blob: 6cbde310abed8df22f9cd6ed80fcc252f4c80f43 [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 Hiroyukia5d76b542007-10-16 01:26:11 -070052#include <linux/page-isolation.h>
Joonsoo Kimeefa864b2014-12-12 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>
Michal Hockod379f012017-02-22 15:42:00 -080058#include <trace/events/oom.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070059#include <linux/prefetch.h>
Lisa Du6e543d52013-09-11 14:22:36 -070060#include <linux/mm_inline.h>
Michal Nazarewicz041d3a82011-12-29 13:09:50 +010061#include <linux/migrate.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>
Ingo Molnar5b3cc152017-02-02 20:43:54 +010064#include <linux/sched/mm.h>
Joonsoo Kim48c96a32014-12-12 16:56:01 -080065#include <linux/page_owner.h>
Mel Gorman0e1cc952015-06-30 14:57:27 -070066#include <linux/kthread.h>
Vladimir Davydov49491482016-07-26 15:24:24 -070067#include <linux/memcontrol.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
Jiang Liu7ee3d4e2013-07-03 15:03:41 -070069#include <asm/sections.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070071#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070072#include "internal.h"
73
Cody P Schaferc8e251f2013-07-03 15:01:29 -070074/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
75static DEFINE_MUTEX(pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -070076#define MIN_PERCPU_PAGELIST_FRACTION (8)
Cody P Schaferc8e251f2013-07-03 15:01:29 -070077
Lee Schermerhorn72812012010-05-26 14:44:56 -070078#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
79DEFINE_PER_CPU(int, numa_node);
80EXPORT_PER_CPU_SYMBOL(numa_node);
81#endif
82
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070083#ifdef CONFIG_HAVE_MEMORYLESS_NODES
84/*
85 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
86 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
87 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
88 * defined in <linux/topology.h>.
89 */
90DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
91EXPORT_PER_CPU_SYMBOL(_numa_mem_);
Joonsoo Kimad2c8142014-10-09 15:26:13 -070092int _node_numa_mem_[MAX_NUMNODES];
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070093#endif
94
Mel Gormanbd233f52017-02-24 14:56:56 -080095/* work_structs for global per-cpu drains */
96DEFINE_MUTEX(pcpu_drain_mutex);
97DEFINE_PER_CPU(struct work_struct, pcpu_drain);
98
Emese Revfy38addce2016-06-20 20:41:19 +020099#ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
Kees Cook58bea4142016-10-19 00:08:04 +0200100volatile unsigned long latent_entropy __latent_entropy;
Emese Revfy38addce2016-06-20 20:41:19 +0200101EXPORT_SYMBOL(latent_entropy);
102#endif
103
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104/*
Christoph Lameter13808912007-10-16 01:25:27 -0700105 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 */
Christoph Lameter13808912007-10-16 01:25:27 -0700107nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
108 [N_POSSIBLE] = NODE_MASK_ALL,
109 [N_ONLINE] = { { [0] = 1UL } },
110#ifndef CONFIG_NUMA
111 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
112#ifdef CONFIG_HIGHMEM
113 [N_HIGH_MEMORY] = { { [0] = 1UL } },
114#endif
Lai Jiangshan20b2f522012-12-12 13:52:00 -0800115#ifdef CONFIG_MOVABLE_NODE
116 [N_MEMORY] = { { [0] = 1UL } },
117#endif
Christoph Lameter13808912007-10-16 01:25:27 -0700118 [N_CPU] = { { [0] = 1UL } },
119#endif /* NUMA */
120};
121EXPORT_SYMBOL(node_states);
122
Jiang Liuc3d5f5f2013-07-03 15:03:14 -0700123/* Protect totalram_pages and zone->managed_pages */
124static DEFINE_SPINLOCK(managed_page_count_lock);
125
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700126unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -0700127unsigned long totalreserve_pages __read_mostly;
Pintu Kumare48322a2014-12-18 16:17:15 -0800128unsigned long totalcma_pages __read_mostly;
Johannes Weinerab8fabd2012-01-10 15:07:42 -0800129
Hugh Dickins1b76b022012-05-11 01:00:07 -0700130int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000131gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -0700133/*
134 * A cached value of the page's pageblock's migratetype, used when the page is
135 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
136 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
137 * Also the migratetype set in the page does not necessarily match the pcplist
138 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
139 * other index - this ensures that it will be put on the correct CMA freelist.
140 */
141static inline int get_pcppage_migratetype(struct page *page)
142{
143 return page->index;
144}
145
146static inline void set_pcppage_migratetype(struct page *page, int migratetype)
147{
148 page->index = migratetype;
149}
150
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800151#ifdef CONFIG_PM_SLEEP
152/*
153 * The following functions are used by the suspend/hibernate code to temporarily
154 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
155 * while devices are suspended. To avoid races with the suspend/hibernate code,
156 * they should always be called with pm_mutex held (gfp_allowed_mask also should
157 * only be modified with pm_mutex held, unless the suspend/hibernate code is
158 * guaranteed not to run in parallel with that modification).
159 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100160
161static gfp_t saved_gfp_mask;
162
163void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800164{
165 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100166 if (saved_gfp_mask) {
167 gfp_allowed_mask = saved_gfp_mask;
168 saved_gfp_mask = 0;
169 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800170}
171
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100172void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800173{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800174 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100175 WARN_ON(saved_gfp_mask);
176 saved_gfp_mask = gfp_allowed_mask;
Mel Gormand0164ad2015-11-06 16:28:21 -0800177 gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800178}
Mel Gormanf90ac392012-01-10 15:07:15 -0800179
180bool pm_suspended_storage(void)
181{
Mel Gormand0164ad2015-11-06 16:28:21 -0800182 if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
Mel Gormanf90ac392012-01-10 15:07:15 -0800183 return false;
184 return true;
185}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800186#endif /* CONFIG_PM_SLEEP */
187
Mel Gormand9c23402007-10-16 01:26:01 -0700188#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -0800189unsigned int pageblock_order __read_mostly;
Mel Gormand9c23402007-10-16 01:26:01 -0700190#endif
191
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800192static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194/*
195 * results with 256, 32 in the lowmem_reserve sysctl:
196 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
197 * 1G machine -> (16M dma, 784M normal, 224M high)
198 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
199 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
Yaowei Bai84109e12015-02-12 15:00:22 -0800200 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100201 *
202 * TBD: should special case ZONE_DMA32 machines here - in those we normally
203 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700205int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800206#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700207 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800208#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700209#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700210 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700211#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700212#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700213 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700214#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700215 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700216};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
218EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
Helge Deller15ad7cd2006-12-06 20:40:36 -0800220static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800221#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700222 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800223#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700224#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700225 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700226#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700227 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700228#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700229 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700230#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700231 "Movable",
Dan Williams033fbae2015-08-09 15:29:06 -0400232#ifdef CONFIG_ZONE_DEVICE
233 "Device",
234#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700235};
236
Vlastimil Babka60f30352016-03-15 14:56:08 -0700237char * const migratetype_names[MIGRATE_TYPES] = {
238 "Unmovable",
239 "Movable",
240 "Reclaimable",
241 "HighAtomic",
242#ifdef CONFIG_CMA
243 "CMA",
244#endif
245#ifdef CONFIG_MEMORY_ISOLATION
246 "Isolate",
247#endif
248};
249
Kirill A. Shutemovf1e61552015-11-06 16:29:50 -0800250compound_page_dtor * const compound_page_dtors[] = {
251 NULL,
252 free_compound_page,
253#ifdef CONFIG_HUGETLB_PAGE
254 free_huge_page,
255#endif
Kirill A. Shutemov9a982252016-01-15 16:54:17 -0800256#ifdef CONFIG_TRANSPARENT_HUGEPAGE
257 free_transhuge_page,
258#endif
Kirill A. Shutemovf1e61552015-11-06 16:29:50 -0800259};
260
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261int min_free_kbytes = 1024;
Han Pingtian42aa83c2014-01-23 15:53:28 -0800262int user_min_free_kbytes = -1;
Johannes Weiner795ae7a2016-03-17 14:19:14 -0700263int watermark_scale_factor = 10;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
Jan Beulich2c85f512009-09-21 17:03:07 -0700265static unsigned long __meminitdata nr_kernel_pages;
266static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700267static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
Tejun Heo0ee332c2011-12-08 10:22:09 -0800269#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
270static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
271static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
272static unsigned long __initdata required_kernelcore;
273static unsigned long __initdata required_movablecore;
274static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Taku Izumi342332e2016-03-15 14:55:22 -0700275static bool mirrored_kernelcore;
Mel Gormanc7132162006-09-27 01:49:43 -0700276
Tejun Heo0ee332c2011-12-08 10:22:09 -0800277/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
278int movable_zone;
279EXPORT_SYMBOL(movable_zone);
280#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700281
Miklos Szeredi418508c2007-05-23 13:57:55 -0700282#if MAX_NUMNODES > 1
283int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700284int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700285EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700286EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700287#endif
288
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700289int page_group_by_mobility_disabled __read_mostly;
290
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700291#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
292static inline void reset_deferred_meminit(pg_data_t *pgdat)
293{
294 pgdat->first_deferred_pfn = ULONG_MAX;
295}
296
297/* Returns true if the struct page for the pfn is uninitialised */
Mel Gorman0e1cc952015-06-30 14:57:27 -0700298static inline bool __meminit early_page_uninitialised(unsigned long pfn)
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700299{
Mel Gormanef70b6f2016-07-14 12:07:23 -0700300 int nid = early_pfn_to_nid(pfn);
301
302 if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700303 return true;
304
305 return false;
306}
307
308/*
309 * Returns false when the remaining initialisation should be deferred until
310 * later in the boot cycle when it can be parallelised.
311 */
312static inline bool update_defer_init(pg_data_t *pgdat,
313 unsigned long pfn, unsigned long zone_end,
314 unsigned long *nr_initialised)
315{
Li Zhang987b3092016-03-17 14:20:16 -0700316 unsigned long max_initialise;
317
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700318 /* Always populate low zones for address-contrained allocations */
319 if (zone_end < pgdat_end_pfn(pgdat))
320 return true;
Li Zhang987b3092016-03-17 14:20:16 -0700321 /*
322 * Initialise at least 2G of a node but also take into account that
323 * two large system hashes that can take up 1GB for 0.25TB/node.
324 */
325 max_initialise = max(2UL << (30 - PAGE_SHIFT),
326 (pgdat->node_spanned_pages >> 8));
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700327
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700328 (*nr_initialised)++;
Li Zhang987b3092016-03-17 14:20:16 -0700329 if ((*nr_initialised > max_initialise) &&
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700330 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
331 pgdat->first_deferred_pfn = pfn;
332 return false;
333 }
334
335 return true;
336}
337#else
338static inline void reset_deferred_meminit(pg_data_t *pgdat)
339{
340}
341
342static inline bool early_page_uninitialised(unsigned long pfn)
343{
344 return false;
345}
346
347static inline bool update_defer_init(pg_data_t *pgdat,
348 unsigned long pfn, unsigned long zone_end,
349 unsigned long *nr_initialised)
350{
351 return true;
352}
353#endif
354
Mel Gorman0b423ca2016-05-19 17:14:27 -0700355/* Return a pointer to the bitmap storing bits affecting a block of pages */
356static inline unsigned long *get_pageblock_bitmap(struct page *page,
357 unsigned long pfn)
358{
359#ifdef CONFIG_SPARSEMEM
360 return __pfn_to_section(pfn)->pageblock_flags;
361#else
362 return page_zone(page)->pageblock_flags;
363#endif /* CONFIG_SPARSEMEM */
364}
365
366static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
367{
368#ifdef CONFIG_SPARSEMEM
369 pfn &= (PAGES_PER_SECTION-1);
370 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
371#else
372 pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
373 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
374#endif /* CONFIG_SPARSEMEM */
375}
376
377/**
378 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
379 * @page: The page within the block of interest
380 * @pfn: The target page frame number
381 * @end_bitidx: The last bit of interest to retrieve
382 * @mask: mask of bits that the caller is interested in
383 *
384 * Return: pageblock_bits flags
385 */
386static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page,
387 unsigned long pfn,
388 unsigned long end_bitidx,
389 unsigned long mask)
390{
391 unsigned long *bitmap;
392 unsigned long bitidx, word_bitidx;
393 unsigned long word;
394
395 bitmap = get_pageblock_bitmap(page, pfn);
396 bitidx = pfn_to_bitidx(page, pfn);
397 word_bitidx = bitidx / BITS_PER_LONG;
398 bitidx &= (BITS_PER_LONG-1);
399
400 word = bitmap[word_bitidx];
401 bitidx += end_bitidx;
402 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
403}
404
405unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
406 unsigned long end_bitidx,
407 unsigned long mask)
408{
409 return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask);
410}
411
412static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
413{
414 return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK);
415}
416
417/**
418 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
419 * @page: The page within the block of interest
420 * @flags: The flags to set
421 * @pfn: The target page frame number
422 * @end_bitidx: The last bit of interest
423 * @mask: mask of bits that the caller is interested in
424 */
425void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
426 unsigned long pfn,
427 unsigned long end_bitidx,
428 unsigned long mask)
429{
430 unsigned long *bitmap;
431 unsigned long bitidx, word_bitidx;
432 unsigned long old_word, word;
433
434 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
435
436 bitmap = get_pageblock_bitmap(page, pfn);
437 bitidx = pfn_to_bitidx(page, pfn);
438 word_bitidx = bitidx / BITS_PER_LONG;
439 bitidx &= (BITS_PER_LONG-1);
440
441 VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
442
443 bitidx += end_bitidx;
444 mask <<= (BITS_PER_LONG - bitidx - 1);
445 flags <<= (BITS_PER_LONG - bitidx - 1);
446
447 word = READ_ONCE(bitmap[word_bitidx]);
448 for (;;) {
449 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
450 if (word == old_word)
451 break;
452 word = old_word;
453 }
454}
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700455
Minchan Kimee6f5092012-07-31 16:43:50 -0700456void set_pageblock_migratetype(struct page *page, int migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700457{
KOSAKI Motohiro5d0f3f72013-11-12 15:08:18 -0800458 if (unlikely(page_group_by_mobility_disabled &&
459 migratetype < MIGRATE_PCPTYPES))
Mel Gorman49255c62009-06-16 15:31:58 -0700460 migratetype = MIGRATE_UNMOVABLE;
461
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700462 set_pageblock_flags_group(page, (unsigned long)migratetype,
463 PB_migrate, PB_migrate_end);
464}
465
Nick Piggin13e74442006-01-06 00:10:58 -0800466#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700467static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700469 int ret = 0;
470 unsigned seq;
471 unsigned long pfn = page_to_pfn(page);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800472 unsigned long sp, start_pfn;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700473
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700474 do {
475 seq = zone_span_seqbegin(zone);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800476 start_pfn = zone->zone_start_pfn;
477 sp = zone->spanned_pages;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800478 if (!zone_spans_pfn(zone, pfn))
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700479 ret = 1;
480 } while (zone_span_seqretry(zone, seq));
481
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800482 if (ret)
Dave Hansen613813e2014-06-04 16:07:27 -0700483 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
484 pfn, zone_to_nid(zone), zone->name,
485 start_pfn, start_pfn + sp);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800486
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700487 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700488}
489
490static int page_is_consistent(struct zone *zone, struct page *page)
491{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700492 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700493 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700495 return 0;
496
497 return 1;
498}
499/*
500 * Temporary debugging check for pages not lying within a given zone.
501 */
502static int bad_range(struct zone *zone, struct page *page)
503{
504 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700506 if (!page_is_consistent(zone, page))
507 return 1;
508
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 return 0;
510}
Nick Piggin13e74442006-01-06 00:10:58 -0800511#else
512static inline int bad_range(struct zone *zone, struct page *page)
513{
514 return 0;
515}
516#endif
517
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700518static void bad_page(struct page *page, const char *reason,
519 unsigned long bad_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800521 static unsigned long resume;
522 static unsigned long nr_shown;
523 static unsigned long nr_unshown;
524
525 /*
526 * Allow a burst of 60 reports, then keep quiet for that minute;
527 * or allow a steady drip of one report per second.
528 */
529 if (nr_shown == 60) {
530 if (time_before(jiffies, resume)) {
531 nr_unshown++;
532 goto out;
533 }
534 if (nr_unshown) {
Vlastimil Babkaff8e8112016-03-15 14:56:24 -0700535 pr_alert(
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800536 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800537 nr_unshown);
538 nr_unshown = 0;
539 }
540 nr_shown = 0;
541 }
542 if (nr_shown++ == 0)
543 resume = jiffies + 60 * HZ;
544
Vlastimil Babkaff8e8112016-03-15 14:56:24 -0700545 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800546 current->comm, page_to_pfn(page));
Vlastimil Babkaff8e8112016-03-15 14:56:24 -0700547 __dump_page(page, reason);
548 bad_flags &= page->flags;
549 if (bad_flags)
550 pr_alert("bad because of flags: %#lx(%pGp)\n",
551 bad_flags, &bad_flags);
Vlastimil Babka4e462112016-03-15 14:56:21 -0700552 dump_page_owner(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700553
Dave Jones4f318882011-10-31 17:07:24 -0700554 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800556out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800557 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800558 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030559 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560}
561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562/*
563 * Higher-order pages are called "compound pages". They are structured thusly:
564 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800565 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800567 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
568 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800570 * The first tail page's ->compound_dtor holds the offset in array of compound
571 * page destructors. See compound_page_dtors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800573 * The first tail page's ->compound_order holds the order of allocation.
