false;
1689}
1690
1691static bool check_new_pcp(struct page *page)
1692{
1693 return check_new_page(page);
1694}
1695#else
1696static bool check_pcp_refill(struct page *page)
1697{
1698 return check_new_page(page);
1699}
1700static bool check_new_pcp(struct page *page)
1701{
1702 return false;
1703}
1704#endif /* CONFIG_DEBUG_VM */
1705
1706static bool check_new_pages(struct page *page, unsigned int order)
1707{
1708 int i;
1709 for (i = 0; i < (1 << order); i++) {
1710 struct page *p = page + i;
1711
1712 if (unlikely(check_new_page(p)))
1713 return true;
1714 }
1715
1716 return false;
1717}
1718
Joonsoo Kim46f24fd2016-07-26 15:23:58 -07001719inline void post_alloc_hook(struct page *page, unsigned int order,
1720 gfp_t gfp_flags)
1721{
1722 set_page_private(page, 0);
1723 set_page_refcounted(page);
1724
1725 arch_alloc_page(page, order);
1726 kernel_map_pages(page, 1 << order, 1);
1727 kernel_poison_pages(page, 1 << order, 1);
1728 kasan_alloc_pages(page, order);
1729 set_page_owner(page, order, gfp_flags);
1730}
1731
Mel Gorman479f8542016-05-19 17:14:35 -07001732static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
Mel Gormanc6038442016-05-19 17:13:38 -07001733 unsigned int alloc_flags)
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001734{
1735 int i;
Laura Abbott1414c7f2016-03-15 14:56:30 -07001736 bool poisoned = true;
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001737
1738 for (i = 0; i < (1 << order); i++) {
1739 struct page *p = page + i;
Laura Abbott1414c7f2016-03-15 14:56:30 -07001740 if (poisoned)
1741 poisoned &= page_is_poisoned(p);
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001742 }
Hugh Dickins689bceb2005-11-21 21:32:20 -08001743
Joonsoo Kim46f24fd2016-07-26 15:23:58 -07001744 post_alloc_hook(page, order, gfp_flags);
Nick Piggin17cf4402006-03-22 00:08:41 -08001745
Laura Abbott1414c7f2016-03-15 14:56:30 -07001746 if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO))
Anisse Astierf4d28972015-06-24 16:56:36 -07001747 for (i = 0; i < (1 << order); i++)
1748 clear_highpage(page + i);
Nick Piggin17cf4402006-03-22 00:08:41 -08001749
1750 if (order && (gfp_flags & __GFP_COMP))
1751 prep_compound_page(page, order);
1752
Vlastimil Babka75379192015-02-11 15:25:38 -08001753 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001754 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
Vlastimil Babka75379192015-02-11 15:25:38 -08001755 * allocate the page. The expectation is that the caller is taking
1756 * steps that will free more memory. The caller should avoid the page
1757 * being used for !PFMEMALLOC purposes.
1758 */
Michal Hocko2f064f32015-08-21 14:11:51 -07001759 if (alloc_flags & ALLOC_NO_WATERMARKS)
1760 set_page_pfmemalloc(page);
1761 else
1762 clear_page_pfmemalloc(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763}
1764
Mel Gorman56fd56b2007-10-16 01:25:58 -07001765/*
1766 * Go through the free lists for the given migratetype and remove
1767 * the smallest available page from the freelists
1768 */
Mel Gorman728ec982009-06-16 15:32:04 -07001769static inline
1770struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -07001771 int migratetype)
1772{
1773 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -07001774 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001775 struct page *page;
1776
1777 /* Find a page of the appropriate size in the preferred list */
1778 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1779 area = &(zone->free_area[current_order]);
Geliang Tanga16601c2016-01-14 15:20:30 -08001780 page = list_first_entry_or_null(&area->free_list[migratetype],
Mel Gorman56fd56b2007-10-16 01:25:58 -07001781 struct page, lru);
Geliang Tanga16601c2016-01-14 15:20:30 -08001782 if (!page)
1783 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001784 list_del(&page->lru);
1785 rmv_page_order(page);
1786 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001787 expand(zone, page, order, current_order, area, migratetype);
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001788 set_pcppage_migratetype(page, migratetype);
Mel Gorman56fd56b2007-10-16 01:25:58 -07001789 return page;
1790 }
1791
1792 return NULL;
1793}
1794
1795
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001796/*
1797 * This array describes the order lists are fallen back to when
1798 * the free lists for the desirable migrate type are depleted
1799 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001800static int fallbacks[MIGRATE_TYPES][4] = {
Mel Gorman974a7862015-11-06 16:28:34 -08001801 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1802 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1803 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
Joonsoo Kimdc676472015-04-14 15:45:15 -07001804#ifdef CONFIG_CMA
Mel Gorman974a7862015-11-06 16:28:34 -08001805 [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001806#endif
Minchan Kim194159f2013-02-22 16:33:58 -08001807#ifdef CONFIG_MEMORY_ISOLATION
Mel Gorman974a7862015-11-06 16:28:34 -08001808 [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001809#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001810};
1811
Joonsoo Kimdc676472015-04-14 15:45:15 -07001812#ifdef CONFIG_CMA
1813static struct page *__rmqueue_cma_fallback(struct zone *zone,
1814 unsigned int order)
1815{
1816 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1817}
1818#else
1819static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1820 unsigned int order) { return NULL; }
1821#endif
1822
Mel Gormanc361be52007-10-16 01:25:51 -07001823/*
1824 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -07001825 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -07001826 * boundary. If alignment is required, use move_freepages_block()
1827 */
Minchan Kim435b4052012-10-08 16:32:16 -07001828int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -07001829 struct page *start_page, struct page *end_page,
1830 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001831{
1832 struct page *page;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08001833 unsigned int order;
Mel Gormand1003132007-10-16 01:26:00 -07001834 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -07001835
1836#ifndef CONFIG_HOLES_IN_ZONE
1837 /*
1838 * page_zone is not safe to call in this context when
1839 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1840 * anyway as we check zone boundaries in move_freepages_block().
1841 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -07001842 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -07001843 */
Mel Gorman97ee4ba2014-10-09 15:28:28 -07001844 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
Mel Gormanc361be52007-10-16 01:25:51 -07001845#endif
1846
1847 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -07001848 /* Make sure we are not inadvertently changing nodes */
Sasha Levin309381fea2014-01-23 15:52:54 -08001849 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
Adam Litke344c7902008-09-02 14:35:38 -07001850
Mel Gormanc361be52007-10-16 01:25:51 -07001851 if (!pfn_valid_within(page_to_pfn(page))) {
1852 page++;
1853 continue;
1854 }
1855
1856 if (!PageBuddy(page)) {
1857 page++;
1858 continue;
1859 }
1860
1861 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001862 list_move(&page->lru,
1863 &zone->free_area[order].free_list[migratetype]);
Mel Gormanc361be52007-10-16 01:25:51 -07001864 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001865 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001866 }
1867
Mel Gormand1003132007-10-16 01:26:00 -07001868 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001869}
1870
Minchan Kimee6f5092012-07-31 16:43:50 -07001871int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001872 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001873{
1874 unsigned long start_pfn, end_pfn;
1875 struct page *start_page, *end_page;
1876
1877 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001878 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001879 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001880 end_page = start_page + pageblock_nr_pages - 1;
1881 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001882
1883 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001884 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001885 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001886 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001887 return 0;
1888
1889 return move_freepages(zone, start_page, end_page, migratetype);
1890}
1891
Mel Gorman2f66a682009-09-21 17:02:31 -07001892static void change_pageblock_range(struct page *pageblock_page,
1893 int start_order, int migratetype)
1894{
1895 int nr_pageblocks = 1 << (start_order - pageblock_order);
1896
1897 while (nr_pageblocks--) {
1898 set_pageblock_migratetype(pageblock_page, migratetype);
1899 pageblock_page += pageblock_nr_pages;
1900 }
1901}
1902
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001903/*
Vlastimil Babka9c0415e2015-02-11 15:28:21 -08001904 * When we are falling back to another migratetype during allocation, try to
1905 * steal extra free pages from the same pageblocks to satisfy further
1906 * allocations, instead of polluting multiple pageblocks.
1907 *
1908 * If we are stealing a relatively large buddy page, it is likely there will
1909 * be more free pages in the pageblock, so try to steal them all. For
1910 * reclaimable and unmovable allocations, we steal regardless of page size,
1911 * as fragmentation caused by those allocations polluting movable pageblocks
1912 * is worse than movable allocations stealing from unmovable and reclaimable
1913 * pageblocks.
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001914 */
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001915static bool can_steal_fallback(unsigned int order, int start_mt)
1916{
1917 /*
1918 * Leaving this order check is intended, although there is
1919 * relaxed order check in next check. The reason is that
1920 * we can actually steal whole pageblock if this condition met,
1921 * but, below check doesn't guarantee it and that is just heuristic
1922 * so could be changed anytime.
1923 */
1924 if (order >= pageblock_order)
1925 return true;
1926
1927 if (order >= pageblock_order / 2 ||
1928 start_mt == MIGRATE_RECLAIMABLE ||
1929 start_mt == MIGRATE_UNMOVABLE ||
1930 page_group_by_mobility_disabled)
1931 return true;
1932
1933 return false;
1934}
1935
1936/*
1937 * This function implements actual steal behaviour. If order is large enough,
1938 * we can steal whole pageblock. If not, we first move freepages in this
1939 * pageblock and check whether half of pages are moved or not. If half of
1940 * pages are moved, we can change migratetype of pageblock and permanently
1941 * use it's pages as requested migratetype in the future.
1942 */
1943static void steal_suitable_fallback(struct zone *zone, struct page *page,
1944 int start_type)
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001945{
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08001946 unsigned int current_order = page_order(page);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001947 int pages;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001948
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001949 /* Take ownership for orders >= pageblock_order */
1950 if (current_order >= pageblock_order) {
1951 change_pageblock_range(page, current_order, start_type);
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001952 return;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001953 }
1954
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001955 pages = move_freepages_block(zone, page, start_type);
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001956
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001957 /* Claim the whole block if over half of it is free */
1958 if (pages >= (1 << (pageblock_order-1)) ||
1959 page_group_by_mobility_disabled)
1960 set_pageblock_migratetype(page, start_type);
1961}
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001962
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001963/*
1964 * Check whether there is a suitable fallback freepage with requested order.
1965 * If only_stealable is true, this function returns fallback_mt only if
1966 * we can steal other freepages all together. This would help to reduce
1967 * fragmentation due to mixed migratetype pages in one pageblock.
1968 */
1969int find_suitable_fallback(struct free_area *area, unsigned int order,
1970 int migratetype, bool only_stealable, bool *can_steal)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001971{
1972 int i;
1973 int fallback_mt;
1974
1975 if (area->nr_free == 0)
1976 return -1;
1977
1978 *can_steal = false;
1979 for (i = 0;; i++) {
1980 fallback_mt = fallbacks[migratetype][i];
Mel Gorman974a7862015-11-06 16:28:34 -08001981 if (fallback_mt == MIGRATE_TYPES)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001982 break;
1983
1984 if (list_empty(&area->free_list[fallback_mt]))
1985 continue;
1986
1987 if (can_steal_fallback(order, migratetype))
1988 *can_steal = true;
1989
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001990 if (!only_stealable)
1991 return fallback_mt;
1992
1993 if (*can_steal)
1994 return fallback_mt;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001995 }
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001996
1997 return -1;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001998}
1999
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002000/*
2001 * Reserve a pageblock for exclusive use of high-order atomic allocations if
2002 * there are no empty page blocks that contain a page with a suitable order
2003 */
2004static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2005 unsigned int alloc_order)
2006{
2007 int mt;
2008 unsigned long max_managed, flags;
2009
2010 /*
2011 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2012 * Check is race-prone but harmless.
2013 */
2014 max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
2015 if (zone->nr_reserved_highatomic >= max_managed)
2016 return;
2017
2018 spin_lock_irqsave(&zone->lock, flags);
2019
2020 /* Recheck the nr_reserved_highatomic limit under the lock */
2021 if (zone->nr_reserved_highatomic >= max_managed)
2022 goto out_unlock;
2023
2024 /* Yoink! */
2025 mt = get_pageblock_migratetype(page);
2026 if (mt != MIGRATE_HIGHATOMIC &&
2027 !is_migrate_isolate(mt) && !is_migrate_cma(mt)) {
2028 zone->nr_reserved_highatomic += pageblock_nr_pages;
2029 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2030 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC);
2031 }
2032
2033out_unlock:
2034 spin_unlock_irqrestore(&zone->lock, flags);
2035}
2036
2037/*
2038 * Used when an allocation is about to fail under memory pressure. This
2039 * potentially hurts the reliability of high-order allocations when under
2040 * intense memory pressure but failed atomic allocations should be easier
2041 * to recover from than an OOM.
2042 */
2043static void unreserve_highatomic_pageblock(const struct alloc_context *ac)
2044{
2045 struct zonelist *zonelist = ac->zonelist;
2046 unsigned long flags;
2047 struct zoneref *z;
2048 struct zone *zone;
2049 struct page *page;
2050 int order;
2051
2052 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2053 ac->nodemask) {
2054 /* Preserve at least one pageblock */
2055 if (zone->nr_reserved_highatomic <= pageblock_nr_pages)
2056 continue;
2057
2058 spin_lock_irqsave(&zone->lock, flags);
2059 for (order = 0; order < MAX_ORDER; order++) {
2060 struct free_area *area = &(zone->free_area[order]);
2061
Geliang Tanga16601c2016-01-14 15:20:30 -08002062 page = list_first_entry_or_null(
2063 &area->free_list[MIGRATE_HIGHATOMIC],
2064 struct page, lru);
2065 if (!page)
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002066 continue;
2067
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002068 /*
2069 * It should never happen but changes to locking could
2070 * inadvertently allow a per-cpu drain to add pages
2071 * to MIGRATE_HIGHATOMIC while unreserving so be safe
2072 * and watch for underflows.
2073 */
2074 zone->nr_reserved_highatomic -= min(pageblock_nr_pages,
2075 zone->nr_reserved_highatomic);
2076
2077 /*
2078 * Convert to ac->migratetype and avoid the normal
2079 * pageblock stealing heuristics. Minimally, the caller
2080 * is doing the work and needs the pages. More
2081 * importantly, if the block was always converted to
2082 * MIGRATE_UNMOVABLE or another type then the number
2083 * of pageblocks that cannot be completely freed
2084 * may increase.
2085 */
2086 set_pageblock_migratetype(page, ac->migratetype);
2087 move_freepages_block(zone, page, ac->migratetype);
2088 spin_unlock_irqrestore(&zone->lock, flags);
2089 return;
2090 }
2091 spin_unlock_irqrestore(&zone->lock, flags);
2092 }
2093}
2094
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002095/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07002096static inline struct page *
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002097__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002098{
Pintu Kumarb8af2942013-09-11 14:20:34 -07002099 struct free_area *area;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002100 unsigned int current_order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002101 struct page *page;
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002102 int fallback_mt;
2103 bool can_steal;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002104
2105 /* Find the largest possible block of pages in the other list */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002106 for (current_order = MAX_ORDER-1;
2107 current_order >= order && current_order <= MAX_ORDER-1;
2108 --current_order) {
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002109 area = &(zone->free_area[current_order]);
2110 fallback_mt = find_suitable_fallback(area, current_order,
Joonsoo Kim2149cda2015-04-14 15:45:21 -07002111 start_migratetype, false, &can_steal);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002112 if (fallback_mt == -1)
2113 continue;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002114
Geliang Tanga16601c2016-01-14 15:20:30 -08002115 page = list_first_entry(&area->free_list[fallback_mt],
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002116 struct page, lru);
2117 if (can_steal)
2118 steal_suitable_fallback(zone, page, start_migratetype);
Mel Gormane0104872007-10-16 01:25:53 -07002119
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002120 /* Remove the page from the freelists */
2121 area->nr_free--;
2122 list_del(&page->lru);
2123 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002124
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002125 expand(zone, page, order, current_order, area,
2126 start_migratetype);
2127 /*
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002128 * The pcppage_migratetype may differ from pageblock's
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002129 * migratetype depending on the decisions in
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002130 * find_suitable_fallback(). This is OK as long as it does not
2131 * differ for MIGRATE_CMA pageblocks. Those can be used as
2132 * fallback only via special __rmqueue_cma_fallback() function
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002133 */
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002134 set_pcppage_migratetype(page, start_migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002135
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002136 trace_mm_page_alloc_extfrag(page, order, current_order,
2137 start_migratetype, fallback_mt);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002138
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002139 return page;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002140 }
2141
Mel Gorman728ec982009-06-16 15:32:04 -07002142 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002143}
2144
Mel Gorman56fd56b2007-10-16 01:25:58 -07002145/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 * Do the hard work of removing an element from the buddy allocator.
2147 * Call me with the zone->lock already held.
2148 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002149static struct page *__rmqueue(struct zone *zone, unsigned int order,
Mel Gorman6ac02062016-01-14 15:20:28 -08002150 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 struct page *page;
2153
Mel Gorman56fd56b2007-10-16 01:25:58 -07002154 page = __rmqueue_smallest(zone, order, migratetype);
Mel Gorman974a7862015-11-06 16:28:34 -08002155 if (unlikely(!page)) {
Joonsoo Kimdc676472015-04-14 15:45:15 -07002156 if (migratetype == MIGRATE_MOVABLE)
2157 page = __rmqueue_cma_fallback(zone, order);
2158
2159 if (!page)
2160 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gorman728ec982009-06-16 15:32:04 -07002161 }
2162
Mel Gorman0d3d0622009-09-21 17:02:44 -07002163 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002164 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165}
2166
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01002167/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 * Obtain a specified number of elements from the buddy allocator, all under
2169 * a single hold of the lock, for efficiency. Add them to the supplied list.
2170 * Returns the number of new pages which were placed at *list.
2171 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01002172static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002173 unsigned long count, struct list_head *list,
Mel Gormanb745bc82014-06-04 16:10:22 -07002174 int migratetype, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175{
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07002176 int i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01002177
Nick Pigginc54ad302006-01-06 00:10:56 -08002178 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 for (i = 0; i < count; ++i) {
Mel Gorman6ac02062016-01-14 15:20:28 -08002180 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08002181 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08002183
Mel Gorman479f8542016-05-19 17:14:35 -07002184 if (unlikely(check_pcp_refill(page)))
2185 continue;
2186
Mel Gorman81eabcb2007-12-17 16:20:05 -08002187 /*
2188 * Split buddy pages returned by expand() are received here
2189 * in physical page order. The page is added to the callers and
2190 * list and the list head then moves forward. From the callers
2191 * perspective, the linked list is ordered by page number in
2192 * some conditions. This is useful for IO devices that can
2193 * merge IO requests if the physical pages are ordered
2194 * properly.
2195 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002196 if (likely(!cold))
Mel Gormane084b2d2009-07-29 15:02:04 -07002197 list_add(&page->lru, list);
2198 else
2199 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08002200 list = &page->lru;
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002201 if (is_migrate_cma(get_pcppage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07002202 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2203 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 }
Mel Gormanf2260e62009-06-16 15:32:13 -07002205 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08002206 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08002207 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208}
2209
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002210#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08002211/*
Christoph Lameter4037d452007-05-09 02:35:14 -07002212 * Called from the vmstat counter updater to drain pagesets of this
2213 * currently executing processor on remote nodes after they have
2214 * expired.
2215 *
Christoph Lameter879336c2006-03-22 00:09:08 -08002216 * Note that this function must be called with the thread pinned to
2217 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08002218 */
Christoph Lameter4037d452007-05-09 02:35:14 -07002219void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002220{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002221 unsigned long flags;
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07002222 int to_drain, batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002223
Christoph Lameter4037d452007-05-09 02:35:14 -07002224 local_irq_save(flags);
Jason Low4db0c3c2015-04-15 16:14:08 -07002225 batch = READ_ONCE(pcp->batch);
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07002226 to_drain = min(pcp->count, batch);
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07002227 if (to_drain > 0) {
2228 free_pcppages_bulk(zone, to_drain, pcp);
2229 pcp->count -= to_drain;
2230 }
Christoph Lameter4037d452007-05-09 02:35:14 -07002231 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002232}
2233#endif
2234
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002235/*
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002236 * Drain pcplists of the indicated processor and zone.
2237 *
2238 * The processor must either be the current processor and the
2239 * thread pinned to the current processor or a processor that
2240 * is not online.
2241 */
2242static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2243{
2244 unsigned long flags;
2245 struct per_cpu_pageset *pset;
2246 struct per_cpu_pages *pcp;
2247
2248 local_irq_save(flags);
2249 pset = per_cpu_ptr(zone->pageset, cpu);
2250
2251 pcp = &pset->pcp;
2252 if (pcp->count) {
2253 free_pcppages_bulk(zone, pcp->count, pcp);
2254 pcp->count = 0;
2255 }
2256 local_irq_restore(flags);
2257}
2258
2259/*
2260 * Drain pcplists of all zones on the indicated processor.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002261 *
2262 * The processor must either be the current processor and the
2263 * thread pinned to the current processor or a processor that
2264 * is not online.
2265 */
2266static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267{
2268 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002270 for_each_populated_zone(zone) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002271 drain_pages_zone(cpu, zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 }
2273}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002275/*
2276 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002277 *
2278 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2279 * the single zone's pages.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002280 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002281void drain_local_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002282{
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002283 int cpu = smp_processor_id();
2284
2285 if (zone)
2286 drain_pages_zone(cpu, zone);
2287 else
2288 drain_pages(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002289}
2290
2291/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002292 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2293 *
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002294 * When zone parameter is non-NULL, spill just the single zone's pages.
2295 *
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002296 * Note that this code is protected against sending an IPI to an offline
2297 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2298 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2299 * nothing keeps CPUs from showing up after we populated the cpumask and
2300 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002301 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002302void drain_all_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002303{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002304 int cpu;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002305
2306 /*
2307 * Allocate in the BSS so we wont require allocation in
2308 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2309 */
2310 static cpumask_t cpus_with_pcps;
2311
2312 /*
2313 * We don't care about racing with CPU hotplug event
2314 * as offline notification will cause the notified
2315 * cpu to drain that CPU pcps and on_each_cpu_mask
2316 * disables preemption as part of its processing
2317 */
2318 for_each_online_cpu(cpu) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002319 struct per_cpu_pageset *pcp;
2320 struct zone *z;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002321 bool has_pcps = false;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002322
2323 if (zone) {
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002324 pcp = per_cpu_ptr(zone->pageset, cpu);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002325 if (pcp->pcp.count)
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002326 has_pcps = true;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002327 } else {
2328 for_each_populated_zone(z) {
2329 pcp = per_cpu_ptr(z->pageset, cpu);
2330 if (pcp->pcp.count) {
2331 has_pcps = true;
2332 break;
2333 }
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002334 }
2335 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002336
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002337 if (has_pcps)
2338 cpumask_set_cpu(cpu, &cpus_with_pcps);
2339 else
2340 cpumask_clear_cpu(cpu, &cpus_with_pcps);
2341 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002342 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
2343 zone, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002344}
2345
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02002346#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347
2348void mark_free_pages(struct zone *zone)
2349{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002350 unsigned long pfn, max_zone_pfn;
2351 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002352 unsigned int order, t;
Geliang Tang86760a22016-01-14 15:20:33 -08002353 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354
Xishi Qiu8080fc02013-09-11 14:21:45 -07002355 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 return;
2357
2358 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002359
Cody P Schafer108bcc92013-02-22 16:35:23 -08002360 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002361 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2362 if (pfn_valid(pfn)) {
Geliang Tang86760a22016-01-14 15:20:33 -08002363 page = pfn_to_page(pfn);
Joonsoo Kimba6b0972016-05-19 17:12:16 -07002364
2365 if (page_zone(page) != zone)
2366 continue;
2367
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07002368 if (!swsusp_page_is_forbidden(page))
2369 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002370 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002372 for_each_migratetype_order(order, t) {
Geliang Tang86760a22016-01-14 15:20:33 -08002373 list_for_each_entry(page,
2374 &zone->free_area[order].free_list[t], lru) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002375 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376
Geliang Tang86760a22016-01-14 15:20:33 -08002377 pfn = page_to_pfn(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002378 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07002379 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002380 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 spin_unlock_irqrestore(&zone->lock, flags);
2383}
Mel Gormane2c55dc2007-10-16 01:25:50 -07002384#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385
2386/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 * Free a 0-order page
Mel Gormanb745bc82014-06-04 16:10:22 -07002388 * cold == true ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002390void free_hot_cold_page(struct page *page, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391{
2392 struct zone *zone = page_zone(page);
2393 struct per_cpu_pages *pcp;
2394 unsigned long flags;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002395 unsigned long pfn = page_to_pfn(page);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002396 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397
Mel Gorman4db75482016-05-19 17:14:32 -07002398 if (!free_pcp_prepare(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -08002399 return;
2400
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002401 migratetype = get_pfnblock_migratetype(page, pfn);
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002402 set_pcppage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07002404 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07002405
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002406 /*
2407 * We only track unmovable, reclaimable and movable on pcp lists.
2408 * Free ISOLATE pages back to the allocator because they are being
2409 * offlined but treat RESERVE as movable pages so we can get those
2410 * areas back if necessary. Otherwise, we may have to free
2411 * excessively into the page allocator
2412 */
2413 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08002414 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002415 free_one_page(zone, page, pfn, 0, migratetype);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002416 goto out;
2417 }
2418 migratetype = MIGRATE_MOVABLE;
2419 }
2420
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002421 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Mel Gormanb745bc82014-06-04 16:10:22 -07002422 if (!cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002423 list_add(&page->lru, &pcp->lists[migratetype]);
Mel Gormanb745bc82014-06-04 16:10:22 -07002424 else
2425 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08002427 if (pcp->count >= pcp->high) {
Jason Low4db0c3c2015-04-15 16:14:08 -07002428 unsigned long batch = READ_ONCE(pcp->batch);
Cody P Schafer998d39c2013-07-03 15:01:32 -07002429 free_pcppages_bulk(zone, batch, pcp);
2430 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08002431 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002432
2433out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435}
2436
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002437/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002438 * Free a list of 0-order pages
2439 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002440void free_hot_cold_page_list(struct list_head *list, bool cold)
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002441{
2442 struct page *page, *next;
2443
2444 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08002445 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002446 free_hot_cold_page(page, cold);
2447 }
2448}
2449
2450/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002451 * split_page takes a non-compound higher-order page, and splits it into
2452 * n (1<<order) sub-pages: page[0..n]
2453 * Each sub-page must be freed individually.
2454 *
2455 * Note: this is probably too low level an operation for use in drivers.
2456 * Please consult with lkml before using this in your driver.