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800574 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800576
Kirill A. Shutemov9a982252016-01-15 16:54:17 -0800577void free_compound_page(struct page *page)
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800578{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700579 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800580}
581
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -0800582void prep_compound_page(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583{
584 int i;
585 int nr_pages = 1 << order;
586
Kirill A. Shutemovf1e61552015-11-06 16:29:50 -0800587 set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700588 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700589 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800590 for (i = 1; i < nr_pages; i++) {
591 struct page *p = page + i;
Youquan Song58a84aa2011-12-08 14:34:18 -0800592 set_page_count(p, 0);
Kirill A. Shutemov1c290f62016-01-15 16:52:07 -0800593 p->mapping = TAIL_MAPPING;
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800594 set_compound_head(p, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 }
Kirill A. Shutemov53f92632016-01-15 16:53:42 -0800596 atomic_set(compound_mapcount_ptr(page), -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597}
598
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800599#ifdef CONFIG_DEBUG_PAGEALLOC
600unsigned int _debug_guardpage_minorder;
Christian Borntraegerea6eabb2016-03-15 14:55:30 -0700601bool _debug_pagealloc_enabled __read_mostly
602 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
Joonsoo Kim505f6d22016-03-17 14:17:56 -0700603EXPORT_SYMBOL(_debug_pagealloc_enabled);
Joonsoo Kime30825f2014-12-12 16:55:49 -0800604bool _debug_guardpage_enabled __read_mostly;
605
Joonsoo Kim031bc572014-12-12 16:55:52 -0800606static int __init early_debug_pagealloc(char *buf)
607{
608 if (!buf)
609 return -EINVAL;
Minfei Huang2a138dc2016-05-20 16:58:13 -0700610 return kstrtobool(buf, &_debug_pagealloc_enabled);
Joonsoo Kim031bc572014-12-12 16:55:52 -0800611}
612early_param("debug_pagealloc", early_debug_pagealloc);
613
Joonsoo Kime30825f2014-12-12 16:55:49 -0800614static bool need_debug_guardpage(void)
615{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800616 /* If we don't use debug_pagealloc, we don't need guard page */
617 if (!debug_pagealloc_enabled())
618 return false;
619
Joonsoo Kimf1c1e9f2016-10-07 16:58:18 -0700620 if (!debug_guardpage_minorder())
621 return false;
622
Joonsoo Kime30825f2014-12-12 16:55:49 -0800623 return true;
624}
625
626static void init_debug_guardpage(void)
627{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800628 if (!debug_pagealloc_enabled())
629 return;
630
Joonsoo Kimf1c1e9f2016-10-07 16:58:18 -0700631 if (!debug_guardpage_minorder())
632 return;
633
Joonsoo Kime30825f2014-12-12 16:55:49 -0800634 _debug_guardpage_enabled = true;
635}
636
637struct page_ext_operations debug_guardpage_ops = {
638 .need = need_debug_guardpage,
639 .init = init_debug_guardpage,
640};
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800641
642static int __init debug_guardpage_minorder_setup(char *buf)
643{
644 unsigned long res;
645
646 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
Joe Perches11705322016-03-17 14:19:50 -0700647 pr_err("Bad debug_guardpage_minorder value\n");
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800648 return 0;
649 }
650 _debug_guardpage_minorder = res;
Joe Perches11705322016-03-17 14:19:50 -0700651 pr_info("Setting debug_guardpage_minorder to %lu\n", res);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800652 return 0;
653}
Joonsoo Kimf1c1e9f2016-10-07 16:58:18 -0700654early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800655
Joonsoo Kimacbc15a2016-10-07 16:58:15 -0700656static inline bool set_page_guard(struct zone *zone, struct page *page,
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800657 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800658{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800659 struct page_ext *page_ext;
660
661 if (!debug_guardpage_enabled())
Joonsoo Kimacbc15a2016-10-07 16:58:15 -0700662 return false;
663
664 if (order >= debug_guardpage_minorder())
665 return false;
Joonsoo Kime30825f2014-12-12 16:55:49 -0800666
667 page_ext = lookup_page_ext(page);
Yang Shif86e4272016-06-03 14:55:38 -0700668 if (unlikely(!page_ext))
Joonsoo Kimacbc15a2016-10-07 16:58:15 -0700669 return false;
Yang Shif86e4272016-06-03 14:55:38 -0700670
Joonsoo Kime30825f2014-12-12 16:55:49 -0800671 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
672
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800673 INIT_LIST_HEAD(&page->lru);
674 set_page_private(page, order);
675 /* Guard pages are not available for any usage */
676 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
Joonsoo Kimacbc15a2016-10-07 16:58:15 -0700677
678 return true;
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800679}
680
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800681static inline void clear_page_guard(struct zone *zone, struct page *page,
682 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800683{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800684 struct page_ext *page_ext;
685
686 if (!debug_guardpage_enabled())
687 return;
688
689 page_ext = lookup_page_ext(page);
Yang Shif86e4272016-06-03 14:55:38 -0700690 if (unlikely(!page_ext))
691 return;
692
Joonsoo Kime30825f2014-12-12 16:55:49 -0800693 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
694
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800695 set_page_private(page, 0);
696 if (!is_migrate_isolate(migratetype))
697 __mod_zone_freepage_state(zone, (1 << order), migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800698}
699#else
Joonsoo Kim980ac162016-10-07 16:58:27 -0700700struct page_ext_operations debug_guardpage_ops;
Joonsoo Kimacbc15a2016-10-07 16:58:15 -0700701static inline bool set_page_guard(struct zone *zone, struct page *page,
702 unsigned int order, int migratetype) { return false; }
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800703static inline void clear_page_guard(struct zone *zone, struct page *page,
704 unsigned int order, int migratetype) {}
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800705#endif
706
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700707static inline void set_page_order(struct page *page, unsigned int order)
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700708{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700709 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000710 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711}
712
713static inline void rmv_page_order(struct page *page)
714{
Nick Piggin676165a2006-04-10 11:21:48 +1000715 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700716 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
718
719/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 * This function checks whether a page is free && is the buddy
721 * we can do coalesce a page and its buddy if
Vlastimil Babka13ad59d2017-02-22 15:41:51 -0800722 * (a) the buddy is not in a hole (check before calling!) &&
Nick Piggin676165a2006-04-10 11:21:48 +1000723 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700724 * (c) a page and its buddy have the same order &&
725 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 *
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700727 * For recording whether a page is in the buddy system, we set ->_mapcount
728 * PAGE_BUDDY_MAPCOUNT_VALUE.
729 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
730 * serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000731 *
732 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700734static inline int page_is_buddy(struct page *page, struct page *buddy,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700735 unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736{
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800737 if (page_is_guard(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700738 if (page_zone_id(page) != page_zone_id(buddy))
739 return 0;
740
Weijie Yang4c5018c2015-02-10 14:11:39 -0800741 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
742
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800743 return 1;
744 }
745
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700746 if (PageBuddy(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700747 /*
748 * zone check is done late to avoid uselessly
749 * calculating zone/node ids for pages that could
750 * never merge.
751 */
752 if (page_zone_id(page) != page_zone_id(buddy))
753 return 0;
754
Weijie Yang4c5018c2015-02-10 14:11:39 -0800755 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
756
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700757 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000758 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700759 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760}
761
762/*
763 * Freeing function for a buddy system allocator.
764 *
765 * The concept of a buddy system is to maintain direct-mapped table
766 * (containing bit values) for memory blocks of various "orders".
767 * The bottom level table contains the map for the smallest allocatable
768 * units of memory (here, pages), and each level above it describes
769 * pairs of units from the levels below, hence, "buddies".
770 * At a high level, all that happens here is marking the table entry
771 * at the bottom level available, and propagating the changes upward
772 * as necessary, plus some accounting needed to play nicely with other
773 * parts of the VM system.
774 * At each level, we keep a list of pages, which are heads of continuous
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700775 * free pages of length of (1 << order) and marked with _mapcount
776 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
777 * field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100779 * other. That is, if we allocate a small block, and both were
780 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100782 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100784 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 */
786
Nick Piggin48db57f2006-01-08 01:00:42 -0800787static inline void __free_one_page(struct page *page,
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700788 unsigned long pfn,
Mel Gormaned0ae212009-06-16 15:32:07 -0700789 struct zone *zone, unsigned int order,
790 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791{
Vlastimil Babka76741e72017-02-22 15:41:48 -0800792 unsigned long combined_pfn;
793 unsigned long uninitialized_var(buddy_pfn);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700794 struct page *buddy;
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700795 unsigned int max_order;
796
797 max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
Cody P Schaferd29bb972013-02-22 16:35:25 -0800799 VM_BUG_ON(!zone_is_initialized(zone));
Kirill A. Shutemov6e9f0d52015-02-11 15:25:50 -0800800 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
Mel Gormaned0ae212009-06-16 15:32:07 -0700802 VM_BUG_ON(migratetype == -1);
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700803 if (likely(!is_migrate_isolate(migratetype)))
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800804 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Mel Gormaned0ae212009-06-16 15:32:07 -0700805
Vlastimil Babka76741e72017-02-22 15:41:48 -0800806 VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800807 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700809continue_merging:
Joonsoo Kim3c605092014-11-13 15:19:21 -0800810 while (order < max_order - 1) {
Vlastimil Babka76741e72017-02-22 15:41:48 -0800811 buddy_pfn = __find_buddy_pfn(pfn, order);
812 buddy = page + (buddy_pfn - pfn);
Vlastimil Babka13ad59d2017-02-22 15:41:51 -0800813
814 if (!pfn_valid_within(buddy_pfn))
815 goto done_merging;
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700816 if (!page_is_buddy(page, buddy, order))
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700817 goto done_merging;
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800818 /*
819 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
820 * merge with it and move up one order.
821 */
822 if (page_is_guard(buddy)) {
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800823 clear_page_guard(zone, buddy, order, migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800824 } else {
825 list_del(&buddy->lru);
826 zone->free_area[order].nr_free--;
827 rmv_page_order(buddy);
828 }
Vlastimil Babka76741e72017-02-22 15:41:48 -0800829 combined_pfn = buddy_pfn & pfn;
830 page = page + (combined_pfn - pfn);
831 pfn = combined_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 order++;
833 }
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700834 if (max_order < MAX_ORDER) {
835 /* If we are here, it means order is >= pageblock_order.
836 * We want to prevent merge between freepages on isolate
837 * pageblock and normal pageblock. Without this, pageblock
838 * isolation could cause incorrect freepage or CMA accounting.
839 *
840 * We don't want to hit this code for the more frequent
841 * low-order merging.
842 */
843 if (unlikely(has_isolate_pageblock(zone))) {
844 int buddy_mt;
845
Vlastimil Babka76741e72017-02-22 15:41:48 -0800846 buddy_pfn = __find_buddy_pfn(pfn, order);
847 buddy = page + (buddy_pfn - pfn);
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700848 buddy_mt = get_pageblock_migratetype(buddy);
849
850 if (migratetype != buddy_mt
851 && (is_migrate_isolate(migratetype) ||
852 is_migrate_isolate(buddy_mt)))
853 goto done_merging;
854 }
855 max_order++;
856 goto continue_merging;
857 }
858
859done_merging:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700861
862 /*
863 * If this is not the largest possible page, check if the buddy
864 * of the next-highest order is free. If it is, it's possible
865 * that pages are being freed that will coalesce soon. In case,
866 * that is happening, add the free page to the tail of the list
867 * so it's less likely to be used soon and more likely to be merged
868 * as a higher order page
869 */
Vlastimil Babka13ad59d2017-02-22 15:41:51 -0800870 if ((order < MAX_ORDER-2) && pfn_valid_within(buddy_pfn)) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700871 struct page *higher_page, *higher_buddy;
Vlastimil Babka76741e72017-02-22 15:41:48 -0800872 combined_pfn = buddy_pfn & pfn;
873 higher_page = page + (combined_pfn - pfn);
874 buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1);
875 higher_buddy = higher_page + (buddy_pfn - combined_pfn);
Tony Luckb4fb8f62017-03-08 09:35:39 -0800876 if (pfn_valid_within(buddy_pfn) &&
877 page_is_buddy(higher_page, higher_buddy, order + 1)) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700878 list_add_tail(&page->lru,
879 &zone->free_area[order].free_list[migratetype]);
880 goto out;
881 }
882 }
883
884 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
885out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 zone->free_area[order].nr_free++;
887}
888
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700889/*
890 * A bad page could be due to a number of fields. Instead of multiple branches,
891 * try and check multiple fields with one check. The caller must do a detailed
892 * check if necessary.
893 */
894static inline bool page_expected_state(struct page *page,
895 unsigned long check_flags)
896{
897 if (unlikely(atomic_read(&page->_mapcount) != -1))
898 return false;
899
900 if (unlikely((unsigned long)page->mapping |
901 page_ref_count(page) |
902#ifdef CONFIG_MEMCG
903 (unsigned long)page->mem_cgroup |
904#endif
905 (page->flags & check_flags)))
906 return false;
907
908 return true;
909}
910
Mel Gormanbb552ac2016-05-19 17:14:18 -0700911static void free_pages_check_bad(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912{
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700913 const char *bad_reason;
914 unsigned long bad_flags;
915
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700916 bad_reason = NULL;
917 bad_flags = 0;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800918
Kirill A. Shutemov53f92632016-01-15 16:53:42 -0800919 if (unlikely(atomic_read(&page->_mapcount) != -1))
Dave Hansenf0b791a2014-01-23 15:52:49 -0800920 bad_reason = "nonzero mapcount";
921 if (unlikely(page->mapping != NULL))
922 bad_reason = "non-NULL mapping";
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700923 if (unlikely(page_ref_count(page) != 0))
Joonsoo Kim0139aa72016-05-19 17:10:49 -0700924 bad_reason = "nonzero _refcount";
Dave Hansenf0b791a2014-01-23 15:52:49 -0800925 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
926 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
927 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
928 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -0800929#ifdef CONFIG_MEMCG
930 if (unlikely(page->mem_cgroup))
931 bad_reason = "page still charged to cgroup";
932#endif
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700933 bad_page(page, bad_reason, bad_flags);
Mel Gormanbb552ac2016-05-19 17:14:18 -0700934}
935
936static inline int free_pages_check(struct page *page)
937{
Mel Gormanda838d42016-05-19 17:14:21 -0700938 if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
Mel Gormanbb552ac2016-05-19 17:14:18 -0700939 return 0;
Mel Gormanbb552ac2016-05-19 17:14:18 -0700940
941 /* Something has gone sideways, find it */
942 free_pages_check_bad(page);
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700943 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944}
945
Mel Gorman4db75482016-05-19 17:14:32 -0700946static int free_tail_pages_check(struct page *head_page, struct page *page)
947{
948 int ret = 1;
949
950 /*
951 * We rely page->lru.next never has bit 0 set, unless the page
952 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
953 */
954 BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
955
956 if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
957 ret = 0;
958 goto out;
959 }
960 switch (page - head_page) {
961 case 1:
962 /* the first tail page: ->mapping is compound_mapcount() */
963 if (unlikely(compound_mapcount(page))) {
964 bad_page(page, "nonzero compound_mapcount", 0);
965 goto out;
966 }
967 break;
968 case 2:
969 /*
970 * the second tail page: ->mapping is
971 * page_deferred_list().next -- ignore value.
972 */
973 break;
974 default:
975 if (page->mapping != TAIL_MAPPING) {
976 bad_page(page, "corrupted mapping in tail page", 0);
977 goto out;
978 }
979 break;
980 }
981 if (unlikely(!PageTail(page))) {
982 bad_page(page, "PageTail not set", 0);
983 goto out;
984 }
985 if (unlikely(compound_head(page) != head_page)) {
986 bad_page(page, "compound_head not consistent", 0);
987 goto out;
988 }
989 ret = 0;
990out:
991 page->mapping = NULL;
992 clear_compound_head(page);
993 return ret;
994}
995
Mel Gormane2769db2016-05-19 17:14:38 -0700996static __always_inline bool free_pages_prepare(struct page *page,
997 unsigned int order, bool check_free)
998{
999 int bad = 0;
1000
1001 VM_BUG_ON_PAGE(PageTail(page), page);
1002
1003 trace_mm_page_free(page, order);
1004 kmemcheck_free_shadow(page, order);
Mel Gormane2769db2016-05-19 17:14:38 -07001005
1006 /*
1007 * Check tail pages before head page information is cleared to
1008 * avoid checking PageCompound for order-0 pages.
1009 */
1010 if (unlikely(order)) {
1011 bool compound = PageCompound(page);
1012 int i;
1013
1014 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1015
Kirill A. Shutemov9a73f612016-07-26 15:25:53 -07001016 if (compound)
1017 ClearPageDoubleMap(page);
Mel Gormane2769db2016-05-19 17:14:38 -07001018 for (i = 1; i < (1 << order); i++) {
1019 if (compound)
1020 bad += free_tail_pages_check(page, page + i);
1021 if (unlikely(free_pages_check(page + i))) {
1022 bad++;
1023 continue;
1024 }
1025 (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1026 }
1027 }
Minchan Kimbda807d2016-07-26 15:23:05 -07001028 if (PageMappingFlags(page))
Mel Gormane2769db2016-05-19 17:14:38 -07001029 page->mapping = NULL;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03001030 if (memcg_kmem_enabled() && PageKmemcg(page))
Vladimir Davydov49491482016-07-26 15:24:24 -07001031 memcg_kmem_uncharge(page, order);
Mel Gormane2769db2016-05-19 17:14:38 -07001032 if (check_free)
1033 bad += free_pages_check(page);
1034 if (bad)
1035 return false;
1036
1037 page_cpupid_reset_last(page);
1038 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1039 reset_page_owner(page, order);
1040
1041 if (!PageHighMem(page)) {
1042 debug_check_no_locks_freed(page_address(page),
1043 PAGE_SIZE << order);
1044 debug_check_no_obj_freed(page_address(page),
1045 PAGE_SIZE << order);
1046 }
1047 arch_free_page(page, order);
1048 kernel_poison_pages(page, 1 << order, 0);
1049 kernel_map_pages(page, 1 << order, 0);
seokhoon.yoon29b52de2016-05-20 16:58:47 -07001050 kasan_free_pages(page, order);
Mel Gormane2769db2016-05-19 17:14:38 -07001051
1052 return true;
1053}
Mel Gorman4db75482016-05-19 17:14:32 -07001054
1055#ifdef CONFIG_DEBUG_VM
1056static inline bool free_pcp_prepare(struct page *page)
1057{
Mel Gormane2769db2016-05-19 17:14:38 -07001058 return free_pages_prepare(page, 0, true);
Mel Gorman4db75482016-05-19 17:14:32 -07001059}
1060
1061static inline bool bulkfree_pcp_prepare(struct page *page)
1062{
1063 return false;
1064}
1065#else
1066static bool free_pcp_prepare(struct page *page)
1067{
Mel Gormane2769db2016-05-19 17:14:38 -07001068 return free_pages_prepare(page, 0, false);
Mel Gorman4db75482016-05-19 17:14:32 -07001069}
1070
1071static bool bulkfree_pcp_prepare(struct page *page)
1072{
1073 return free_pages_check(page);
1074}
1075#endif /* CONFIG_DEBUG_VM */
1076
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001078 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -07001080 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 *
1082 * If the zone was previously in an "all pages pinned" state then look to
1083 * see if this freeing clears that state.