2457 */
2458void split_page(struct page *page, unsigned int order)
2459{
2460 int i;
2461
Sasha Levin309381fea2014-01-23 15:52:54 -08002462 VM_BUG_ON_PAGE(PageCompound(page), page);
2463 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002464
2465#ifdef CONFIG_KMEMCHECK
2466 /*
2467 * Split shadow pages too, because free(page[0]) would
2468 * otherwise free the whole shadow.
2469 */
2470 if (kmemcheck_page_is_tracked(page))
2471 split_page(virt_to_page(page[0].shadow), order);
2472#endif
2473
Joonsoo Kima9627bc2016-07-26 15:23:49 -07002474 for (i = 1; i < (1 << order); i++)
Nick Piggin7835e982006-03-22 00:08:40 -08002475 set_page_refcounted(page + i);
Joonsoo Kima9627bc2016-07-26 15:23:49 -07002476 split_page_owner(page, order);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002477}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07002478EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002479
Joonsoo Kim3c605092014-11-13 15:19:21 -08002480int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07002481{
Mel Gorman748446b2010-05-24 14:32:27 -07002482 unsigned long watermark;
2483 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002484 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07002485
2486 BUG_ON(!PageBuddy(page));
2487
2488 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002489 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07002490
Minchan Kim194159f2013-02-22 16:33:58 -08002491 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002492 /* Obey watermarks as if the page was being allocated */
2493 watermark = low_wmark_pages(zone) + (1 << order);
2494 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2495 return 0;
2496
Mel Gorman8fb74b92013-01-11 14:32:16 -08002497 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002498 }
Mel Gorman748446b2010-05-24 14:32:27 -07002499
2500 /* Remove page from free list */
2501 list_del(&page->lru);
2502 zone->free_area[order].nr_free--;
2503 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002504
zhong jiang400bc7f2016-07-28 15:45:07 -07002505 /*
2506 * Set the pageblock if the isolated page is at least half of a
2507 * pageblock
2508 */
Mel Gorman748446b2010-05-24 14:32:27 -07002509 if (order >= pageblock_order - 1) {
2510 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002511 for (; page < endpage; page += pageblock_nr_pages) {
2512 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08002513 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002514 set_pageblock_migratetype(page,
2515 MIGRATE_MOVABLE);
2516 }
Mel Gorman748446b2010-05-24 14:32:27 -07002517 }
2518
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002519
Mel Gorman8fb74b92013-01-11 14:32:16 -08002520 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002521}
2522
2523/*
Mel Gorman060e7412016-05-19 17:13:27 -07002524 * Update NUMA hit/miss statistics
2525 *
2526 * Must be called with interrupts disabled.
2527 *
2528 * When __GFP_OTHER_NODE is set assume the node of the preferred
2529 * zone is the local node. This is useful for daemons who allocate
2530 * memory on behalf of other processes.
2531 */
2532static inline void zone_statistics(struct zone *preferred_zone, struct zone *z,
2533 gfp_t flags)
2534{
2535#ifdef CONFIG_NUMA
2536 int local_nid = numa_node_id();
2537 enum zone_stat_item local_stat = NUMA_LOCAL;
2538
2539 if (unlikely(flags & __GFP_OTHER_NODE)) {
2540 local_stat = NUMA_OTHER;
2541 local_nid = preferred_zone->node;
2542 }
2543
2544 if (z->node == local_nid) {
2545 __inc_zone_state(z, NUMA_HIT);
2546 __inc_zone_state(z, local_stat);
2547 } else {
2548 __inc_zone_state(z, NUMA_MISS);
2549 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
2550 }
2551#endif
2552}
2553
2554/*
Vlastimil Babka75379192015-02-11 15:25:38 -08002555 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07002557static inline
2558struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002559 struct zone *zone, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07002560 gfp_t gfp_flags, unsigned int alloc_flags,
2561 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562{
2563 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08002564 struct page *page;
Mel Gormanb745bc82014-06-04 16:10:22 -07002565 bool cold = ((gfp_flags & __GFP_COLD) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566
Nick Piggin48db57f2006-01-08 01:00:42 -08002567 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002569 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 local_irq_save(flags);
Mel Gorman479f8542016-05-19 17:14:35 -07002572 do {
2573 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2574 list = &pcp->lists[migratetype];
2575 if (list_empty(list)) {
2576 pcp->count += rmqueue_bulk(zone, 0,
2577 pcp->batch, list,
2578 migratetype, cold);
2579 if (unlikely(list_empty(list)))
2580 goto failed;
2581 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07002582
Mel Gorman479f8542016-05-19 17:14:35 -07002583 if (cold)
2584 page = list_last_entry(list, struct page, lru);
2585 else
2586 page = list_first_entry(list, struct page, lru);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002587
Vlastimil Babka83b93552016-06-03 14:55:52 -07002588 list_del(&page->lru);
2589 pcp->count--;
2590
2591 } while (check_new_pcp(page));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002592 } else {
Michal Hocko0f352e52016-03-17 14:19:32 -07002593 /*
2594 * We most definitely don't want callers attempting to
2595 * allocate greater than order-1 page units with __GFP_NOFAIL.
2596 */
2597 WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 spin_lock_irqsave(&zone->lock, flags);
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002599
Mel Gorman479f8542016-05-19 17:14:35 -07002600 do {
2601 page = NULL;
2602 if (alloc_flags & ALLOC_HARDER) {
2603 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
2604 if (page)
2605 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2606 }
2607 if (!page)
2608 page = __rmqueue(zone, order, migratetype);
2609 } while (page && check_new_pages(page, order));
Nick Piggina74609f2006-01-06 00:11:20 -08002610 spin_unlock(&zone->lock);
2611 if (!page)
2612 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07002613 __mod_zone_freepage_state(zone, -(1 << order),
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002614 get_pcppage_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 }
2616
Mel Gorman16709d12016-07-28 15:46:56 -07002617 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07002618 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002619 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620
Sasha Levin309381fea2014-01-23 15:52:54 -08002621 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08002623
2624failed:
2625 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002626 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627}
2628
Akinobu Mita933e3122006-12-08 02:39:45 -08002629#ifdef CONFIG_FAIL_PAGE_ALLOC
2630
Akinobu Mitab2588c42011-07-26 16:09:03 -07002631static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08002632 struct fault_attr attr;
2633
Viresh Kumar621a5f72015-09-26 15:04:07 -07002634 bool ignore_gfp_highmem;
Mel Gorman71baba42015-11-06 16:28:28 -08002635 bool ignore_gfp_reclaim;
Akinobu Mita54114992007-07-15 23:40:23 -07002636 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08002637} fail_page_alloc = {
2638 .attr = FAULT_ATTR_INITIALIZER,
Mel Gorman71baba42015-11-06 16:28:28 -08002639 .ignore_gfp_reclaim = true,
Viresh Kumar621a5f72015-09-26 15:04:07 -07002640 .ignore_gfp_highmem = true,
Akinobu Mita54114992007-07-15 23:40:23 -07002641 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08002642};
2643
2644static int __init setup_fail_page_alloc(char *str)
2645{
2646 return setup_fault_attr(&fail_page_alloc.attr, str);
2647}
2648__setup("fail_page_alloc=", setup_fail_page_alloc);
2649
Gavin Shandeaf3862012-07-31 16:41:51 -07002650static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002651{
Akinobu Mita54114992007-07-15 23:40:23 -07002652 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07002653 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002654 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07002655 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002656 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002657 return false;
Mel Gorman71baba42015-11-06 16:28:28 -08002658 if (fail_page_alloc.ignore_gfp_reclaim &&
2659 (gfp_mask & __GFP_DIRECT_RECLAIM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002660 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002661
2662 return should_fail(&fail_page_alloc.attr, 1 << order);
2663}
2664
2665#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2666
2667static int __init fail_page_alloc_debugfs(void)
2668{
Al Virof4ae40a2011-07-24 04:33:43 -04002669 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08002670 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08002671
Akinobu Mitadd48c082011-08-03 16:21:01 -07002672 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2673 &fail_page_alloc.attr);
2674 if (IS_ERR(dir))
2675 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002676
Akinobu Mitab2588c42011-07-26 16:09:03 -07002677 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
Mel Gorman71baba42015-11-06 16:28:28 -08002678 &fail_page_alloc.ignore_gfp_reclaim))
Akinobu Mitab2588c42011-07-26 16:09:03 -07002679 goto fail;
2680 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2681 &fail_page_alloc.ignore_gfp_highmem))
2682 goto fail;
2683 if (!debugfs_create_u32("min-order", mode, dir,
2684 &fail_page_alloc.min_order))
2685 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08002686
Akinobu Mitab2588c42011-07-26 16:09:03 -07002687 return 0;
2688fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07002689 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002690
Akinobu Mitab2588c42011-07-26 16:09:03 -07002691 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08002692}
2693
2694late_initcall(fail_page_alloc_debugfs);
2695
2696#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2697
2698#else /* CONFIG_FAIL_PAGE_ALLOC */
2699
Gavin Shandeaf3862012-07-31 16:41:51 -07002700static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002701{
Gavin Shandeaf3862012-07-31 16:41:51 -07002702 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002703}
2704
2705#endif /* CONFIG_FAIL_PAGE_ALLOC */
2706
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707/*
Mel Gorman97a16fc2015-11-06 16:28:40 -08002708 * Return true if free base pages are above 'mark'. For high-order checks it
2709 * will return true of the order-0 watermark is reached and there is at least
2710 * one free page of a suitable size. Checking now avoids taking the zone lock
2711 * to check in the allocation paths if no pages are free.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 */
Michal Hocko86a294a2016-05-20 16:57:12 -07002713bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2714 int classzone_idx, unsigned int alloc_flags,
2715 long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716{
Christoph Lameterd23ad422007-02-10 01:43:02 -08002717 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 int o;
Mel Gormanc6038442016-05-19 17:13:38 -07002719 const bool alloc_harder = (alloc_flags & ALLOC_HARDER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002721 /* free_pages may go negative - that's OK */
Michal Hockodf0a6da2012-01-10 15:08:02 -08002722 free_pages -= (1 << order) - 1;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002723
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002724 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 min -= min / 2;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002726
2727 /*
2728 * If the caller does not have rights to ALLOC_HARDER then subtract
2729 * the high-atomic reserves. This will over-estimate the size of the
2730 * atomic reserve but it avoids a search.
2731 */
Mel Gorman97a16fc2015-11-06 16:28:40 -08002732 if (likely(!alloc_harder))
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002733 free_pages -= z->nr_reserved_highatomic;
2734 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 min -= min / 4;
Mel Gormane2b19192015-11-06 16:28:09 -08002736
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002737#ifdef CONFIG_CMA
2738 /* If allocation can't use CMA areas don't use free CMA pages */
2739 if (!(alloc_flags & ALLOC_CMA))
Mel Gorman97a16fc2015-11-06 16:28:40 -08002740 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002741#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002742
Mel Gorman97a16fc2015-11-06 16:28:40 -08002743 /*
2744 * Check watermarks for an order-0 allocation request. If these
2745 * are not met, then a high-order request also cannot go ahead
2746 * even if a suitable page happened to be free.
2747 */
2748 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08002749 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750
Mel Gorman97a16fc2015-11-06 16:28:40 -08002751 /* If this is an order-0 request then the watermark is fine */
2752 if (!order)
2753 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754
Mel Gorman97a16fc2015-11-06 16:28:40 -08002755 /* For a high-order request, check at least one suitable page is free */
2756 for (o = order; o < MAX_ORDER; o++) {
2757 struct free_area *area = &z->free_area[o];
2758 int mt;
2759
2760 if (!area->nr_free)
2761 continue;
2762
2763 if (alloc_harder)
2764 return true;
2765
2766 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
2767 if (!list_empty(&area->free_list[mt]))
2768 return true;
2769 }
2770
2771#ifdef CONFIG_CMA
2772 if ((alloc_flags & ALLOC_CMA) &&
2773 !list_empty(&area->free_list[MIGRATE_CMA])) {
2774 return true;
2775 }
2776#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 }
Mel Gorman97a16fc2015-11-06 16:28:40 -08002778 return false;
Mel Gorman88f5acf2011-01-13 15:45:41 -08002779}
2780
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002781bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Mel Gormanc6038442016-05-19 17:13:38 -07002782 int classzone_idx, unsigned int alloc_flags)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002783{
2784 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2785 zone_page_state(z, NR_FREE_PAGES));
2786}
2787
Mel Gorman48ee5f32016-05-19 17:14:07 -07002788static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
2789 unsigned long mark, int classzone_idx, unsigned int alloc_flags)
2790{
2791 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2792 long cma_pages = 0;
2793
2794#ifdef CONFIG_CMA
2795 /* If allocation can't use CMA areas don't use free CMA pages */
2796 if (!(alloc_flags & ALLOC_CMA))
2797 cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES);
2798#endif
2799
2800 /*
2801 * Fast check for order-0 only. If this fails then the reserves
2802 * need to be calculated. There is a corner case where the check
2803 * passes but only the high-order atomic reserve are free. If
2804 * the caller is !atomic then it'll uselessly search the free
2805 * list. That corner case is then slower but it is harmless.
2806 */
2807 if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx])
2808 return true;
2809
2810 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2811 free_pages);
2812}
2813
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002814bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
Mel Gormane2b19192015-11-06 16:28:09 -08002815 unsigned long mark, int classzone_idx)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002816{
2817 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2818
2819 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2820 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2821
Mel Gormane2b19192015-11-06 16:28:09 -08002822 return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002823 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824}
2825
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002826#ifdef CONFIG_NUMA
David Rientjes957f8222012-10-08 16:33:24 -07002827static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2828{
Mel Gorman5f7a75a2014-06-04 16:07:15 -07002829 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2830 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07002831}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002832#else /* CONFIG_NUMA */
David Rientjes957f8222012-10-08 16:33:24 -07002833static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2834{
2835 return true;
2836}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002837#endif /* CONFIG_NUMA */
2838
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002839/*
Paul Jackson0798e512006-12-06 20:31:38 -08002840 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002841 * a page.
2842 */
2843static struct page *
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002844get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2845 const struct alloc_context *ac)
Martin Hicks753ee722005-06-21 17:14:41 -07002846{
Mel Gormanc33d6c02016-05-19 17:14:10 -07002847 struct zoneref *z = ac->preferred_zoneref;
Mel Gorman5117f452009-06-16 15:31:59 -07002848 struct zone *zone;
Mel Gorman3b8c0be2016-07-28 15:46:53 -07002849 struct pglist_data *last_pgdat_dirty_limit = NULL;
2850
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002851 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002852 * Scan zonelist, looking for a zone with enough free.
Vladimir Davydov344736f2014-10-20 15:50:30 +04002853 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002854 */
Mel Gormanc33d6c02016-05-19 17:14:10 -07002855 for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002856 ac->nodemask) {
Mel Gormanbe06af02016-05-19 17:13:47 -07002857 struct page *page;
Johannes Weinere085dbc2013-09-11 14:20:46 -07002858 unsigned long mark;
2859
Mel Gorman664eedd2014-06-04 16:10:08 -07002860 if (cpusets_enabled() &&
2861 (alloc_flags & ALLOC_CPUSET) &&
Vlastimil Babka002f2902016-05-19 17:14:30 -07002862 !__cpuset_zone_allowed(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07002863 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08002864 /*
2865 * When allocating a page cache page for writing, we
Mel Gorman281e3722016-07-28 15:46:11 -07002866 * want to get it from a node that is within its dirty
2867 * limit, such that no single node holds more than its
Johannes Weinera756cf52012-01-10 15:07:49 -08002868 * proportional share of globally allowed dirty pages.
Mel Gorman281e3722016-07-28 15:46:11 -07002869 * The dirty limits take into account the node's
Johannes Weinera756cf52012-01-10 15:07:49 -08002870 * lowmem reserves and high watermark so that kswapd
2871 * should be able to balance it without having to
2872 * write pages from its LRU list.
2873 *
Johannes Weinera756cf52012-01-10 15:07:49 -08002874 * XXX: For now, allow allocations to potentially
Mel Gorman281e3722016-07-28 15:46:11 -07002875 * exceed the per-node dirty limit in the slowpath
Mel Gormanc9ab0c42015-11-06 16:28:12 -08002876 * (spread_dirty_pages unset) before going into reclaim,
Johannes Weinera756cf52012-01-10 15:07:49 -08002877 * which is important when on a NUMA setup the allowed
Mel Gorman281e3722016-07-28 15:46:11 -07002878 * nodes are together not big enough to reach the
Johannes Weinera756cf52012-01-10 15:07:49 -08002879 * global limit. The proper fix for these situations
Mel Gorman281e3722016-07-28 15:46:11 -07002880 * will require awareness of nodes in the
Johannes Weinera756cf52012-01-10 15:07:49 -08002881 * dirty-throttling and the flusher threads.
2882 */
Mel Gorman3b8c0be2016-07-28 15:46:53 -07002883 if (ac->spread_dirty_pages) {
2884 if (last_pgdat_dirty_limit == zone->zone_pgdat)
2885 continue;
2886
2887 if (!node_dirty_ok(zone->zone_pgdat)) {
2888 last_pgdat_dirty_limit = zone->zone_pgdat;
2889 continue;
2890 }
2891 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002892
Johannes Weinere085dbc2013-09-11 14:20:46 -07002893 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gorman48ee5f32016-05-19 17:14:07 -07002894 if (!zone_watermark_fast(zone, order, mark,
Mel Gorman93ea9962016-05-19 17:14:13 -07002895 ac_classzone_idx(ac), alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07002896 int ret;
2897
Mel Gorman5dab2912014-06-04 16:10:14 -07002898 /* Checked here to keep the fast path fast */
2899 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2900 if (alloc_flags & ALLOC_NO_WATERMARKS)
2901 goto try_this_zone;
2902
Mel Gormana5f5f912016-07-28 15:46:32 -07002903 if (node_reclaim_mode == 0 ||
Mel Gormanc33d6c02016-05-19 17:14:10 -07002904 !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
Mel Gormancd38b112011-07-25 17:12:29 -07002905 continue;
2906
Mel Gormana5f5f912016-07-28 15:46:32 -07002907 ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
Mel Gormanfa5e0842009-06-16 15:33:22 -07002908 switch (ret) {
Mel Gormana5f5f912016-07-28 15:46:32 -07002909 case NODE_RECLAIM_NOSCAN:
Mel Gormanfa5e0842009-06-16 15:33:22 -07002910 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07002911 continue;
Mel Gormana5f5f912016-07-28 15:46:32 -07002912 case NODE_RECLAIM_FULL:
Mel Gormanfa5e0842009-06-16 15:33:22 -07002913 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002914 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002915 default:
2916 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002917 if (zone_watermark_ok(zone, order, mark,
Mel Gorman93ea9962016-05-19 17:14:13 -07002918 ac_classzone_idx(ac), alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002919 goto try_this_zone;
2920
Mel Gormanfed27192013-04-29 15:07:57 -07002921 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002922 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002923 }
2924
Mel Gormanfa5e0842009-06-16 15:33:22 -07002925try_this_zone:
Mel Gormanc33d6c02016-05-19 17:14:10 -07002926 page = buffered_rmqueue(ac->preferred_zoneref->zone, zone, order,
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002927 gfp_mask, alloc_flags, ac->migratetype);
Vlastimil Babka75379192015-02-11 15:25:38 -08002928 if (page) {
Mel Gorman479f8542016-05-19 17:14:35 -07002929 prep_new_page(page, order, gfp_mask, alloc_flags);
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002930
2931 /*
2932 * If this is a high-order atomic allocation then check
2933 * if the pageblock should be reserved for the future
2934 */
2935 if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
2936 reserve_highatomic_pageblock(page, zone, order);
2937
Vlastimil Babka75379192015-02-11 15:25:38 -08002938 return page;
2939 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002940 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002941
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002942 return NULL;
Martin Hicks753ee722005-06-21 17:14:41 -07002943}
2944
David Rientjes29423e772011-03-22 16:30:47 -07002945/*
2946 * Large machines with many possible nodes should not always dump per-node
2947 * meminfo in irq context.
2948 */
2949static inline bool should_suppress_show_mem(void)
2950{
2951 bool ret = false;
2952
2953#if NODES_SHIFT > 8
2954 ret = in_interrupt();
2955#endif
2956 return ret;
2957}
2958
Dave Hansena238ab52011-05-24 17:12:16 -07002959static DEFINE_RATELIMIT_STATE(nopage_rs,
2960 DEFAULT_RATELIMIT_INTERVAL,
2961 DEFAULT_RATELIMIT_BURST);
2962
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08002963void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...)
Dave Hansena238ab52011-05-24 17:12:16 -07002964{
Dave Hansena238ab52011-05-24 17:12:16 -07002965 unsigned int filter = SHOW_MEM_FILTER_NODES;
2966
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08002967 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2968 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002969 return;
2970
2971 /*
2972 * This documents exceptions given to allocations in certain
2973 * contexts that are allowed to allocate outside current's set
2974 * of allowed nodes.
2975 */
2976 if (!(gfp_mask & __GFP_NOMEMALLOC))
2977 if (test_thread_flag(TIF_MEMDIE) ||
2978 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2979 filter &= ~SHOW_MEM_FILTER_NODES;
Mel Gormand0164ad2015-11-06 16:28:21 -08002980 if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
Dave Hansena238ab52011-05-24 17:12:16 -07002981 filter &= ~SHOW_MEM_FILTER_NODES;
2982
2983 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002984 struct va_format vaf;
2985 va_list args;
2986
Dave Hansena238ab52011-05-24 17:12:16 -07002987 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002988
2989 vaf.fmt = fmt;
2990 vaf.va = &args;
2991
2992 pr_warn("%pV", &vaf);
2993
Dave Hansena238ab52011-05-24 17:12:16 -07002994 va_end(args);
2995 }
2996
Vlastimil Babkac5c990e2016-03-15 14:56:02 -07002997 pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n",
2998 current->comm, order, gfp_mask, &gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002999 dump_stack();
3000 if (!should_suppress_show_mem())
3001 show_mem(filter);
3002}
3003
Mel Gorman11e33f62009-06-16 15:31:57 -07003004static inline struct page *
3005__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003006 const struct alloc_context *ac, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07003007{
David Rientjes6e0fc462015-09-08 15:00:36 -07003008 struct oom_control oc = {
3009 .zonelist = ac->zonelist,
3010 .nodemask = ac->nodemask,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07003011 .memcg = NULL,
David Rientjes6e0fc462015-09-08 15:00:36 -07003012 .gfp_mask = gfp_mask,
3013 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07003014 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016
Johannes Weiner9879de72015-01-26 12:58:32 -08003017 *did_some_progress = 0;
3018
Johannes Weiner9879de72015-01-26 12:58:32 -08003019 /*
Johannes Weinerdc564012015-06-24 16:57:19 -07003020 * Acquire the oom lock. If that fails, somebody else is
3021 * making progress for us.
Johannes Weiner9879de72015-01-26 12:58:32 -08003022 */
Johannes Weinerdc564012015-06-24 16:57:19 -07003023 if (!mutex_trylock(&oom_lock)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08003024 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07003025 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026 return NULL;
3027 }
Jens Axboe6b1de912005-11-17 21:35:02 +01003028
Mel Gorman11e33f62009-06-16 15:31:57 -07003029 /*
3030 * Go through the zonelist yet one more time, keep very high watermark
3031 * here, this is only to catch a parallel oom killing, we must fail if
3032 * we're still under heavy pressure.
3033 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003034 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
3035 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003036 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07003037 goto out;
3038
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08003039 if (!(gfp_mask & __GFP_NOFAIL)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08003040 /* Coredumps can quickly deplete all memory reserves */
3041 if (current->flags & PF_DUMPCORE)
3042 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08003043 /* The OOM killer will not help higher order allocs */
3044 if (order > PAGE_ALLOC_COSTLY_ORDER)
3045 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07003046 /* The OOM killer does not needlessly kill tasks for lowmem */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003047 if (ac->high_zoneidx < ZONE_NORMAL)
David Rientjes03668b32010-08-09 17:18:54 -07003048 goto out;
Johannes Weiner90839052015-06-24 16:57:21 -07003049 if (pm_suspended_storage())
3050 goto out;
Michal Hocko3da88fb2016-05-19 17:13:09 -07003051 /*
3052 * XXX: GFP_NOFS allocations should rather fail than rely on
3053 * other request to make a forward progress.
3054 * We are in an unfortunate situation where out_of_memory cannot
3055 * do much for this context but let's try it to at least get
3056 * access to memory reserved if the current task is killed (see
3057 * out_of_memory). Once filesystems are ready to handle allocation
3058 * failures more gracefully we should just bail out here.
3059 */
3060
David Rientjes4167e9b2015-04-14 15:46:55 -07003061 /* The OOM killer may not free memory on a specific node */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08003062 if (gfp_mask & __GFP_THISNODE)
3063 goto out;
3064 }
Mel Gorman11e33f62009-06-16 15:31:57 -07003065 /* Exhausted what can be done so it's blamo time */
Michal Hocko5020e282016-01-14 15:20:36 -08003066 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
Michal Hockoc32b3cb2015-02-11 15:26:24 -08003067 *did_some_progress = 1;
Michal Hocko5020e282016-01-14 15:20:36 -08003068
3069 if (gfp_mask & __GFP_NOFAIL) {
3070 page = get_page_from_freelist(gfp_mask, order,
3071 ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac);
3072 /*
3073 * fallback to ignore cpuset restriction if our nodes
3074 * are depleted
3075 */
3076 if (!page)
3077 page = get_page_from_freelist(gfp_mask, order,
3078 ALLOC_NO_WATERMARKS, ac);
3079 }
3080 }
Mel Gorman11e33f62009-06-16 15:31:57 -07003081out:
Johannes Weinerdc564012015-06-24 16:57:19 -07003082 mutex_unlock(&oom_lock);
Mel Gorman11e33f62009-06-16 15:31:57 -07003083 return page;
3084}
3085
Michal Hocko33c2d212016-05-20 16:57:06 -07003086/*
3087 * Maximum number of compaction retries wit a progress before OOM
3088 * killer is consider as the only way to move forward.