1084 *
1085 * And clear the zone's pages_scanned counter, to hold off the "all pages are
1086 * pinned" detection logic.
1087 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001088static void free_pcppages_bulk(struct zone *zone, int count,
1089 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090{
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001091 int migratetype = 0;
Mel Gormana6f9edd62009-09-21 17:03:20 -07001092 int batch_free = 0;
Mel Gorman374ad052017-02-24 14:56:38 -08001093 unsigned long nr_scanned, flags;
Mel Gorman37779992016-05-19 17:13:58 -07001094 bool isolated_pageblocks;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001095
Mel Gorman374ad052017-02-24 14:56:38 -08001096 spin_lock_irqsave(&zone->lock, flags);
Mel Gorman37779992016-05-19 17:13:58 -07001097 isolated_pageblocks = has_isolate_pageblock(zone);
Mel Gorman599d0c92016-07-28 15:45:31 -07001098 nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED);
Mel Gorman0d5d8232014-08-06 16:07:16 -07001099 if (nr_scanned)
Mel Gorman599d0c92016-07-28 15:45:31 -07001100 __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -07001101
Mel Gormane5b31ac2016-05-19 17:14:24 -07001102 while (count) {
Nick Piggin48db57f2006-01-08 01:00:42 -08001103 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001104 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -08001105
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001106 /*
Mel Gormana6f9edd62009-09-21 17:03:20 -07001107 * Remove pages from lists in a round-robin fashion. A
1108 * batch_free count is maintained that is incremented when an
1109 * empty list is encountered. This is so more pages are freed
1110 * off fuller lists instead of spinning excessively around empty
1111 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001112 */
1113 do {
Mel Gormana6f9edd62009-09-21 17:03:20 -07001114 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001115 if (++migratetype == MIGRATE_PCPTYPES)
1116 migratetype = 0;
1117 list = &pcp->lists[migratetype];
1118 } while (list_empty(list));
1119
Namhyung Kim1d168712011-03-22 16:32:45 -07001120 /* This is the only non-empty list. Free them all. */
1121 if (batch_free == MIGRATE_PCPTYPES)
Mel Gormane5b31ac2016-05-19 17:14:24 -07001122 batch_free = count;
Namhyung Kim1d168712011-03-22 16:32:45 -07001123
Mel Gormana6f9edd62009-09-21 17:03:20 -07001124 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -07001125 int mt; /* migratetype of the to-be-freed page */
1126
Geliang Tanga16601c2016-01-14 15:20:30 -08001127 page = list_last_entry(list, struct page, lru);
Mel Gormana6f9edd62009-09-21 17:03:20 -07001128 /* must delete as __free_one_page list manipulates */
1129 list_del(&page->lru);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -07001130
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001131 mt = get_pcppage_migratetype(page);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -07001132 /* MIGRATE_ISOLATE page should not go to pcplists */
1133 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1134 /* Pageblock could have been isolated meanwhile */
Mel Gorman37779992016-05-19 17:13:58 -07001135 if (unlikely(isolated_pageblocks))
Joonsoo Kim51bb1a42014-11-13 15:19:14 -08001136 mt = get_pageblock_migratetype(page);
Joonsoo Kim51bb1a42014-11-13 15:19:14 -08001137
Mel Gorman4db75482016-05-19 17:14:32 -07001138 if (bulkfree_pcp_prepare(page))
1139 continue;
1140
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001141 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -07001142 trace_mm_page_pcpu_drain(page, 0, mt);
Mel Gormane5b31ac2016-05-19 17:14:24 -07001143 } while (--count && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 }
Mel Gorman374ad052017-02-24 14:56:38 -08001145 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146}
1147
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001148static void free_one_page(struct zone *zone,
1149 struct page *page, unsigned long pfn,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001150 unsigned int order,
Mel Gormaned0ae212009-06-16 15:32:07 -07001151 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -08001152{
Mel Gorman374ad052017-02-24 14:56:38 -08001153 unsigned long nr_scanned, flags;
1154 spin_lock_irqsave(&zone->lock, flags);
1155 __count_vm_events(PGFREE, 1 << order);
Mel Gorman599d0c92016-07-28 15:45:31 -07001156 nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED);
Mel Gorman0d5d8232014-08-06 16:07:16 -07001157 if (nr_scanned)
Mel Gorman599d0c92016-07-28 15:45:31 -07001158 __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -07001159
Joonsoo Kimad53f922014-11-13 15:19:11 -08001160 if (unlikely(has_isolate_pageblock(zone) ||
1161 is_migrate_isolate(migratetype))) {
1162 migratetype = get_pfnblock_migratetype(page, pfn);
Joonsoo Kimad53f922014-11-13 15:19:11 -08001163 }
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001164 __free_one_page(page, pfn, zone, order, migratetype);
Mel Gorman374ad052017-02-24 14:56:38 -08001165 spin_unlock_irqrestore(&zone->lock, flags);
Nick Piggin48db57f2006-01-08 01:00:42 -08001166}
1167
Robin Holt1e8ce832015-06-30 14:56:45 -07001168static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1169 unsigned long zone, int nid)
1170{
Robin Holt1e8ce832015-06-30 14:56:45 -07001171 set_page_links(page, zone, nid, pfn);
Robin Holt1e8ce832015-06-30 14:56:45 -07001172 init_page_count(page);
1173 page_mapcount_reset(page);
1174 page_cpupid_reset_last(page);
Robin Holt1e8ce832015-06-30 14:56:45 -07001175
Robin Holt1e8ce832015-06-30 14:56:45 -07001176 INIT_LIST_HEAD(&page->lru);
1177#ifdef WANT_PAGE_VIRTUAL
1178 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1179 if (!is_highmem_idx(zone))
1180 set_page_address(page, __va(pfn << PAGE_SHIFT));
1181#endif
1182}
1183
1184static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
1185 int nid)
1186{
1187 return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
1188}
1189
Mel Gorman7e18adb2015-06-30 14:57:05 -07001190#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1191static void init_reserved_page(unsigned long pfn)
1192{
1193 pg_data_t *pgdat;
1194 int nid, zid;
1195
1196 if (!early_page_uninitialised(pfn))
1197 return;
1198
1199 nid = early_pfn_to_nid(pfn);
1200 pgdat = NODE_DATA(nid);
1201
1202 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1203 struct zone *zone = &pgdat->node_zones[zid];
1204
1205 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1206 break;
1207 }
1208 __init_single_pfn(pfn, zid, nid);
1209}
1210#else
1211static inline void init_reserved_page(unsigned long pfn)
1212{
1213}
1214#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1215
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001216/*
1217 * Initialised pages do not have PageReserved set. This function is
1218 * called for each range allocated by the bootmem allocator and
1219 * marks the pages PageReserved. The remaining valid pages are later
1220 * sent to the buddy page allocator.
1221 */
Stefan Bader4b50bcc2016-05-20 16:58:38 -07001222void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001223{
1224 unsigned long start_pfn = PFN_DOWN(start);
1225 unsigned long end_pfn = PFN_UP(end);
1226
Mel Gorman7e18adb2015-06-30 14:57:05 -07001227 for (; start_pfn < end_pfn; start_pfn++) {
1228 if (pfn_valid(start_pfn)) {
1229 struct page *page = pfn_to_page(start_pfn);
1230
1231 init_reserved_page(start_pfn);
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -08001232
1233 /* Avoid false-positive PageTail() */
1234 INIT_LIST_HEAD(&page->lru);
1235
Mel Gorman7e18adb2015-06-30 14:57:05 -07001236 SetPageReserved(page);
1237 }
1238 }
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001239}
1240
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001241static void __free_pages_ok(struct page *page, unsigned int order)
1242{
Minchan Kim95e34412012-10-08 16:32:11 -07001243 int migratetype;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001244 unsigned long pfn = page_to_pfn(page);
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001245
Mel Gormane2769db2016-05-19 17:14:38 -07001246 if (!free_pages_prepare(page, order, true))
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001247 return;
1248
Mel Gormancfc47a22014-06-04 16:10:19 -07001249 migratetype = get_pfnblock_migratetype(page, pfn);
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001250 free_one_page(page_zone(page), page, pfn, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251}
1252
Li Zhang949698a2016-05-19 17:11:37 -07001253static void __init __free_pages_boot_core(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -08001254{
Johannes Weinerc3993072012-01-10 15:08:10 -08001255 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001256 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -08001257 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -08001258
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001259 prefetchw(p);
1260 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1261 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -08001262 __ClearPageReserved(p);
1263 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -08001264 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001265 __ClearPageReserved(p);
1266 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -08001267
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001268 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -08001269 set_page_refcounted(page);
1270 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -08001271}
1272
Mel Gorman75a592a2015-06-30 14:56:59 -07001273#if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1274 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
Mel Gorman7ace9912015-08-06 15:46:13 -07001275
Mel Gorman75a592a2015-06-30 14:56:59 -07001276static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1277
1278int __meminit early_pfn_to_nid(unsigned long pfn)
1279{
Mel Gorman7ace9912015-08-06 15:46:13 -07001280 static DEFINE_SPINLOCK(early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -07001281 int nid;
1282
Mel Gorman7ace9912015-08-06 15:46:13 -07001283 spin_lock(&early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -07001284 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
Mel Gorman7ace9912015-08-06 15:46:13 -07001285 if (nid < 0)
Mel Gormane4568d32016-07-14 12:07:20 -07001286 nid = first_online_node;
Mel Gorman7ace9912015-08-06 15:46:13 -07001287 spin_unlock(&early_pfn_lock);
1288
1289 return nid;
Mel Gorman75a592a2015-06-30 14:56:59 -07001290}
1291#endif
1292
1293#ifdef CONFIG_NODES_SPAN_OTHER_NODES
1294static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1295 struct mminit_pfnnid_cache *state)
1296{
1297 int nid;
1298
1299 nid = __early_pfn_to_nid(pfn, state);
1300 if (nid >= 0 && nid != node)
1301 return false;
1302 return true;
1303}
1304
1305/* Only safe to use early in boot when initialisation is single-threaded */
1306static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1307{
1308 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1309}
1310
1311#else
1312
1313static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1314{
1315 return true;
1316}
1317static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1318 struct mminit_pfnnid_cache *state)
1319{
1320 return true;
1321}
1322#endif
1323
1324
Mel Gorman0e1cc952015-06-30 14:57:27 -07001325void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001326 unsigned int order)
1327{
1328 if (early_page_uninitialised(pfn))
1329 return;
Li Zhang949698a2016-05-19 17:11:37 -07001330 return __free_pages_boot_core(page, order);
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001331}
1332
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001333/*
1334 * Check that the whole (or subset of) a pageblock given by the interval of
1335 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1336 * with the migration of free compaction scanner. The scanners then need to
1337 * use only pfn_valid_within() check for arches that allow holes within
1338 * pageblocks.
1339 *
1340 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1341 *
1342 * It's possible on some configurations to have a setup like node0 node1 node0
1343 * i.e. it's possible that all pages within a zones range of pages do not
1344 * belong to a single zone. We assume that a border between node0 and node1
1345 * can occur within a single pageblock, but not a node0 node1 node0
1346 * interleaving within a single pageblock. It is therefore sufficient to check
1347 * the first and last page of a pageblock and avoid checking each individual
1348 * page in a pageblock.
1349 */
1350struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1351 unsigned long end_pfn, struct zone *zone)
1352{
1353 struct page *start_page;
1354 struct page *end_page;
1355
1356 /* end_pfn is one past the range we are checking */
1357 end_pfn--;
1358
1359 if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1360 return NULL;
1361
1362 start_page = pfn_to_page(start_pfn);
1363
1364 if (page_zone(start_page) != zone)
1365 return NULL;
1366
1367 end_page = pfn_to_page(end_pfn);
1368
1369 /* This gives a shorter code than deriving page_zone(end_page) */
1370 if (page_zone_id(start_page) != page_zone_id(end_page))
1371 return NULL;
1372
1373 return start_page;
1374}
1375
1376void set_zone_contiguous(struct zone *zone)
1377{
1378 unsigned long block_start_pfn = zone->zone_start_pfn;
1379 unsigned long block_end_pfn;
1380
1381 block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1382 for (; block_start_pfn < zone_end_pfn(zone);
1383 block_start_pfn = block_end_pfn,
1384 block_end_pfn += pageblock_nr_pages) {
1385
1386 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1387
1388 if (!__pageblock_pfn_to_page(block_start_pfn,
1389 block_end_pfn, zone))
1390 return;
1391 }
1392
1393 /* We confirm that there is no hole */
1394 zone->contiguous = true;
1395}
1396
1397void clear_zone_contiguous(struct zone *zone)
1398{
1399 zone->contiguous = false;
1400}
1401
Mel Gorman7e18adb2015-06-30 14:57:05 -07001402#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
Mel Gorman0e1cc952015-06-30 14:57:27 -07001403static void __init deferred_free_range(struct page *page,
Mel Gormana4de83d2015-06-30 14:57:16 -07001404 unsigned long pfn, int nr_pages)
1405{
1406 int i;
1407
1408 if (!page)
1409 return;
1410
1411 /* Free a large naturally-aligned chunk if possible */
Xishi Qiue7801492016-10-07 16:58:09 -07001412 if (nr_pages == pageblock_nr_pages &&
1413 (pfn & (pageblock_nr_pages - 1)) == 0) {
Mel Gormanac5d2532015-06-30 14:57:20 -07001414 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Xishi Qiue7801492016-10-07 16:58:09 -07001415 __free_pages_boot_core(page, pageblock_order);
Mel Gormana4de83d2015-06-30 14:57:16 -07001416 return;
1417 }
1418
Xishi Qiue7801492016-10-07 16:58:09 -07001419 for (i = 0; i < nr_pages; i++, page++, pfn++) {
1420 if ((pfn & (pageblock_nr_pages - 1)) == 0)
1421 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Li Zhang949698a2016-05-19 17:11:37 -07001422 __free_pages_boot_core(page, 0);
Xishi Qiue7801492016-10-07 16:58:09 -07001423 }
Mel Gormana4de83d2015-06-30 14:57:16 -07001424}
1425
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001426/* Completion tracking for deferred_init_memmap() threads */
1427static atomic_t pgdat_init_n_undone __initdata;
1428static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1429
1430static inline void __init pgdat_init_report_one_done(void)
1431{
1432 if (atomic_dec_and_test(&pgdat_init_n_undone))
1433 complete(&pgdat_init_all_done_comp);
1434}
Mel Gorman0e1cc952015-06-30 14:57:27 -07001435
Mel Gorman7e18adb2015-06-30 14:57:05 -07001436/* Initialise remaining memory on a node */
Mel Gorman0e1cc952015-06-30 14:57:27 -07001437static int __init deferred_init_memmap(void *data)
Mel Gorman7e18adb2015-06-30 14:57:05 -07001438{
Mel Gorman0e1cc952015-06-30 14:57:27 -07001439 pg_data_t *pgdat = data;
1440 int nid = pgdat->node_id;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001441 struct mminit_pfnnid_cache nid_init_state = { };
1442 unsigned long start = jiffies;
1443 unsigned long nr_pages = 0;
1444 unsigned long walk_start, walk_end;
1445 int i, zid;
1446 struct zone *zone;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001447 unsigned long first_init_pfn = pgdat->first_deferred_pfn;
Mel Gorman0e1cc952015-06-30 14:57:27 -07001448 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001449
Mel Gorman0e1cc952015-06-30 14:57:27 -07001450 if (first_init_pfn == ULONG_MAX) {
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001451 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001452 return 0;
1453 }
1454
1455 /* Bind memory initialisation thread to a local node if possible */
1456 if (!cpumask_empty(cpumask))
1457 set_cpus_allowed_ptr(current, cpumask);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001458
1459 /* Sanity check boundaries */
1460 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1461 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1462 pgdat->first_deferred_pfn = ULONG_MAX;
1463
1464 /* Only the highest zone is deferred so find it */
1465 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1466 zone = pgdat->node_zones + zid;
1467 if (first_init_pfn < zone_end_pfn(zone))
1468 break;
1469 }
1470
1471 for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1472 unsigned long pfn, end_pfn;
Mel Gorman54608c32015-06-30 14:57:09 -07001473 struct page *page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001474 struct page *free_base_page = NULL;
1475 unsigned long free_base_pfn = 0;
1476 int nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001477
1478 end_pfn = min(walk_end, zone_end_pfn(zone));
1479 pfn = first_init_pfn;
1480 if (pfn < walk_start)
1481 pfn = walk_start;
1482 if (pfn < zone->zone_start_pfn)
1483 pfn = zone->zone_start_pfn;
1484
1485 for (; pfn < end_pfn; pfn++) {
Mel Gorman54608c32015-06-30 14:57:09 -07001486 if (!pfn_valid_within(pfn))
Mel Gormana4de83d2015-06-30 14:57:16 -07001487 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001488
Mel Gorman54608c32015-06-30 14:57:09 -07001489 /*
1490 * Ensure pfn_valid is checked every
Xishi Qiue7801492016-10-07 16:58:09 -07001491 * pageblock_nr_pages for memory holes
Mel Gorman54608c32015-06-30 14:57:09 -07001492 */
Xishi Qiue7801492016-10-07 16:58:09 -07001493 if ((pfn & (pageblock_nr_pages - 1)) == 0) {
Mel Gorman54608c32015-06-30 14:57:09 -07001494 if (!pfn_valid(pfn)) {
1495 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001496 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001497 }
1498 }
1499
1500 if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1501 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001502 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001503 }
1504
1505 /* Minimise pfn page lookups and scheduler checks */
Xishi Qiue7801492016-10-07 16:58:09 -07001506 if (page && (pfn & (pageblock_nr_pages - 1)) != 0) {
Mel Gorman54608c32015-06-30 14:57:09 -07001507 page++;
1508 } else {
Mel Gormana4de83d2015-06-30 14:57:16 -07001509 nr_pages += nr_to_free;
1510 deferred_free_range(free_base_page,
1511 free_base_pfn, nr_to_free);
1512 free_base_page = NULL;
1513 free_base_pfn = nr_to_free = 0;
1514
Mel Gorman54608c32015-06-30 14:57:09 -07001515 page = pfn_to_page(pfn);
1516 cond_resched();
1517 }
Mel Gorman7e18adb2015-06-30 14:57:05 -07001518
1519 if (page->flags) {
1520 VM_BUG_ON(page_zone(page) != zone);
Mel Gormana4de83d2015-06-30 14:57:16 -07001521 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001522 }
1523
1524 __init_single_page(page, pfn, zid, nid);
Mel Gormana4de83d2015-06-30 14:57:16 -07001525 if (!free_base_page) {
1526 free_base_page = page;
1527 free_base_pfn = pfn;
1528 nr_to_free = 0;
1529 }
1530 nr_to_free++;
1531
1532 /* Where possible, batch up pages for a single free */
1533 continue;
1534free_range:
1535 /* Free the current block of pages to allocator */
1536 nr_pages += nr_to_free;
1537 deferred_free_range(free_base_page, free_base_pfn,
1538 nr_to_free);
1539 free_base_page = NULL;
1540 free_base_pfn = nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001541 }
Xishi Qiue7801492016-10-07 16:58:09 -07001542 /* Free the last block of pages to allocator */
1543 nr_pages += nr_to_free;
1544 deferred_free_range(free_base_page, free_base_pfn, nr_to_free);
Mel Gormana4de83d2015-06-30 14:57:16 -07001545
Mel Gorman7e18adb2015-06-30 14:57:05 -07001546 first_init_pfn = max(end_pfn, first_init_pfn);
1547 }
1548
1549 /* Sanity check that the next zone really is unpopulated */
1550 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1551
Mel Gorman0e1cc952015-06-30 14:57:27 -07001552 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
Mel Gorman7e18adb2015-06-30 14:57:05 -07001553 jiffies_to_msecs(jiffies - start));
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001554
1555 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001556 return 0;
1557}
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001558#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
Mel Gorman0e1cc952015-06-30 14:57:27 -07001559
1560void __init page_alloc_init_late(void)
1561{
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001562 struct zone *zone;
1563
1564#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
Mel Gorman0e1cc952015-06-30 14:57:27 -07001565 int nid;
1566
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001567 /* There will be num_node_state(N_MEMORY) threads */
1568 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
Mel Gorman0e1cc952015-06-30 14:57:27 -07001569 for_each_node_state(nid, N_MEMORY) {
Mel Gorman0e1cc952015-06-30 14:57:27 -07001570 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1571 }
1572
1573 /* Block until all are initialised */
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001574 wait_for_completion(&pgdat_init_all_done_comp);
Mel Gorman4248b0d2015-08-06 15:46:20 -07001575
1576 /* Reinit limits that are based on free pages after the kernel is up */
1577 files_maxfiles_init();
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001578#endif
1579
1580 for_each_populated_zone(zone)
1581 set_zone_contiguous(zone);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001582}
Mel Gorman7e18adb2015-06-30 14:57:05 -07001583
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001584#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +08001585/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001586void __init init_cma_reserved_pageblock(struct page *page)
1587{
1588 unsigned i = pageblock_nr_pages;
1589 struct page *p = page;
1590
1591 do {
1592 __ClearPageReserved(p);
1593 set_page_count(p, 0);
1594 } while (++p, --i);
1595
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001596 set_pageblock_migratetype(page, MIGRATE_CMA);
Michal Nazarewiczdc783272014-07-02 15:22:35 -07001597
1598 if (pageblock_order >= MAX_ORDER) {
1599 i = pageblock_nr_pages;
1600 p = page;
1601 do {
1602 set_page_refcounted(p);
1603 __free_pages(p, MAX_ORDER - 1);
1604 p += MAX_ORDER_NR_PAGES;
1605 } while (i -= MAX_ORDER_NR_PAGES);
1606 } else {
1607 set_page_refcounted(page);
1608 __free_pages(page, pageblock_order);
1609 }
1610
Jiang Liu3dcc0572013-07-03 15:03:21 -07001611 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001612}
1613#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614
1615/*
1616 * The order of subdivision here is critical for the IO subsystem.