3089 */
3090#define MAX_COMPACT_RETRIES 16
3091
Mel Gorman56de7262010-05-24 14:32:30 -07003092#ifdef CONFIG_COMPACTION
3093/* Try memory compaction for high-order allocations before reclaim */
3094static struct page *
3095__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07003096 unsigned int alloc_flags, const struct alloc_context *ac,
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003097 enum compact_priority prio, enum compact_result *compact_result)
Mel Gorman56de7262010-05-24 14:32:30 -07003098{
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003099 struct page *page;
Vlastimil Babka53853e22014-10-09 15:27:02 -07003100
Mel Gorman66199712012-01-12 17:19:41 -08003101 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07003102 return NULL;
3103
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003104 current->flags |= PF_MEMALLOC;
Michal Hockoc5d01d02016-05-20 16:56:53 -07003105 *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
Vlastimil Babkac3486f52016-07-28 15:49:30 -07003106 prio);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003107 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07003108
Michal Hockoc5d01d02016-05-20 16:56:53 -07003109 if (*compact_result <= COMPACT_INACTIVE)
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003110 return NULL;
Mel Gorman56de7262010-05-24 14:32:30 -07003111
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003112 /*
3113 * At least in one zone compaction wasn't deferred or skipped, so let's
3114 * count a compaction stall
3115 */
3116 count_vm_event(COMPACTSTALL);
3117
Vlastimil Babka31a6c192016-07-28 15:49:13 -07003118 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003119
3120 if (page) {
3121 struct zone *zone = page_zone(page);
3122
3123 zone->compact_blockskip_flush = false;
3124 compaction_defer_reset(zone, order, true);
3125 count_vm_event(COMPACTSUCCESS);
3126 return page;
3127 }
3128
3129 /*
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003130 * It's bad if compaction run occurs and fails. The most likely reason
3131 * is that pages exist, but not enough to satisfy watermarks.
3132 */
3133 count_vm_event(COMPACTFAIL);
3134
3135 cond_resched();
3136
Mel Gorman56de7262010-05-24 14:32:30 -07003137 return NULL;
3138}
Michal Hocko33c2d212016-05-20 16:57:06 -07003139
3140static inline bool
Michal Hocko86a294a2016-05-20 16:57:12 -07003141should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003142 enum compact_result compact_result,
3143 enum compact_priority *compact_priority,
Michal Hocko33c2d212016-05-20 16:57:06 -07003144 int compaction_retries)
3145{
Michal Hocko7854ea62016-05-20 16:57:09 -07003146 int max_retries = MAX_COMPACT_RETRIES;
3147
Michal Hocko33c2d212016-05-20 16:57:06 -07003148 if (!order)
3149 return false;
3150
3151 /*
3152 * compaction considers all the zone as desperately out of memory
3153 * so it doesn't really make much sense to retry except when the
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003154 * failure could be caused by insufficient priority
Michal Hocko33c2d212016-05-20 16:57:06 -07003155 */
3156 if (compaction_failed(compact_result)) {
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003157 if (*compact_priority > MIN_COMPACT_PRIORITY) {
3158 (*compact_priority)--;
Michal Hocko33c2d212016-05-20 16:57:06 -07003159 return true;
3160 }
3161 return false;
3162 }
3163
3164 /*
Michal Hocko7854ea62016-05-20 16:57:09 -07003165 * make sure the compaction wasn't deferred or didn't bail out early
3166 * due to locks contention before we declare that we should give up.
Michal Hocko86a294a2016-05-20 16:57:12 -07003167 * But do not retry if the given zonelist is not suitable for
3168 * compaction.
Michal Hocko33c2d212016-05-20 16:57:06 -07003169 */
Michal Hocko7854ea62016-05-20 16:57:09 -07003170 if (compaction_withdrawn(compact_result))
Michal Hocko86a294a2016-05-20 16:57:12 -07003171 return compaction_zonelist_suitable(ac, order, alloc_flags);
Michal Hocko7854ea62016-05-20 16:57:09 -07003172
3173 /*
3174 * !costly requests are much more important than __GFP_REPEAT
3175 * costly ones because they are de facto nofail and invoke OOM
3176 * killer to move on while costly can fail and users are ready
3177 * to cope with that. 1/4 retries is rather arbitrary but we
3178 * would need much more detailed feedback from compaction to
3179 * make a better decision.
3180 */
3181 if (order > PAGE_ALLOC_COSTLY_ORDER)
3182 max_retries /= 4;
3183 if (compaction_retries <= max_retries)
3184 return true;
Michal Hocko33c2d212016-05-20 16:57:06 -07003185
3186 return false;
3187}
Mel Gorman56de7262010-05-24 14:32:30 -07003188#else
3189static inline struct page *
3190__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07003191 unsigned int alloc_flags, const struct alloc_context *ac,
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003192 enum compact_priority prio, enum compact_result *compact_result)
Mel Gorman56de7262010-05-24 14:32:30 -07003193{
Michal Hocko33c2d212016-05-20 16:57:06 -07003194 *compact_result = COMPACT_SKIPPED;
Mel Gorman56de7262010-05-24 14:32:30 -07003195 return NULL;
3196}
Michal Hocko33c2d212016-05-20 16:57:06 -07003197
3198static inline bool
Michal Hocko86a294a2016-05-20 16:57:12 -07003199should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
3200 enum compact_result compact_result,
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003201 enum compact_priority *compact_priority,
Michal Hocko33c2d212016-05-20 16:57:06 -07003202 int compaction_retries)
3203{
Michal Hocko31e49bf2016-05-20 16:57:15 -07003204 struct zone *zone;
3205 struct zoneref *z;
3206
3207 if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
3208 return false;
3209
3210 /*
3211 * There are setups with compaction disabled which would prefer to loop
3212 * inside the allocator rather than hit the oom killer prematurely.
3213 * Let's give them a good hope and keep retrying while the order-0
3214 * watermarks are OK.
3215 */
3216 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3217 ac->nodemask) {
3218 if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
3219 ac_classzone_idx(ac), alloc_flags))
3220 return true;
3221 }
Michal Hocko33c2d212016-05-20 16:57:06 -07003222 return false;
3223}
Mel Gorman56de7262010-05-24 14:32:30 -07003224#endif /* CONFIG_COMPACTION */
3225
Marek Szyprowskibba90712012-01-25 12:09:52 +01003226/* Perform direct synchronous page reclaim */
3227static int
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003228__perform_reclaim(gfp_t gfp_mask, unsigned int order,
3229 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003230{
Mel Gorman11e33f62009-06-16 15:31:57 -07003231 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01003232 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07003233
3234 cond_resched();
3235
3236 /* We now go into synchronous reclaim */
3237 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003238 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07003239 lockdep_set_current_reclaim_state(gfp_mask);
3240 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003241 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07003242
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003243 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
3244 ac->nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07003245
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003246 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07003247 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003248 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07003249
3250 cond_resched();
3251
Marek Szyprowskibba90712012-01-25 12:09:52 +01003252 return progress;
3253}
3254
3255/* The really slow allocator path where we enter direct reclaim */
3256static inline struct page *
3257__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07003258 unsigned int alloc_flags, const struct alloc_context *ac,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003259 unsigned long *did_some_progress)
Marek Szyprowskibba90712012-01-25 12:09:52 +01003260{
3261 struct page *page = NULL;
3262 bool drained = false;
3263
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003264 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003265 if (unlikely(!(*did_some_progress)))
3266 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07003267
Mel Gorman9ee493c2010-09-09 16:38:18 -07003268retry:
Vlastimil Babka31a6c192016-07-28 15:49:13 -07003269 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003270
3271 /*
3272 * If an allocation failed after direct reclaim, it could be because
Mel Gorman0aaa29a2015-11-06 16:28:37 -08003273 * pages are pinned on the per-cpu lists or in high alloc reserves.
3274 * Shrink them them and try again
Mel Gorman9ee493c2010-09-09 16:38:18 -07003275 */
3276 if (!page && !drained) {
Mel Gorman0aaa29a2015-11-06 16:28:37 -08003277 unreserve_highatomic_pageblock(ac);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08003278 drain_all_pages(NULL);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003279 drained = true;
3280 goto retry;
3281 }
3282
Mel Gorman11e33f62009-06-16 15:31:57 -07003283 return page;
3284}
3285
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003286static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003287{
3288 struct zoneref *z;
3289 struct zone *zone;
Mel Gormane1a55632016-07-28 15:46:26 -07003290 pg_data_t *last_pgdat = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07003291
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003292 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
Mel Gormane1a55632016-07-28 15:46:26 -07003293 ac->high_zoneidx, ac->nodemask) {
3294 if (last_pgdat != zone->zone_pgdat)
Mel Gorman52e9f872016-07-28 15:46:29 -07003295 wakeup_kswapd(zone, order, ac->high_zoneidx);
Mel Gormane1a55632016-07-28 15:46:26 -07003296 last_pgdat = zone->zone_pgdat;
3297 }
Mel Gorman11e33f62009-06-16 15:31:57 -07003298}
3299
Mel Gormanc6038442016-05-19 17:13:38 -07003300static inline unsigned int
Peter Zijlstra341ce062009-06-16 15:32:02 -07003301gfp_to_alloc_flags(gfp_t gfp_mask)
3302{
Mel Gormanc6038442016-05-19 17:13:38 -07003303 unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
Peter Zijlstra341ce062009-06-16 15:32:02 -07003304
Mel Gormana56f57f2009-06-16 15:32:02 -07003305 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07003306 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07003307
Peter Zijlstra341ce062009-06-16 15:32:02 -07003308 /*
3309 * The caller may dip into page reserves a bit more if the caller
3310 * cannot run direct reclaim, or if the caller has realtime scheduling
3311 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
Mel Gormand0164ad2015-11-06 16:28:21 -08003312 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
Peter Zijlstra341ce062009-06-16 15:32:02 -07003313 */
Namhyung Kime6223a32010-10-26 14:21:59 -07003314 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07003315
Mel Gormand0164ad2015-11-06 16:28:21 -08003316 if (gfp_mask & __GFP_ATOMIC) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003317 /*
David Rientjesb104a352014-07-30 16:08:24 -07003318 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3319 * if it can't schedule.
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003320 */
David Rientjesb104a352014-07-30 16:08:24 -07003321 if (!(gfp_mask & __GFP_NOMEMALLOC))
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003322 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07003323 /*
David Rientjesb104a352014-07-30 16:08:24 -07003324 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
Vladimir Davydov344736f2014-10-20 15:50:30 +04003325 * comment for __cpuset_node_allowed().
Peter Zijlstra341ce062009-06-16 15:32:02 -07003326 */
3327 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003328 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07003329 alloc_flags |= ALLOC_HARDER;
3330
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07003331#ifdef CONFIG_CMA
David Rientjes43e7a342014-10-09 15:27:25 -07003332 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07003333 alloc_flags |= ALLOC_CMA;
3334#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07003335 return alloc_flags;
3336}
3337
Mel Gorman072bb0a2012-07-31 16:43:58 -07003338bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
3339{
Vlastimil Babka31a6c192016-07-28 15:49:13 -07003340 if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
3341 return false;
3342
3343 if (gfp_mask & __GFP_MEMALLOC)
3344 return true;
3345 if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
3346 return true;
3347 if (!in_interrupt() &&
3348 ((current->flags & PF_MEMALLOC) ||
3349 unlikely(test_thread_flag(TIF_MEMDIE))))
3350 return true;
3351
3352 return false;
Mel Gorman072bb0a2012-07-31 16:43:58 -07003353}
3354
Michal Hocko0a0337e2016-05-20 16:57:00 -07003355/*
3356 * Maximum number of reclaim retries without any progress before OOM killer
3357 * is consider as the only way to move forward.
3358 */
3359#define MAX_RECLAIM_RETRIES 16
3360
3361/*
3362 * Checks whether it makes sense to retry the reclaim to make a forward progress
3363 * for the given allocation request.
3364 * The reclaim feedback represented by did_some_progress (any progress during
Michal Hocko7854ea62016-05-20 16:57:09 -07003365 * the last reclaim round) and no_progress_loops (number of reclaim rounds without
3366 * any progress in a row) is considered as well as the reclaimable pages on the
3367 * applicable zone list (with a backoff mechanism which is a function of
3368 * no_progress_loops).
Michal Hocko0a0337e2016-05-20 16:57:00 -07003369 *
3370 * Returns true if a retry is viable or false to enter the oom path.
3371 */
3372static inline bool
3373should_reclaim_retry(gfp_t gfp_mask, unsigned order,
3374 struct alloc_context *ac, int alloc_flags,
Michal Hocko7854ea62016-05-20 16:57:09 -07003375 bool did_some_progress, int no_progress_loops)
Michal Hocko0a0337e2016-05-20 16:57:00 -07003376{
3377 struct zone *zone;
3378 struct zoneref *z;
3379
3380 /*
3381 * Make sure we converge to OOM if we cannot make any progress
3382 * several times in the row.
3383 */
3384 if (no_progress_loops > MAX_RECLAIM_RETRIES)
3385 return false;
3386
Michal Hocko0a0337e2016-05-20 16:57:00 -07003387 /*
Mel Gormanbca67592016-07-28 15:47:05 -07003388 * Keep reclaiming pages while there is a chance this will lead
3389 * somewhere. If none of the target zones can satisfy our allocation
3390 * request even if all reclaimable pages are considered then we are
3391 * screwed and have to go OOM.
Michal Hocko0a0337e2016-05-20 16:57:00 -07003392 */
3393 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3394 ac->nodemask) {
3395 unsigned long available;
Michal Hockoede37712016-05-20 16:57:03 -07003396 unsigned long reclaimable;
Michal Hocko0a0337e2016-05-20 16:57:00 -07003397
Mel Gorman5a1c84b2016-07-28 15:47:31 -07003398 available = reclaimable = zone_reclaimable_pages(zone);
Michal Hocko0a0337e2016-05-20 16:57:00 -07003399 available -= DIV_ROUND_UP(no_progress_loops * available,
3400 MAX_RECLAIM_RETRIES);
Mel Gorman5a1c84b2016-07-28 15:47:31 -07003401 available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
Michal Hocko0a0337e2016-05-20 16:57:00 -07003402
3403 /*
3404 * Would the allocation succeed if we reclaimed the whole
Mel Gorman5a1c84b2016-07-28 15:47:31 -07003405 * available?
Michal Hocko0a0337e2016-05-20 16:57:00 -07003406 */
Mel Gorman5a1c84b2016-07-28 15:47:31 -07003407 if (__zone_watermark_ok(zone, order, min_wmark_pages(zone),
3408 ac_classzone_idx(ac), alloc_flags, available)) {
Michal Hockoede37712016-05-20 16:57:03 -07003409 /*
3410 * If we didn't make any progress and have a lot of
3411 * dirty + writeback pages then we should wait for
3412 * an IO to complete to slow down the reclaim and
3413 * prevent from pre mature OOM
3414 */
3415 if (!did_some_progress) {
Mel Gorman11fb9982016-07-28 15:46:20 -07003416 unsigned long write_pending;
Michal Hockoede37712016-05-20 16:57:03 -07003417
Mel Gorman5a1c84b2016-07-28 15:47:31 -07003418 write_pending = zone_page_state_snapshot(zone,
3419 NR_ZONE_WRITE_PENDING);
Michal Hockoede37712016-05-20 16:57:03 -07003420
Mel Gorman11fb9982016-07-28 15:46:20 -07003421 if (2 * write_pending > reclaimable) {
Michal Hockoede37712016-05-20 16:57:03 -07003422 congestion_wait(BLK_RW_ASYNC, HZ/10);
3423 return true;
3424 }
3425 }
Mel Gorman5a1c84b2016-07-28 15:47:31 -07003426
Michal Hockoede37712016-05-20 16:57:03 -07003427 /*
3428 * Memory allocation/reclaim might be called from a WQ
3429 * context and the current implementation of the WQ
3430 * concurrency control doesn't recognize that
3431 * a particular WQ is congested if the worker thread is
3432 * looping without ever sleeping. Therefore we have to
3433 * do a short sleep here rather than calling
3434 * cond_resched().
3435 */
3436 if (current->flags & PF_WQ_WORKER)
3437 schedule_timeout_uninterruptible(1);
3438 else
3439 cond_resched();
3440
Michal Hocko0a0337e2016-05-20 16:57:00 -07003441 return true;
3442 }
3443 }
3444
3445 return false;
3446}
3447
Mel Gorman11e33f62009-06-16 15:31:57 -07003448static inline struct page *
3449__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003450 struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003451{
Mel Gormand0164ad2015-11-06 16:28:21 -08003452 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
Mel Gorman11e33f62009-06-16 15:31:57 -07003453 struct page *page = NULL;
Mel Gormanc6038442016-05-19 17:13:38 -07003454 unsigned int alloc_flags;
Mel Gorman11e33f62009-06-16 15:31:57 -07003455 unsigned long did_some_progress;
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003456 enum compact_priority compact_priority = DEF_COMPACT_PRIORITY;
Michal Hockoc5d01d02016-05-20 16:56:53 -07003457 enum compact_result compact_result;
Michal Hocko33c2d212016-05-20 16:57:06 -07003458 int compaction_retries = 0;
Michal Hocko0a0337e2016-05-20 16:57:00 -07003459 int no_progress_loops = 0;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003460
Christoph Lameter952f3b52006-12-06 20:33:26 -08003461 /*
Mel Gorman72807a72009-06-16 15:32:18 -07003462 * In the slowpath, we sanity check order to avoid ever trying to
3463 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3464 * be using allocators in order of preference for an area that is
3465 * too large.
3466 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07003467 if (order >= MAX_ORDER) {
3468 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07003469 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07003470 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003471
Christoph Lameter952f3b52006-12-06 20:33:26 -08003472 /*
Mel Gormand0164ad2015-11-06 16:28:21 -08003473 * We also sanity check to catch abuse of atomic reserves being used by
3474 * callers that are not in atomic context.
3475 */
3476 if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
3477 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
3478 gfp_mask &= ~__GFP_ATOMIC;
3479
Vlastimil Babka31a6c192016-07-28 15:49:13 -07003480 /*
3481 * The fast path uses conservative alloc_flags to succeed only until
3482 * kswapd needs to be woken up, and to avoid the cost of setting up
3483 * alloc_flags precisely. So we do that now.
3484 */
3485 alloc_flags = gfp_to_alloc_flags(gfp_mask);
3486
Mel Gormand0164ad2015-11-06 16:28:21 -08003487 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003488 wake_all_kswapds(order, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003489
Paul Jackson9bf22292005-09-06 15:18:12 -07003490 /*
Vlastimil Babka23771232016-07-28 15:49:16 -07003491 * The adjusted alloc_flags might result in immediate success, so try
3492 * that first
3493 */
3494 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3495 if (page)
3496 goto got_pg;
3497
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003498 /*
3499 * For costly allocations, try direct compaction first, as it's likely
3500 * that we have enough base pages and don't need to reclaim. Don't try
3501 * that for allocations that are allowed to ignore watermarks, as the
3502 * ALLOC_NO_WATERMARKS attempt didn't yet happen.
3503 */
3504 if (can_direct_reclaim && order > PAGE_ALLOC_COSTLY_ORDER &&
3505 !gfp_pfmemalloc_allowed(gfp_mask)) {
3506 page = __alloc_pages_direct_compact(gfp_mask, order,
3507 alloc_flags, ac,
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003508 INIT_COMPACT_PRIORITY,
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003509 &compact_result);
3510 if (page)
3511 goto got_pg;
3512
Vlastimil Babka3eb27712016-07-28 15:49:22 -07003513 /*
3514 * Checks for costly allocations with __GFP_NORETRY, which
3515 * includes THP page fault allocations
3516 */
3517 if (gfp_mask & __GFP_NORETRY) {
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003518 /*
3519 * If compaction is deferred for high-order allocations,
3520 * it is because sync compaction recently failed. If
3521 * this is the case and the caller requested a THP
3522 * allocation, we do not want to heavily disrupt the
3523 * system, so we fail the allocation instead of entering
3524 * direct reclaim.
3525 */
3526 if (compact_result == COMPACT_DEFERRED)
3527 goto nopage;
3528
3529 /*
Vlastimil Babka3eb27712016-07-28 15:49:22 -07003530 * Looks like reclaim/compaction is worth trying, but
3531 * sync compaction could be very expensive, so keep
Vlastimil Babka25160352016-07-28 15:49:25 -07003532 * using async compaction.
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003533 */
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003534 compact_priority = INIT_COMPACT_PRIORITY;
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003535 }
3536 }
Vlastimil Babka23771232016-07-28 15:49:16 -07003537
3538retry:
3539 /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
3540 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3541 wake_all_kswapds(order, ac);
3542
3543 if (gfp_pfmemalloc_allowed(gfp_mask))
3544 alloc_flags = ALLOC_NO_WATERMARKS;
3545
3546 /*
Mel Gormane46e7b72016-06-03 14:56:01 -07003547 * Reset the zonelist iterators if memory policies can be ignored.
3548 * These allocations are high priority and system rather than user
3549 * orientated.
3550 */
Vlastimil Babka23771232016-07-28 15:49:16 -07003551 if (!(alloc_flags & ALLOC_CPUSET) || (alloc_flags & ALLOC_NO_WATERMARKS)) {
Mel Gormane46e7b72016-06-03 14:56:01 -07003552 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
3553 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
3554 ac->high_zoneidx, ac->nodemask);
3555 }
3556
Vlastimil Babka23771232016-07-28 15:49:16 -07003557 /* Attempt with potentially adjusted zonelist and alloc_flags */
Vlastimil Babka31a6c192016-07-28 15:49:13 -07003558 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003559 if (page)
3560 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561
Mel Gormand0164ad2015-11-06 16:28:21 -08003562 /* Caller is not willing to reclaim, we can't balance anything */
3563 if (!can_direct_reclaim) {
David Rientjesaed0a0e2014-01-21 15:51:12 -08003564 /*
Michal Hocko33d53102016-01-14 15:19:05 -08003565 * All existing users of the __GFP_NOFAIL are blockable, so warn
3566 * of any new users that actually allow this type of allocation
3567 * to fail.
David Rientjesaed0a0e2014-01-21 15:51:12 -08003568 */
3569 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570 goto nopage;
David Rientjesaed0a0e2014-01-21 15:51:12 -08003571 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572
Peter Zijlstra341ce062009-06-16 15:32:02 -07003573 /* Avoid recursion of direct reclaim */
Michal Hocko33d53102016-01-14 15:19:05 -08003574 if (current->flags & PF_MEMALLOC) {
3575 /*
3576 * __GFP_NOFAIL request from this context is rather bizarre
3577 * because we cannot reclaim anything and only can loop waiting
3578 * for somebody to do a work for us.
3579 */
3580 if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3581 cond_resched();
3582 goto retry;
3583 }
Peter Zijlstra341ce062009-06-16 15:32:02 -07003584 goto nopage;
Michal Hocko33d53102016-01-14 15:19:05 -08003585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003586
David Rientjes6583bb62009-07-29 15:02:06 -07003587 /* Avoid allocations with no watermarks from looping endlessly */
3588 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3589 goto nopage;
3590
David Rientjes8fe78042014-08-06 16:07:54 -07003591
Mel Gorman11e33f62009-06-16 15:31:57 -07003592 /* Try direct reclaim and then allocating */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003593 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3594 &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07003595 if (page)
3596 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003597
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003598 /* Try direct compaction and then allocating */
3599 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003600 compact_priority, &compact_result);
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003601 if (page)
3602 goto got_pg;
3603
3604 if (order && compaction_made_progress(compact_result))
3605 compaction_retries++;
3606
Johannes Weiner90839052015-06-24 16:57:21 -07003607 /* Do not loop if specifically requested */
3608 if (gfp_mask & __GFP_NORETRY)
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003609 goto nopage;
Johannes Weiner90839052015-06-24 16:57:21 -07003610
Michal Hocko0a0337e2016-05-20 16:57:00 -07003611 /*
3612 * Do not retry costly high order allocations unless they are
3613 * __GFP_REPEAT
3614 */
3615 if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_REPEAT))
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003616 goto nopage;
Michal Hocko0a0337e2016-05-20 16:57:00 -07003617
Michal Hocko7854ea62016-05-20 16:57:09 -07003618 /*
3619 * Costly allocations might have made a progress but this doesn't mean
3620 * their order will become available due to high fragmentation so
3621 * always increment the no progress counter for them
3622 */
3623 if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
Michal Hocko0a0337e2016-05-20 16:57:00 -07003624 no_progress_loops = 0;
Michal Hocko7854ea62016-05-20 16:57:09 -07003625 else
Michal Hocko0a0337e2016-05-20 16:57:00 -07003626 no_progress_loops++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627
Michal Hocko0a0337e2016-05-20 16:57:00 -07003628 if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
Michal Hocko7854ea62016-05-20 16:57:09 -07003629 did_some_progress > 0, no_progress_loops))
Michal Hocko0a0337e2016-05-20 16:57:00 -07003630 goto retry;
3631
Michal Hocko33c2d212016-05-20 16:57:06 -07003632 /*
3633 * It doesn't make any sense to retry for the compaction if the order-0
3634 * reclaim is not able to make any progress because the current
3635 * implementation of the compaction depends on the sufficient amount
3636 * of free memory (see __compaction_suitable)
3637 */
3638 if (did_some_progress > 0 &&
Michal Hocko86a294a2016-05-20 16:57:12 -07003639 should_compact_retry(ac, order, alloc_flags,
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003640 compact_result, &compact_priority,
Michal Hocko86a294a2016-05-20 16:57:12 -07003641 compaction_retries))
Michal Hocko33c2d212016-05-20 16:57:06 -07003642 goto retry;
3643
Johannes Weiner90839052015-06-24 16:57:21 -07003644 /* Reclaim has failed us, start killing things */
3645 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3646 if (page)
3647 goto got_pg;
3648
3649 /* Retry as long as the OOM killer is making progress */
Michal Hocko0a0337e2016-05-20 16:57:00 -07003650 if (did_some_progress) {
3651 no_progress_loops = 0;
Johannes Weiner90839052015-06-24 16:57:21 -07003652 goto retry;
Michal Hocko0a0337e2016-05-20 16:57:00 -07003653 }
Johannes Weiner90839052015-06-24 16:57:21 -07003654
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07003656 warn_alloc_failed(gfp_mask, order, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657got_pg:
Mel Gorman072bb0a2012-07-31 16:43:58 -07003658 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659}
Mel Gorman11e33f62009-06-16 15:31:57 -07003660
3661/*
3662 * This is the 'heart' of the zoned buddy allocator.
3663 */
3664struct page *
3665__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3666 struct zonelist *zonelist, nodemask_t *nodemask)
3667{
Mel Gorman5bb1b162016-05-19 17:13:50 -07003668 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003669 unsigned int cpuset_mems_cookie;
Mel Gormane6cbd7f2016-07-28 15:46:50 -07003670 unsigned int alloc_flags = ALLOC_WMARK_LOW;
Mel Gorman83d4ca82016-05-19 17:13:56 -07003671 gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003672 struct alloc_context ac = {
3673 .high_zoneidx = gfp_zone(gfp_mask),
Mel Gorman682a3382016-05-19 17:13:30 -07003674 .zonelist = zonelist,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003675 .nodemask = nodemask,
3676 .migratetype = gfpflags_to_migratetype(gfp_mask),
3677 };
Mel Gorman11e33f62009-06-16 15:31:57 -07003678
Mel Gorman682a3382016-05-19 17:13:30 -07003679 if (cpusets_enabled()) {
Mel Gorman83d4ca82016-05-19 17:13:56 -07003680 alloc_mask |= __GFP_HARDWALL;
Mel Gorman682a3382016-05-19 17:13:30 -07003681 alloc_flags |= ALLOC_CPUSET;
3682 if (!ac.nodemask)
3683 ac.nodemask = &cpuset_current_mems_allowed;
3684 }
3685
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003686 gfp_mask &= gfp_allowed_mask;
3687
Mel Gorman11e33f62009-06-16 15:31:57 -07003688 lockdep_trace_alloc(gfp_mask);
3689
Mel Gormand0164ad2015-11-06 16:28:21 -08003690 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
Mel Gorman11e33f62009-06-16 15:31:57 -07003691
3692 if (should_fail_alloc_page(gfp_mask, order))
3693 return NULL;
3694
3695 /*
3696 * Check the zones suitable for the gfp_mask contain at least one
3697 * valid zone. It's possible to have an empty zonelist as a result
David Rientjes4167e9b2015-04-14 15:46:55 -07003698 * of __GFP_THISNODE and a memoryless node
Mel Gorman11e33f62009-06-16 15:31:57 -07003699 */
3700 if (unlikely(!zonelist->_zonerefs->zone))
3701 return NULL;
3702
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003703 if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
Vlastimil Babka21bb9bd2014-10-09 15:26:51 -07003704 alloc_flags |= ALLOC_CMA;
3705
Mel Gormancc9a6c82012-03-21 16:34:11 -07003706retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07003707 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003708
Mel Gormanc9ab0c42015-11-06 16:28:12 -08003709 /* Dirty zone balancing only done in the fast path */
3710 ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE);
3711
Mel Gormane46e7b72016-06-03 14:56:01 -07003712 /*
3713 * The preferred zone is used for statistics but crucially it is
3714 * also used as the starting point for the zonelist iterator. It
3715 * may get reset for allocations that ignore memory policies.
3716 */
Mel Gormanc33d6c02016-05-19 17:14:10 -07003717 ac.preferred_zoneref = first_zones_zonelist(ac.zonelist,
3718 ac.high_zoneidx, ac.nodemask);
3719 if (!ac.preferred_zoneref) {
Mel Gorman5bb1b162016-05-19 17:13:50 -07003720 page = NULL;
Mel Gorman4fcb0972016-05-19 17:14:01 -07003721 goto no_zone;
Mel Gorman5bb1b162016-05-19 17:13:50 -07003722 }
3723
Mel Gorman5117f452009-06-16 15:31:59 -07003724 /* First allocation attempt */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003725 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
Mel Gorman4fcb0972016-05-19 17:14:01 -07003726 if (likely(page))
3727 goto out;
Andrew Morton91fbdc02015-02-11 15:25:04 -08003728
Mel Gorman4fcb0972016-05-19 17:14:01 -07003729 /*
3730 * Runtime PM, block IO and its error handling path can deadlock
3731 * because I/O on the device might not complete.
3732 */
3733 alloc_mask = memalloc_noio_flags(gfp_mask);
3734 ac.spread_dirty_pages = false;
Mel Gorman11e33f62009-06-16 15:31:57 -07003735
Mel Gorman47415262016-05-19 17:14:44 -07003736 /*
3737 * Restore the original nodemask if it was potentially replaced with
3738 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
3739 */
3740 if (cpusets_enabled())
3741 ac.nodemask = nodemask;
Mel Gorman4fcb0972016-05-19 17:14:01 -07003742 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
Xishi Qiu23f086f2015-02-11 15:25:07 -08003743
Mel Gorman4fcb0972016-05-19 17:14:01 -07003744no_zone:
Mel Gormancc9a6c82012-03-21 16:34:11 -07003745 /*
3746 * When updating a task's mems_allowed, it is possible to race with
3747 * parallel threads in such a way that an allocation can fail while
3748 * the mask is being updated. If a page allocation is about to fail,
3749 * check if the cpuset changed during allocation and if so, retry.
3750 */
Mel Gorman83d4ca82016-05-19 17:13:56 -07003751 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) {
3752 alloc_mask = gfp_mask;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003753 goto retry_cpuset;
Mel Gorman83d4ca82016-05-19 17:13:56 -07003754 }
Mel Gormancc9a6c82012-03-21 16:34:11 -07003755
Mel Gorman4fcb0972016-05-19 17:14:01 -07003756out:
Vladimir Davydovc4159a72016-08-08 23:03:12 +03003757 if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page &&
3758 unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) {
3759 __free_pages(page, order);
3760 page = NULL;
Vladimir Davydov49491482016-07-26 15:24:24 -07003761 }
3762
Mel Gorman4fcb0972016-05-19 17:14:01 -07003763 if (kmemcheck_enabled && page)
3764 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3765
3766 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
3767
Mel Gorman11e33f62009-06-16 15:31:57 -07003768 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769}
Mel Gormand2391712009-06-16 15:31:52 -07003770EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771
3772/*
3773 * Common helper functions.
3774 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08003775unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003776{
Akinobu Mita945a1112009-09-21 17:01:47 -07003777 struct page *page;
3778
3779 /*
3780 * __get_free_pages() returns a 32-bit address, which cannot represent
3781 * a highmem page
3782 */
3783 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3784
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785 page = alloc_pages(gfp_mask, order);
3786 if (!page)
3787 return 0;
3788 return (unsigned long) page_address(page);
3789}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003790EXPORT_SYMBOL(__get_free_pages);
3791
Harvey Harrison920c7a52008-02-04 22:29:26 -08003792unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793{
Akinobu Mita945a1112009-09-21 17:01:47 -07003794 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003795}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003796EXPORT_SYMBOL(get_zeroed_page);
3797
Harvey Harrison920c7a52008-02-04 22:29:26 -08003798void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003799{
Nick Pigginb5810032005-10-29 18:16:12 -07003800 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003801 if (order == 0)
Mel Gormanb745bc82014-06-04 16:10:22 -07003802 free_hot_cold_page(page, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003803 else
3804 __free_pages_ok(page, order);
3805 }
3806}
3807
3808EXPORT_SYMBOL(__free_pages);
3809
Harvey Harrison920c7a52008-02-04 22:29:26 -08003810void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811{
3812 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07003813 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814 __free_pages(virt_to_page((void *)addr), order);
3815 }
3816}
3817
3818EXPORT_SYMBOL(free_pages);
3819
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003820/*
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003821 * Page Fragment:
3822 * An arbitrary-length arbitrary-offset area of memory which resides
3823 * within a 0 or higher order page. Multiple fragments within that page
3824 * are individually refcounted, in the page's reference counter.
3825 *
3826 * The page_frag functions below provide a simple allocation framework for
3827 * page fragments. This is used by the network stack and network device
3828 * drivers to provide a backing region of memory for use as either an
3829 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3830 */
3831static struct page *__page_frag_refill(struct page_frag_cache *nc,
3832 gfp_t gfp_mask)
3833{
3834 struct page *page = NULL;
3835 gfp_t gfp = gfp_mask;
3836
3837#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3838 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3839 __GFP_NOMEMALLOC;
3840 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3841 PAGE_FRAG_CACHE_MAX_ORDER);
3842 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3843#endif
3844 if (unlikely(!page))
3845 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3846
3847 nc->va = page ? page_address(page) : NULL;
3848
3849 return page;
3850}
3851
3852void *__alloc_page_frag(struct page_frag_cache *nc,
3853 unsigned int fragsz, gfp_t gfp_mask)
3854{
3855 unsigned int size = PAGE_SIZE;
3856 struct page *page;
3857 int offset;
3858
3859 if (unlikely(!nc->va)) {
3860refill:
3861 page = __page_frag_refill(nc, gfp_mask);
3862 if (!page)
3863 return NULL;
3864
3865#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3866 /* if size can vary use size else just use PAGE_SIZE */
3867 size = nc->size;
3868#endif
3869 /* Even if we own the page, we do not use atomic_set().
3870 * This would break get_page_unless_zero() users.
3871 */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07003872 page_ref_add(page, size - 1);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003873
3874 /* reset page count bias and offset to start of new frag */
Michal Hocko2f064f32015-08-21 14:11:51 -07003875 nc->pfmemalloc = page_is_pfmemalloc(page);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003876 nc->pagecnt_bias = size;
3877 nc->offset = size;
3878 }
3879
3880 offset = nc->offset - fragsz;
3881 if (unlikely(offset < 0)) {
3882 page = virt_to_page(nc->va);
3883
Joonsoo Kimfe896d12016-03-17 14:19:26 -07003884 if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003885 goto refill;
3886
3887#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3888 /* if size can vary use size else just use PAGE_SIZE */
3889 size = nc->size;
3890#endif
3891 /* OK, page count is 0, we can safely set it */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07003892 set_page_count(page, size);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003893
3894 /* reset page count bias and offset to start of new frag */
3895 nc->pagecnt_bias = size;
3896 offset = size - fragsz;
3897 }
3898
3899 nc->pagecnt_bias--;
3900 nc->offset = offset;
3901
3902 return nc->va + offset;
3903}
3904EXPORT_SYMBOL(__alloc_page_frag);
3905
3906/*
3907 * Frees a page fragment allocated out of either a compound or order 0 page.
3908 */
3909void __free_page_frag(void *addr)
3910{
3911 struct page *page = virt_to_head_page(addr);
3912
3913 if (unlikely(put_page_testzero(page)))
3914 __free_pages_ok(page, compound_order(page));
3915}
3916EXPORT_SYMBOL(__free_page_frag);
3917
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08003918static void *make_alloc_exact(unsigned long addr, unsigned int order,
3919 size_t size)
Andi Kleenee85c2e2011-05-11 15:13:34 -07003920{
3921 if (addr) {
3922 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3923 unsigned long used = addr + PAGE_ALIGN(size);
3924
3925 split_page(virt_to_page((void *)addr), order);
3926 while (used < alloc_end) {
3927 free_page(used);
3928 used += PAGE_SIZE;
3929 }
3930 }
3931 return (void *)addr;
3932}
3933
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003934/**
3935 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3936 * @size: the number of bytes to allocate
3937 * @gfp_mask: GFP flags for the allocation
3938 *
3939 * This function is similar to alloc_pages(), except that it allocates the
3940 * minimum number of pages to satisfy the request. alloc_pages() can only
3941 * allocate memory in power-of-two pages.
3942 *
3943 * This function is also limited by MAX_ORDER.
3944 *
3945 * Memory allocated by this function must be released by free_pages_exact().
3946 */
3947void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3948{
3949 unsigned int order = get_order(size);
3950 unsigned long addr;
3951
3952 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07003953 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003954}
3955EXPORT_SYMBOL(alloc_pages_exact);
3956
3957/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07003958 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3959 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07003960 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07003961 * @size: the number of bytes to allocate
3962 * @gfp_mask: GFP flags for the allocation
3963 *
3964 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3965 * back.
Andi Kleenee85c2e2011-05-11 15:13:34 -07003966 */
Fabian Fredericke1931812014-08-06 16:04:59 -07003967void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
Andi Kleenee85c2e2011-05-11 15:13:34 -07003968{
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08003969 unsigned int order = get_order(size);
Andi Kleenee85c2e2011-05-11 15:13:34 -07003970 struct page *p = alloc_pages_node(nid, gfp_mask, order);
3971 if (!p)
3972 return NULL;
3973 return make_alloc_exact((unsigned long)page_address(p), order, size);
3974}
Andi Kleenee85c2e2011-05-11 15:13:34 -07003975
3976/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003977 * free_pages_exact - release memory allocated via alloc_pages_exact()
3978 * @virt: the value returned by alloc_pages_exact.
3979 * @size: size of allocation, same value as passed to alloc_pages_exact().
3980 *
3981 * Release the memory allocated by a previous call to alloc_pages_exact.
3982 */
3983void free_pages_exact(void *virt, size_t size)
3984{
3985 unsigned long addr = (unsigned long)virt;
3986 unsigned long end = addr + PAGE_ALIGN(size);
3987
3988 while (addr < end) {
3989 free_page(addr);
3990 addr += PAGE_SIZE;
3991 }
3992}
3993EXPORT_SYMBOL(free_pages_exact);
3994
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003995/**
3996 * nr_free_zone_pages - count number of pages beyond high watermark
3997 * @offset: The zone index of the highest zone
3998 *
3999 * nr_free_zone_pages() counts the number of counts pages which are beyond the
4000 * high watermark within all zones at or below a given zone index. For each
4001 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07004002 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004003 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004004static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005{
Mel Gormandd1a2392008-04-28 02:12:17 -07004006 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07004007 struct zone *zone;
4008
Martin J. Blighe310fd42005-07-29 22:59:18 -07004009 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004010 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004011
Mel Gorman0e884602008-04-28 02:12:14 -07004012 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004013
Mel Gorman54a6eb52008-04-28 02:12:16 -07004014 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08004015 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07004016 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07004017 if (size > high)
4018 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019 }
4020
4021 return sum;
4022}
4023
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004024/**
4025 * nr_free_buffer_pages - count number of pages beyond high watermark
4026 *
4027 * nr_free_buffer_pages() counts the number of pages which are beyond the high
4028 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004029 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004030unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004031{
Al Viroaf4ca452005-10-21 02:55:38 -04004032 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004033}
Meelap Shahc2f1a552007-07-17 04:04:39 -07004034EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004035
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004036/**
4037 * nr_free_pagecache_pages - count number of pages beyond high watermark
4038 *
4039 * nr_free_pagecache_pages() counts the number of pages which are beyond the
4040 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004041 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004042unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004043{
Mel Gorman2a1e2742007-07-17 04:03:12 -07004044 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07004046
4047static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004048{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08004049 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08004050 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004051}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004052
Igor Redkod02bd272016-03-17 14:19:05 -07004053long si_mem_available(void)
4054{
4055 long available;
4056 unsigned long pagecache;
4057 unsigned long wmark_low = 0;
4058 unsigned long pages[NR_LRU_LISTS];
4059 struct zone *zone;
4060 int lru;
4061
4062 for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
Mel Gorman2f95ff92016-08-11 15:32:57 -07004063 pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
Igor Redkod02bd272016-03-17 14:19:05 -07004064
4065 for_each_zone(zone)
4066 wmark_low += zone->watermark[WMARK_LOW];
4067
4068 /*
4069 * Estimate the amount of memory available for userspace allocations,
4070 * without causing swapping.
4071 */
4072 available = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
4073
4074 /*
4075 * Not all the page cache can be freed, otherwise the system will
4076 * start swapping. Assume at least half of the page cache, or the
4077 * low watermark worth of cache, needs to stay.
4078 */
4079 pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
4080 pagecache -= min(pagecache / 2, wmark_low);
4081 available += pagecache;
4082
4083 /*
4084 * Part of the reclaimable slab consists of items that are in use,
4085 * and cannot be freed. Cap this estimate at the low watermark.
4086 */
4087 available += global_page_state(NR_SLAB_RECLAIMABLE) -
4088 min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low);
4089
4090 if (available < 0)
4091 available = 0;
4092 return available;
4093}
4094EXPORT_SYMBOL_GPL(si_mem_available);
4095
Linus Torvalds1da177e2005-04-16 15:20:36 -07004096void si_meminfo(struct sysinfo *val)
4097{
4098 val->totalram = totalram_pages;
Mel Gorman11fb9982016-07-28 15:46:20 -07004099 val->sharedram = global_node_page_state(NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08004100 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004101 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004102 val->totalhigh = totalhigh_pages;
4103 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004104 val->mem_unit = PAGE_SIZE;
4105}
4106
4107EXPORT_SYMBOL(si_meminfo);
4108
4109#ifdef CONFIG_NUMA
4110void si_meminfo_node(struct sysinfo *val, int nid)
4111{
Jiang Liucdd91a72013-07-03 15:03:27 -07004112 int zone_type; /* needs to be signed */
4113 unsigned long managed_pages = 0;
Joonsoo Kimfc2bd792016-05-19 17:12:23 -07004114 unsigned long managed_highpages = 0;
4115 unsigned long free_highpages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004116 pg_data_t *pgdat = NODE_DATA(nid);
4117
Jiang Liucdd91a72013-07-03 15:03:27 -07004118 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
4119 managed_pages += pgdat->node_zones[zone_type].managed_pages;
4120 val->totalram = managed_pages;
Mel Gorman11fb9982016-07-28 15:46:20 -07004121 val->sharedram = node_page_state(pgdat, NR_SHMEM);
Mel Gorman75ef7182016-07-28 15:45:24 -07004122 val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004123#ifdef CONFIG_HIGHMEM
Joonsoo Kimfc2bd792016-05-19 17:12:23 -07004124 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
4125 struct zone *zone = &pgdat->node_zones[zone_type];
4126
4127 if (is_highmem(zone)) {
4128 managed_highpages += zone->managed_pages;
4129 free_highpages += zone_page_state(zone, NR_FREE_PAGES);
4130 }
4131 }
4132 val->totalhigh = managed_highpages;
4133 val->freehigh = free_highpages;
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004134#else
Joonsoo Kimfc2bd792016-05-19 17:12:23 -07004135 val->totalhigh = managed_highpages;
4136 val->freehigh = free_highpages;
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004137#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004138 val->mem_unit = PAGE_SIZE;
4139}
4140#endif
4141
David Rientjesddd588b2011-03-22 16:30:46 -07004142/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07004143 * Determine whether the node should be displayed or not, depending on whether
4144 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07004145 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07004146bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07004147{
4148 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07004149 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07004150
4151 if (!(flags & SHOW_MEM_FILTER_NODES))
4152 goto out;
4153
Mel Gormancc9a6c82012-03-21 16:34:11 -07004154 do {
Mel Gormand26914d2014-04-03 14:47:24 -07004155 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07004156 ret = !node_isset(nid, cpuset_current_mems_allowed);
Mel Gormand26914d2014-04-03 14:47:24 -07004157 } while (read_mems_allowed_retry(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07004158out:
4159 return ret;
4160}
4161
Linus Torvalds1da177e2005-04-16 15:20:36 -07004162#define K(x) ((x) << (PAGE_SHIFT-10))
4163
Rabin Vincent377e4f12012-12-11 16:00:24 -08004164static void show_migration_types(unsigned char type)
4165{
4166 static const char types[MIGRATE_TYPES] = {
4167 [MIGRATE_UNMOVABLE] = 'U',
Rabin Vincent377e4f12012-12-11 16:00:24 -08004168 [MIGRATE_MOVABLE] = 'M',
Vlastimil Babka475a2f92015-12-11 13:40:29 -08004169 [MIGRATE_RECLAIMABLE] = 'E',
4170 [MIGRATE_HIGHATOMIC] = 'H',
Rabin Vincent377e4f12012-12-11 16:00:24 -08004171#ifdef CONFIG_CMA
4172 [MIGRATE_CMA] = 'C',
4173#endif
Minchan Kim194159f2013-02-22 16:33:58 -08004174#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08004175 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08004176#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08004177 };
4178 char tmp[MIGRATE_TYPES + 1];
4179 char *p = tmp;
4180 int i;
4181
4182 for (i = 0; i < MIGRATE_TYPES; i++) {
4183 if (type & (1 << i))
4184 *p++ = types[i];
4185 }
4186
4187 *p = '\0';
4188 printk("(%s) ", tmp);
4189}
4190
Linus Torvalds1da177e2005-04-16 15:20:36 -07004191/*
4192 * Show free area list (used inside shift_scroll-lock stuff)
4193 * We also calculate the percentage fragmentation. We do this by counting the
4194 * memory on each free list with the exception of the first item on the list.
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004195 *
4196 * Bits in @filter:
4197 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4198 * cpuset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004199 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07004200void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004201{
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004202 unsigned long free_pcp = 0;
Jes Sorensenc7241912006-09-27 01:50:05 -07004203 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004204 struct zone *zone;
Mel Gorman599d0c92016-07-28 15:45:31 -07004205 pg_data_t *pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004206
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07004207 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07004208 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07004209 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004210
Konstantin Khlebnikov761b0672015-04-14 15:45:32 -07004211 for_each_online_cpu(cpu)
4212 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004213 }
4214
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07004215 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4216 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004217 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4218 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004219 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004220 " free:%lu free_pcp:%lu free_cma:%lu\n",
Mel Gorman599d0c92016-07-28 15:45:31 -07004221 global_node_page_state(NR_ACTIVE_ANON),
4222 global_node_page_state(NR_INACTIVE_ANON),
4223 global_node_page_state(NR_ISOLATED_ANON),
4224 global_node_page_state(NR_ACTIVE_FILE),
4225 global_node_page_state(NR_INACTIVE_FILE),
4226 global_node_page_state(NR_ISOLATED_FILE),
4227 global_node_page_state(NR_UNEVICTABLE),
Mel Gorman11fb9982016-07-28 15:46:20 -07004228 global_node_page_state(NR_FILE_DIRTY),
4229 global_node_page_state(NR_WRITEBACK),
4230 global_node_page_state(NR_UNSTABLE_NFS),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07004231 global_page_state(NR_SLAB_RECLAIMABLE),
4232 global_page_state(NR_SLAB_UNRECLAIMABLE),
Mel Gorman50658e22016-07-28 15:46:14 -07004233 global_node_page_state(NR_FILE_MAPPED),
Mel Gorman11fb9982016-07-28 15:46:20 -07004234 global_node_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08004235 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004236 global_page_state(NR_BOUNCE),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004237 global_page_state(NR_FREE_PAGES),
4238 free_pcp,
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004239 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004240
Mel Gorman599d0c92016-07-28 15:45:31 -07004241 for_each_online_pgdat(pgdat) {
4242 printk("Node %d"
4243 " active_anon:%lukB"
4244 " inactive_anon:%lukB"
4245 " active_file:%lukB"
4246 " inactive_file:%lukB"
4247 " unevictable:%lukB"
4248 " isolated(anon):%lukB"
4249 " isolated(file):%lukB"
Mel Gorman50658e22016-07-28 15:46:14 -07004250 " mapped:%lukB"
Mel Gorman11fb9982016-07-28 15:46:20 -07004251 " dirty:%lukB"
4252 " writeback:%lukB"
4253 " shmem:%lukB"
4254#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4255 " shmem_thp: %lukB"
4256 " shmem_pmdmapped: %lukB"
4257 " anon_thp: %lukB"
4258#endif
4259 " writeback_tmp:%lukB"
4260 " unstable:%lukB"
Minchan Kim33e077b2016-07-28 15:47:14 -07004261 " pages_scanned:%lu"
Mel Gorman599d0c92016-07-28 15:45:31 -07004262 " all_unreclaimable? %s"
4263 "\n",
4264 pgdat->node_id,
4265 K(node_page_state(pgdat, NR_ACTIVE_ANON)),
4266 K(node_page_state(pgdat, NR_INACTIVE_ANON)),
4267 K(node_page_state(pgdat, NR_ACTIVE_FILE)),
4268 K(node_page_state(pgdat, NR_INACTIVE_FILE)),
4269 K(node_page_state(pgdat, NR_UNEVICTABLE)),
4270 K(node_page_state(pgdat, NR_ISOLATED_ANON)),
4271 K(node_page_state(pgdat, NR_ISOLATED_FILE)),
Mel Gorman50658e22016-07-28 15:46:14 -07004272 K(node_page_state(pgdat, NR_FILE_MAPPED)),
Mel Gorman11fb9982016-07-28 15:46:20 -07004273 K(node_page_state(pgdat, NR_FILE_DIRTY)),
4274 K(node_page_state(pgdat, NR_WRITEBACK)),
4275#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4276 K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR),
4277 K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)
4278 * HPAGE_PMD_NR),
4279 K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR),
4280#endif
4281 K(node_page_state(pgdat, NR_SHMEM)),
4282 K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
4283 K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
Minchan Kim33e077b2016-07-28 15:47:14 -07004284 node_page_state(pgdat, NR_PAGES_SCANNED),
Mel Gorman599d0c92016-07-28 15:45:31 -07004285 !pgdat_reclaimable(pgdat) ? "yes" : "no");
4286 }
4287
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07004288 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004289 int i;
4290
David Rientjes7bf02ea2011-05-24 17:11:16 -07004291 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07004292 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004293
4294 free_pcp = 0;
4295 for_each_online_cpu(cpu)
4296 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
4297
Linus Torvalds1da177e2005-04-16 15:20:36 -07004298 show_node(zone);
4299 printk("%s"
4300 " free:%lukB"
4301 " min:%lukB"
4302 " low:%lukB"
4303 " high:%lukB"
Minchan Kim71c799f2016-07-28 15:47:26 -07004304 " active_anon:%lukB"
4305 " inactive_anon:%lukB"
4306 " active_file:%lukB"
4307 " inactive_file:%lukB"
4308 " unevictable:%lukB"
Mel Gorman5a1c84b2016-07-28 15:47:31 -07004309 " writepending:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07004310 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08004311 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004312 " mlocked:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004313 " slab_reclaimable:%lukB"
4314 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07004315 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004316 " pagetables:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004317 " bounce:%lukB"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004318 " free_pcp:%lukB"
4319 " local_pcp:%ukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004320 " free_cma:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07004321 "\n",
4322 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08004323 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07004324 K(min_wmark_pages(zone)),
4325 K(low_wmark_pages(zone)),
4326 K(high_wmark_pages(zone)),
Minchan Kim71c799f2016-07-28 15:47:26 -07004327 K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
4328 K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
4329 K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
4330 K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
4331 K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
Mel Gorman5a1c84b2016-07-28 15:47:31 -07004332 K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07004333 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08004334 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004335 K(zone_page_state(zone, NR_MLOCK)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004336 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
4337 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
Andy Lutomirskid30dd8b2016-07-28 15:48:14 -07004338 zone_page_state(zone, NR_KERNEL_STACK_KB),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004339 K(zone_page_state(zone, NR_PAGETABLE)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004340 K(zone_page_state(zone, NR_BOUNCE)),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004341 K(free_pcp),
4342 K(this_cpu_read(zone->pageset->pcp.count)),
Minchan Kim33e077b2016-07-28 15:47:14 -07004343 K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004344 printk("lowmem_reserve[]:");
4345 for (i = 0; i < MAX_NR_ZONES; i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07004346 printk(" %ld", zone->lowmem_reserve[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004347 printk("\n");
4348 }
4349
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07004350 for_each_populated_zone(zone) {
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004351 unsigned int order;
4352 unsigned long nr[MAX_ORDER], flags, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08004353 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004354
David Rientjes7bf02ea2011-05-24 17:11:16 -07004355 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07004356 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004357 show_node(zone);
4358 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004359
4360 spin_lock_irqsave(&zone->lock, flags);
4361 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08004362 struct free_area *area = &zone->free_area[order];
4363 int type;
4364
4365 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07004366 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08004367
4368 types[order] = 0;
4369 for (type = 0; type < MIGRATE_TYPES; type++) {
4370 if (!list_empty(&area->free_list[type]))
4371 types[order] |= 1 << type;
4372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004373 }
4374 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08004375 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07004376 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08004377 if (nr[order])
4378 show_migration_types(types[order]);
4379 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004380 printk("= %lukB\n", K(total));
4381 }
4382
David Rientjes949f7ec2013-04-29 15:07:48 -07004383 hugetlb_show_meminfo();
4384
Mel Gorman11fb9982016-07-28 15:46:20 -07004385 printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
Larry Woodmane6f36022008-02-04 22:29:30 -08004386
Linus Torvalds1da177e2005-04-16 15:20:36 -07004387 show_swap_cache_info();
4388}
4389
Mel Gorman19770b32008-04-28 02:12:18 -07004390static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
4391{
4392 zoneref->zone = zone;
4393 zoneref->zone_idx = zone_idx(zone);
4394}
4395
Linus Torvalds1da177e2005-04-16 15:20:36 -07004396/*
4397 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08004398 *
4399 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004400 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004401static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004402 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004403{
Christoph Lameter1a932052006-01-06 00:11:16 -08004404 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004405 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08004406
4407 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004408 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08004409 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08004410 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07004411 zoneref_set_zone(zone,
4412 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08004413 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004414 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004415 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004416
Christoph Lameter070f8032006-01-06 00:11:19 -08004417 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004418}
4419
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004420
4421/*
4422 * zonelist_order:
4423 * 0 = automatic detection of better ordering.