1617 * Please do not alter this order without good reasons and regression
1618 * testing. Specifically, as large blocks of memory are subdivided,
1619 * the order in which smaller blocks are delivered depends on the order
1620 * they're subdivided in this function. This is the primary factor
1621 * influencing the order in which pages are delivered to the IO
1622 * subsystem according to empirical testing, and this is also justified
1623 * by considering the behavior of a buddy system containing a single
1624 * large block of memory acted on by a series of small allocations.
1625 * This behavior is a critical factor in sglist merging's success.
1626 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +01001627 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 */
Nick Piggin085cc7d2006-01-06 00:11:01 -08001629static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001630 int low, int high, struct free_area *area,
1631 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632{
1633 unsigned long size = 1 << high;
1634
1635 while (high > low) {
1636 area--;
1637 high--;
1638 size >>= 1;
Sasha Levin309381fea2014-01-23 15:52:54 -08001639 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08001640
Joonsoo Kimacbc15a2016-10-07 16:58:15 -07001641 /*
1642 * Mark as guard pages (or page), that will allow to
1643 * merge back to allocator when buddy will be freed.
1644 * Corresponding page table entries will not be touched,
1645 * pages will stay not present in virtual address space
1646 */
1647 if (set_page_guard(zone, &page[size], high, migratetype))
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08001648 continue;
Joonsoo Kimacbc15a2016-10-07 16:58:15 -07001649
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001650 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 area->nr_free++;
1652 set_page_order(&page[size], high);
1653 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654}
1655
Vlastimil Babka4e611802016-05-19 17:14:41 -07001656static void check_new_page_bad(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657{
Vlastimil Babka4e611802016-05-19 17:14:41 -07001658 const char *bad_reason = NULL;
1659 unsigned long bad_flags = 0;
Dave Hansenf0b791a2014-01-23 15:52:49 -08001660
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001661 if (unlikely(atomic_read(&page->_mapcount) != -1))
Dave Hansenf0b791a2014-01-23 15:52:49 -08001662 bad_reason = "nonzero mapcount";
1663 if (unlikely(page->mapping != NULL))
1664 bad_reason = "non-NULL mapping";
Joonsoo Kimfe896d12016-03-17 14:19:26 -07001665 if (unlikely(page_ref_count(page) != 0))
Dave Hansenf0b791a2014-01-23 15:52:49 -08001666 bad_reason = "nonzero _count";
Naoya Horiguchif4c18e62015-08-06 15:47:08 -07001667 if (unlikely(page->flags & __PG_HWPOISON)) {
1668 bad_reason = "HWPoisoned (hardware-corrupted)";
1669 bad_flags = __PG_HWPOISON;
Naoya Horiguchie570f562016-05-20 16:58:50 -07001670 /* Don't complain about hwpoisoned pages */
1671 page_mapcount_reset(page); /* remove PageBuddy */
1672 return;
Naoya Horiguchif4c18e62015-08-06 15:47:08 -07001673 }
Dave Hansenf0b791a2014-01-23 15:52:49 -08001674 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1675 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1676 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1677 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -08001678#ifdef CONFIG_MEMCG
1679 if (unlikely(page->mem_cgroup))
1680 bad_reason = "page still charged to cgroup";
1681#endif
Vlastimil Babka4e611802016-05-19 17:14:41 -07001682 bad_page(page, bad_reason, bad_flags);
1683}
1684
1685/*
1686 * This page is about to be returned from the page allocator
1687 */
1688static inline int check_new_page(struct page *page)
1689{
1690 if (likely(page_expected_state(page,
1691 PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
1692 return 0;
1693
1694 check_new_page_bad(page);
1695 return 1;
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001696}
1697
Laura Abbott1414c7f2016-03-15 14:56:30 -07001698static inline bool free_pages_prezeroed(bool poisoned)
1699{
1700 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1701 page_poisoning_enabled() && poisoned;
1702}
1703
Mel Gorman479f8542016-05-19 17:14:35 -07001704#ifdef CONFIG_DEBUG_VM
1705static bool check_pcp_refill(struct page *page)
1706{
1707 return false;
1708}
1709
1710static bool check_new_pcp(struct page *page)
1711{
1712 return check_new_page(page);
1713}
1714#else
1715static bool check_pcp_refill(struct page *page)
1716{
1717 return check_new_page(page);
1718}
1719static bool check_new_pcp(struct page *page)
1720{
1721 return false;
1722}
1723#endif /* CONFIG_DEBUG_VM */
1724
1725static bool check_new_pages(struct page *page, unsigned int order)
1726{
1727 int i;
1728 for (i = 0; i < (1 << order); i++) {
1729 struct page *p = page + i;
1730
1731 if (unlikely(check_new_page(p)))
1732 return true;
1733 }
1734
1735 return false;
1736}
1737
Joonsoo Kim46f24fd2016-07-26 15:23:58 -07001738inline void post_alloc_hook(struct page *page, unsigned int order,
1739 gfp_t gfp_flags)
1740{
1741 set_page_private(page, 0);
1742 set_page_refcounted(page);
1743
1744 arch_alloc_page(page, order);
1745 kernel_map_pages(page, 1 << order, 1);
1746 kernel_poison_pages(page, 1 << order, 1);
1747 kasan_alloc_pages(page, order);
1748 set_page_owner(page, order, gfp_flags);
1749}
1750
Mel Gorman479f8542016-05-19 17:14:35 -07001751static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
Mel Gormanc6038442016-05-19 17:13:38 -07001752 unsigned int alloc_flags)
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001753{
1754 int i;
Laura Abbott1414c7f2016-03-15 14:56:30 -07001755 bool poisoned = true;
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001756
1757 for (i = 0; i < (1 << order); i++) {
1758 struct page *p = page + i;
Laura Abbott1414c7f2016-03-15 14:56:30 -07001759 if (poisoned)
1760 poisoned &= page_is_poisoned(p);
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001761 }
Hugh Dickins689bceb2005-11-21 21:32:20 -08001762
Joonsoo Kim46f24fd2016-07-26 15:23:58 -07001763 post_alloc_hook(page, order, gfp_flags);
Nick Piggin17cf4402006-03-22 00:08:41 -08001764
Laura Abbott1414c7f2016-03-15 14:56:30 -07001765 if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO))
Anisse Astierf4d28972015-06-24 16:56:36 -07001766 for (i = 0; i < (1 << order); i++)
1767 clear_highpage(page + i);
Nick Piggin17cf4402006-03-22 00:08:41 -08001768
1769 if (order && (gfp_flags & __GFP_COMP))
1770 prep_compound_page(page, order);
1771
Vlastimil Babka75379192015-02-11 15:25:38 -08001772 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001773 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
Vlastimil Babka75379192015-02-11 15:25:38 -08001774 * allocate the page. The expectation is that the caller is taking
1775 * steps that will free more memory. The caller should avoid the page
1776 * being used for !PFMEMALLOC purposes.
1777 */
Michal Hocko2f064f32015-08-21 14:11:51 -07001778 if (alloc_flags & ALLOC_NO_WATERMARKS)
1779 set_page_pfmemalloc(page);
1780 else
1781 clear_page_pfmemalloc(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782}
1783
Mel Gorman56fd56b2007-10-16 01:25:58 -07001784/*
1785 * Go through the free lists for the given migratetype and remove
1786 * the smallest available page from the freelists
1787 */
Mel Gorman728ec982009-06-16 15:32:04 -07001788static inline
1789struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -07001790 int migratetype)
1791{
1792 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -07001793 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001794 struct page *page;
1795
1796 /* Find a page of the appropriate size in the preferred list */
1797 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1798 area = &(zone->free_area[current_order]);
Geliang Tanga16601c2016-01-14 15:20:30 -08001799 page = list_first_entry_or_null(&area->free_list[migratetype],
Mel Gorman56fd56b2007-10-16 01:25:58 -07001800 struct page, lru);
Geliang Tanga16601c2016-01-14 15:20:30 -08001801 if (!page)
1802 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001803 list_del(&page->lru);
1804 rmv_page_order(page);
1805 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001806 expand(zone, page, order, current_order, area, migratetype);
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001807 set_pcppage_migratetype(page, migratetype);
Mel Gorman56fd56b2007-10-16 01:25:58 -07001808 return page;
1809 }
1810
1811 return NULL;
1812}
1813
1814
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001815/*
1816 * This array describes the order lists are fallen back to when
1817 * the free lists for the desirable migrate type are depleted
1818 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001819static int fallbacks[MIGRATE_TYPES][4] = {
Mel Gorman974a7862015-11-06 16:28:34 -08001820 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1821 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1822 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
Joonsoo Kimdc676472015-04-14 15:45:15 -07001823#ifdef CONFIG_CMA
Mel Gorman974a7862015-11-06 16:28:34 -08001824 [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001825#endif
Minchan Kim194159f2013-02-22 16:33:58 -08001826#ifdef CONFIG_MEMORY_ISOLATION
Mel Gorman974a7862015-11-06 16:28:34 -08001827 [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001828#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001829};
1830
Joonsoo Kimdc676472015-04-14 15:45:15 -07001831#ifdef CONFIG_CMA
1832static struct page *__rmqueue_cma_fallback(struct zone *zone,
1833 unsigned int order)
1834{
1835 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1836}
1837#else
1838static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1839 unsigned int order) { return NULL; }
1840#endif
1841
Mel Gormanc361be52007-10-16 01:25:51 -07001842/*
1843 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -07001844 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -07001845 * boundary. If alignment is required, use move_freepages_block()
1846 */
Minchan Kim435b4052012-10-08 16:32:16 -07001847int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -07001848 struct page *start_page, struct page *end_page,
1849 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001850{
1851 struct page *page;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08001852 unsigned int order;
Mel Gormand1003132007-10-16 01:26:00 -07001853 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -07001854
1855#ifndef CONFIG_HOLES_IN_ZONE
1856 /*
1857 * page_zone is not safe to call in this context when
1858 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1859 * anyway as we check zone boundaries in move_freepages_block().
1860 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -07001861 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -07001862 */
Mel Gorman97ee4ba2014-10-09 15:28:28 -07001863 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
Mel Gormanc361be52007-10-16 01:25:51 -07001864#endif
1865
1866 for (page = start_page; page <= end_page;) {
1867 if (!pfn_valid_within(page_to_pfn(page))) {
1868 page++;
1869 continue;
1870 }
1871
Ard Biesheuvelf073bdc2017-01-10 16:58:00 -08001872 /* Make sure we are not inadvertently changing nodes */
1873 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
1874
Mel Gormanc361be52007-10-16 01:25:51 -07001875 if (!PageBuddy(page)) {
1876 page++;
1877 continue;
1878 }
1879
1880 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001881 list_move(&page->lru,
1882 &zone->free_area[order].free_list[migratetype]);
Mel Gormanc361be52007-10-16 01:25:51 -07001883 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001884 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001885 }
1886
Mel Gormand1003132007-10-16 01:26:00 -07001887 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001888}
1889
Minchan Kimee6f5092012-07-31 16:43:50 -07001890int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001891 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001892{
1893 unsigned long start_pfn, end_pfn;
1894 struct page *start_page, *end_page;
1895
1896 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001897 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001898 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001899 end_page = start_page + pageblock_nr_pages - 1;
1900 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001901
1902 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001903 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001904 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001905 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001906 return 0;
1907
1908 return move_freepages(zone, start_page, end_page, migratetype);
1909}
1910
Mel Gorman2f66a682009-09-21 17:02:31 -07001911static void change_pageblock_range(struct page *pageblock_page,
1912 int start_order, int migratetype)
1913{
1914 int nr_pageblocks = 1 << (start_order - pageblock_order);
1915
1916 while (nr_pageblocks--) {
1917 set_pageblock_migratetype(pageblock_page, migratetype);
1918 pageblock_page += pageblock_nr_pages;
1919 }
1920}
1921
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001922/*
Vlastimil Babka9c0415e2015-02-11 15:28:21 -08001923 * When we are falling back to another migratetype during allocation, try to
1924 * steal extra free pages from the same pageblocks to satisfy further
1925 * allocations, instead of polluting multiple pageblocks.
1926 *
1927 * If we are stealing a relatively large buddy page, it is likely there will
1928 * be more free pages in the pageblock, so try to steal them all. For
1929 * reclaimable and unmovable allocations, we steal regardless of page size,
1930 * as fragmentation caused by those allocations polluting movable pageblocks
1931 * is worse than movable allocations stealing from unmovable and reclaimable
1932 * pageblocks.
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001933 */
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001934static bool can_steal_fallback(unsigned int order, int start_mt)
1935{
1936 /*
1937 * Leaving this order check is intended, although there is
1938 * relaxed order check in next check. The reason is that
1939 * we can actually steal whole pageblock if this condition met,
1940 * but, below check doesn't guarantee it and that is just heuristic
1941 * so could be changed anytime.
1942 */
1943 if (order >= pageblock_order)
1944 return true;
1945
1946 if (order >= pageblock_order / 2 ||
1947 start_mt == MIGRATE_RECLAIMABLE ||
1948 start_mt == MIGRATE_UNMOVABLE ||
1949 page_group_by_mobility_disabled)
1950 return true;
1951
1952 return false;
1953}
1954
1955/*
1956 * This function implements actual steal behaviour. If order is large enough,
1957 * we can steal whole pageblock. If not, we first move freepages in this
1958 * pageblock and check whether half of pages are moved or not. If half of
1959 * pages are moved, we can change migratetype of pageblock and permanently
1960 * use it's pages as requested migratetype in the future.
1961 */
1962static void steal_suitable_fallback(struct zone *zone, struct page *page,
1963 int start_type)
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001964{
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08001965 unsigned int current_order = page_order(page);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001966 int pages;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001967
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001968 /* Take ownership for orders >= pageblock_order */
1969 if (current_order >= pageblock_order) {
1970 change_pageblock_range(page, current_order, start_type);
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001971 return;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001972 }
1973
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001974 pages = move_freepages_block(zone, page, start_type);
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001975
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001976 /* Claim the whole block if over half of it is free */
1977 if (pages >= (1 << (pageblock_order-1)) ||
1978 page_group_by_mobility_disabled)
1979 set_pageblock_migratetype(page, start_type);
1980}
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001981
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001982/*
1983 * Check whether there is a suitable fallback freepage with requested order.
1984 * If only_stealable is true, this function returns fallback_mt only if
1985 * we can steal other freepages all together. This would help to reduce
1986 * fragmentation due to mixed migratetype pages in one pageblock.