4424 * 1 = order by ([node] distance, -zonetype)
4425 * 2 = order by (-zonetype, [node] distance)
4426 *
4427 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4428 * the same zonelist. So only NUMA can configure this param.
4429 */
4430#define ZONELIST_ORDER_DEFAULT 0
4431#define ZONELIST_ORDER_NODE 1
4432#define ZONELIST_ORDER_ZONE 2
4433
4434/* zonelist order in the kernel.
4435 * set_zonelist_order() will set this to NODE or ZONE.
4436 */
4437static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
4438static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
4439
4440
Linus Torvalds1da177e2005-04-16 15:20:36 -07004441#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004442/* The value user specified ....changed by config */
4443static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4444/* string for sysctl */
4445#define NUMA_ZONELIST_ORDER_LEN 16
4446char numa_zonelist_order[16] = "default";
4447
4448/*
4449 * interface for configure zonelist ordering.
4450 * command line option "numa_zonelist_order"
4451 * = "[dD]efault - default, automatic configuration.
4452 * = "[nN]ode - order by node locality, then by zone within node
4453 * = "[zZ]one - order by zone, then by locality within zone
4454 */
4455
4456static int __parse_numa_zonelist_order(char *s)
4457{
4458 if (*s == 'd' || *s == 'D') {
4459 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4460 } else if (*s == 'n' || *s == 'N') {
4461 user_zonelist_order = ZONELIST_ORDER_NODE;
4462 } else if (*s == 'z' || *s == 'Z') {
4463 user_zonelist_order = ZONELIST_ORDER_ZONE;
4464 } else {
Joe Perches11705322016-03-17 14:19:50 -07004465 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004466 return -EINVAL;
4467 }
4468 return 0;
4469}
4470
4471static __init int setup_numa_zonelist_order(char *s)
4472{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08004473 int ret;
4474
4475 if (!s)
4476 return 0;
4477
4478 ret = __parse_numa_zonelist_order(s);
4479 if (ret == 0)
4480 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
4481
4482 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004483}
4484early_param("numa_zonelist_order", setup_numa_zonelist_order);
4485
4486/*
4487 * sysctl handler for numa_zonelist_order
4488 */
Joe Perchescccad5b2014-06-06 14:38:09 -07004489int numa_zonelist_order_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004490 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004491 loff_t *ppos)
4492{
4493 char saved_string[NUMA_ZONELIST_ORDER_LEN];
4494 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01004495 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004496
Andi Kleen443c6f12009-12-23 21:00:47 +01004497 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07004498 if (write) {
4499 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
4500 ret = -EINVAL;
4501 goto out;
4502 }
4503 strcpy(saved_string, (char *)table->data);
4504 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004505 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004506 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01004507 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004508 if (write) {
4509 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07004510
4511 ret = __parse_numa_zonelist_order((char *)table->data);
4512 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004513 /*
4514 * bogus value. restore saved string
4515 */
Chen Gangdacbde02013-07-03 15:02:35 -07004516 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004517 NUMA_ZONELIST_ORDER_LEN);
4518 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004519 } else if (oldval != user_zonelist_order) {
4520 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07004521 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004522 mutex_unlock(&zonelists_mutex);
4523 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004524 }
Andi Kleen443c6f12009-12-23 21:00:47 +01004525out:
4526 mutex_unlock(&zl_order_mutex);
4527 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004528}
4529
4530
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004531#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004532static int node_load[MAX_NUMNODES];
4533
Linus Torvalds1da177e2005-04-16 15:20:36 -07004534/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07004535 * 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 -07004536 * @node: node whose fallback list we're appending
4537 * @used_node_mask: nodemask_t of already used nodes
4538 *
4539 * We use a number of factors to determine which is the next node that should
4540 * appear on a given node's fallback list. The node should not have appeared
4541 * already in @node's fallback list, and it should be the next closest node
4542 * according to the distance array (which contains arbitrary distance values
4543 * from each node to each node in the system), and should also prefer nodes
4544 * with no CPUs, since presumably they'll have very little allocation pressure
4545 * on them otherwise.
4546 * It returns -1 if no node is found.
4547 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004548static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004549{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01004550 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004551 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08004552 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10304553 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004554
Linus Torvalds4cf808e2006-02-17 20:38:21 +01004555 /* Use the local node if we haven't already */
4556 if (!node_isset(node, *used_node_mask)) {
4557 node_set(node, *used_node_mask);
4558 return node;
4559 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004560
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004561 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004562
4563 /* Don't want a node to appear more than once */
4564 if (node_isset(n, *used_node_mask))
4565 continue;
4566
Linus Torvalds1da177e2005-04-16 15:20:36 -07004567 /* Use the distance array to find the distance */
4568 val = node_distance(node, n);
4569
Linus Torvalds4cf808e2006-02-17 20:38:21 +01004570 /* Penalize nodes under us ("prefer the next node") */
4571 val += (n < node);
4572
Linus Torvalds1da177e2005-04-16 15:20:36 -07004573 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10304574 tmp = cpumask_of_node(n);
4575 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004576 val += PENALTY_FOR_NODE_WITH_CPUS;
4577
4578 /* Slight preference for less loaded node */
4579 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4580 val += node_load[n];
4581
4582 if (val < min_val) {
4583 min_val = val;
4584 best_node = n;
4585 }
4586 }
4587
4588 if (best_node >= 0)
4589 node_set(best_node, *used_node_mask);
4590
4591 return best_node;
4592}
4593
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004594
4595/*
4596 * Build zonelists ordered by node and zones within node.
4597 * This results in maximum locality--normal zone overflows into local
4598 * DMA zone, if any--but risks exhausting DMA zone.
4599 */
4600static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004601{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004602 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004603 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004604
Mel Gorman54a6eb52008-04-28 02:12:16 -07004605 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07004606 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07004607 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004608 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07004609 zonelist->_zonerefs[j].zone = NULL;
4610 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004611}
4612
4613/*
Christoph Lameter523b9452007-10-16 01:25:37 -07004614 * Build gfp_thisnode zonelists
4615 */
4616static void build_thisnode_zonelists(pg_data_t *pgdat)
4617{
Christoph Lameter523b9452007-10-16 01:25:37 -07004618 int j;
4619 struct zonelist *zonelist;
4620
Mel Gorman54a6eb52008-04-28 02:12:16 -07004621 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004622 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07004623 zonelist->_zonerefs[j].zone = NULL;
4624 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07004625}
4626
4627/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004628 * Build zonelists ordered by zone and nodes within zones.
4629 * This results in conserving DMA zone[s] until all Normal memory is
4630 * exhausted, but results in overflowing to remote node while memory
4631 * may still exist in local DMA zone.
4632 */
4633static int node_order[MAX_NUMNODES];
4634
4635static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4636{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004637 int pos, j, node;
4638 int zone_type; /* needs to be signed */
4639 struct zone *z;
4640 struct zonelist *zonelist;
4641
Mel Gorman54a6eb52008-04-28 02:12:16 -07004642 zonelist = &pgdat->node_zonelists[0];
4643 pos = 0;
4644 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4645 for (j = 0; j < nr_nodes; j++) {
4646 node = node_order[j];
4647 z = &NODE_DATA(node)->node_zones[zone_type];
4648 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07004649 zoneref_set_zone(z,
4650 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004651 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004652 }
4653 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004654 }
Mel Gormandd1a2392008-04-28 02:12:17 -07004655 zonelist->_zonerefs[pos].zone = NULL;
4656 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004657}
4658
Mel Gorman31939132014-10-09 15:28:30 -07004659#if defined(CONFIG_64BIT)
4660/*
4661 * Devices that require DMA32/DMA are relatively rare and do not justify a
4662 * penalty to every machine in case the specialised case applies. Default
4663 * to Node-ordering on 64-bit NUMA machines
4664 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004665static int default_zonelist_order(void)
4666{
Mel Gorman31939132014-10-09 15:28:30 -07004667 return ZONELIST_ORDER_NODE;
4668}
4669#else
4670/*
4671 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4672 * by the kernel. If processes running on node 0 deplete the low memory zone
4673 * then reclaim will occur more frequency increasing stalls and potentially
4674 * be easier to OOM if a large percentage of the zone is under writeback or
4675 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4676 * Hence, default to zone ordering on 32-bit.
4677 */
4678static int default_zonelist_order(void)
4679{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004680 return ZONELIST_ORDER_ZONE;
4681}
Mel Gorman31939132014-10-09 15:28:30 -07004682#endif /* CONFIG_64BIT */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004683
4684static void set_zonelist_order(void)
4685{
4686 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4687 current_zonelist_order = default_zonelist_order();
4688 else
4689 current_zonelist_order = user_zonelist_order;
4690}
4691
4692static void build_zonelists(pg_data_t *pgdat)
4693{
Yaowei Baic00eb152016-01-14 15:19:00 -08004694 int i, node, load;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004695 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004696 int local_node, prev_node;
4697 struct zonelist *zonelist;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004698 unsigned int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004699
4700 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07004701 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004702 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07004703 zonelist->_zonerefs[0].zone = NULL;
4704 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004705 }
4706
4707 /* NUMA-aware ordering of nodes */
4708 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004709 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004710 prev_node = local_node;
4711 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004712
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004713 memset(node_order, 0, sizeof(node_order));
Yaowei Baic00eb152016-01-14 15:19:00 -08004714 i = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004715
Linus Torvalds1da177e2005-04-16 15:20:36 -07004716 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4717 /*
4718 * We don't want to pressure a particular node.
4719 * So adding penalty to the first node in same
4720 * distance group to make it round-robin.
4721 */
David Rientjes957f8222012-10-08 16:33:24 -07004722 if (node_distance(local_node, node) !=
4723 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004724 node_load[node] = load;
4725
Linus Torvalds1da177e2005-04-16 15:20:36 -07004726 prev_node = node;
4727 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004728 if (order == ZONELIST_ORDER_NODE)
4729 build_zonelists_in_node_order(pgdat, node);
4730 else
Yaowei Baic00eb152016-01-14 15:19:00 -08004731 node_order[i++] = node; /* remember order */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004733
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004734 if (order == ZONELIST_ORDER_ZONE) {
4735 /* calculate node order -- i.e., DMA last! */
Yaowei Baic00eb152016-01-14 15:19:00 -08004736 build_zonelists_in_zone_order(pgdat, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004737 }
Christoph Lameter523b9452007-10-16 01:25:37 -07004738
4739 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004740}
4741
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004742#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4743/*
4744 * Return node id of node used for "local" allocations.
4745 * I.e., first node id of first zone in arg node's generic zonelist.
4746 * Used for initializing percpu 'numa_mem', which is used primarily
4747 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4748 */
4749int local_memory_node(int node)
4750{
Mel Gormanc33d6c02016-05-19 17:14:10 -07004751 struct zoneref *z;
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004752
Mel Gormanc33d6c02016-05-19 17:14:10 -07004753 z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004754 gfp_zone(GFP_KERNEL),
Mel Gormanc33d6c02016-05-19 17:14:10 -07004755 NULL);
4756 return z->zone->node;
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004757}
4758#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004759
Joonsoo Kim6423aa82016-08-10 16:27:49 -07004760static void setup_min_unmapped_ratio(void);
4761static void setup_min_slab_ratio(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004762#else /* CONFIG_NUMA */
4763
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004764static void set_zonelist_order(void)
4765{
4766 current_zonelist_order = ZONELIST_ORDER_ZONE;
4767}
4768
4769static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004770{
Christoph Lameter19655d32006-09-25 23:31:19 -07004771 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07004772 enum zone_type j;
4773 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004774
4775 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004776
Mel Gorman54a6eb52008-04-28 02:12:16 -07004777 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004778 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004779
Mel Gorman54a6eb52008-04-28 02:12:16 -07004780 /*
4781 * Now we build the zonelist so that it contains the zones
4782 * of all the other nodes.
4783 * We don't want to pressure a particular node, so when
4784 * building the zones for node N, we make sure that the
4785 * zones coming right after the local ones are those from
4786 * node N+1 (modulo N)
4787 */
4788 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4789 if (!node_online(node))
4790 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004791 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004792 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07004793 for (node = 0; node < local_node; node++) {
4794 if (!node_online(node))
4795 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004796 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004797 }
4798
Mel Gormandd1a2392008-04-28 02:12:17 -07004799 zonelist->_zonerefs[j].zone = NULL;
4800 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004801}
4802
4803#endif /* CONFIG_NUMA */
4804
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004805/*
4806 * Boot pageset table. One per cpu which is going to be used for all
4807 * zones and all nodes. The parameters will be set in such a way
4808 * that an item put on a list will immediately be handed over to
4809 * the buddy list. This is safe since pageset manipulation is done
4810 * with interrupts disabled.
4811 *
4812 * The boot_pagesets must be kept even after bootup is complete for
4813 * unused processors and/or zones. They do play a role for bootstrapping
4814 * hotplugged processors.
4815 *
4816 * zoneinfo_show() and maybe other functions do
4817 * not check if the processor is online before following the pageset pointer.
4818 * Other parts of the kernel may not check if the zone is available.
4819 */
4820static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4821static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07004822static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004823
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004824/*
4825 * Global mutex to protect against size modification of zonelists
4826 * as well as to serialize pageset setup for the new populated zone.
4827 */
4828DEFINE_MUTEX(zonelists_mutex);
4829
Rusty Russell9b1a4d32008-07-28 12:16:30 -05004830/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07004831static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004832{
Yasunori Goto68113782006-06-23 02:03:11 -07004833 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004834 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07004835 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004836
Bo Liu7f9cfb32009-08-18 14:11:19 -07004837#ifdef CONFIG_NUMA
4838 memset(node_load, 0, sizeof(node_load));
4839#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07004840
4841 if (self && !node_online(self->node_id)) {
4842 build_zonelists(self);
Jiang Liu9adb62a2012-07-31 16:43:28 -07004843 }
4844
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004845 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07004846 pg_data_t *pgdat = NODE_DATA(nid);
4847
4848 build_zonelists(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004849 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004850
4851 /*
4852 * Initialize the boot_pagesets that are going to be used
4853 * for bootstrapping processors. The real pagesets for
4854 * each zone will be allocated later when the per cpu
4855 * allocator is available.
4856 *
4857 * boot_pagesets are used also for bootstrapping offline
4858 * cpus if the system is already booted because the pagesets
4859 * are needed to initialize allocators on a specific cpu too.
4860 * F.e. the percpu allocator needs the page allocator which
4861 * needs the percpu allocator in order to allocate its pagesets
4862 * (a chicken-egg dilemma).
4863 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004864 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004865 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4866
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004867#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4868 /*
4869 * We now know the "local memory node" for each node--
4870 * i.e., the node of the first zone in the generic zonelist.
4871 * Set up numa_mem percpu variable for on-line cpus. During
4872 * boot, only the boot cpu should be on-line; we'll init the
4873 * secondary cpus' numa_mem as they come on-line. During
4874 * node/memory hotplug, we'll fixup all on-line cpus.
4875 */
4876 if (cpu_online(cpu))
4877 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4878#endif
4879 }
4880
Yasunori Goto68113782006-06-23 02:03:11 -07004881 return 0;
4882}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004883
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004884static noinline void __init
4885build_all_zonelists_init(void)
4886{
4887 __build_all_zonelists(NULL);
4888 mminit_verify_zonelist();
4889 cpuset_init_current_mems_allowed();
4890}
4891
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004892/*
4893 * Called with zonelists_mutex held always
4894 * unless system_state == SYSTEM_BOOTING.
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004895 *
4896 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4897 * [we're only called with non-NULL zone through __meminit paths] and
4898 * (2) call of __init annotated helper build_all_zonelists_init
4899 * [protected by SYSTEM_BOOTING].
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004900 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07004901void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07004902{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004903 set_zonelist_order();
4904
Yasunori Goto68113782006-06-23 02:03:11 -07004905 if (system_state == SYSTEM_BOOTING) {
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004906 build_all_zonelists_init();
Yasunori Goto68113782006-06-23 02:03:11 -07004907 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08004908#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07004909 if (zone)
4910 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08004911#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07004912 /* we have to stop all cpus to guarantee there is no user
4913 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07004914 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07004915 /* cpuset refresh routine should be here */
4916 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07004917 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004918 /*
4919 * Disable grouping by mobility if the number of pages in the
4920 * system is too low to allow the mechanism to work. It would be
4921 * more accurate, but expensive to check per-zone. This check is
4922 * made on memory-hotadd so a system can start with mobility
4923 * disabled and enable it later
4924 */
Mel Gormand9c23402007-10-16 01:26:01 -07004925 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004926 page_group_by_mobility_disabled = 1;
4927 else
4928 page_group_by_mobility_disabled = 0;
4929
Joe Perches756a025f02016-03-17 14:19:47 -07004930 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
4931 nr_online_nodes,
4932 zonelist_order_name[current_zonelist_order],
4933 page_group_by_mobility_disabled ? "off" : "on",
4934 vm_total_pages);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004935#ifdef CONFIG_NUMA
Anton Blanchardf88dfff2014-12-10 15:42:53 -08004936 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004937#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004938}
4939
4940/*
4941 * Helper functions to size the waitqueue hash table.
4942 * Essentially these want to choose hash table sizes sufficiently
4943 * large so that collisions trying to wait on pages are rare.
4944 * But in fact, the number of active page waitqueues on typical
4945 * systems is ridiculously low, less than 200. So this is even
4946 * conservative, even though it seems large.
4947 *
4948 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4949 * waitqueues, i.e. the size of the waitq table given the number of pages.
4950 */
4951#define PAGES_PER_WAITQUEUE 256
4952
Yasunori Gotocca448f2006-06-23 02:03:10 -07004953#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07004954static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004955{
4956 unsigned long size = 1;
4957
4958 pages /= PAGES_PER_WAITQUEUE;
4959
4960 while (size < pages)
4961 size <<= 1;
4962
4963 /*
4964 * Once we have dozens or even hundreds of threads sleeping
4965 * on IO we've got bigger problems than wait queue collision.
4966 * Limit the size of the wait table to a reasonable size.
4967 */
4968 size = min(size, 4096UL);
4969
4970 return max(size, 4UL);
4971}
Yasunori Gotocca448f2006-06-23 02:03:10 -07004972#else
4973/*
4974 * A zone's size might be changed by hot-add, so it is not possible to determine
4975 * a suitable size for its wait_table. So we use the maximum size now.
4976 *
4977 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4978 *
4979 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4980 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4981 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4982 *
4983 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4984 * or more by the traditional way. (See above). It equals:
4985 *
4986 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4987 * ia64(16K page size) : = ( 8G + 4M)byte.
4988 * powerpc (64K page size) : = (32G +16M)byte.
4989 */
4990static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4991{
4992 return 4096UL;
4993}
4994#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004995
4996/*
4997 * This is an integer logarithm so that shifts can be used later
4998 * to extract the more random high bits from the multiplicative
4999 * hash function before the remainder is taken.
5000 */
5001static inline unsigned long wait_table_bits(unsigned long size)
5002{
5003 return ffz(~size);
5004}
5005
Mel Gorman56fd56b2007-10-16 01:25:58 -07005006/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005007 * Initially all pages are reserved - free ones are freed
5008 * up by free_all_bootmem() once the early boot process is
5009 * done. Non-atomic initialization, single-pass.
5010 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01005011void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08005012 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005013{
Dan Williams4b94ffd2016-01-15 16:56:22 -08005014 struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn));
Andy Whitcroft29751f62005-06-23 00:08:00 -07005015 unsigned long end_pfn = start_pfn + size;
Dan Williams4b94ffd2016-01-15 16:56:22 -08005016 pg_data_t *pgdat = NODE_DATA(nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -07005017 unsigned long pfn;
Mel Gorman3a80a7f2015-06-30 14:57:02 -07005018 unsigned long nr_initialised = 0;
Taku Izumi342332e2016-03-15 14:55:22 -07005019#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5020 struct memblock_region *r = NULL, *tmp;
5021#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005022
Hugh Dickins22b31ee2009-01-06 14:40:09 -08005023 if (highest_memmap_pfn < end_pfn - 1)
5024 highest_memmap_pfn = end_pfn - 1;
5025
Dan Williams4b94ffd2016-01-15 16:56:22 -08005026 /*
5027 * Honor reservation requested by the driver for this ZONE_DEVICE
5028 * memory
5029 */
5030 if (altmap && start_pfn == altmap->base_pfn)
5031 start_pfn += altmap->reserve;
5032
Greg Ungerercbe8dd42006-01-12 01:05:24 -08005033 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08005034 /*
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005035 * There can be holes in boot-time mem_map[]s handed to this
5036 * function. They do not exist on hotplugged memory.
Dave Hansena2f3aa022007-01-10 23:15:30 -08005037 */
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005038 if (context != MEMMAP_EARLY)
5039 goto not_early;
5040
5041 if (!early_pfn_valid(pfn))
5042 continue;
5043 if (!early_pfn_in_nid(pfn, nid))
5044 continue;
5045 if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised))
5046 break;
Taku Izumi342332e2016-03-15 14:55:22 -07005047
5048#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005049 /*
5050 * If not mirrored_kernelcore and ZONE_MOVABLE exists, range
5051 * from zone_movable_pfn[nid] to end of each node should be
5052 * ZONE_MOVABLE not ZONE_NORMAL. skip it.
5053 */
5054 if (!mirrored_kernelcore && zone_movable_pfn[nid])
5055 if (zone == ZONE_NORMAL && pfn >= zone_movable_pfn[nid])
5056 continue;
Taku Izumi342332e2016-03-15 14:55:22 -07005057
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005058 /*
5059 * Check given memblock attribute by firmware which can affect
5060 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
5061 * mirrored, it's an overlapped memmap init. skip it.
5062 */
5063 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
5064 if (!r || pfn >= memblock_region_memory_end_pfn(r)) {
5065 for_each_memblock(memory, tmp)
5066 if (pfn < memblock_region_memory_end_pfn(tmp))
5067 break;
5068 r = tmp;
Taku Izumi342332e2016-03-15 14:55:22 -07005069 }
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005070 if (pfn >= memblock_region_memory_base_pfn(r) &&
5071 memblock_is_mirror(r)) {
5072 /* already initialized as NORMAL */
5073 pfn = memblock_region_memory_end_pfn(r);
5074 continue;
5075 }
Dave Hansena2f3aa022007-01-10 23:15:30 -08005076 }
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005077#endif
Mel Gormanac5d2532015-06-30 14:57:20 -07005078
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005079not_early:
Mel Gormanac5d2532015-06-30 14:57:20 -07005080 /*
5081 * Mark the block movable so that blocks are reserved for
5082 * movable at startup. This will force kernel allocations
5083 * to reserve their blocks rather than leaking throughout
5084 * the address space during boot when many long-lived
Mel Gorman974a7862015-11-06 16:28:34 -08005085 * kernel allocations are made.
Mel Gormanac5d2532015-06-30 14:57:20 -07005086 *
5087 * bitmap is created for zone's valid pfn range. but memmap
5088 * can be created for invalid pages (for alignment)
5089 * check here not to call set_pageblock_migratetype() against
5090 * pfn out of zone.
5091 */
5092 if (!(pfn & (pageblock_nr_pages - 1))) {
5093 struct page *page = pfn_to_page(pfn);
5094
5095 __init_single_page(page, pfn, zone, nid);
5096 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5097 } else {
5098 __init_single_pfn(pfn, zone, nid);
5099 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005100 }
5101}
5102
Andi Kleen1e548de2008-02-04 22:29:26 -08005103static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005104{
Mel Gorman7aeb09f2014-06-04 16:10:21 -07005105 unsigned int order, t;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07005106 for_each_migratetype_order(order, t) {
5107 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005108 zone->free_area[order].nr_free = 0;
5109 }
5110}
5111
5112#ifndef __HAVE_ARCH_MEMMAP_INIT
5113#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08005114 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005115#endif
5116
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005117static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005118{
David Howells3a6be872009-05-06 16:03:03 -07005119#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005120 int batch;
5121
5122 /*
5123 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07005124 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005125 *
5126 * OK, so we don't know how big the cache is. So guess.
5127 */
Jiang Liub40da042013-02-22 16:33:52 -08005128 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07005129 if (batch * PAGE_SIZE > 512 * 1024)
5130 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005131 batch /= 4; /* We effectively *= 4 below */
5132 if (batch < 1)
5133 batch = 1;
5134
5135 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11005136 * Clamp the batch to a 2^n - 1 value. Having a power
5137 * of 2 value was found to be more likely to have
5138 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005139 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11005140 * For example if 2 tasks are alternately allocating
5141 * batches of pages, one task can end up with a lot
5142 * of pages of one half of the possible page colors
5143 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005144 */
David Howells91552032009-05-06 16:03:02 -07005145 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07005146
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005147 return batch;
David Howells3a6be872009-05-06 16:03:03 -07005148
5149#else
5150 /* The deferral and batching of frees should be suppressed under NOMMU
5151 * conditions.
5152 *
5153 * The problem is that NOMMU needs to be able to allocate large chunks
5154 * of contiguous memory as there's no hardware page translation to
5155 * assemble apparent contiguous memory from discontiguous pages.
5156 *
5157 * Queueing large contiguous runs of pages for batching, however,
5158 * causes the pages to actually be freed in smaller chunks. As there
5159 * can be a significant delay between the individual batches being
5160 * recycled, this leads to the once large chunks of space being
5161 * fragmented and becoming unavailable for high-order allocations.