1987 */
1988int find_suitable_fallback(struct free_area *area, unsigned int order,
1989 int migratetype, bool only_stealable, bool *can_steal)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001990{
1991 int i;
1992 int fallback_mt;
1993
1994 if (area->nr_free == 0)
1995 return -1;
1996
1997 *can_steal = false;
1998 for (i = 0;; i++) {
1999 fallback_mt = fallbacks[migratetype][i];
Mel Gorman974a7862015-11-06 16:28:34 -08002000 if (fallback_mt == MIGRATE_TYPES)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002001 break;
2002
2003 if (list_empty(&area->free_list[fallback_mt]))
2004 continue;
2005
2006 if (can_steal_fallback(order, migratetype))
2007 *can_steal = true;
2008
Joonsoo Kim2149cda2015-04-14 15:45:21 -07002009 if (!only_stealable)
2010 return fallback_mt;
2011
2012 if (*can_steal)
2013 return fallback_mt;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07002014 }
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002015
2016 return -1;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07002017}
2018
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002019/*
2020 * Reserve a pageblock for exclusive use of high-order atomic allocations if
2021 * there are no empty page blocks that contain a page with a suitable order
2022 */
2023static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2024 unsigned int alloc_order)
2025{
2026 int mt;
2027 unsigned long max_managed, flags;
2028
2029 /*
2030 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2031 * Check is race-prone but harmless.
2032 */
2033 max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
2034 if (zone->nr_reserved_highatomic >= max_managed)
2035 return;
2036
2037 spin_lock_irqsave(&zone->lock, flags);
2038
2039 /* Recheck the nr_reserved_highatomic limit under the lock */
2040 if (zone->nr_reserved_highatomic >= max_managed)
2041 goto out_unlock;
2042
2043 /* Yoink! */
2044 mt = get_pageblock_migratetype(page);
2045 if (mt != MIGRATE_HIGHATOMIC &&
2046 !is_migrate_isolate(mt) && !is_migrate_cma(mt)) {
2047 zone->nr_reserved_highatomic += pageblock_nr_pages;
2048 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2049 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC);
2050 }
2051
2052out_unlock:
2053 spin_unlock_irqrestore(&zone->lock, flags);
2054}
2055
2056/*
2057 * Used when an allocation is about to fail under memory pressure. This
2058 * potentially hurts the reliability of high-order allocations when under
2059 * intense memory pressure but failed atomic allocations should be easier
2060 * to recover from than an OOM.
Minchan Kim29fac032016-12-12 16:42:14 -08002061 *
2062 * If @force is true, try to unreserve a pageblock even though highatomic
2063 * pageblock is exhausted.
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002064 */
Minchan Kim29fac032016-12-12 16:42:14 -08002065static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2066 bool force)
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002067{
2068 struct zonelist *zonelist = ac->zonelist;
2069 unsigned long flags;
2070 struct zoneref *z;
2071 struct zone *zone;
2072 struct page *page;
2073 int order;
Minchan Kim04c87162016-12-12 16:42:11 -08002074 bool ret;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002075
2076 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2077 ac->nodemask) {
Minchan Kim29fac032016-12-12 16:42:14 -08002078 /*
2079 * Preserve at least one pageblock unless memory pressure
2080 * is really high.
2081 */
2082 if (!force && zone->nr_reserved_highatomic <=
2083 pageblock_nr_pages)
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002084 continue;
2085
2086 spin_lock_irqsave(&zone->lock, flags);
2087 for (order = 0; order < MAX_ORDER; order++) {
2088 struct free_area *area = &(zone->free_area[order]);
2089
Geliang Tanga16601c2016-01-14 15:20:30 -08002090 page = list_first_entry_or_null(
2091 &area->free_list[MIGRATE_HIGHATOMIC],
2092 struct page, lru);
2093 if (!page)
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002094 continue;
2095
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002096 /*
Minchan Kim4855e4a2016-12-12 16:42:08 -08002097 * In page freeing path, migratetype change is racy so
2098 * we can counter several free pages in a pageblock
2099 * in this loop althoug we changed the pageblock type
2100 * from highatomic to ac->migratetype. So we should
2101 * adjust the count once.
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002102 */
Minchan Kim4855e4a2016-12-12 16:42:08 -08002103 if (get_pageblock_migratetype(page) ==
2104 MIGRATE_HIGHATOMIC) {
2105 /*
2106 * It should never happen but changes to
2107 * locking could inadvertently allow a per-cpu
2108 * drain to add pages to MIGRATE_HIGHATOMIC
2109 * while unreserving so be safe and watch for
2110 * underflows.
2111 */
2112 zone->nr_reserved_highatomic -= min(
2113 pageblock_nr_pages,
2114 zone->nr_reserved_highatomic);
2115 }
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002116
2117 /*
2118 * Convert to ac->migratetype and avoid the normal
2119 * pageblock stealing heuristics. Minimally, the caller
2120 * is doing the work and needs the pages. More
2121 * importantly, if the block was always converted to
2122 * MIGRATE_UNMOVABLE or another type then the number
2123 * of pageblocks that cannot be completely freed
2124 * may increase.
2125 */
2126 set_pageblock_migratetype(page, ac->migratetype);
Minchan Kim04c87162016-12-12 16:42:11 -08002127 ret = move_freepages_block(zone, page, ac->migratetype);
Minchan Kim29fac032016-12-12 16:42:14 -08002128 if (ret) {
2129 spin_unlock_irqrestore(&zone->lock, flags);
2130 return ret;
2131 }
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002132 }
2133 spin_unlock_irqrestore(&zone->lock, flags);
2134 }
Minchan Kim04c87162016-12-12 16:42:11 -08002135
2136 return false;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002137}
2138
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002139/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07002140static inline struct page *
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002141__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002142{
Pintu Kumarb8af2942013-09-11 14:20:34 -07002143 struct free_area *area;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002144 unsigned int current_order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002145 struct page *page;
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002146 int fallback_mt;
2147 bool can_steal;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002148
2149 /* Find the largest possible block of pages in the other list */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002150 for (current_order = MAX_ORDER-1;
2151 current_order >= order && current_order <= MAX_ORDER-1;
2152 --current_order) {
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002153 area = &(zone->free_area[current_order]);
2154 fallback_mt = find_suitable_fallback(area, current_order,
Joonsoo Kim2149cda2015-04-14 15:45:21 -07002155 start_migratetype, false, &can_steal);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002156 if (fallback_mt == -1)
2157 continue;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002158
Geliang Tanga16601c2016-01-14 15:20:30 -08002159 page = list_first_entry(&area->free_list[fallback_mt],
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002160 struct page, lru);
Minchan Kim88ed3652016-12-12 16:42:05 -08002161 if (can_steal &&
2162 get_pageblock_migratetype(page) != MIGRATE_HIGHATOMIC)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002163 steal_suitable_fallback(zone, page, start_migratetype);
Mel Gormane0104872007-10-16 01:25:53 -07002164
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002165 /* Remove the page from the freelists */
2166 area->nr_free--;
2167 list_del(&page->lru);
2168 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002169
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002170 expand(zone, page, order, current_order, area,
2171 start_migratetype);
2172 /*
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002173 * The pcppage_migratetype may differ from pageblock's
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002174 * migratetype depending on the decisions in
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002175 * find_suitable_fallback(). This is OK as long as it does not
2176 * differ for MIGRATE_CMA pageblocks. Those can be used as
2177 * fallback only via special __rmqueue_cma_fallback() function
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002178 */
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002179 set_pcppage_migratetype(page, start_migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002180
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002181 trace_mm_page_alloc_extfrag(page, order, current_order,
2182 start_migratetype, fallback_mt);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002183
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002184 return page;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002185 }
2186
Mel Gorman728ec982009-06-16 15:32:04 -07002187 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002188}
2189
Mel Gorman56fd56b2007-10-16 01:25:58 -07002190/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 * Do the hard work of removing an element from the buddy allocator.
2192 * Call me with the zone->lock already held.
2193 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002194static struct page *__rmqueue(struct zone *zone, unsigned int order,
Mel Gorman6ac02062016-01-14 15:20:28 -08002195 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 struct page *page;
2198
Mel Gorman56fd56b2007-10-16 01:25:58 -07002199 page = __rmqueue_smallest(zone, order, migratetype);
Mel Gorman974a7862015-11-06 16:28:34 -08002200 if (unlikely(!page)) {
Joonsoo Kimdc676472015-04-14 15:45:15 -07002201 if (migratetype == MIGRATE_MOVABLE)
2202 page = __rmqueue_cma_fallback(zone, order);
2203
2204 if (!page)
2205 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gorman728ec982009-06-16 15:32:04 -07002206 }
2207
Mel Gorman0d3d0622009-09-21 17:02:44 -07002208 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002209 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210}
2211
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01002212/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 * Obtain a specified number of elements from the buddy allocator, all under
2214 * a single hold of the lock, for efficiency. Add them to the supplied list.
2215 * Returns the number of new pages which were placed at *list.
2216 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01002217static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002218 unsigned long count, struct list_head *list,
Mel Gormanb745bc82014-06-04 16:10:22 -07002219 int migratetype, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220{
Mel Gormana6de7342016-12-12 16:44:41 -08002221 int i, alloced = 0;
Mel Gorman374ad052017-02-24 14:56:38 -08002222 unsigned long flags;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01002223
Mel Gorman374ad052017-02-24 14:56:38 -08002224 spin_lock_irqsave(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 for (i = 0; i < count; ++i) {
Mel Gorman6ac02062016-01-14 15:20:28 -08002226 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08002227 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08002229
Mel Gorman479f8542016-05-19 17:14:35 -07002230 if (unlikely(check_pcp_refill(page)))
2231 continue;
2232
Mel Gorman81eabcb2007-12-17 16:20:05 -08002233 /*
2234 * Split buddy pages returned by expand() are received here
2235 * in physical page order. The page is added to the callers and
2236 * list and the list head then moves forward. From the callers
2237 * perspective, the linked list is ordered by page number in
2238 * some conditions. This is useful for IO devices that can
2239 * merge IO requests if the physical pages are ordered
2240 * properly.
2241 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002242 if (likely(!cold))
Mel Gormane084b2d2009-07-29 15:02:04 -07002243 list_add(&page->lru, list);
2244 else
2245 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08002246 list = &page->lru;
Mel Gormana6de7342016-12-12 16:44:41 -08002247 alloced++;
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002248 if (is_migrate_cma(get_pcppage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07002249 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2250 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 }
Mel Gormana6de7342016-12-12 16:44:41 -08002252
2253 /*
2254 * i pages were removed from the buddy list even if some leak due
2255 * to check_pcp_refill failing so adjust NR_FREE_PAGES based
2256 * on i. Do not confuse with 'alloced' which is the number of
2257 * pages added to the pcp list.
2258 */
Mel Gormanf2260e62009-06-16 15:32:13 -07002259 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Mel Gorman374ad052017-02-24 14:56:38 -08002260 spin_unlock_irqrestore(&zone->lock, flags);
Mel Gormana6de7342016-12-12 16:44:41 -08002261 return alloced;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262}
2263
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002264#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08002265/*
Christoph Lameter4037d452007-05-09 02:35:14 -07002266 * Called from the vmstat counter updater to drain pagesets of this
2267 * currently executing processor on remote nodes after they have
2268 * expired.
2269 *
Christoph Lameter879336c2006-03-22 00:09:08 -08002270 * Note that this function must be called with the thread pinned to
2271 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08002272 */
Christoph Lameter4037d452007-05-09 02:35:14 -07002273void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002274{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002275 unsigned long flags;
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07002276 int to_drain, batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002277
Christoph Lameter4037d452007-05-09 02:35:14 -07002278 local_irq_save(flags);
Jason Low4db0c3c2015-04-15 16:14:08 -07002279 batch = READ_ONCE(pcp->batch);
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07002280 to_drain = min(pcp->count, batch);
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07002281 if (to_drain > 0) {
2282 free_pcppages_bulk(zone, to_drain, pcp);
2283 pcp->count -= to_drain;
2284 }
Christoph Lameter4037d452007-05-09 02:35:14 -07002285 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002286}
2287#endif
2288
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002289/*
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002290 * Drain pcplists of the indicated processor and zone.
2291 *
2292 * The processor must either be the current processor and the
2293 * thread pinned to the current processor or a processor that
2294 * is not online.
2295 */
2296static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2297{
2298 unsigned long flags;
2299 struct per_cpu_pageset *pset;
2300 struct per_cpu_pages *pcp;
2301
2302 local_irq_save(flags);
2303 pset = per_cpu_ptr(zone->pageset, cpu);
2304
2305 pcp = &pset->pcp;
2306 if (pcp->count) {
2307 free_pcppages_bulk(zone, pcp->count, pcp);
2308 pcp->count = 0;
2309 }
2310 local_irq_restore(flags);
2311}
2312
2313/*
2314 * Drain pcplists of all zones on the indicated processor.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002315 *
2316 * The processor must either be the current processor and the
2317 * thread pinned to the current processor or a processor that
2318 * is not online.
2319 */
2320static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321{
2322 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002324 for_each_populated_zone(zone) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002325 drain_pages_zone(cpu, zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 }
2327}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002329/*
2330 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002331 *
2332 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2333 * the single zone's pages.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002334 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002335void drain_local_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002336{
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002337 int cpu = smp_processor_id();
2338
2339 if (zone)
2340 drain_pages_zone(cpu, zone);
2341 else
2342 drain_pages(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002343}
2344
Mel Gorman0ccce3b2017-02-24 14:56:32 -08002345static void drain_local_pages_wq(struct work_struct *work)
2346{
Michal Hockoa459eeb2017-02-24 14:56:35 -08002347 /*
2348 * drain_all_pages doesn't use proper cpu hotplug protection so
2349 * we can race with cpu offline when the WQ can move this from
2350 * a cpu pinned worker to an unbound one. We can operate on a different
2351 * cpu which is allright but we also have to make sure to not move to
2352 * a different one.
2353 */
2354 preempt_disable();
Mel Gorman0ccce3b2017-02-24 14:56:32 -08002355 drain_local_pages(NULL);
Michal Hockoa459eeb2017-02-24 14:56:35 -08002356 preempt_enable();
Mel Gorman0ccce3b2017-02-24 14:56:32 -08002357}
2358
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002359/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002360 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2361 *
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002362 * When zone parameter is non-NULL, spill just the single zone's pages.
2363 *
Mel Gorman0ccce3b2017-02-24 14:56:32 -08002364 * Note that this can be extremely slow as the draining happens in a workqueue.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002365 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002366void drain_all_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002367{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002368 int cpu;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002369
2370 /*
2371 * Allocate in the BSS so we wont require allocation in
2372 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2373 */
2374 static cpumask_t cpus_with_pcps;
2375
Mel Gorman0ccce3b2017-02-24 14:56:32 -08002376 /* Workqueues cannot recurse */
2377 if (current->flags & PF_WQ_WORKER)
2378 return;
2379
Mel Gormanbd233f52017-02-24 14:56:56 -08002380 /*
2381 * Do not drain if one is already in progress unless it's specific to
2382 * a zone. Such callers are primarily CMA and memory hotplug and need
2383 * the drain to be complete when the call returns.
2384 */
2385 if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) {
2386 if (!zone)
2387 return;
2388 mutex_lock(&pcpu_drain_mutex);
2389 }
Mel Gorman0ccce3b2017-02-24 14:56:32 -08002390
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002391 /*
2392 * We don't care about racing with CPU hotplug event
2393 * as offline notification will cause the notified
2394 * cpu to drain that CPU pcps and on_each_cpu_mask
2395 * disables preemption as part of its processing
2396 */
2397 for_each_online_cpu(cpu) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002398 struct per_cpu_pageset *pcp;
2399 struct zone *z;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002400 bool has_pcps = false;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002401
2402 if (zone) {
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002403 pcp = per_cpu_ptr(zone->pageset, cpu);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002404 if (pcp->pcp.count)
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002405 has_pcps = true;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002406 } else {
2407 for_each_populated_zone(z) {
2408 pcp = per_cpu_ptr(z->pageset, cpu);
2409 if (pcp->pcp.count) {
2410 has_pcps = true;
2411 break;
2412 }
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002413 }
2414 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002415
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002416 if (has_pcps)
2417 cpumask_set_cpu(cpu, &cpus_with_pcps);
2418 else
2419 cpumask_clear_cpu(cpu, &cpus_with_pcps);
2420 }
Mel Gorman0ccce3b2017-02-24 14:56:32 -08002421
Mel Gormanbd233f52017-02-24 14:56:56 -08002422 for_each_cpu(cpu, &cpus_with_pcps) {
2423 struct work_struct *work = per_cpu_ptr(&pcpu_drain, cpu);
2424 INIT_WORK(work, drain_local_pages_wq);
2425 schedule_work_on(cpu, work);
Mel Gorman0ccce3b2017-02-24 14:56:32 -08002426 }
Mel Gormanbd233f52017-02-24 14:56:56 -08002427 for_each_cpu(cpu, &cpus_with_pcps)
2428 flush_work(per_cpu_ptr(&pcpu_drain, cpu));
2429
2430 mutex_unlock(&pcpu_drain_mutex);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002431}
2432
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02002433#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434
2435void mark_free_pages(struct zone *zone)
2436{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002437 unsigned long pfn, max_zone_pfn;
2438 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002439 unsigned int order, t;
Geliang Tang86760a22016-01-14 15:20:33 -08002440 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
Xishi Qiu8080fc02013-09-11 14:21:45 -07002442 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 return;
2444
2445 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002446
Cody P Schafer108bcc92013-02-22 16:35:23 -08002447 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002448 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2449 if (pfn_valid(pfn)) {
Geliang Tang86760a22016-01-14 15:20:33 -08002450 page = pfn_to_page(pfn);
Joonsoo Kimba6b0972016-05-19 17:12:16 -07002451
2452 if (page_zone(page) != zone)
2453 continue;
2454
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07002455 if (!swsusp_page_is_forbidden(page))
2456 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002457 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002459 for_each_migratetype_order(order, t) {
Geliang Tang86760a22016-01-14 15:20:33 -08002460 list_for_each_entry(page,
2461 &zone->free_area[order].free_list[t], lru) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002462 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
Geliang Tang86760a22016-01-14 15:20:33 -08002464 pfn = page_to_pfn(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002465 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07002466 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002467 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002468 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 spin_unlock_irqrestore(&zone->lock, flags);
2470}
Mel Gormane2c55dc2007-10-16 01:25:50 -07002471#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472
2473/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 * Free a 0-order page
Mel Gormanb745bc82014-06-04 16:10:22 -07002475 * cold == true ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002477void free_hot_cold_page(struct page *page, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478{
2479 struct zone *zone = page_zone(page);
2480 struct per_cpu_pages *pcp;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002481 unsigned long pfn = page_to_pfn(page);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002482 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483
Mel Gorman374ad052017-02-24 14:56:38 -08002484 if (in_interrupt()) {
2485 __free_pages_ok(page, 0);
2486 return;
2487 }
2488
Mel Gorman4db75482016-05-19 17:14:32 -07002489 if (!free_pcp_prepare(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -08002490 return;
2491
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002492 migratetype = get_pfnblock_migratetype(page, pfn);
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002493 set_pcppage_migratetype(page, migratetype);
Mel Gorman374ad052017-02-24 14:56:38 -08002494 preempt_disable();
Mel Gormanda456f12009-06-16 15:32:08 -07002495
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002496 /*
2497 * We only track unmovable, reclaimable and movable on pcp lists.
2498 * Free ISOLATE pages back to the allocator because they are being
2499 * offlined but treat RESERVE as movable pages so we can get those
2500 * areas back if necessary. Otherwise, we may have to free
2501 * excessively into the page allocator
2502 */
2503 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08002504 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002505 free_one_page(zone, page, pfn, 0, migratetype);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002506 goto out;
2507 }
2508 migratetype = MIGRATE_MOVABLE;
2509 }
2510
Mel Gorman374ad052017-02-24 14:56:38 -08002511 __count_vm_event(PGFREE);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002512 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Mel Gormanb745bc82014-06-04 16:10:22 -07002513 if (!cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002514 list_add(&page->lru, &pcp->lists[migratetype]);
Mel Gormanb745bc82014-06-04 16:10:22 -07002515 else
2516 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08002518 if (pcp->count >= pcp->high) {
Jason Low4db0c3c2015-04-15 16:14:08 -07002519 unsigned long batch = READ_ONCE(pcp->batch);
Cody P Schafer998d39c2013-07-03 15:01:32 -07002520 free_pcppages_bulk(zone, batch, pcp);
2521 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08002522 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002523
2524out:
Mel Gorman374ad052017-02-24 14:56:38 -08002525 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526}
2527
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002528/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002529 * Free a list of 0-order pages
2530 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002531void free_hot_cold_page_list(struct list_head *list, bool cold)
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002532{
2533 struct page *page, *next;
2534
2535 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08002536 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002537 free_hot_cold_page(page, cold);
2538 }
2539}
2540
2541/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002542 * split_page takes a non-compound higher-order page, and splits it into
2543 * n (1<<order) sub-pages: page[0..n]
2544 * Each sub-page must be freed individually.
2545 *
2546 * Note: this is probably too low level an operation for use in drivers.
2547 * Please consult with lkml before using this in your driver.
2548 */
2549void split_page(struct page *page, unsigned int order)
2550{
2551 int i;
2552
Sasha Levin309381fea2014-01-23 15:52:54 -08002553 VM_BUG_ON_PAGE(PageCompound(page), page);
2554 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002555
2556#ifdef CONFIG_KMEMCHECK
2557 /*
2558 * Split shadow pages too, because free(page[0]) would
2559 * otherwise free the whole shadow.
2560 */
2561 if (kmemcheck_page_is_tracked(page))
2562 split_page(virt_to_page(page[0].shadow), order);
2563#endif
2564
Joonsoo Kima9627bc2016-07-26 15:23:49 -07002565 for (i = 1; i < (1 << order); i++)
Nick Piggin7835e982006-03-22 00:08:40 -08002566 set_page_refcounted(page + i);
Joonsoo Kima9627bc2016-07-26 15:23:49 -07002567 split_page_owner(page, order);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002568}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07002569EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002570
Joonsoo Kim3c605092014-11-13 15:19:21 -08002571int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07002572{
Mel Gorman748446b2010-05-24 14:32:27 -07002573 unsigned long watermark;
2574 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002575 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07002576
2577 BUG_ON(!PageBuddy(page));
2578
2579 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002580 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07002581
Minchan Kim194159f2013-02-22 16:33:58 -08002582 if (!is_migrate_isolate(mt)) {
Vlastimil Babka8348faf2016-10-07 16:58:00 -07002583 /*
2584 * Obey watermarks as if the page was being allocated. We can
2585 * emulate a high-order watermark check with a raised order-0
2586 * watermark, because we already know our high-order page
2587 * exists.
2588 */
2589 watermark = min_wmark_pages(zone) + (1UL << order);
Vlastimil Babka984fdba2016-10-07 16:57:57 -07002590 if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA))
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002591 return 0;
2592
Mel Gorman8fb74b92013-01-11 14:32:16 -08002593 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002594 }
Mel Gorman748446b2010-05-24 14:32:27 -07002595
2596 /* Remove page from free list */
2597 list_del(&page->lru);
2598 zone->free_area[order].nr_free--;
2599 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002600
zhong jiang400bc7f2016-07-28 15:45:07 -07002601 /*
2602 * Set the pageblock if the isolated page is at least half of a
2603 * pageblock
2604 */
Mel Gorman748446b2010-05-24 14:32:27 -07002605 if (order >= pageblock_order - 1) {
2606 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002607 for (; page < endpage; page += pageblock_nr_pages) {
2608 int mt = get_pageblock_migratetype(page);
Minchan Kim88ed3652016-12-12 16:42:05 -08002609 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)
2610 && mt != MIGRATE_HIGHATOMIC)
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002611 set_pageblock_migratetype(page,
2612 MIGRATE_MOVABLE);
2613 }
Mel Gorman748446b2010-05-24 14:32:27 -07002614 }
2615
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002616
Mel Gorman8fb74b92013-01-11 14:32:16 -08002617 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002618}
2619
2620/*
Mel Gorman060e7412016-05-19 17:13:27 -07002621 * Update NUMA hit/miss statistics
2622 *
2623 * Must be called with interrupts disabled.
Mel Gorman060e7412016-05-19 17:13:27 -07002624 */
Michal Hocko41b61672017-01-10 16:57:42 -08002625static inline void zone_statistics(struct zone *preferred_zone, struct zone *z)
Mel Gorman060e7412016-05-19 17:13:27 -07002626{
2627#ifdef CONFIG_NUMA
Mel Gorman060e7412016-05-19 17:13:27 -07002628 enum zone_stat_item local_stat = NUMA_LOCAL;
2629
Michal Hocko2df26632017-01-10 16:57:39 -08002630 if (z->node != numa_node_id())
Mel Gorman060e7412016-05-19 17:13:27 -07002631 local_stat = NUMA_OTHER;
Mel Gorman060e7412016-05-19 17:13:27 -07002632
Michal Hocko2df26632017-01-10 16:57:39 -08002633 if (z->node == preferred_zone->node)
Mel Gorman060e7412016-05-19 17:13:27 -07002634 __inc_zone_state(z, NUMA_HIT);
Michal Hocko2df26632017-01-10 16:57:39 -08002635 else {
Mel Gorman060e7412016-05-19 17:13:27 -07002636 __inc_zone_state(z, NUMA_MISS);
2637 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
2638 }
Michal Hocko2df26632017-01-10 16:57:39 -08002639 __inc_zone_state(z, local_stat);
Mel Gorman060e7412016-05-19 17:13:27 -07002640#endif
2641}
2642
Mel Gorman066b2392017-02-24 14:56:26 -08002643/* Remove page from the per-cpu list, caller must protect the list */
2644static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype,
2645 bool cold, struct per_cpu_pages *pcp,
2646 struct list_head *list)
2647{
2648 struct page *page;
2649
Mel Gorman374ad052017-02-24 14:56:38 -08002650 VM_BUG_ON(in_interrupt());
2651
Mel Gorman066b2392017-02-24 14:56:26 -08002652 do {
2653 if (list_empty(list)) {
2654 pcp->count += rmqueue_bulk(zone, 0,
2655 pcp->batch, list,
2656 migratetype, cold);
2657 if (unlikely(list_empty(list)))
2658 return NULL;
2659 }
2660
2661 if (cold)
2662 page = list_last_entry(list, struct page, lru);
2663 else
2664 page = list_first_entry(list, struct page, lru);
2665
2666 list_del(&page->lru);
2667 pcp->count--;
2668 } while (check_new_pcp(page));
2669
2670 return page;
2671}
2672
2673/* Lock and remove page from the per-cpu list */
2674static struct page *rmqueue_pcplist(struct zone *preferred_zone,
2675 struct zone *zone, unsigned int order,
2676 gfp_t gfp_flags, int migratetype)
2677{
2678 struct per_cpu_pages *pcp;
2679 struct list_head *list;
2680 bool cold = ((gfp_flags & __GFP_COLD) != 0);
2681 struct page *page;
Mel Gorman066b2392017-02-24 14:56:26 -08002682
Mel Gorman374ad052017-02-24 14:56:38 -08002683 preempt_disable();
Mel Gorman066b2392017-02-24 14:56:26 -08002684 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2685 list = &pcp->lists[migratetype];
2686 page = __rmqueue_pcplist(zone, migratetype, cold, pcp, list);
2687 if (page) {
2688 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
2689 zone_statistics(preferred_zone, zone);
2690 }
Mel Gorman374ad052017-02-24 14:56:38 -08002691 preempt_enable();
Mel Gorman066b2392017-02-24 14:56:26 -08002692 return page;
2693}
2694
Mel Gorman060e7412016-05-19 17:13:27 -07002695/*
Vlastimil Babka75379192015-02-11 15:25:38 -08002696 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07002698static inline
Mel Gorman066b2392017-02-24 14:56:26 -08002699struct page *rmqueue(struct zone *preferred_zone,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002700 struct zone *zone, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07002701 gfp_t gfp_flags, unsigned int alloc_flags,
2702 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703{
2704 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08002705 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706
Mel Gorman374ad052017-02-24 14:56:38 -08002707 if (likely(order == 0) && !in_interrupt()) {
Mel Gorman066b2392017-02-24 14:56:26 -08002708 page = rmqueue_pcplist(preferred_zone, zone, order,
2709 gfp_flags, migratetype);
2710 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 }
2712
Mel Gorman066b2392017-02-24 14:56:26 -08002713 /*
2714 * We most definitely don't want callers attempting to
2715 * allocate greater than order-1 page units with __GFP_NOFAIL.
2716 */
2717 WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
2718 spin_lock_irqsave(&zone->lock, flags);
2719
2720 do {
2721 page = NULL;
2722 if (alloc_flags & ALLOC_HARDER) {
2723 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
2724 if (page)
2725 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2726 }
2727 if (!page)
2728 page = __rmqueue(zone, order, migratetype);
2729 } while (page && check_new_pages(page, order));
2730 spin_unlock(&zone->lock);
2731 if (!page)
2732 goto failed;
2733 __mod_zone_freepage_state(zone, -(1 << order),
2734 get_pcppage_migratetype(page));
2735
Mel Gorman16709d12016-07-28 15:46:56 -07002736 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
Michal Hocko41b61672017-01-10 16:57:42 -08002737 zone_statistics(preferred_zone, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08002738 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739
Mel Gorman066b2392017-02-24 14:56:26 -08002740out:
2741 VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08002743
2744failed:
2745 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002746 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747}
2748
Akinobu Mita933e3122006-12-08 02:39:45 -08002749#ifdef CONFIG_FAIL_PAGE_ALLOC
2750
Akinobu Mitab2588c42011-07-26 16:09:03 -07002751static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08002752 struct fault_attr attr;
2753
Viresh Kumar621a5f72015-09-26 15:04:07 -07002754 bool ignore_gfp_highmem;
Mel Gorman71baba42015-11-06 16:28:28 -08002755 bool ignore_gfp_reclaim;
Akinobu Mita54114992007-07-15 23:40:23 -07002756 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08002757} fail_page_alloc = {
2758 .attr = FAULT_ATTR_INITIALIZER,
Mel Gorman71baba42015-11-06 16:28:28 -08002759 .ignore_gfp_reclaim = true,
Viresh Kumar621a5f72015-09-26 15:04:07 -07002760 .ignore_gfp_highmem = true,
Akinobu Mita54114992007-07-15 23:40:23 -07002761 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08002762};
2763
2764static int __init setup_fail_page_alloc(char *str)
2765{
2766 return setup_fault_attr(&fail_page_alloc.attr, str);
2767}
2768__setup("fail_page_alloc=", setup_fail_page_alloc);
2769
Gavin Shandeaf3862012-07-31 16:41:51 -07002770static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002771{
Akinobu Mita54114992007-07-15 23:40:23 -07002772 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07002773 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002774 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07002775 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002776 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002777 return false;
Mel Gorman71baba42015-11-06 16:28:28 -08002778 if (fail_page_alloc.ignore_gfp_reclaim &&
2779 (gfp_mask & __GFP_DIRECT_RECLAIM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002780 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002781
2782 return should_fail(&fail_page_alloc.attr, 1 << order);
2783}
2784
2785#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2786
2787static int __init fail_page_alloc_debugfs(void)
2788{
Al Virof4ae40a2011-07-24 04:33:43 -04002789 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08002790 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08002791
Akinobu Mitadd48c082011-08-03 16:21:01 -07002792 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2793 &fail_page_alloc.attr);
2794 if (IS_ERR(dir))
2795 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002796
Akinobu Mitab2588c42011-07-26 16:09:03 -07002797 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
Mel Gorman71baba42015-11-06 16:28:28 -08002798 &fail_page_alloc.ignore_gfp_reclaim))
Akinobu Mitab2588c42011-07-26 16:09:03 -07002799 goto fail;
2800 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2801 &fail_page_alloc.ignore_gfp_highmem))
2802 goto fail;
2803 if (!debugfs_create_u32("min-order", mode, dir,
2804 &fail_page_alloc.min_order))
2805 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08002806
Akinobu Mitab2588c42011-07-26 16:09:03 -07002807 return 0;
2808fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07002809 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002810
Akinobu Mitab2588c42011-07-26 16:09:03 -07002811 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08002812}
2813
2814late_initcall(fail_page_alloc_debugfs);
2815
2816#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2817
2818#else /* CONFIG_FAIL_PAGE_ALLOC */
2819
Gavin Shandeaf3862012-07-31 16:41:51 -07002820static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002821{
Gavin Shandeaf3862012-07-31 16:41:51 -07002822 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002823}
2824
2825#endif /* CONFIG_FAIL_PAGE_ALLOC */
2826
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827/*
Mel Gorman97a16fc2015-11-06 16:28:40 -08002828 * Return true if free base pages are above 'mark'. For high-order checks it
2829 * will return true of the order-0 watermark is reached and there is at least
2830 * one free page of a suitable size. Checking now avoids taking the zone lock
2831 * to check in the allocation paths if no pages are free.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 */
Michal Hocko86a294a2016-05-20 16:57:12 -07002833bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2834 int classzone_idx, unsigned int alloc_flags,
2835 long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836{
Christoph Lameterd23ad422007-02-10 01:43:02 -08002837 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 int o;
Mel Gormanc6038442016-05-19 17:13:38 -07002839 const bool alloc_harder = (alloc_flags & ALLOC_HARDER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002841 /* free_pages may go negative - that's OK */
Michal Hockodf0a6da2012-01-10 15:08:02 -08002842 free_pages -= (1 << order) - 1;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002843
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002844 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 min -= min / 2;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002846
2847 /*
2848 * If the caller does not have rights to ALLOC_HARDER then subtract
2849 * the high-atomic reserves. This will over-estimate the size of the
2850 * atomic reserve but it avoids a search.
2851 */
Mel Gorman97a16fc2015-11-06 16:28:40 -08002852 if (likely(!alloc_harder))
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002853 free_pages -= z->nr_reserved_highatomic;
2854 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 min -= min / 4;
Mel Gormane2b19192015-11-06 16:28:09 -08002856
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002857#ifdef CONFIG_CMA
2858 /* If allocation can't use CMA areas don't use free CMA pages */
2859 if (!(alloc_flags & ALLOC_CMA))
Mel Gorman97a16fc2015-11-06 16:28:40 -08002860 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002861#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002862
Mel Gorman97a16fc2015-11-06 16:28:40 -08002863 /*
2864 * Check watermarks for an order-0 allocation request. If these
2865 * are not met, then a high-order request also cannot go ahead
2866 * even if a suitable page happened to be free.
2867 */
2868 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08002869 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870
Mel Gorman97a16fc2015-11-06 16:28:40 -08002871 /* If this is an order-0 request then the watermark is fine */
2872 if (!order)
2873 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874
Mel Gorman97a16fc2015-11-06 16:28:40 -08002875 /* For a high-order request, check at least one suitable page is free */
2876 for (o = order; o < MAX_ORDER; o++) {
2877 struct free_area *area = &z->free_area[o];
2878 int mt;
2879
2880 if (!area->nr_free)
2881 continue;
2882
2883 if (alloc_harder)
2884 return true;
2885
2886 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
2887 if (!list_empty(&area->free_list[mt]))
2888 return true;
2889 }
2890
2891#ifdef CONFIG_CMA
2892 if ((alloc_flags & ALLOC_CMA) &&
2893 !list_empty(&area->free_list[MIGRATE_CMA])) {
2894 return true;
2895 }
2896#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 }
Mel Gorman97a16fc2015-11-06 16:28:40 -08002898 return false;
Mel Gorman88f5acf2011-01-13 15:45:41 -08002899}
2900
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002901bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Mel Gormanc6038442016-05-19 17:13:38 -07002902 int classzone_idx, unsigned int alloc_flags)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002903{
2904 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2905 zone_page_state(z, NR_FREE_PAGES));
2906}
2907
Mel Gorman48ee5f32016-05-19 17:14:07 -07002908static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
2909 unsigned long mark, int classzone_idx, unsigned int alloc_flags)
2910{
2911 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2912 long cma_pages = 0;
2913
2914#ifdef CONFIG_CMA
2915 /* If allocation can't use CMA areas don't use free CMA pages */
2916 if (!(alloc_flags & ALLOC_CMA))
2917 cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES);
2918#endif
2919
2920 /*
2921 * Fast check for order-0 only. If this fails then the reserves
2922 * need to be calculated. There is a corner case where the check
2923 * passes but only the high-order atomic reserve are free. If
2924 * the caller is !atomic then it'll uselessly search the free
2925 * list. That corner case is then slower but it is harmless.
2926 */
2927 if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx])
2928 return true;
2929
2930 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2931 free_pages);
2932}
2933
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002934bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
Mel Gormane2b19192015-11-06 16:28:09 -08002935 unsigned long mark, int classzone_idx)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002936{
2937 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2938
2939 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2940 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2941
Mel Gormane2b19192015-11-06 16:28:09 -08002942 return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002943 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944}
2945
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002946#ifdef CONFIG_NUMA
David Rientjes957f8222012-10-08 16:33:24 -07002947static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2948{
Gavin Shane02dc012017-02-24 14:59:33 -08002949 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <=
Mel Gorman5f7a75a2014-06-04 16:07:15 -07002950 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07002951}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002952#else /* CONFIG_NUMA */
David Rientjes957f8222012-10-08 16:33:24 -07002953static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2954{
2955 return true;
2956}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002957#endif /* CONFIG_NUMA */
2958
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002959/*
Paul Jackson0798e512006-12-06 20:31:38 -08002960 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002961 * a page.