5162 */
5163 return 0;
5164#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005165}
5166
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005167/*
5168 * pcp->high and pcp->batch values are related and dependent on one another:
5169 * ->batch must never be higher then ->high.
5170 * The following function updates them in a safe manner without read side
5171 * locking.
5172 *
5173 * Any new users of pcp->batch and pcp->high should ensure they can cope with
5174 * those fields changing asynchronously (acording the the above rule).
5175 *
5176 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5177 * outside of boot time (or some other assurance that no concurrent updaters
5178 * exist).
5179 */
5180static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
5181 unsigned long batch)
5182{
5183 /* start with a fail safe value for batch */
5184 pcp->batch = 1;
5185 smp_wmb();
5186
5187 /* Update high, then batch, in order */
5188 pcp->high = high;
5189 smp_wmb();
5190
5191 pcp->batch = batch;
5192}
5193
Cody P Schafer36640332013-07-03 15:01:40 -07005194/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07005195static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
5196{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005197 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07005198}
5199
Cody P Schafer88c90db2013-07-03 15:01:35 -07005200static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07005201{
5202 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07005203 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07005204
Magnus Damm1c6fe942005-10-26 01:58:59 -07005205 memset(p, 0, sizeof(*p));
5206
Christoph Lameter3dfa5722008-02-04 22:29:19 -08005207 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07005208 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07005209 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
5210 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07005211}
5212
Cody P Schafer88c90db2013-07-03 15:01:35 -07005213static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
5214{
5215 pageset_init(p);
5216 pageset_set_batch(p, batch);
5217}
5218
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005219/*
Cody P Schafer36640332013-07-03 15:01:40 -07005220 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005221 * to the value high for the pageset p.
5222 */
Cody P Schafer36640332013-07-03 15:01:40 -07005223static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005224 unsigned long high)
5225{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005226 unsigned long batch = max(1UL, high / 4);
5227 if ((high / 4) > (PAGE_SHIFT * 8))
5228 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005229
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005230 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005231}
5232
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005233static void pageset_set_high_and_batch(struct zone *zone,
5234 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005235{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005236 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07005237 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005238 (zone->managed_pages /
5239 percpu_pagelist_fraction));
5240 else
5241 pageset_set_batch(pcp, zone_batchsize(zone));
5242}
5243
Cody P Schafer169f6c12013-07-03 15:01:41 -07005244static void __meminit zone_pageset_init(struct zone *zone, int cpu)
5245{
5246 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
5247
5248 pageset_init(pcp);
5249 pageset_set_high_and_batch(zone, pcp);
5250}
5251
Jiang Liu4ed7e022012-07-31 16:43:35 -07005252static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07005253{
5254 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07005255 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005256 for_each_possible_cpu(cpu)
5257 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07005258}
5259
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005260/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005261 * Allocate per cpu pagesets and initialize them.
5262 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07005263 */
Al Viro78d99552005-12-15 09:18:25 +00005264void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005265{
Mel Gormanb4911ea2016-08-04 15:31:49 -07005266 struct pglist_data *pgdat;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005267 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005268
Wu Fengguang319774e2010-05-24 14:32:49 -07005269 for_each_populated_zone(zone)
5270 setup_zone_pageset(zone);
Mel Gormanb4911ea2016-08-04 15:31:49 -07005271
5272 for_each_online_pgdat(pgdat)
5273 pgdat->per_cpu_nodestats =
5274 alloc_percpu(struct per_cpu_nodestat);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005275}
5276
Fabian Frederickbd721ea2016-08-02 14:03:33 -07005277static noinline __ref
Yasunori Gotocca448f2006-06-23 02:03:10 -07005278int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07005279{
5280 int i;
Yasunori Gotocca448f2006-06-23 02:03:10 -07005281 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07005282
5283 /*
5284 * The per-page waitqueue mechanism uses hashed waitqueues
5285 * per zone.
5286 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07005287 zone->wait_table_hash_nr_entries =
5288 wait_table_hash_nr_entries(zone_size_pages);
5289 zone->wait_table_bits =
5290 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07005291 alloc_size = zone->wait_table_hash_nr_entries
5292 * sizeof(wait_queue_head_t);
5293
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07005294 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07005295 zone->wait_table = (wait_queue_head_t *)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005296 memblock_virt_alloc_node_nopanic(
5297 alloc_size, zone->zone_pgdat->node_id);
Yasunori Gotocca448f2006-06-23 02:03:10 -07005298 } else {
5299 /*
5300 * This case means that a zone whose size was 0 gets new memory
5301 * via memory hot-add.
5302 * But it may be the case that a new node was hot-added. In
5303 * this case vmalloc() will not be able to use this new node's
5304 * memory - this wait_table must be initialized to use this new
5305 * node itself as well.
5306 * To use this new node's memory, further consideration will be
5307 * necessary.
5308 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07005309 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07005310 }
5311 if (!zone->wait_table)
5312 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07005313
Pintu Kumarb8af2942013-09-11 14:20:34 -07005314 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07005315 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07005316
5317 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07005318}
5319
Matt Tolentinoc09b4242006-01-17 07:03:44 +01005320static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07005321{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005322 /*
5323 * per cpu subsystem is not up at this point. The following code
5324 * relies on the ability of the linker to provide the
5325 * offset of a (static) per cpu variable into the per cpu area.
5326 */
5327 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07005328
Xishi Qiub38a8722013-11-12 15:07:20 -08005329 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005330 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
5331 zone->name, zone->present_pages,
5332 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07005333}
5334
Jiang Liu4ed7e022012-07-31 16:43:35 -07005335int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07005336 unsigned long zone_start_pfn,
Yaowei Baib171e402015-11-05 18:47:06 -08005337 unsigned long size)
Dave Hansened8ece22005-10-29 18:16:50 -07005338{
5339 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07005340 int ret;
5341 ret = zone_wait_table_init(zone, size);
5342 if (ret)
5343 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07005344 pgdat->nr_zones = zone_idx(zone) + 1;
5345
Dave Hansened8ece22005-10-29 18:16:50 -07005346 zone->zone_start_pfn = zone_start_pfn;
5347
Mel Gorman708614e2008-07-23 21:26:51 -07005348 mminit_dprintk(MMINIT_TRACE, "memmap_init",
5349 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5350 pgdat->node_id,
5351 (unsigned long)zone_idx(zone),
5352 zone_start_pfn, (zone_start_pfn + size));
5353
Andi Kleen1e548de2008-02-04 22:29:26 -08005354 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07005355
5356 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07005357}
5358
Tejun Heo0ee332c2011-12-08 10:22:09 -08005359#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07005360#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
Mel Gorman8a942fd2015-06-30 14:56:55 -07005361
Mel Gormanc7132162006-09-27 01:49:43 -07005362/*
5363 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
Mel Gormanc7132162006-09-27 01:49:43 -07005364 */
Mel Gorman8a942fd2015-06-30 14:56:55 -07005365int __meminit __early_pfn_to_nid(unsigned long pfn,
5366 struct mminit_pfnnid_cache *state)
Mel Gormanc7132162006-09-27 01:49:43 -07005367{
Tejun Heoc13291a2011-07-12 10:46:30 +02005368 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07005369 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07005370
Mel Gorman8a942fd2015-06-30 14:56:55 -07005371 if (state->last_start <= pfn && pfn < state->last_end)
5372 return state->last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005373
Yinghai Lue76b63f2013-09-11 14:22:17 -07005374 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
5375 if (nid != -1) {
Mel Gorman8a942fd2015-06-30 14:56:55 -07005376 state->last_start = start_pfn;
5377 state->last_end = end_pfn;
5378 state->last_nid = nid;
Yinghai Lue76b63f2013-09-11 14:22:17 -07005379 }
5380
5381 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005382}
5383#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5384
Mel Gormanc7132162006-09-27 01:49:43 -07005385/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08005386 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005387 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08005388 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07005389 *
Zhang Zhen7d018172014-06-04 16:10:53 -07005390 * If an architecture guarantees that all ranges registered contain no holes
5391 * and may be freed, this this function may be used instead of calling
5392 * memblock_free_early_nid() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07005393 */
Tejun Heoc13291a2011-07-12 10:46:30 +02005394void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07005395{
Tejun Heoc13291a2011-07-12 10:46:30 +02005396 unsigned long start_pfn, end_pfn;
5397 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005398
Tejun Heoc13291a2011-07-12 10:46:30 +02005399 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
5400 start_pfn = min(start_pfn, max_low_pfn);
5401 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005402
Tejun Heoc13291a2011-07-12 10:46:30 +02005403 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005404 memblock_free_early_nid(PFN_PHYS(start_pfn),
5405 (end_pfn - start_pfn) << PAGE_SHIFT,
5406 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005407 }
5408}
5409
5410/**
5411 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005412 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07005413 *
Zhang Zhen7d018172014-06-04 16:10:53 -07005414 * If an architecture guarantees that all ranges registered contain no holes and may
5415 * be freed, this function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07005416 */
5417void __init sparse_memory_present_with_active_regions(int nid)
5418{
Tejun Heoc13291a2011-07-12 10:46:30 +02005419 unsigned long start_pfn, end_pfn;
5420 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005421
Tejun Heoc13291a2011-07-12 10:46:30 +02005422 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
5423 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005424}
5425
5426/**
5427 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005428 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5429 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5430 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07005431 *
5432 * It returns the start and end page frame of a node based on information
Zhang Zhen7d018172014-06-04 16:10:53 -07005433 * provided by memblock_set_node(). If called for a node
Mel Gormanc7132162006-09-27 01:49:43 -07005434 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005435 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07005436 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005437void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005438 unsigned long *start_pfn, unsigned long *end_pfn)
5439{
Tejun Heoc13291a2011-07-12 10:46:30 +02005440 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005441 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02005442
Mel Gormanc7132162006-09-27 01:49:43 -07005443 *start_pfn = -1UL;
5444 *end_pfn = 0;
5445
Tejun Heoc13291a2011-07-12 10:46:30 +02005446 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5447 *start_pfn = min(*start_pfn, this_start_pfn);
5448 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005449 }
5450
Christoph Lameter633c0662007-10-16 01:25:37 -07005451 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07005452 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005453}
5454
5455/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07005456 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5457 * assumption is made that zones within a node are ordered in monotonic
5458 * increasing memory addresses so that the "highest" populated zone is used
5459 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005460static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005461{
5462 int zone_index;
5463 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5464 if (zone_index == ZONE_MOVABLE)
5465 continue;
5466
5467 if (arch_zone_highest_possible_pfn[zone_index] >
5468 arch_zone_lowest_possible_pfn[zone_index])
5469 break;
5470 }
5471
5472 VM_BUG_ON(zone_index == -1);
5473 movable_zone = zone_index;
5474}
5475
5476/*
5477 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03005478 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005479 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5480 * in each node depending on the size of each node and how evenly kernelcore
5481 * is distributed. This helper function adjusts the zone ranges
5482 * provided by the architecture for a given node by using the end of the
5483 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5484 * zones within a node are in order of monotonic increases memory addresses
5485 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005486static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005487 unsigned long zone_type,
5488 unsigned long node_start_pfn,
5489 unsigned long node_end_pfn,
5490 unsigned long *zone_start_pfn,
5491 unsigned long *zone_end_pfn)
5492{
5493 /* Only adjust if ZONE_MOVABLE is on this node */
5494 if (zone_movable_pfn[nid]) {
5495 /* Size ZONE_MOVABLE */
5496 if (zone_type == ZONE_MOVABLE) {
5497 *zone_start_pfn = zone_movable_pfn[nid];
5498 *zone_end_pfn = min(node_end_pfn,
5499 arch_zone_highest_possible_pfn[movable_zone]);
5500
Mel Gorman2a1e2742007-07-17 04:03:12 -07005501 /* Check if this whole range is within ZONE_MOVABLE */
5502 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5503 *zone_start_pfn = *zone_end_pfn;
5504 }
5505}
5506
5507/*
Mel Gormanc7132162006-09-27 01:49:43 -07005508 * Return the number of pages a zone spans in a node, including holes
5509 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5510 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005511static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005512 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005513 unsigned long node_start_pfn,
5514 unsigned long node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005515 unsigned long *zone_start_pfn,
5516 unsigned long *zone_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005517 unsigned long *ignored)
5518{
Xishi Qiub5685e92015-09-08 15:04:16 -07005519 /* When hotadd a new node from cpu_up(), the node should be empty */
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005520 if (!node_start_pfn && !node_end_pfn)
5521 return 0;
5522
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005523 /* Get the start and end of the zone */
Taku Izumid91749c2016-03-15 14:55:18 -07005524 *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5525 *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07005526 adjust_zone_range_for_zone_movable(nid, zone_type,
5527 node_start_pfn, node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005528 zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005529
5530 /* Check that this node has pages within the zone's required range */
Taku Izumid91749c2016-03-15 14:55:18 -07005531 if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07005532 return 0;
5533
5534 /* Move the zone boundaries inside the node if necessary */
Taku Izumid91749c2016-03-15 14:55:18 -07005535 *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
5536 *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005537
5538 /* Return the spanned pages */
Taku Izumid91749c2016-03-15 14:55:18 -07005539 return *zone_end_pfn - *zone_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005540}
5541
5542/*
5543 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005544 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07005545 */
Yinghai Lu32996252009-12-15 17:59:02 -08005546unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005547 unsigned long range_start_pfn,
5548 unsigned long range_end_pfn)
5549{
Tejun Heo96e907d2011-07-12 10:46:29 +02005550 unsigned long nr_absent = range_end_pfn - range_start_pfn;
5551 unsigned long start_pfn, end_pfn;
5552 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07005553
Tejun Heo96e907d2011-07-12 10:46:29 +02005554 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5555 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5556 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5557 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005558 }
Tejun Heo96e907d2011-07-12 10:46:29 +02005559 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07005560}
5561
5562/**
5563 * absent_pages_in_range - Return number of page frames in holes within a range
5564 * @start_pfn: The start PFN to start searching for holes
5565 * @end_pfn: The end PFN to stop searching for holes
5566 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005567 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07005568 */
5569unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5570 unsigned long end_pfn)
5571{
5572 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5573}
5574
5575/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005576static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005577 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005578 unsigned long node_start_pfn,
5579 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005580 unsigned long *ignored)
5581{
Tejun Heo96e907d2011-07-12 10:46:29 +02005582 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5583 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07005584 unsigned long zone_start_pfn, zone_end_pfn;
Taku Izumi342332e2016-03-15 14:55:22 -07005585 unsigned long nr_absent;
Mel Gorman9c7cd682006-09-27 01:49:58 -07005586
Xishi Qiub5685e92015-09-08 15:04:16 -07005587 /* When hotadd a new node from cpu_up(), the node should be empty */
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005588 if (!node_start_pfn && !node_end_pfn)
5589 return 0;
5590
Tejun Heo96e907d2011-07-12 10:46:29 +02005591 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5592 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07005593
Mel Gorman2a1e2742007-07-17 04:03:12 -07005594 adjust_zone_range_for_zone_movable(nid, zone_type,
5595 node_start_pfn, node_end_pfn,
5596 &zone_start_pfn, &zone_end_pfn);
Taku Izumi342332e2016-03-15 14:55:22 -07005597 nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5598
5599 /*
5600 * ZONE_MOVABLE handling.
5601 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5602 * and vice versa.
5603 */
5604 if (zone_movable_pfn[nid]) {
5605 if (mirrored_kernelcore) {
5606 unsigned long start_pfn, end_pfn;
5607 struct memblock_region *r;
5608
5609 for_each_memblock(memory, r) {
5610 start_pfn = clamp(memblock_region_memory_base_pfn(r),
5611 zone_start_pfn, zone_end_pfn);
5612 end_pfn = clamp(memblock_region_memory_end_pfn(r),
5613 zone_start_pfn, zone_end_pfn);
5614
5615 if (zone_type == ZONE_MOVABLE &&
5616 memblock_is_mirror(r))
5617 nr_absent += end_pfn - start_pfn;
5618
5619 if (zone_type == ZONE_NORMAL &&
5620 !memblock_is_mirror(r))
5621 nr_absent += end_pfn - start_pfn;
5622 }
5623 } else {
5624 if (zone_type == ZONE_NORMAL)
5625 nr_absent += node_end_pfn - zone_movable_pfn[nid];
5626 }
5627 }
5628
5629 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07005630}
Mel Gorman0e0b8642006-09-27 01:49:56 -07005631
Tejun Heo0ee332c2011-12-08 10:22:09 -08005632#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005633static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005634 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005635 unsigned long node_start_pfn,
5636 unsigned long node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005637 unsigned long *zone_start_pfn,
5638 unsigned long *zone_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005639 unsigned long *zones_size)
5640{
Taku Izumid91749c2016-03-15 14:55:18 -07005641 unsigned int zone;
5642
5643 *zone_start_pfn = node_start_pfn;
5644 for (zone = 0; zone < zone_type; zone++)
5645 *zone_start_pfn += zones_size[zone];
5646
5647 *zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
5648
Mel Gormanc7132162006-09-27 01:49:43 -07005649 return zones_size[zone_type];
5650}
5651
Paul Mundt6ea6e682007-07-15 23:38:20 -07005652static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005653 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005654 unsigned long node_start_pfn,
5655 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005656 unsigned long *zholes_size)
5657{
5658 if (!zholes_size)
5659 return 0;
5660
5661 return zholes_size[zone_type];
5662}
Yinghai Lu20e69262013-03-01 14:51:27 -08005663
Tejun Heo0ee332c2011-12-08 10:22:09 -08005664#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005665
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005666static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005667 unsigned long node_start_pfn,
5668 unsigned long node_end_pfn,
5669 unsigned long *zones_size,
5670 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07005671{
Gu Zhengfebd5942015-06-24 16:57:02 -07005672 unsigned long realtotalpages = 0, totalpages = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005673 enum zone_type i;
5674
Gu Zhengfebd5942015-06-24 16:57:02 -07005675 for (i = 0; i < MAX_NR_ZONES; i++) {
5676 struct zone *zone = pgdat->node_zones + i;
Taku Izumid91749c2016-03-15 14:55:18 -07005677 unsigned long zone_start_pfn, zone_end_pfn;
Gu Zhengfebd5942015-06-24 16:57:02 -07005678 unsigned long size, real_size;
Mel Gormanc7132162006-09-27 01:49:43 -07005679
Gu Zhengfebd5942015-06-24 16:57:02 -07005680 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5681 node_start_pfn,
5682 node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005683 &zone_start_pfn,
5684 &zone_end_pfn,
Gu Zhengfebd5942015-06-24 16:57:02 -07005685 zones_size);
5686 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005687 node_start_pfn, node_end_pfn,
5688 zholes_size);
Taku Izumid91749c2016-03-15 14:55:18 -07005689 if (size)
5690 zone->zone_start_pfn = zone_start_pfn;
5691 else
5692 zone->zone_start_pfn = 0;
Gu Zhengfebd5942015-06-24 16:57:02 -07005693 zone->spanned_pages = size;
5694 zone->present_pages = real_size;
5695
5696 totalpages += size;
5697 realtotalpages += real_size;
5698 }
5699
5700 pgdat->node_spanned_pages = totalpages;
Mel Gormanc7132162006-09-27 01:49:43 -07005701 pgdat->node_present_pages = realtotalpages;
5702 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5703 realtotalpages);
5704}
5705
Mel Gorman835c1342007-10-16 01:25:47 -07005706#ifndef CONFIG_SPARSEMEM
5707/*
5708 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07005709 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5710 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07005711 * round what is now in bits to nearest long in bits, then return it in
5712 * bytes.
5713 */
Linus Torvalds7c455122013-02-18 09:58:02 -08005714static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005715{
5716 unsigned long usemapsize;
5717
Linus Torvalds7c455122013-02-18 09:58:02 -08005718 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005719 usemapsize = roundup(zonesize, pageblock_nr_pages);
5720 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07005721 usemapsize *= NR_PAGEBLOCK_BITS;
5722 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5723
5724 return usemapsize / 8;
5725}
5726
5727static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08005728 struct zone *zone,
5729 unsigned long zone_start_pfn,
5730 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005731{
Linus Torvalds7c455122013-02-18 09:58:02 -08005732 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07005733 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08005734 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005735 zone->pageblock_flags =
5736 memblock_virt_alloc_node_nopanic(usemapsize,
5737 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07005738}
5739#else
Linus Torvalds7c455122013-02-18 09:58:02 -08005740static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5741 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07005742#endif /* CONFIG_SPARSEMEM */
5743
Mel Gormand9c23402007-10-16 01:26:01 -07005744#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08005745
Mel Gormand9c23402007-10-16 01:26:01 -07005746/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07005747void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07005748{
Andrew Morton955c1cd2012-05-29 15:06:31 -07005749 unsigned int order;
5750
Mel Gormand9c23402007-10-16 01:26:01 -07005751 /* Check that pageblock_nr_pages has not already been setup */
5752 if (pageblock_order)
5753 return;
5754
Andrew Morton955c1cd2012-05-29 15:06:31 -07005755 if (HPAGE_SHIFT > PAGE_SHIFT)
5756 order = HUGETLB_PAGE_ORDER;
5757 else
5758 order = MAX_ORDER - 1;
5759
Mel Gormand9c23402007-10-16 01:26:01 -07005760 /*
5761 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07005762 * This value may be variable depending on boot parameters on IA64 and
5763 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07005764 */
5765 pageblock_order = order;
5766}
5767#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5768
Mel Gormanba72cb82007-11-28 16:21:13 -08005769/*
5770 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07005771 * is unused as pageblock_order is set at compile-time. See
5772 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5773 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08005774 */
Chen Gang15ca2202013-09-11 14:20:27 -07005775void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08005776{
Mel Gormanba72cb82007-11-28 16:21:13 -08005777}
Mel Gormand9c23402007-10-16 01:26:01 -07005778
5779#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5780
Jiang Liu01cefae2012-12-12 13:52:19 -08005781static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5782 unsigned long present_pages)
5783{
5784 unsigned long pages = spanned_pages;
5785
5786 /*
5787 * Provide a more accurate estimation if there are holes within
5788 * the zone and SPARSEMEM is in use. If there are holes within the
5789 * zone, each populated memory region may cost us one or two extra
5790 * memmap pages due to alignment because memmap pages for each
5791 * populated regions may not naturally algined on page boundary.
5792 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5793 */
5794 if (spanned_pages > present_pages + (present_pages >> 4) &&
5795 IS_ENABLED(CONFIG_SPARSEMEM))
5796 pages = present_pages;
5797
5798 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5799}
5800
Linus Torvalds1da177e2005-04-16 15:20:36 -07005801/*
5802 * Set up the zone data structures:
5803 * - mark all pages reserved
5804 * - mark all memory queues empty
5805 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07005806 *
5807 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005808 */
Wei Yang7f3eb552015-09-08 14:59:50 -07005809static void __paginginit free_area_init_core(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005810{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07005811 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07005812 int nid = pgdat->node_id;
Yasunori Goto718127c2006-06-23 02:03:10 -07005813 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005814
Dave Hansen208d54e2005-10-29 18:16:52 -07005815 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01005816#ifdef CONFIG_NUMA_BALANCING
5817 spin_lock_init(&pgdat->numabalancing_migrate_lock);
5818 pgdat->numabalancing_migrate_nr_pages = 0;
5819 pgdat->numabalancing_migrate_next_window = jiffies;
5820#endif
Kirill A. Shutemova3d0a9182016-02-02 16:57:08 -08005821#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5822 spin_lock_init(&pgdat->split_queue_lock);
5823 INIT_LIST_HEAD(&pgdat->split_queue);
5824 pgdat->split_queue_len = 0;
5825#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005826 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07005827 init_waitqueue_head(&pgdat->pfmemalloc_wait);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07005828#ifdef CONFIG_COMPACTION
5829 init_waitqueue_head(&pgdat->kcompactd_wait);
5830#endif
Joonsoo Kimeefa8642014-12-12 16:55:46 -08005831 pgdat_page_ext_init(pgdat);
Mel Gormana52633d2016-07-28 15:45:28 -07005832 spin_lock_init(&pgdat->lru_lock);
Mel Gormana9dd0a82016-07-28 15:46:02 -07005833 lruvec_init(node_lruvec(pgdat));
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01005834
Linus Torvalds1da177e2005-04-16 15:20:36 -07005835 for (j = 0; j < MAX_NR_ZONES; j++) {
5836 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08005837 unsigned long size, realsize, freesize, memmap_pages;
Taku Izumid91749c2016-03-15 14:55:18 -07005838 unsigned long zone_start_pfn = zone->zone_start_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005839
Gu Zhengfebd5942015-06-24 16:57:02 -07005840 size = zone->spanned_pages;
5841 realsize = freesize = zone->present_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005842
Mel Gorman0e0b8642006-09-27 01:49:56 -07005843 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08005844 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07005845 * is used by this zone for memmap. This affects the watermark
5846 * and per-cpu initialisations
5847 */
Jiang Liu01cefae2012-12-12 13:52:19 -08005848 memmap_pages = calc_memmap_size(size, realsize);
Zhong Hongboba914f42014-12-12 16:56:21 -08005849 if (!is_highmem_idx(j)) {
5850 if (freesize >= memmap_pages) {
5851 freesize -= memmap_pages;
5852 if (memmap_pages)
5853 printk(KERN_DEBUG
5854 " %s zone: %lu pages used for memmap\n",
5855 zone_names[j], memmap_pages);
5856 } else
Joe Perches11705322016-03-17 14:19:50 -07005857 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
Zhong Hongboba914f42014-12-12 16:56:21 -08005858 zone_names[j], memmap_pages, freesize);
5859 }
Mel Gorman0e0b8642006-09-27 01:49:56 -07005860
Christoph Lameter62672762007-02-10 01:43:07 -08005861 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08005862 if (j == 0 && freesize > dma_reserve) {
5863 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07005864 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08005865 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07005866 }
5867
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07005868 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08005869 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08005870 /* Charge for highmem memmap if there are enough kernel pages */
5871 else if (nr_kernel_pages > memmap_pages * 2)
5872 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08005873 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005874
Jiang Liu9feedc92012-12-12 13:52:12 -08005875 /*
5876 * Set an approximate value for lowmem here, it will be adjusted
5877 * when the bootmem allocator frees pages into the buddy system.
5878 * And all highmem pages will be managed by the buddy system.
5879 */
5880 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07005881#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07005882 zone->node = nid;
Christoph Lameter96146342006-07-03 00:24:13 -07005883#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005884 zone->name = zone_names[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -07005885 zone->zone_pgdat = pgdat;
Mel Gormana52633d2016-07-28 15:45:28 -07005886 spin_lock_init(&zone->lock);
5887 zone_seqlock_init(zone);
Dave Hansened8ece22005-10-29 18:16:50 -07005888 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005889
Linus Torvalds1da177e2005-04-16 15:20:36 -07005890 if (!size)
5891 continue;
5892
Andrew Morton955c1cd2012-05-29 15:06:31 -07005893 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08005894 setup_usemap(pgdat, zone, zone_start_pfn, size);
Yaowei Baib171e402015-11-05 18:47:06 -08005895 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
Yasunori Goto718127c2006-06-23 02:03:10 -07005896 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07005897 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005898 }
5899}
5900
Fabian Frederickbd721ea2016-08-02 14:03:33 -07005901static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005902{
Tony Luckb0aeba72015-11-10 10:09:47 -08005903 unsigned long __maybe_unused start = 0;
Laura Abbotta1c34a32015-11-05 18:48:46 -08005904 unsigned long __maybe_unused offset = 0;
5905
Linus Torvalds1da177e2005-04-16 15:20:36 -07005906 /* Skip empty nodes */
5907 if (!pgdat->node_spanned_pages)
5908 return;
5909
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005910#ifdef CONFIG_FLAT_NODE_MEM_MAP
Tony Luckb0aeba72015-11-10 10:09:47 -08005911 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
5912 offset = pgdat->node_start_pfn - start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005913 /* ia64 gets its own node_mem_map, before this, without bootmem */
5914 if (!pgdat->node_mem_map) {
Tony Luckb0aeba72015-11-10 10:09:47 -08005915 unsigned long size, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005916 struct page *map;
5917
Bob Piccoe984bb42006-05-20 15:00:31 -07005918 /*
5919 * The zone's endpoints aren't required to be MAX_ORDER
5920 * aligned but the node_mem_map endpoints must be in order
5921 * for the buddy allocator to function correctly.
5922 */
Cody P Schafer108bcc92013-02-22 16:35:23 -08005923 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07005924 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5925 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07005926 map = alloc_remap(pgdat->node_id, size);
5927 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005928 map = memblock_virt_alloc_node_nopanic(size,
5929 pgdat->node_id);
Laura Abbotta1c34a32015-11-05 18:48:46 -08005930 pgdat->node_mem_map = map + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005931 }
Roman Zippel12d810c2007-05-31 00:40:54 -07005932#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07005933 /*
5934 * With no DISCONTIG, the global mem_map is just set as node 0's
5935 */
Mel Gormanc7132162006-09-27 01:49:43 -07005936 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005937 mem_map = NODE_DATA(0)->node_mem_map;
Laura Abbotta1c34a32015-11-05 18:48:46 -08005938#if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
Mel Gormanc7132162006-09-27 01:49:43 -07005939 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Laura Abbotta1c34a32015-11-05 18:48:46 -08005940 mem_map -= offset;
Tejun Heo0ee332c2011-12-08 10:22:09 -08005941#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005942 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005943#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005944#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005945}
5946
Johannes Weiner9109fb72008-07-23 21:27:20 -07005947void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5948 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005949{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005950 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005951 unsigned long start_pfn = 0;
5952 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07005953
Minchan Kim88fdf752012-07-31 16:46:14 -07005954 /* pg_data_t should be reset to zero when it's allocated */
Mel Gorman38087d92016-07-28 15:45:49 -07005955 WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07005956
Mel Gorman3a80a7f2015-06-30 14:57:02 -07005957 reset_deferred_meminit(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005958 pgdat->node_id = nid;
5959 pgdat->node_start_pfn = node_start_pfn;
Mel Gorman75ef7182016-07-28 15:45:24 -07005960 pgdat->per_cpu_nodestats = NULL;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005961#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5962 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
Juergen Gross8d29e182015-02-11 15:26:01 -08005963 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
Zhen Lei4ada0c52015-09-08 15:04:19 -07005964 (u64)start_pfn << PAGE_SHIFT,
5965 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
Taku Izumid91749c2016-03-15 14:55:18 -07005966#else
5967 start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005968#endif
5969 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5970 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005971
5972 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07005973#ifdef CONFIG_FLAT_NODE_MEM_MAP
5974 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5975 nid, (unsigned long)pgdat,
5976 (unsigned long)pgdat->node_mem_map);
5977#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005978
Wei Yang7f3eb552015-09-08 14:59:50 -07005979 free_area_init_core(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005980}
5981
Tejun Heo0ee332c2011-12-08 10:22:09 -08005982#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07005983
5984#if MAX_NUMNODES > 1
5985/*
5986 * Figure out the number of possible node ids.
5987 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07005988void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07005989{
Wei Yang904a9552015-09-08 14:59:48 -07005990 unsigned int highest;
Miklos Szeredi418508c2007-05-23 13:57:55 -07005991
Wei Yang904a9552015-09-08 14:59:48 -07005992 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
Miklos Szeredi418508c2007-05-23 13:57:55 -07005993 nr_node_ids = highest + 1;
5994}
Miklos Szeredi418508c2007-05-23 13:57:55 -07005995#endif
5996
Mel Gormanc7132162006-09-27 01:49:43 -07005997/**
Tejun Heo1e019792011-07-12 09:45:34 +02005998 * node_map_pfn_alignment - determine the maximum internode alignment
5999 *
6000 * This function should be called after node map is populated and sorted.
6001 * It calculates the maximum power of two alignment which can distinguish
6002 * all the nodes.
6003 *
6004 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6005 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
6006 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
6007 * shifted, 1GiB is enough and this function will indicate so.
6008 *
6009 * This is used to test whether pfn -> nid mapping of the chosen memory
6010 * model has fine enough granularity to avoid incorrect mapping for the
6011 * populated node map.
6012 *
6013 * Returns the determined alignment in pfn's. 0 if there is no alignment
6014 * requirement (single node).
6015 */
6016unsigned long __init node_map_pfn_alignment(void)
6017{
6018 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02006019 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02006020 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02006021 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02006022
Tejun Heoc13291a2011-07-12 10:46:30 +02006023 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02006024 if (!start || last_nid < 0 || last_nid == nid) {
6025 last_nid = nid;
6026 last_end = end;
6027 continue;
6028 }
6029
6030 /*
6031 * Start with a mask granular enough to pin-point to the
6032 * start pfn and tick off bits one-by-one until it becomes
6033 * too coarse to separate the current node from the last.
6034 */
6035 mask = ~((1 << __ffs(start)) - 1);
6036 while (mask && last_end <= (start & (mask << 1)))
6037 mask <<= 1;
6038
6039 /* accumulate all internode masks */
6040 accl_mask |= mask;
6041 }
6042
6043 /* convert mask to number of pages */
6044 return ~accl_mask + 1;
6045}
6046
Mel Gormana6af2bc2007-02-10 01:42:57 -08006047/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07006048static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07006049{
Mel Gormana6af2bc2007-02-10 01:42:57 -08006050 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02006051 unsigned long start_pfn;
6052 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00006053
Tejun Heoc13291a2011-07-12 10:46:30 +02006054 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
6055 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07006056
Mel Gormana6af2bc2007-02-10 01:42:57 -08006057 if (min_pfn == ULONG_MAX) {
Joe Perches11705322016-03-17 14:19:50 -07006058 pr_warn("Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08006059 return 0;
6060 }
6061
6062 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07006063}
6064
6065/**
6066 * find_min_pfn_with_active_regions - Find the minimum PFN registered
6067 *
6068 * It returns the minimum PFN based on information provided via
Zhang Zhen7d018172014-06-04 16:10:53 -07006069 * memblock_set_node().
Mel Gormanc7132162006-09-27 01:49:43 -07006070 */
6071unsigned long __init find_min_pfn_with_active_regions(void)
6072{
6073 return find_min_pfn_for_node(MAX_NUMNODES);
6074}
6075
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006076/*
6077 * early_calculate_totalpages()
6078 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006079 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006080 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07006081static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07006082{
Mel Gorman7e63efe2007-07-17 04:03:15 -07006083 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02006084 unsigned long start_pfn, end_pfn;
6085 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07006086
Tejun Heoc13291a2011-07-12 10:46:30 +02006087 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
6088 unsigned long pages = end_pfn - start_pfn;
6089
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006090 totalpages += pages;
6091 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006092 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006093 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07006094 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07006095}
6096
Mel Gorman2a1e2742007-07-17 04:03:12 -07006097/*
6098 * Find the PFN the Movable zone begins in each node. Kernel memory
6099 * is spread evenly between nodes as long as the nodes have enough
6100 * memory. When they don't, some nodes will have more kernelcore than
6101 * others
6102 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07006103static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07006104{
6105 int i, nid;
6106 unsigned long usable_startpfn;
6107 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07006108 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006109 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006110 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006111 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07006112 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006113
6114 /* Need to find movable_zone earlier when movable_node is specified. */
6115 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07006116
Mel Gorman7e63efe2007-07-17 04:03:15 -07006117 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006118 * If movable_node is specified, ignore kernelcore and movablecore
6119 * options.
6120 */
6121 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07006122 for_each_memblock(memory, r) {
6123 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006124 continue;
6125
Emil Medve136199f2014-04-07 15:37:52 -07006126 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006127
Emil Medve136199f2014-04-07 15:37:52 -07006128 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006129 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6130 min(usable_startpfn, zone_movable_pfn[nid]) :
6131 usable_startpfn;
6132 }
6133
6134 goto out2;
6135 }
6136
6137 /*
Taku Izumi342332e2016-03-15 14:55:22 -07006138 * If kernelcore=mirror is specified, ignore movablecore option
6139 */
6140 if (mirrored_kernelcore) {
6141 bool mem_below_4gb_not_mirrored = false;
6142
6143 for_each_memblock(memory, r) {
6144 if (memblock_is_mirror(r))
6145 continue;
6146
6147 nid = r->nid;
6148
6149 usable_startpfn = memblock_region_memory_base_pfn(r);
6150
6151 if (usable_startpfn < 0x100000) {
6152 mem_below_4gb_not_mirrored = true;
6153 continue;
6154 }
6155
6156 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6157 min(usable_startpfn, zone_movable_pfn[nid]) :
6158 usable_startpfn;
6159 }
6160
6161 if (mem_below_4gb_not_mirrored)
6162 pr_warn("This configuration results in unmirrored kernel memory.");
6163
6164 goto out2;
6165 }
6166
6167 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006168 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07006169 * kernelcore that corresponds so that memory usable for
6170 * any allocation type is evenly spread. If both kernelcore
6171 * and movablecore are specified, then the value of kernelcore
6172 * will be used for required_kernelcore if it's greater than
6173 * what movablecore would have allowed.
6174 */
6175 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07006176 unsigned long corepages;
6177
6178 /*
6179 * Round-up so that ZONE_MOVABLE is at least as large as what
6180 * was requested by the user
6181 */
6182 required_movablecore =
6183 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
Xishi Qiu9fd745d2015-11-05 18:48:11 -08006184 required_movablecore = min(totalpages, required_movablecore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07006185 corepages = totalpages - required_movablecore;
6186
6187 required_kernelcore = max(required_kernelcore, corepages);
6188 }
6189
Xishi Qiubde304b2015-11-05 18:48:56 -08006190 /*
6191 * If kernelcore was not specified or kernelcore size is larger
6192 * than totalpages, there is no ZONE_MOVABLE.
6193 */
6194 if (!required_kernelcore || required_kernelcore >= totalpages)
Yinghai Lu66918dc2009-06-30 11:41:37 -07006195 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07006196
6197 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07006198 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
6199
6200restart:
6201 /* Spread kernelcore memory as evenly as possible throughout nodes */
6202 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006203 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02006204 unsigned long start_pfn, end_pfn;
6205
Mel Gorman2a1e2742007-07-17 04:03:12 -07006206 /*
6207 * Recalculate kernelcore_node if the division per node
6208 * now exceeds what is necessary to satisfy the requested
6209 * amount of memory for the kernel
6210 */
6211 if (required_kernelcore < kernelcore_node)
6212 kernelcore_node = required_kernelcore / usable_nodes;
6213
6214 /*
6215 * As the map is walked, we track how much memory is usable
6216 * by the kernel using kernelcore_remaining. When it is
6217 * 0, the rest of the node is usable by ZONE_MOVABLE
6218 */
6219 kernelcore_remaining = kernelcore_node;
6220
6221 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02006222 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07006223 unsigned long size_pages;
6224
Tejun Heoc13291a2011-07-12 10:46:30 +02006225 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07006226 if (start_pfn >= end_pfn)
6227 continue;
6228
6229 /* Account for what is only usable for kernelcore */
6230 if (start_pfn < usable_startpfn) {
6231 unsigned long kernel_pages;
6232 kernel_pages = min(end_pfn, usable_startpfn)
6233 - start_pfn;
6234
6235 kernelcore_remaining -= min(kernel_pages,
6236 kernelcore_remaining);
6237 required_kernelcore -= min(kernel_pages,
6238 required_kernelcore);
6239
6240 /* Continue if range is now fully accounted */
6241 if (end_pfn <= usable_startpfn) {
6242
6243 /*
6244 * Push zone_movable_pfn to the end so
6245 * that if we have to rebalance
6246 * kernelcore across nodes, we will
6247 * not double account here
6248 */
6249 zone_movable_pfn[nid] = end_pfn;
6250 continue;
6251 }
6252 start_pfn = usable_startpfn;
6253 }
6254
6255 /*
6256 * The usable PFN range for ZONE_MOVABLE is from
6257 * start_pfn->end_pfn. Calculate size_pages as the
6258 * number of pages used as kernelcore
6259 */
6260 size_pages = end_pfn - start_pfn;
6261 if (size_pages > kernelcore_remaining)
6262 size_pages = kernelcore_remaining;
6263 zone_movable_pfn[nid] = start_pfn + size_pages;
6264
6265 /*
6266 * Some kernelcore has been met, update counts and
6267 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07006268 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07006269 */
6270 required_kernelcore -= min(required_kernelcore,
6271 size_pages);
6272 kernelcore_remaining -= size_pages;
6273 if (!kernelcore_remaining)
6274 break;
6275 }
6276 }
6277
6278 /*
6279 * If there is still required_kernelcore, we do another pass with one
6280 * less node in the count. This will push zone_movable_pfn[nid] further
6281 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07006282 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07006283 */
6284 usable_nodes--;
6285 if (usable_nodes && required_kernelcore > usable_nodes)
6286 goto restart;
6287
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006288out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07006289 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6290 for (nid = 0; nid < MAX_NUMNODES; nid++)
6291 zone_movable_pfn[nid] =
6292 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07006293
Yinghai Lu20e69262013-03-01 14:51:27 -08006294out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07006295 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006296 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07006297}
6298
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006299/* Any regular or high memory on that node ? */
6300static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006301{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006302 enum zone_type zone_type;
6303
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006304 if (N_MEMORY == N_NORMAL_MEMORY)
6305 return;
6306
6307 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006308 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08006309 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006310 node_set_state(nid, N_HIGH_MEMORY);
6311 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
6312 zone_type <= ZONE_NORMAL)
6313 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08006314 break;
6315 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006316 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006317}
6318
Mel Gormanc7132162006-09-27 01:49:43 -07006319/**
6320 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006321 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07006322 *
6323 * This will call free_area_init_node() for each active node in the system.
Zhang Zhen7d018172014-06-04 16:10:53 -07006324 * Using the page ranges provided by memblock_set_node(), the size of each
Mel Gormanc7132162006-09-27 01:49:43 -07006325 * zone in each node and their holes is calculated. If the maximum PFN
6326 * between two adjacent zones match, it is assumed that the zone is empty.
6327 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6328 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6329 * starts where the previous one ended. For example, ZONE_DMA32 starts
6330 * at arch_max_dma_pfn.
6331 */
6332void __init free_area_init_nodes(unsigned long *max_zone_pfn)
6333{
Tejun Heoc13291a2011-07-12 10:46:30 +02006334 unsigned long start_pfn, end_pfn;
6335 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08006336
Mel Gormanc7132162006-09-27 01:49:43 -07006337 /* Record where the zone boundaries are */
6338 memset(arch_zone_lowest_possible_pfn, 0,
6339 sizeof(arch_zone_lowest_possible_pfn));
6340 memset(arch_zone_highest_possible_pfn, 0,
6341 sizeof(arch_zone_highest_possible_pfn));
Oliver O'Halloran90cae1f2016-07-26 15:22:17 -07006342
6343 start_pfn = find_min_pfn_with_active_regions();
6344
6345 for (i = 0; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07006346 if (i == ZONE_MOVABLE)
6347 continue;
Oliver O'Halloran90cae1f2016-07-26 15:22:17 -07006348
6349 end_pfn = max(max_zone_pfn[i], start_pfn);
6350 arch_zone_lowest_possible_pfn[i] = start_pfn;
6351 arch_zone_highest_possible_pfn[i] = end_pfn;
6352
6353 start_pfn = end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07006354 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07006355 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
6356 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
6357
6358 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6359 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07006360 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07006361
Mel Gormanc7132162006-09-27 01:49:43 -07006362 /* Print out the zone ranges */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006363 pr_info("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07006364 for (i = 0; i < MAX_NR_ZONES; i++) {
6365 if (i == ZONE_MOVABLE)
6366 continue;
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006367 pr_info(" %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08006368 if (arch_zone_lowest_possible_pfn[i] ==
6369 arch_zone_highest_possible_pfn[i])
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006370 pr_cont("empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08006371 else
Juergen Gross8d29e182015-02-11 15:26:01 -08006372 pr_cont("[mem %#018Lx-%#018Lx]\n",
6373 (u64)arch_zone_lowest_possible_pfn[i]
6374 << PAGE_SHIFT,
6375 ((u64)arch_zone_highest_possible_pfn[i]
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07006376 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07006377 }
6378
6379 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006380 pr_info("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07006381 for (i = 0; i < MAX_NUMNODES; i++) {
6382 if (zone_movable_pfn[i])
Juergen Gross8d29e182015-02-11 15:26:01 -08006383 pr_info(" Node %d: %#018Lx\n", i,
6384 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07006385 }
Mel Gormanc7132162006-09-27 01:49:43 -07006386
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07006387 /* Print out the early node map */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006388 pr_info("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02006389 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Juergen Gross8d29e182015-02-11 15:26:01 -08006390 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
6391 (u64)start_pfn << PAGE_SHIFT,
6392 ((u64)end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07006393
6394 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07006395 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08006396 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07006397 for_each_online_node(nid) {
6398 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07006399 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07006400 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006401
6402 /* Any memory on that node */
6403 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006404 node_set_state(nid, N_MEMORY);
6405 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07006406 }
6407}
Mel Gorman2a1e2742007-07-17 04:03:12 -07006408
Mel Gorman7e63efe2007-07-17 04:03:15 -07006409static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07006410{
6411 unsigned long long coremem;
6412 if (!p)
6413 return -EINVAL;
6414
6415 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07006416 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07006417
Mel Gorman7e63efe2007-07-17 04:03:15 -07006418 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07006419 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
6420
6421 return 0;
6422}
Mel Gormaned7ed362007-07-17 04:03:14 -07006423
Mel Gorman7e63efe2007-07-17 04:03:15 -07006424/*
6425 * kernelcore=size sets the amount of memory for use for allocations that
6426 * cannot be reclaimed or migrated.
6427 */
6428static int __init cmdline_parse_kernelcore(char *p)
6429{
Taku Izumi342332e2016-03-15 14:55:22 -07006430 /* parse kernelcore=mirror */
6431 if (parse_option_str(p, "mirror")) {
6432 mirrored_kernelcore = true;
6433 return 0;
6434 }
6435
Mel Gorman7e63efe2007-07-17 04:03:15 -07006436 return cmdline_parse_core(p, &required_kernelcore);
6437}
6438
6439/*
6440 * movablecore=size sets the amount of memory for use for allocations that
6441 * can be reclaimed or migrated.
6442 */
6443static int __init cmdline_parse_movablecore(char *p)
6444{
6445 return cmdline_parse_core(p, &required_movablecore);
6446}
6447
Mel Gormaned7ed362007-07-17 04:03:14 -07006448early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07006449early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07006450
Tejun Heo0ee332c2011-12-08 10:22:09 -08006451#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07006452
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006453void adjust_managed_page_count(struct page *page, long count)
6454{
6455 spin_lock(&managed_page_count_lock);
6456 page_zone(page)->managed_pages += count;
6457 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07006458#ifdef CONFIG_HIGHMEM
6459 if (PageHighMem(page))
6460 totalhigh_pages += count;
6461#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006462 spin_unlock(&managed_page_count_lock);
6463}
Jiang Liu3dcc0572013-07-03 15:03:21 -07006464EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006465
Jiang Liu11199692013-07-03 15:02:48 -07006466unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07006467{
Jiang Liu11199692013-07-03 15:02:48 -07006468 void *pos;
6469 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07006470
Jiang Liu11199692013-07-03 15:02:48 -07006471 start = (void *)PAGE_ALIGN((unsigned long)start);
6472 end = (void *)((unsigned long)end & PAGE_MASK);
6473 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07006474 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07006475 memset(pos, poison, PAGE_SIZE);
6476 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07006477 }
6478
6479 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07006480 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07006481 s, pages << (PAGE_SHIFT - 10), start, end);
6482
6483 return pages;
6484}
Jiang Liu11199692013-07-03 15:02:48 -07006485EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07006486
Jiang Liucfa11e02013-04-29 15:07:00 -07006487#ifdef CONFIG_HIGHMEM
6488void free_highmem_page(struct page *page)
6489{
6490 __free_reserved_page(page);
6491 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07006492 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07006493 totalhigh_pages++;
6494}
6495#endif
6496
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006497
6498void __init mem_init_print_info(const char *str)
6499{
6500 unsigned long physpages, codesize, datasize, rosize, bss_size;
6501 unsigned long init_code_size, init_data_size;
6502
6503 physpages = get_num_physpages();
6504 codesize = _etext - _stext;
6505 datasize = _edata - _sdata;
6506 rosize = __end_rodata - __start_rodata;
6507 bss_size = __bss_stop - __bss_start;
6508 init_data_size = __init_end - __init_begin;
6509 init_code_size = _einittext - _sinittext;
6510
6511 /*
6512 * Detect special cases and adjust section sizes accordingly:
6513 * 1) .init.* may be embedded into .data sections
6514 * 2) .init.text.* may be out of [__init_begin, __init_end],
6515 * please refer to arch/tile/kernel/vmlinux.lds.S.
6516 * 3) .rodata.* may be embedded into .text or .data sections.
6517 */
6518#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07006519 do { \
6520 if (start <= pos && pos < end && size > adj) \
6521 size -= adj; \
6522 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006523
6524 adj_init_size(__init_begin, __init_end, init_data_size,
6525 _sinittext, init_code_size);
6526 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
6527 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
6528 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
6529 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
6530
6531#undef adj_init_size
6532
Joe Perches756a025f02016-03-17 14:19:47 -07006533 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 -07006534#ifdef CONFIG_HIGHMEM
Joe Perches756a025f02016-03-17 14:19:47 -07006535 ", %luK highmem"
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006536#endif
Joe Perches756a025f02016-03-17 14:19:47 -07006537 "%s%s)\n",
6538 nr_free_pages() << (PAGE_SHIFT - 10),
6539 physpages << (PAGE_SHIFT - 10),
6540 codesize >> 10, datasize >> 10, rosize >> 10,
6541 (init_data_size + init_code_size) >> 10, bss_size >> 10,
6542 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10),
6543 totalcma_pages << (PAGE_SHIFT - 10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006544#ifdef CONFIG_HIGHMEM
Joe Perches756a025f02016-03-17 14:19:47 -07006545 totalhigh_pages << (PAGE_SHIFT - 10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006546#endif
Joe Perches756a025f02016-03-17 14:19:47 -07006547 str ? ", " : "", str ? str : "");
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006548}
6549
Mel Gorman0e0b8642006-09-27 01:49:56 -07006550/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006551 * set_dma_reserve - set the specified number of pages reserved in the first zone
6552 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07006553 *
Yaowei Bai013110a2015-09-08 15:04:10 -07006554 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
Mel Gorman0e0b8642006-09-27 01:49:56 -07006555 * In the DMA zone, a significant percentage may be consumed by kernel image
6556 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006557 * function may optionally be used to account for unfreeable pages in the
6558 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6559 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07006560 */
6561void __init set_dma_reserve(unsigned long new_dma_reserve)
6562{
6563 dma_reserve = new_dma_reserve;
6564}
6565
Linus Torvalds1da177e2005-04-16 15:20:36 -07006566void __init free_area_init(unsigned long *zones_size)
6567{
Johannes Weiner9109fb72008-07-23 21:27:20 -07006568 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006569 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
6570}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006571
Linus Torvalds1da177e2005-04-16 15:20:36 -07006572static int page_alloc_cpu_notify(struct notifier_block *self,
6573 unsigned long action, void *hcpu)
6574{
6575 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006576
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07006577 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07006578 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006579 drain_pages(cpu);
6580
6581 /*
6582 * Spill the event counters of the dead processor
6583 * into the current processors event counters.
6584 * This artificially elevates the count of the current
6585 * processor.
6586 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07006587 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006588
6589 /*
6590 * Zero the differential counters of the dead processor
6591 * so that the vm statistics are consistent.
6592 *
6593 * This is only okay since the processor is dead and cannot
6594 * race with what we are doing.
6595 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07006596 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006597 }
6598 return NOTIFY_OK;
6599}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006600
6601void __init page_alloc_init(void)
6602{
6603 hotcpu_notifier(page_alloc_cpu_notify, 0);
6604}
6605
6606/*
Yaowei Bai34b10062015-09-08 15:04:13 -07006607 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006608 * or min_free_kbytes changes.
6609 */
6610static void calculate_totalreserve_pages(void)
6611{
6612 struct pglist_data *pgdat;
6613 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006614 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006615
6616 for_each_online_pgdat(pgdat) {
Mel Gorman281e3722016-07-28 15:46:11 -07006617
6618 pgdat->totalreserve_pages = 0;
6619
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006620 for (i = 0; i < MAX_NR_ZONES; i++) {
6621 struct zone *zone = pgdat->node_zones + i;
Mel Gorman3484b2d2014-08-06 16:07:14 -07006622 long max = 0;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006623
6624 /* Find valid and maximum lowmem_reserve in the zone */
6625 for (j = i; j < MAX_NR_ZONES; j++) {
6626 if (zone->lowmem_reserve[j] > max)
6627 max = zone->lowmem_reserve[j];
6628 }
6629
Mel Gorman41858962009-06-16 15:32:12 -07006630 /* we treat the high watermark as reserved pages. */
6631 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006632
Jiang Liub40da042013-02-22 16:33:52 -08006633 if (max > zone->managed_pages)
6634 max = zone->managed_pages;
Johannes Weinera8d01432016-01-14 15:20:15 -08006635
Mel Gorman281e3722016-07-28 15:46:11 -07006636 pgdat->totalreserve_pages += max;
Johannes Weinera8d01432016-01-14 15:20:15 -08006637
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006638 reserve_pages += max;
6639 }
6640 }
6641 totalreserve_pages = reserve_pages;
6642}
6643
6644/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07006645 * setup_per_zone_lowmem_reserve - called whenever
Yaowei Bai34b10062015-09-08 15:04:13 -07006646 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
Linus Torvalds1da177e2005-04-16 15:20:36 -07006647 * has a correct pages reserved value, so an adequate number of
6648 * pages are left in the zone after a successful __alloc_pages().