2962 */
2963static struct page *
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002964get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2965 const struct alloc_context *ac)
Martin Hicks753ee722005-06-21 17:14:41 -07002966{
Mel Gormanc33d6c02016-05-19 17:14:10 -07002967 struct zoneref *z = ac->preferred_zoneref;
Mel Gorman5117f452009-06-16 15:31:59 -07002968 struct zone *zone;
Mel Gorman3b8c0be2016-07-28 15:46:53 -07002969 struct pglist_data *last_pgdat_dirty_limit = NULL;
2970
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002971 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002972 * Scan zonelist, looking for a zone with enough free.
Vladimir Davydov344736f2014-10-20 15:50:30 +04002973 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002974 */
Mel Gormanc33d6c02016-05-19 17:14:10 -07002975 for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002976 ac->nodemask) {
Mel Gormanbe06af02016-05-19 17:13:47 -07002977 struct page *page;
Johannes Weinere085dbc2013-09-11 14:20:46 -07002978 unsigned long mark;
2979
Mel Gorman664eedd2014-06-04 16:10:08 -07002980 if (cpusets_enabled() &&
2981 (alloc_flags & ALLOC_CPUSET) &&
Vlastimil Babka002f2902016-05-19 17:14:30 -07002982 !__cpuset_zone_allowed(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07002983 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08002984 /*
2985 * When allocating a page cache page for writing, we
Mel Gorman281e3722016-07-28 15:46:11 -07002986 * want to get it from a node that is within its dirty
2987 * limit, such that no single node holds more than its
Johannes Weinera756cf52012-01-10 15:07:49 -08002988 * proportional share of globally allowed dirty pages.
Mel Gorman281e3722016-07-28 15:46:11 -07002989 * The dirty limits take into account the node's
Johannes Weinera756cf52012-01-10 15:07:49 -08002990 * lowmem reserves and high watermark so that kswapd
2991 * should be able to balance it without having to
2992 * write pages from its LRU list.
2993 *
Johannes Weinera756cf52012-01-10 15:07:49 -08002994 * XXX: For now, allow allocations to potentially
Mel Gorman281e3722016-07-28 15:46:11 -07002995 * exceed the per-node dirty limit in the slowpath
Mel Gormanc9ab0c42015-11-06 16:28:12 -08002996 * (spread_dirty_pages unset) before going into reclaim,
Johannes Weinera756cf52012-01-10 15:07:49 -08002997 * which is important when on a NUMA setup the allowed
Mel Gorman281e3722016-07-28 15:46:11 -07002998 * nodes are together not big enough to reach the
Johannes Weinera756cf52012-01-10 15:07:49 -08002999 * global limit. The proper fix for these situations
Mel Gorman281e3722016-07-28 15:46:11 -07003000 * will require awareness of nodes in the
Johannes Weinera756cf52012-01-10 15:07:49 -08003001 * dirty-throttling and the flusher threads.
3002 */
Mel Gorman3b8c0be2016-07-28 15:46:53 -07003003 if (ac->spread_dirty_pages) {
3004 if (last_pgdat_dirty_limit == zone->zone_pgdat)
3005 continue;
3006
3007 if (!node_dirty_ok(zone->zone_pgdat)) {
3008 last_pgdat_dirty_limit = zone->zone_pgdat;
3009 continue;
3010 }
3011 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003012
Johannes Weinere085dbc2013-09-11 14:20:46 -07003013 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gorman48ee5f32016-05-19 17:14:07 -07003014 if (!zone_watermark_fast(zone, order, mark,
Mel Gorman93ea9962016-05-19 17:14:13 -07003015 ac_classzone_idx(ac), alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07003016 int ret;
3017
Mel Gorman5dab2912014-06-04 16:10:14 -07003018 /* Checked here to keep the fast path fast */
3019 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
3020 if (alloc_flags & ALLOC_NO_WATERMARKS)
3021 goto try_this_zone;
3022
Mel Gormana5f5f912016-07-28 15:46:32 -07003023 if (node_reclaim_mode == 0 ||
Mel Gormanc33d6c02016-05-19 17:14:10 -07003024 !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
Mel Gormancd38b112011-07-25 17:12:29 -07003025 continue;
3026
Mel Gormana5f5f912016-07-28 15:46:32 -07003027 ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
Mel Gormanfa5e0842009-06-16 15:33:22 -07003028 switch (ret) {
Mel Gormana5f5f912016-07-28 15:46:32 -07003029 case NODE_RECLAIM_NOSCAN:
Mel Gormanfa5e0842009-06-16 15:33:22 -07003030 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07003031 continue;
Mel Gormana5f5f912016-07-28 15:46:32 -07003032 case NODE_RECLAIM_FULL:
Mel Gormanfa5e0842009-06-16 15:33:22 -07003033 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07003034 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07003035 default:
3036 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07003037 if (zone_watermark_ok(zone, order, mark,
Mel Gorman93ea9962016-05-19 17:14:13 -07003038 ac_classzone_idx(ac), alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07003039 goto try_this_zone;
3040
Mel Gormanfed27192013-04-29 15:07:57 -07003041 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08003042 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003043 }
3044
Mel Gormanfa5e0842009-06-16 15:33:22 -07003045try_this_zone:
Mel Gorman066b2392017-02-24 14:56:26 -08003046 page = rmqueue(ac->preferred_zoneref->zone, zone, order,
Mel Gorman0aaa29a2015-11-06 16:28:37 -08003047 gfp_mask, alloc_flags, ac->migratetype);
Vlastimil Babka75379192015-02-11 15:25:38 -08003048 if (page) {
Mel Gorman479f8542016-05-19 17:14:35 -07003049 prep_new_page(page, order, gfp_mask, alloc_flags);
Mel Gorman0aaa29a2015-11-06 16:28:37 -08003050
3051 /*
3052 * If this is a high-order atomic allocation then check
3053 * if the pageblock should be reserved for the future
3054 */
3055 if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
3056 reserve_highatomic_pageblock(page, zone, order);
3057
Vlastimil Babka75379192015-02-11 15:25:38 -08003058 return page;
3059 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003060 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003061
Mel Gorman4ffeaf32014-08-06 16:07:22 -07003062 return NULL;
Martin Hicks753ee722005-06-21 17:14:41 -07003063}
3064
David Rientjes29423e772011-03-22 16:30:47 -07003065/*
3066 * Large machines with many possible nodes should not always dump per-node
3067 * meminfo in irq context.
3068 */
3069static inline bool should_suppress_show_mem(void)
3070{
3071 bool ret = false;
3072
3073#if NODES_SHIFT > 8
3074 ret = in_interrupt();
3075#endif
3076 return ret;
3077}
3078
Michal Hocko9af744d2017-02-22 15:46:16 -08003079static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask)
Dave Hansena238ab52011-05-24 17:12:16 -07003080{
Dave Hansena238ab52011-05-24 17:12:16 -07003081 unsigned int filter = SHOW_MEM_FILTER_NODES;
Michal Hockoaa187502017-02-22 15:41:45 -08003082 static DEFINE_RATELIMIT_STATE(show_mem_rs, HZ, 1);
Dave Hansena238ab52011-05-24 17:12:16 -07003083
Michal Hockoaa187502017-02-22 15:41:45 -08003084 if (should_suppress_show_mem() || !__ratelimit(&show_mem_rs))
Dave Hansena238ab52011-05-24 17:12:16 -07003085 return;
3086
3087 /*
3088 * This documents exceptions given to allocations in certain
3089 * contexts that are allowed to allocate outside current's set
3090 * of allowed nodes.
3091 */
3092 if (!(gfp_mask & __GFP_NOMEMALLOC))
3093 if (test_thread_flag(TIF_MEMDIE) ||
3094 (current->flags & (PF_MEMALLOC | PF_EXITING)))
3095 filter &= ~SHOW_MEM_FILTER_NODES;
Mel Gormand0164ad2015-11-06 16:28:21 -08003096 if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
Dave Hansena238ab52011-05-24 17:12:16 -07003097 filter &= ~SHOW_MEM_FILTER_NODES;
3098
Michal Hocko9af744d2017-02-22 15:46:16 -08003099 show_mem(filter, nodemask);
Michal Hockoaa187502017-02-22 15:41:45 -08003100}
3101
Michal Hockoa8e99252017-02-22 15:46:10 -08003102void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
Michal Hockoaa187502017-02-22 15:41:45 -08003103{
3104 struct va_format vaf;
3105 va_list args;
3106 static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL,
3107 DEFAULT_RATELIMIT_BURST);
3108
3109 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
3110 debug_guardpage_minorder() > 0)
3111 return;
3112
Michal Hocko7877cdc2016-10-07 17:01:55 -07003113 pr_warn("%s: ", current->comm);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07003114
Michal Hocko7877cdc2016-10-07 17:01:55 -07003115 va_start(args, fmt);
3116 vaf.fmt = fmt;
3117 vaf.va = &args;
3118 pr_cont("%pV", &vaf);
3119 va_end(args);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07003120
David Rientjes685dbf62017-02-22 15:46:28 -08003121 pr_cont(", mode:%#x(%pGg), nodemask=", gfp_mask, &gfp_mask);
3122 if (nodemask)
3123 pr_cont("%*pbl\n", nodemask_pr_args(nodemask));
3124 else
3125 pr_cont("(null)\n");
3126
Michal Hockoa8e99252017-02-22 15:46:10 -08003127 cpuset_print_current_mems_allowed();
Joe Perches3ee9a4f2011-10-31 17:08:35 -07003128
Dave Hansena238ab52011-05-24 17:12:16 -07003129 dump_stack();
David Rientjes685dbf62017-02-22 15:46:28 -08003130 warn_alloc_show_mem(gfp_mask, nodemask);
Dave Hansena238ab52011-05-24 17:12:16 -07003131}
3132
Mel Gorman11e33f62009-06-16 15:31:57 -07003133static inline struct page *
Michal Hocko6c18ba72017-02-22 15:46:25 -08003134__alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order,
3135 unsigned int alloc_flags,
3136 const struct alloc_context *ac)
3137{
3138 struct page *page;
3139
3140 page = get_page_from_freelist(gfp_mask, order,
3141 alloc_flags|ALLOC_CPUSET, ac);
3142 /*
3143 * fallback to ignore cpuset restriction if our nodes
3144 * are depleted
3145 */
3146 if (!page)
3147 page = get_page_from_freelist(gfp_mask, order,
3148 alloc_flags, ac);
3149
3150 return page;
3151}
3152
3153static inline struct page *
Mel Gorman11e33f62009-06-16 15:31:57 -07003154__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003155 const struct alloc_context *ac, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07003156{
David Rientjes6e0fc462015-09-08 15:00:36 -07003157 struct oom_control oc = {
3158 .zonelist = ac->zonelist,
3159 .nodemask = ac->nodemask,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07003160 .memcg = NULL,
David Rientjes6e0fc462015-09-08 15:00:36 -07003161 .gfp_mask = gfp_mask,
3162 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07003163 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165
Johannes Weiner9879de72015-01-26 12:58:32 -08003166 *did_some_progress = 0;
3167
Johannes Weiner9879de72015-01-26 12:58:32 -08003168 /*
Johannes Weinerdc564012015-06-24 16:57:19 -07003169 * Acquire the oom lock. If that fails, somebody else is
3170 * making progress for us.
Johannes Weiner9879de72015-01-26 12:58:32 -08003171 */
Johannes Weinerdc564012015-06-24 16:57:19 -07003172 if (!mutex_trylock(&oom_lock)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08003173 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07003174 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003175 return NULL;
3176 }
Jens Axboe6b1de912005-11-17 21:35:02 +01003177
Mel Gorman11e33f62009-06-16 15:31:57 -07003178 /*
3179 * Go through the zonelist yet one more time, keep very high watermark
3180 * here, this is only to catch a parallel oom killing, we must fail if
3181 * we're still under heavy pressure.
3182 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003183 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
3184 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003185 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07003186 goto out;
3187
Michal Hocko06ad2762017-02-22 15:46:22 -08003188 /* Coredumps can quickly deplete all memory reserves */
3189 if (current->flags & PF_DUMPCORE)
3190 goto out;
3191 /* The OOM killer will not help higher order allocs */
3192 if (order > PAGE_ALLOC_COSTLY_ORDER)
3193 goto out;
3194 /* The OOM killer does not needlessly kill tasks for lowmem */
3195 if (ac->high_zoneidx < ZONE_NORMAL)
3196 goto out;
3197 if (pm_suspended_storage())
3198 goto out;
3199 /*
3200 * XXX: GFP_NOFS allocations should rather fail than rely on
3201 * other request to make a forward progress.
3202 * We are in an unfortunate situation where out_of_memory cannot
3203 * do much for this context but let's try it to at least get
3204 * access to memory reserved if the current task is killed (see
3205 * out_of_memory). Once filesystems are ready to handle allocation
3206 * failures more gracefully we should just bail out here.
3207 */
Michal Hocko3da88fb2016-05-19 17:13:09 -07003208
Michal Hocko06ad2762017-02-22 15:46:22 -08003209 /* The OOM killer may not free memory on a specific node */
3210 if (gfp_mask & __GFP_THISNODE)
3211 goto out;
3212
Mel Gorman11e33f62009-06-16 15:31:57 -07003213 /* Exhausted what can be done so it's blamo time */
Michal Hocko5020e282016-01-14 15:20:36 -08003214 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
Michal Hockoc32b3cb2015-02-11 15:26:24 -08003215 *did_some_progress = 1;
Michal Hocko5020e282016-01-14 15:20:36 -08003216
Michal Hocko6c18ba72017-02-22 15:46:25 -08003217 /*
3218 * Help non-failing allocations by giving them access to memory
3219 * reserves
3220 */
3221 if (gfp_mask & __GFP_NOFAIL)
3222 page = __alloc_pages_cpuset_fallback(gfp_mask, order,
Michal Hocko5020e282016-01-14 15:20:36 -08003223 ALLOC_NO_WATERMARKS, ac);
Michal Hocko5020e282016-01-14 15:20:36 -08003224 }
Mel Gorman11e33f62009-06-16 15:31:57 -07003225out:
Johannes Weinerdc564012015-06-24 16:57:19 -07003226 mutex_unlock(&oom_lock);
Mel Gorman11e33f62009-06-16 15:31:57 -07003227 return page;
3228}
3229
Michal Hocko33c2d212016-05-20 16:57:06 -07003230/*
3231 * Maximum number of compaction retries wit a progress before OOM
3232 * killer is consider as the only way to move forward.
3233 */
3234#define MAX_COMPACT_RETRIES 16
3235
Mel Gorman56de7262010-05-24 14:32:30 -07003236#ifdef CONFIG_COMPACTION
3237/* Try memory compaction for high-order allocations before reclaim */
3238static struct page *
3239__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07003240 unsigned int alloc_flags, const struct alloc_context *ac,
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003241 enum compact_priority prio, enum compact_result *compact_result)
Mel Gorman56de7262010-05-24 14:32:30 -07003242{
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003243 struct page *page;
Vlastimil Babka53853e22014-10-09 15:27:02 -07003244
Mel Gorman66199712012-01-12 17:19:41 -08003245 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07003246 return NULL;
3247
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003248 current->flags |= PF_MEMALLOC;
Michal Hockoc5d01d02016-05-20 16:56:53 -07003249 *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
Vlastimil Babkac3486f52016-07-28 15:49:30 -07003250 prio);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003251 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07003252
Michal Hockoc5d01d02016-05-20 16:56:53 -07003253 if (*compact_result <= COMPACT_INACTIVE)
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003254 return NULL;
Mel Gorman56de7262010-05-24 14:32:30 -07003255
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003256 /*
3257 * At least in one zone compaction wasn't deferred or skipped, so let's
3258 * count a compaction stall
3259 */
3260 count_vm_event(COMPACTSTALL);
3261
Vlastimil Babka31a6c192016-07-28 15:49:13 -07003262 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003263
3264 if (page) {
3265 struct zone *zone = page_zone(page);
3266
3267 zone->compact_blockskip_flush = false;
3268 compaction_defer_reset(zone, order, true);
3269 count_vm_event(COMPACTSUCCESS);
3270 return page;
3271 }
3272
3273 /*
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003274 * It's bad if compaction run occurs and fails. The most likely reason
3275 * is that pages exist, but not enough to satisfy watermarks.
3276 */
3277 count_vm_event(COMPACTFAIL);
3278
3279 cond_resched();
3280
Mel Gorman56de7262010-05-24 14:32:30 -07003281 return NULL;
3282}
Michal Hocko33c2d212016-05-20 16:57:06 -07003283
Vlastimil Babka32508452016-10-07 17:00:28 -07003284static inline bool
3285should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
3286 enum compact_result compact_result,
3287 enum compact_priority *compact_priority,
Vlastimil Babkad9436492016-10-07 17:00:31 -07003288 int *compaction_retries)
Vlastimil Babka32508452016-10-07 17:00:28 -07003289{
3290 int max_retries = MAX_COMPACT_RETRIES;
Vlastimil Babkac2033b02016-10-07 17:00:34 -07003291 int min_priority;
Michal Hocko65190cf2017-02-22 15:42:03 -08003292 bool ret = false;
3293 int retries = *compaction_retries;
3294 enum compact_priority priority = *compact_priority;
Vlastimil Babka32508452016-10-07 17:00:28 -07003295
3296 if (!order)
3297 return false;
3298
Vlastimil Babkad9436492016-10-07 17:00:31 -07003299 if (compaction_made_progress(compact_result))
3300 (*compaction_retries)++;
3301
Vlastimil Babka32508452016-10-07 17:00:28 -07003302 /*
3303 * compaction considers all the zone as desperately out of memory
3304 * so it doesn't really make much sense to retry except when the
3305 * failure could be caused by insufficient priority
3306 */
Vlastimil Babkad9436492016-10-07 17:00:31 -07003307 if (compaction_failed(compact_result))
3308 goto check_priority;
Vlastimil Babka32508452016-10-07 17:00:28 -07003309
3310 /*
3311 * make sure the compaction wasn't deferred or didn't bail out early
3312 * due to locks contention before we declare that we should give up.