6649 */
6650static void setup_per_zone_lowmem_reserve(void)
6651{
6652 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006653 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006654
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08006655 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006656 for (j = 0; j < MAX_NR_ZONES; j++) {
6657 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08006658 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006659
6660 zone->lowmem_reserve[j] = 0;
6661
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006662 idx = j;
6663 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006664 struct zone *lower_zone;
6665
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006666 idx--;
6667
Linus Torvalds1da177e2005-04-16 15:20:36 -07006668 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6669 sysctl_lowmem_reserve_ratio[idx] = 1;
6670
6671 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08006672 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07006673 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08006674 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006675 }
6676 }
6677 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006678
6679 /* update totalreserve_pages */
6680 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006681}
6682
Mel Gormancfd3da12011-04-25 21:36:42 +00006683static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006684{
6685 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6686 unsigned long lowmem_pages = 0;
6687 struct zone *zone;
6688 unsigned long flags;
6689
6690 /* Calculate total number of !ZONE_HIGHMEM pages */
6691 for_each_zone(zone) {
6692 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08006693 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006694 }
6695
6696 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07006697 u64 tmp;
6698
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006699 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08006700 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07006701 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006702 if (is_highmem(zone)) {
6703 /*
Nick Piggin669ed172005-11-13 16:06:45 -08006704 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6705 * need highmem pages, so cap pages_min to a small
6706 * value here.
6707 *
Mel Gorman41858962009-06-16 15:32:12 -07006708 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Yaowei Bai42ff2702015-04-14 15:47:14 -07006709 * deltas control asynch page reclaim, and so should
Nick Piggin669ed172005-11-13 16:06:45 -08006710 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006711 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08006712 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006713
Jiang Liub40da042013-02-22 16:33:52 -08006714 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08006715 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07006716 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006717 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08006718 /*
6719 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07006720 * proportionate to the zone's size.
6721 */
Mel Gorman41858962009-06-16 15:32:12 -07006722 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006723 }
6724
Johannes Weiner795ae7a2016-03-17 14:19:14 -07006725 /*
6726 * Set the kswapd watermarks distance according to the
6727 * scale factor in proportion to available memory, but
6728 * ensure a minimum size on small systems.
6729 */
6730 tmp = max_t(u64, tmp >> 2,
6731 mult_frac(zone->managed_pages,
6732 watermark_scale_factor, 10000));
6733
6734 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp;
6735 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006736
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006737 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006738 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006739
6740 /* update totalreserve_pages */
6741 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006742}
6743
Mel Gormancfd3da12011-04-25 21:36:42 +00006744/**
6745 * setup_per_zone_wmarks - called when min_free_kbytes changes
6746 * or when memory is hot-{added|removed}
6747 *
6748 * Ensures that the watermark[min,low,high] values for each zone are set
6749 * correctly with respect to min_free_kbytes.
6750 */
6751void setup_per_zone_wmarks(void)
6752{
6753 mutex_lock(&zonelists_mutex);
6754 __setup_per_zone_wmarks();
6755 mutex_unlock(&zonelists_mutex);
6756}
6757
Randy Dunlap55a44622009-09-21 17:01:20 -07006758/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07006759 * Initialise min_free_kbytes.
6760 *
6761 * For small machines we want it small (128k min). For large machines
6762 * we want it large (64MB max). But it is not linear, because network
6763 * bandwidth does not increase linearly with machine size. We use
6764 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006765 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07006766 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6767 *
6768 * which yields
6769 *
6770 * 16MB: 512k
6771 * 32MB: 724k
6772 * 64MB: 1024k
6773 * 128MB: 1448k
6774 * 256MB: 2048k
6775 * 512MB: 2896k
6776 * 1024MB: 4096k
6777 * 2048MB: 5792k
6778 * 4096MB: 8192k
6779 * 8192MB: 11584k
6780 * 16384MB: 16384k
6781 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07006782int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006783{
6784 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07006785 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006786
6787 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07006788 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006789
Michal Hocko5f127332013-07-08 16:00:40 -07006790 if (new_min_free_kbytes > user_min_free_kbytes) {
6791 min_free_kbytes = new_min_free_kbytes;
6792 if (min_free_kbytes < 128)
6793 min_free_kbytes = 128;
6794 if (min_free_kbytes > 65536)
6795 min_free_kbytes = 65536;
6796 } else {
6797 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6798 new_min_free_kbytes, user_min_free_kbytes);
6799 }
Minchan Kimbc75d332009-06-16 15:32:48 -07006800 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07006801 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006802 setup_per_zone_lowmem_reserve();
Joonsoo Kim6423aa82016-08-10 16:27:49 -07006803
6804#ifdef CONFIG_NUMA
6805 setup_min_unmapped_ratio();
6806 setup_min_slab_ratio();
6807#endif
6808
Linus Torvalds1da177e2005-04-16 15:20:36 -07006809 return 0;
6810}
Jason Baronbc22af72016-05-05 16:22:12 -07006811core_initcall(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006812
6813/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07006814 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07006815 * that we can call two helper functions whenever min_free_kbytes
6816 * changes.
6817 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006818int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006819 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006820{
Han Pingtianda8c7572014-01-23 15:53:17 -08006821 int rc;
6822
6823 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6824 if (rc)
6825 return rc;
6826
Michal Hocko5f127332013-07-08 16:00:40 -07006827 if (write) {
6828 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07006829 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07006830 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006831 return 0;
6832}
6833
Johannes Weiner795ae7a2016-03-17 14:19:14 -07006834int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
6835 void __user *buffer, size_t *length, loff_t *ppos)
6836{
6837 int rc;
6838
6839 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6840 if (rc)
6841 return rc;
6842
6843 if (write)
6844 setup_per_zone_wmarks();
6845
6846 return 0;
6847}
6848
Christoph Lameter96146342006-07-03 00:24:13 -07006849#ifdef CONFIG_NUMA
Joonsoo Kim6423aa82016-08-10 16:27:49 -07006850static void setup_min_unmapped_ratio(void)
Christoph Lameter96146342006-07-03 00:24:13 -07006851{
Joonsoo Kim6423aa82016-08-10 16:27:49 -07006852 pg_data_t *pgdat;
Christoph Lameter96146342006-07-03 00:24:13 -07006853 struct zone *zone;
Christoph Lameter96146342006-07-03 00:24:13 -07006854
Mel Gormana5f5f912016-07-28 15:46:32 -07006855 for_each_online_pgdat(pgdat)
Joonsoo Kim81cbcbc2016-08-10 16:27:46 -07006856 pgdat->min_unmapped_pages = 0;
Mel Gormana5f5f912016-07-28 15:46:32 -07006857
Christoph Lameter96146342006-07-03 00:24:13 -07006858 for_each_zone(zone)
Mel Gormana5f5f912016-07-28 15:46:32 -07006859 zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07006860 sysctl_min_unmapped_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07006861}
Christoph Lameter0ff38492006-09-25 23:31:52 -07006862
Joonsoo Kim6423aa82016-08-10 16:27:49 -07006863
6864int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006865 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07006866{
Christoph Lameter0ff38492006-09-25 23:31:52 -07006867 int rc;
6868
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006869 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07006870 if (rc)
6871 return rc;
6872
Joonsoo Kim6423aa82016-08-10 16:27:49 -07006873 setup_min_unmapped_ratio();
6874
6875 return 0;
6876}
6877
6878static void setup_min_slab_ratio(void)
6879{
6880 pg_data_t *pgdat;
6881 struct zone *zone;
6882
Mel Gormana5f5f912016-07-28 15:46:32 -07006883 for_each_online_pgdat(pgdat)
6884 pgdat->min_slab_pages = 0;
6885
Christoph Lameter0ff38492006-09-25 23:31:52 -07006886 for_each_zone(zone)
Mel Gormana5f5f912016-07-28 15:46:32 -07006887 zone->zone_pgdat->min_slab_pages += (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07006888 sysctl_min_slab_ratio) / 100;
Joonsoo Kim6423aa82016-08-10 16:27:49 -07006889}
6890
6891int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
6892 void __user *buffer, size_t *length, loff_t *ppos)
6893{
6894 int rc;
6895
6896 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6897 if (rc)
6898 return rc;
6899
6900 setup_min_slab_ratio();
6901
Christoph Lameter0ff38492006-09-25 23:31:52 -07006902 return 0;
6903}
Christoph Lameter96146342006-07-03 00:24:13 -07006904#endif
6905
Linus Torvalds1da177e2005-04-16 15:20:36 -07006906/*
6907 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6908 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6909 * whenever sysctl_lowmem_reserve_ratio changes.
6910 *
6911 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07006912 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07006913 * if in function of the boot time zone sizes.
6914 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006915int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006916 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006917{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006918 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006919 setup_per_zone_lowmem_reserve();
6920 return 0;
6921}
6922
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006923/*
6924 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07006925 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6926 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006927 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006928int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006929 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006930{
6931 struct zone *zone;
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006932 int old_percpu_pagelist_fraction;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006933 int ret;
6934
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006935 mutex_lock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006936 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6937
6938 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
6939 if (!write || ret < 0)
6940 goto out;
6941
6942 /* Sanity checking to avoid pcp imbalance */
6943 if (percpu_pagelist_fraction &&
6944 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6945 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6946 ret = -EINVAL;
6947 goto out;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006948 }
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006949
6950 /* No change? */
6951 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6952 goto out;
6953
6954 for_each_populated_zone(zone) {
6955 unsigned int cpu;
6956
6957 for_each_possible_cpu(cpu)
6958 pageset_set_high_and_batch(zone,
6959 per_cpu_ptr(zone->pageset, cpu));
6960 }
6961out:
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006962 mutex_unlock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006963 return ret;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006964}
6965
Rasmus Villemoesa9919c72015-06-24 16:56:28 -07006966#ifdef CONFIG_NUMA
David S. Millerf034b5d2006-08-24 03:08:07 -07006967int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006968
Linus Torvalds1da177e2005-04-16 15:20:36 -07006969static int __init set_hashdist(char *str)
6970{
6971 if (!str)
6972 return 0;
6973 hashdist = simple_strtoul(str, &str, 0);
6974 return 1;
6975}
6976__setup("hashdist=", set_hashdist);
6977#endif
6978
6979/*
6980 * allocate a large system hash table from bootmem
6981 * - it is assumed that the hash table must contain an exact power-of-2
6982 * quantity of entries
6983 * - limit is the number of hash buckets, not the total allocation size
6984 */
6985void *__init alloc_large_system_hash(const char *tablename,
6986 unsigned long bucketsize,
6987 unsigned long numentries,
6988 int scale,
6989 int flags,
6990 unsigned int *_hash_shift,
6991 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00006992 unsigned long low_limit,
6993 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006994{
Tim Bird31fe62b2012-05-23 13:33:35 +00006995 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006996 unsigned long log2qty, size;
6997 void *table = NULL;
6998
6999 /* allow the kernel cmdline to have a say */
7000 if (!numentries) {
7001 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08007002 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07007003
7004 /* It isn't necessary when PAGE_SIZE >= 1MB */
7005 if (PAGE_SHIFT < 20)
7006 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007007
7008 /* limit to 1 bucket per 2^scale bytes of low memory */
7009 if (scale > PAGE_SHIFT)
7010 numentries >>= (scale - PAGE_SHIFT);
7011 else
7012 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08007013
7014 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07007015 if (unlikely(flags & HASH_SMALL)) {
7016 /* Makes no sense without HASH_EARLY */
7017 WARN_ON(!(flags & HASH_EARLY));
7018 if (!(numentries >> *_hash_shift)) {
7019 numentries = 1UL << *_hash_shift;
7020 BUG_ON(!numentries);
7021 }
7022 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08007023 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007024 }
John Hawkes6e692ed2006-03-25 03:08:02 -08007025 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007026
7027 /* limit allocation size to 1/16 total memory by default */
7028 if (max == 0) {
7029 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
7030 do_div(max, bucketsize);
7031 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08007032 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007033
Tim Bird31fe62b2012-05-23 13:33:35 +00007034 if (numentries < low_limit)
7035 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007036 if (numentries > max)
7037 numentries = max;
7038
David Howellsf0d1b0b2006-12-08 02:37:49 -08007039 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007040
7041 do {
7042 size = bucketsize << log2qty;
7043 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08007044 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007045 else if (hashdist)
7046 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
7047 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07007048 /*
7049 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07007050 * some pages at the end of hash table which
7051 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07007052 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01007053 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07007054 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01007055 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
7056 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007057 }
7058 } while (!table && size > PAGE_SIZE && --log2qty);
7059
7060 if (!table)
7061 panic("Failed to allocate %s hash table\n", tablename);
7062
Joe Perches11705322016-03-17 14:19:50 -07007063 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7064 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007065
7066 if (_hash_shift)
7067 *_hash_shift = log2qty;
7068 if (_hash_mask)
7069 *_hash_mask = (1 << log2qty) - 1;
7070
7071 return table;
7072}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08007073
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07007074/*
Minchan Kim80934512012-07-31 16:43:01 -07007075 * This function checks whether pageblock includes unmovable pages or not.
7076 * If @count is not zero, it is okay to include less @count unmovable pages
7077 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07007078 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07007079 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
7080 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07007081 */
Wen Congyangb023f462012-12-11 16:00:45 -08007082bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
7083 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007084{
7085 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01007086 int mt;
7087
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007088 /*
7089 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07007090 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007091 */
7092 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07007093 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01007094 mt = get_pageblock_migratetype(page);
7095 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07007096 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007097
7098 pfn = page_to_pfn(page);
7099 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
7100 unsigned long check = pfn + iter;
7101
Namhyung Kim29723fc2011-02-25 14:44:25 -08007102 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007103 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08007104
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007105 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07007106
7107 /*
7108 * Hugepages are not in LRU lists, but they're movable.
7109 * We need not scan over tail pages bacause we don't
7110 * handle each tail page individually in migration.
7111 */
7112 if (PageHuge(page)) {
7113 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
7114 continue;
7115 }
7116
Minchan Kim97d255c2012-07-31 16:42:59 -07007117 /*
7118 * We can't use page_count without pin a page
7119 * because another CPU can free compound page.
7120 * This check already skips compound tails of THP
Joonsoo Kim0139aa72016-05-19 17:10:49 -07007121 * because their page->_refcount is zero at all time.
Minchan Kim97d255c2012-07-31 16:42:59 -07007122 */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07007123 if (!page_ref_count(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007124 if (PageBuddy(page))
7125 iter += (1 << page_order(page)) - 1;
7126 continue;
7127 }
Minchan Kim97d255c2012-07-31 16:42:59 -07007128
Wen Congyangb023f462012-12-11 16:00:45 -08007129 /*
7130 * The HWPoisoned page may be not in buddy system, and
7131 * page_count() is not 0.
7132 */
7133 if (skip_hwpoisoned_pages && PageHWPoison(page))
7134 continue;
7135
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007136 if (!PageLRU(page))
7137 found++;
7138 /*
Johannes Weiner6b4f7792014-12-12 16:56:13 -08007139 * If there are RECLAIMABLE pages, we need to check
7140 * it. But now, memory offline itself doesn't call
7141 * shrink_node_slabs() and it still to be fixed.
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007142 */
7143 /*
7144 * If the page is not RAM, page_count()should be 0.
7145 * we don't need more check. This is an _used_ not-movable page.
7146 *
7147 * The problematic thing here is PG_reserved pages. PG_reserved
7148 * is set to both of a memory hole page and a _used_ kernel
7149 * page at boot.
7150 */
7151 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07007152 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007153 }
Minchan Kim80934512012-07-31 16:43:01 -07007154 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007155}
7156
7157bool is_pageblock_removable_nolock(struct page *page)
7158{
Michal Hocko656a0702012-01-20 14:33:58 -08007159 struct zone *zone;
7160 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08007161
7162 /*
7163 * We have to be careful here because we are iterating over memory
7164 * sections which are not zone aware so we might end up outside of
7165 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08007166 * We have to take care about the node as well. If the node is offline
7167 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08007168 */
Michal Hocko656a0702012-01-20 14:33:58 -08007169 if (!node_online(page_to_nid(page)))
7170 return false;
7171
7172 zone = page_zone(page);
7173 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08007174 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08007175 return false;
7176
Wen Congyangb023f462012-12-11 16:00:45 -08007177 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07007178}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007179
Vlastimil Babka080fe202016-02-05 15:36:41 -08007180#if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007181
7182static unsigned long pfn_max_align_down(unsigned long pfn)
7183{
7184 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
7185 pageblock_nr_pages) - 1);
7186}
7187
7188static unsigned long pfn_max_align_up(unsigned long pfn)
7189{
7190 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
7191 pageblock_nr_pages));
7192}
7193
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007194/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007195static int __alloc_contig_migrate_range(struct compact_control *cc,
7196 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007197{
7198 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07007199 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007200 unsigned long pfn = start;
7201 unsigned int tries = 0;
7202 int ret = 0;
7203
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08007204 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007205
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007206 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007207 if (fatal_signal_pending(current)) {
7208 ret = -EINTR;
7209 break;
7210 }
7211
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007212 if (list_empty(&cc->migratepages)) {
7213 cc->nr_migratepages = 0;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07007214 pfn = isolate_migratepages_range(cc, pfn, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007215 if (!pfn) {
7216 ret = -EINTR;
7217 break;
7218 }
7219 tries = 0;
7220 } else if (++tries == 5) {
7221 ret = ret < 0 ? ret : -EBUSY;
7222 break;
7223 }
7224
Minchan Kimbeb51ea2012-10-08 16:33:51 -07007225 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
7226 &cc->migratepages);
7227 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07007228
Hugh Dickins9c620e22013-02-22 16:35:14 -08007229 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07007230 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007231 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08007232 if (ret < 0) {
7233 putback_movable_pages(&cc->migratepages);
7234 return ret;
7235 }
7236 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007237}
7238
7239/**
7240 * alloc_contig_range() -- tries to allocate given range of pages
7241 * @start: start PFN to allocate
7242 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02007243 * @migratetype: migratetype of the underlaying pageblocks (either
7244 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
7245 * in range must have the same migratetype and it must
7246 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007247 *
7248 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7249 * aligned, however it's the caller's responsibility to guarantee that
7250 * we are the only thread that changes migrate type of pageblocks the
7251 * pages fall in.
7252 *
7253 * The PFN range must belong to a single zone.
7254 *
7255 * Returns zero on success or negative error code. On success all
7256 * pages which PFN is in [start, end) are allocated for the caller and
7257 * need to be freed with free_contig_range().
7258 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02007259int alloc_contig_range(unsigned long start, unsigned long end,
7260 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007261{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007262 unsigned long outer_start, outer_end;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08007263 unsigned int order;
7264 int ret = 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007265
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007266 struct compact_control cc = {
7267 .nr_migratepages = 0,
7268 .order = -1,
7269 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07007270 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007271 .ignore_skip_hint = true,
7272 };
7273 INIT_LIST_HEAD(&cc.migratepages);
7274
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007275 /*
7276 * What we do here is we mark all pageblocks in range as
7277 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7278 * have different sizes, and due to the way page allocator
7279 * work, we align the range to biggest of the two pages so
7280 * that page allocator won't try to merge buddies from
7281 * different pageblocks and change MIGRATE_ISOLATE to some
7282 * other migration type.
7283 *
7284 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7285 * migrate the pages from an unaligned range (ie. pages that
7286 * we are interested in). This will put all the pages in
7287 * range back to page allocator as MIGRATE_ISOLATE.
7288 *
7289 * When this is done, we take the pages in range from page
7290 * allocator removing them from the buddy system. This way
7291 * page allocator will never consider using them.
7292 *
7293 * This lets us mark the pageblocks back as
7294 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7295 * aligned range but not in the unaligned, original range are
7296 * put back to page allocator so that buddy can use them.
7297 */
7298
7299 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08007300 pfn_max_align_up(end), migratetype,
7301 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007302 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07007303 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007304
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007305 /*
7306 * In case of -EBUSY, we'd like to know which page causes problem.
7307 * So, just fall through. We will check it in test_pages_isolated().
7308 */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007309 ret = __alloc_contig_migrate_range(&cc, start, end);
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007310 if (ret && ret != -EBUSY)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007311 goto done;
7312
7313 /*
7314 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7315 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7316 * more, all pages in [start, end) are free in page allocator.
7317 * What we are going to do is to allocate all pages from
7318 * [start, end) (that is remove them from page allocator).
7319 *
7320 * The only problem is that pages at the beginning and at the
7321 * end of interesting range may be not aligned with pages that
7322 * page allocator holds, ie. they can be part of higher order
7323 * pages. Because of this, we reserve the bigger range and
7324 * once this is done free the pages we are not interested in.
7325 *
7326 * We don't have to hold zone->lock here because the pages are
7327 * isolated thus they won't get removed from buddy.
7328 */
7329
7330 lru_add_drain_all();
Vlastimil Babka510f5502014-12-10 15:43:07 -08007331 drain_all_pages(cc.zone);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007332
7333 order = 0;
7334 outer_start = start;
7335 while (!PageBuddy(pfn_to_page(outer_start))) {
7336 if (++order >= MAX_ORDER) {
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007337 outer_start = start;
7338 break;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007339 }
7340 outer_start &= ~0UL << order;
7341 }
7342
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007343 if (outer_start != start) {
7344 order = page_order(pfn_to_page(outer_start));
7345
7346 /*
7347 * outer_start page could be small order buddy page and
7348 * it doesn't include start page. Adjust outer_start
7349 * in this case to report failed page properly
7350 * on tracepoint in test_pages_isolated()
7351 */
7352 if (outer_start + (1UL << order) <= start)
7353 outer_start = start;
7354 }
7355
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007356 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08007357 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewiczdae803e2014-11-13 15:19:27 -08007358 pr_info("%s: [%lx, %lx) PFNs busy\n",
7359 __func__, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007360 ret = -EBUSY;
7361 goto done;
7362 }
7363
Marek Szyprowski49f223a2012-01-25 12:49:24 +01007364 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007365 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007366 if (!outer_end) {
7367 ret = -EBUSY;
7368 goto done;
7369 }
7370
7371 /* Free head and tail (if any) */
7372 if (start != outer_start)
7373 free_contig_range(outer_start, start - outer_start);
7374 if (end != outer_end)
7375 free_contig_range(end, outer_end - end);
7376
7377done:
7378 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02007379 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007380 return ret;
7381}
7382
7383void free_contig_range(unsigned long pfn, unsigned nr_pages)
7384{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08007385 unsigned int count = 0;
7386
7387 for (; nr_pages--; pfn++) {
7388 struct page *page = pfn_to_page(pfn);
7389
7390 count += page_count(page) != 1;
7391 __free_page(page);
7392 }
7393 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007394}
7395#endif
7396
Jiang Liu4ed7e022012-07-31 16:43:35 -07007397#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07007398/*
7399 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7400 * page high values need to be recalulated.
7401 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07007402void __meminit zone_pcp_update(struct zone *zone)
7403{
Cody P Schafer0a647f32013-07-03 15:01:33 -07007404 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007405 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07007406 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07007407 pageset_set_high_and_batch(zone,
7408 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007409 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07007410}
7411#endif
7412
Jiang Liu340175b2012-07-31 16:43:32 -07007413void zone_pcp_reset(struct zone *zone)
7414{
7415 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07007416 int cpu;
7417 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07007418
7419 /* avoid races with drain_pages() */
7420 local_irq_save(flags);
7421 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07007422 for_each_online_cpu(cpu) {
7423 pset = per_cpu_ptr(zone->pageset, cpu);
7424 drain_zonestat(zone, pset);
7425 }
Jiang Liu340175b2012-07-31 16:43:32 -07007426 free_percpu(zone->pageset);
7427 zone->pageset = &boot_pageset;
7428 }
7429 local_irq_restore(flags);
7430}
7431
Wen Congyang6dcd73d2012-12-11 16:01:01 -08007432#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007433/*
Joonsoo Kimb9eb6312016-05-19 17:12:06 -07007434 * All pages in the range must be in a single zone and isolated
7435 * before calling this.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007436 */
7437void
7438__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
7439{
7440 struct page *page;
7441 struct zone *zone;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07007442 unsigned int order, i;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007443 unsigned long pfn;
7444 unsigned long flags;
7445 /* find the first valid pfn */
7446 for (pfn = start_pfn; pfn < end_pfn; pfn++)
7447 if (pfn_valid(pfn))
7448 break;
7449 if (pfn == end_pfn)
7450 return;
7451 zone = page_zone(pfn_to_page(pfn));
7452 spin_lock_irqsave(&zone->lock, flags);
7453 pfn = start_pfn;
7454 while (pfn < end_pfn) {
7455 if (!pfn_valid(pfn)) {
7456 pfn++;
7457 continue;
7458 }
7459 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08007460 /*
7461 * The HWPoisoned page may be not in buddy system, and
7462 * page_count() is not 0.
7463 */
7464 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
7465 pfn++;
7466 SetPageReserved(page);
7467 continue;
7468 }
7469
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007470 BUG_ON(page_count(page));
7471 BUG_ON(!PageBuddy(page));
7472 order = page_order(page);
7473#ifdef CONFIG_DEBUG_VM
Joe Perches11705322016-03-17 14:19:50 -07007474 pr_info("remove from free list %lx %d %lx\n",
7475 pfn, 1 << order, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007476#endif
7477 list_del(&page->lru);
7478 rmv_page_order(page);
7479 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007480 for (i = 0; i < (1 << order); i++)
7481 SetPageReserved((page+i));
7482 pfn += (1 << order);
7483 }
7484 spin_unlock_irqrestore(&zone->lock, flags);
7485}
7486#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007487
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007488bool is_free_buddy_page(struct page *page)
7489{
7490 struct zone *zone = page_zone(page);
7491 unsigned long pfn = page_to_pfn(page);
7492 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07007493 unsigned int order;
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007494
7495 spin_lock_irqsave(&zone->lock, flags);
7496 for (order = 0; order < MAX_ORDER; order++) {
7497 struct page *page_head = page - (pfn & ((1 << order) - 1));
7498
7499 if (PageBuddy(page_head) && page_order(page_head) >= order)
7500 break;
7501 }
7502 spin_unlock_irqrestore(&zone->lock, flags);
7503
7504 return order < MAX_ORDER;
7505}