3313 * But do not retry if the given zonelist is not suitable for
3314 * compaction.
3315 */
Michal Hocko65190cf2017-02-22 15:42:03 -08003316 if (compaction_withdrawn(compact_result)) {
3317 ret = compaction_zonelist_suitable(ac, order, alloc_flags);
3318 goto out;
3319 }
Vlastimil Babka32508452016-10-07 17:00:28 -07003320
3321 /*
3322 * !costly requests are much more important than __GFP_REPEAT
3323 * costly ones because they are de facto nofail and invoke OOM
3324 * killer to move on while costly can fail and users are ready
3325 * to cope with that. 1/4 retries is rather arbitrary but we
3326 * would need much more detailed feedback from compaction to
3327 * make a better decision.
3328 */
3329 if (order > PAGE_ALLOC_COSTLY_ORDER)
3330 max_retries /= 4;
Michal Hocko65190cf2017-02-22 15:42:03 -08003331 if (*compaction_retries <= max_retries) {
3332 ret = true;
3333 goto out;
3334 }
Vlastimil Babka32508452016-10-07 17:00:28 -07003335
Vlastimil Babkad9436492016-10-07 17:00:31 -07003336 /*
3337 * Make sure there are attempts at the highest priority if we exhausted
3338 * all retries or failed at the lower priorities.
3339 */
3340check_priority:
Vlastimil Babkac2033b02016-10-07 17:00:34 -07003341 min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
3342 MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
Michal Hocko65190cf2017-02-22 15:42:03 -08003343
Vlastimil Babkac2033b02016-10-07 17:00:34 -07003344 if (*compact_priority > min_priority) {
Vlastimil Babkad9436492016-10-07 17:00:31 -07003345 (*compact_priority)--;
3346 *compaction_retries = 0;
Michal Hocko65190cf2017-02-22 15:42:03 -08003347 ret = true;
Vlastimil Babkad9436492016-10-07 17:00:31 -07003348 }
Michal Hocko65190cf2017-02-22 15:42:03 -08003349out:
3350 trace_compact_retry(order, priority, compact_result, retries, max_retries, ret);
3351 return ret;
Vlastimil Babka32508452016-10-07 17:00:28 -07003352}
Mel Gorman56de7262010-05-24 14:32:30 -07003353#else
3354static inline struct page *
3355__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07003356 unsigned int alloc_flags, const struct alloc_context *ac,
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003357 enum compact_priority prio, enum compact_result *compact_result)
Mel Gorman56de7262010-05-24 14:32:30 -07003358{
Michal Hocko33c2d212016-05-20 16:57:06 -07003359 *compact_result = COMPACT_SKIPPED;
Mel Gorman56de7262010-05-24 14:32:30 -07003360 return NULL;
3361}
Michal Hocko33c2d212016-05-20 16:57:06 -07003362
3363static inline bool
Michal Hocko86a294a2016-05-20 16:57:12 -07003364should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
3365 enum compact_result compact_result,
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003366 enum compact_priority *compact_priority,
Vlastimil Babkad9436492016-10-07 17:00:31 -07003367 int *compaction_retries)
Michal Hocko33c2d212016-05-20 16:57:06 -07003368{
Michal Hocko31e49bf2016-05-20 16:57:15 -07003369 struct zone *zone;
3370 struct zoneref *z;
3371
3372 if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
3373 return false;
3374
3375 /*
3376 * There are setups with compaction disabled which would prefer to loop
3377 * inside the allocator rather than hit the oom killer prematurely.
3378 * Let's give them a good hope and keep retrying while the order-0
3379 * watermarks are OK.
3380 */
3381 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3382 ac->nodemask) {
3383 if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
3384 ac_classzone_idx(ac), alloc_flags))
3385 return true;
3386 }
Michal Hocko33c2d212016-05-20 16:57:06 -07003387 return false;
3388}
Vlastimil Babka32508452016-10-07 17:00:28 -07003389#endif /* CONFIG_COMPACTION */
Mel Gorman56de7262010-05-24 14:32:30 -07003390
Marek Szyprowskibba90712012-01-25 12:09:52 +01003391/* Perform direct synchronous page reclaim */
3392static int
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003393__perform_reclaim(gfp_t gfp_mask, unsigned int order,
3394 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003395{
Mel Gorman11e33f62009-06-16 15:31:57 -07003396 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01003397 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07003398
3399 cond_resched();
3400
3401 /* We now go into synchronous reclaim */
3402 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003403 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07003404 lockdep_set_current_reclaim_state(gfp_mask);
3405 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003406 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07003407
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003408 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
3409 ac->nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07003410
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003411 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07003412 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003413 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07003414
3415 cond_resched();
3416
Marek Szyprowskibba90712012-01-25 12:09:52 +01003417 return progress;
3418}
3419
3420/* The really slow allocator path where we enter direct reclaim */
3421static inline struct page *
3422__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07003423 unsigned int alloc_flags, const struct alloc_context *ac,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003424 unsigned long *did_some_progress)
Marek Szyprowskibba90712012-01-25 12:09:52 +01003425{
3426 struct page *page = NULL;
3427 bool drained = false;
3428
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003429 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003430 if (unlikely(!(*did_some_progress)))
3431 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07003432
Mel Gorman9ee493c2010-09-09 16:38:18 -07003433retry:
Vlastimil Babka31a6c192016-07-28 15:49:13 -07003434 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003435
3436 /*
3437 * If an allocation failed after direct reclaim, it could be because
Mel Gorman0aaa29a2015-11-06 16:28:37 -08003438 * pages are pinned on the per-cpu lists or in high alloc reserves.
3439 * Shrink them them and try again
Mel Gorman9ee493c2010-09-09 16:38:18 -07003440 */
3441 if (!page && !drained) {
Minchan Kim29fac032016-12-12 16:42:14 -08003442 unreserve_highatomic_pageblock(ac, false);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08003443 drain_all_pages(NULL);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003444 drained = true;
3445 goto retry;
3446 }
3447
Mel Gorman11e33f62009-06-16 15:31:57 -07003448 return page;
3449}
3450
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003451static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003452{
3453 struct zoneref *z;
3454 struct zone *zone;
Mel Gormane1a55632016-07-28 15:46:26 -07003455 pg_data_t *last_pgdat = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07003456
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003457 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
Mel Gormane1a55632016-07-28 15:46:26 -07003458 ac->high_zoneidx, ac->nodemask) {
3459 if (last_pgdat != zone->zone_pgdat)
Mel Gorman52e9f872016-07-28 15:46:29 -07003460 wakeup_kswapd(zone, order, ac->high_zoneidx);
Mel Gormane1a55632016-07-28 15:46:26 -07003461 last_pgdat = zone->zone_pgdat;
3462 }
Mel Gorman11e33f62009-06-16 15:31:57 -07003463}
3464
Mel Gormanc6038442016-05-19 17:13:38 -07003465static inline unsigned int
Peter Zijlstra341ce062009-06-16 15:32:02 -07003466gfp_to_alloc_flags(gfp_t gfp_mask)
3467{
Mel Gormanc6038442016-05-19 17:13:38 -07003468 unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
Peter Zijlstra341ce062009-06-16 15:32:02 -07003469
Mel Gormana56f57f2009-06-16 15:32:02 -07003470 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07003471 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07003472
Peter Zijlstra341ce062009-06-16 15:32:02 -07003473 /*
3474 * The caller may dip into page reserves a bit more if the caller
3475 * cannot run direct reclaim, or if the caller has realtime scheduling
3476 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
Mel Gormand0164ad2015-11-06 16:28:21 -08003477 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
Peter Zijlstra341ce062009-06-16 15:32:02 -07003478 */
Namhyung Kime6223a32010-10-26 14:21:59 -07003479 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07003480
Mel Gormand0164ad2015-11-06 16:28:21 -08003481 if (gfp_mask & __GFP_ATOMIC) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003482 /*
David Rientjesb104a352014-07-30 16:08:24 -07003483 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3484 * if it can't schedule.
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003485 */
David Rientjesb104a352014-07-30 16:08:24 -07003486 if (!(gfp_mask & __GFP_NOMEMALLOC))
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003487 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07003488 /*
David Rientjesb104a352014-07-30 16:08:24 -07003489 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
Vladimir Davydov344736f2014-10-20 15:50:30 +04003490 * comment for __cpuset_node_allowed().
Peter Zijlstra341ce062009-06-16 15:32:02 -07003491 */
3492 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003493 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07003494 alloc_flags |= ALLOC_HARDER;
3495
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07003496#ifdef CONFIG_CMA
David Rientjes43e7a342014-10-09 15:27:25 -07003497 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07003498 alloc_flags |= ALLOC_CMA;
3499#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07003500 return alloc_flags;
3501}
3502
Mel Gorman072bb0a2012-07-31 16:43:58 -07003503bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
3504{
Vlastimil Babka31a6c192016-07-28 15:49:13 -07003505 if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
3506 return false;
3507
3508 if (gfp_mask & __GFP_MEMALLOC)
3509 return true;
3510 if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
3511 return true;
3512 if (!in_interrupt() &&
3513 ((current->flags & PF_MEMALLOC) ||
3514 unlikely(test_thread_flag(TIF_MEMDIE))))
3515 return true;
3516
3517 return false;
Mel Gorman072bb0a2012-07-31 16:43:58 -07003518}
3519
Michal Hocko0a0337e2016-05-20 16:57:00 -07003520/*
3521 * Maximum number of reclaim retries without any progress before OOM killer
3522 * is consider as the only way to move forward.
3523 */
3524#define MAX_RECLAIM_RETRIES 16
3525
3526/*
3527 * Checks whether it makes sense to retry the reclaim to make a forward progress
3528 * for the given allocation request.
3529 * The reclaim feedback represented by did_some_progress (any progress during
Michal Hocko7854ea62016-05-20 16:57:09 -07003530 * the last reclaim round) and no_progress_loops (number of reclaim rounds without
3531 * any progress in a row) is considered as well as the reclaimable pages on the
3532 * applicable zone list (with a backoff mechanism which is a function of
3533 * no_progress_loops).
Michal Hocko0a0337e2016-05-20 16:57:00 -07003534 *
3535 * Returns true if a retry is viable or false to enter the oom path.
3536 */
3537static inline bool
3538should_reclaim_retry(gfp_t gfp_mask, unsigned order,
3539 struct alloc_context *ac, int alloc_flags,
Vlastimil Babka423b4522016-10-07 17:00:40 -07003540 bool did_some_progress, int *no_progress_loops)
Michal Hocko0a0337e2016-05-20 16:57:00 -07003541{
3542 struct zone *zone;
3543 struct zoneref *z;
3544
3545 /*
Vlastimil Babka423b4522016-10-07 17:00:40 -07003546 * Costly allocations might have made a progress but this doesn't mean
3547 * their order will become available due to high fragmentation so
3548 * always increment the no progress counter for them
3549 */
3550 if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
3551 *no_progress_loops = 0;
3552 else
3553 (*no_progress_loops)++;
3554
3555 /*
Michal Hocko0a0337e2016-05-20 16:57:00 -07003556 * Make sure we converge to OOM if we cannot make any progress
3557 * several times in the row.
3558 */
Minchan Kim04c87162016-12-12 16:42:11 -08003559 if (*no_progress_loops > MAX_RECLAIM_RETRIES) {
3560 /* Before OOM, exhaust highatomic_reserve */
Minchan Kim29fac032016-12-12 16:42:14 -08003561 return unreserve_highatomic_pageblock(ac, true);
Minchan Kim04c87162016-12-12 16:42:11 -08003562 }
Michal Hocko0a0337e2016-05-20 16:57:00 -07003563
Michal Hocko0a0337e2016-05-20 16:57:00 -07003564 /*
Mel Gormanbca67592016-07-28 15:47:05 -07003565 * Keep reclaiming pages while there is a chance this will lead
3566 * somewhere. If none of the target zones can satisfy our allocation
3567 * request even if all reclaimable pages are considered then we are
3568 * screwed and have to go OOM.
Michal Hocko0a0337e2016-05-20 16:57:00 -07003569 */
3570 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3571 ac->nodemask) {
3572 unsigned long available;
Michal Hockoede37712016-05-20 16:57:03 -07003573 unsigned long reclaimable;
Michal Hockod379f012017-02-22 15:42:00 -08003574 unsigned long min_wmark = min_wmark_pages(zone);
3575 bool wmark;
Michal Hocko0a0337e2016-05-20 16:57:00 -07003576
Mel Gorman5a1c84b2016-07-28 15:47:31 -07003577 available = reclaimable = zone_reclaimable_pages(zone);
Vlastimil Babka423b4522016-10-07 17:00:40 -07003578 available -= DIV_ROUND_UP((*no_progress_loops) * available,
Michal Hocko0a0337e2016-05-20 16:57:00 -07003579 MAX_RECLAIM_RETRIES);
Mel Gorman5a1c84b2016-07-28 15:47:31 -07003580 available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
Michal Hocko0a0337e2016-05-20 16:57:00 -07003581
3582 /*
3583 * Would the allocation succeed if we reclaimed the whole
Mel Gorman5a1c84b2016-07-28 15:47:31 -07003584 * available?
Michal Hocko0a0337e2016-05-20 16:57:00 -07003585 */
Michal Hockod379f012017-02-22 15:42:00 -08003586 wmark = __zone_watermark_ok(zone, order, min_wmark,
3587 ac_classzone_idx(ac), alloc_flags, available);
3588 trace_reclaim_retry_zone(z, order, reclaimable,
3589 available, min_wmark, *no_progress_loops, wmark);
3590 if (wmark) {
Michal Hockoede37712016-05-20 16:57:03 -07003591 /*
3592 * If we didn't make any progress and have a lot of
3593 * dirty + writeback pages then we should wait for
3594 * an IO to complete to slow down the reclaim and
3595 * prevent from pre mature OOM
3596 */
3597 if (!did_some_progress) {
Mel Gorman11fb9982016-07-28 15:46:20 -07003598 unsigned long write_pending;
Michal Hockoede37712016-05-20 16:57:03 -07003599
Mel Gorman5a1c84b2016-07-28 15:47:31 -07003600 write_pending = zone_page_state_snapshot(zone,
3601 NR_ZONE_WRITE_PENDING);
Michal Hockoede37712016-05-20 16:57:03 -07003602
Mel Gorman11fb9982016-07-28 15:46:20 -07003603 if (2 * write_pending > reclaimable) {
Michal Hockoede37712016-05-20 16:57:03 -07003604 congestion_wait(BLK_RW_ASYNC, HZ/10);
3605 return true;
3606 }
3607 }
Mel Gorman5a1c84b2016-07-28 15:47:31 -07003608
Michal Hockoede37712016-05-20 16:57:03 -07003609 /*
3610 * Memory allocation/reclaim might be called from a WQ
3611 * context and the current implementation of the WQ
3612 * concurrency control doesn't recognize that
3613 * a particular WQ is congested if the worker thread is
3614 * looping without ever sleeping. Therefore we have to
3615 * do a short sleep here rather than calling
3616 * cond_resched().
3617 */
3618 if (current->flags & PF_WQ_WORKER)
3619 schedule_timeout_uninterruptible(1);
3620 else
3621 cond_resched();
3622
Michal Hocko0a0337e2016-05-20 16:57:00 -07003623 return true;
3624 }
3625 }
3626
3627 return false;
3628}
3629
Mel Gorman11e33f62009-06-16 15:31:57 -07003630static inline struct page *
3631__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003632 struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003633{
Mel Gormand0164ad2015-11-06 16:28:21 -08003634 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
Mel Gorman11e33f62009-06-16 15:31:57 -07003635 struct page *page = NULL;
Mel Gormanc6038442016-05-19 17:13:38 -07003636 unsigned int alloc_flags;
Mel Gorman11e33f62009-06-16 15:31:57 -07003637 unsigned long did_some_progress;
Vlastimil Babka5ce9bfe2017-01-24 15:18:38 -08003638 enum compact_priority compact_priority;
Michal Hockoc5d01d02016-05-20 16:56:53 -07003639 enum compact_result compact_result;
Vlastimil Babka5ce9bfe2017-01-24 15:18:38 -08003640 int compaction_retries;
3641 int no_progress_loops;
Michal Hocko63f53de2016-10-07 17:01:58 -07003642 unsigned long alloc_start = jiffies;
3643 unsigned int stall_timeout = 10 * HZ;
Vlastimil Babka5ce9bfe2017-01-24 15:18:38 -08003644 unsigned int cpuset_mems_cookie;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003645
Christoph Lameter952f3b52006-12-06 20:33:26 -08003646 /*
Mel Gorman72807a72009-06-16 15:32:18 -07003647 * In the slowpath, we sanity check order to avoid ever trying to
3648 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3649 * be using allocators in order of preference for an area that is
3650 * too large.
3651 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07003652 if (order >= MAX_ORDER) {
3653 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07003654 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07003655 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003656
Christoph Lameter952f3b52006-12-06 20:33:26 -08003657 /*
Mel Gormand0164ad2015-11-06 16:28:21 -08003658 * We also sanity check to catch abuse of atomic reserves being used by
3659 * callers that are not in atomic context.
3660 */
3661 if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
3662 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
3663 gfp_mask &= ~__GFP_ATOMIC;
3664
Vlastimil Babka5ce9bfe2017-01-24 15:18:38 -08003665retry_cpuset:
3666 compaction_retries = 0;
3667 no_progress_loops = 0;
3668 compact_priority = DEF_COMPACT_PRIORITY;
3669 cpuset_mems_cookie = read_mems_allowed_begin();
Michal Hocko9a67f642017-02-22 15:46:19 -08003670
3671 /*
3672 * The fast path uses conservative alloc_flags to succeed only until
3673 * kswapd needs to be woken up, and to avoid the cost of setting up
3674 * alloc_flags precisely. So we do that now.
3675 */
3676 alloc_flags = gfp_to_alloc_flags(gfp_mask);
3677
Vlastimil Babkae47483b2017-01-24 15:18:41 -08003678 /*
3679 * We need to recalculate the starting point for the zonelist iterator
3680 * because we might have used different nodemask in the fast path, or
3681 * there was a cpuset modification and we are retrying - otherwise we
3682 * could end up iterating over non-eligible zones endlessly.
3683 */
3684 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
3685 ac->high_zoneidx, ac->nodemask);
3686 if (!ac->preferred_zoneref->zone)
3687 goto nopage;
3688
Mel Gormand0164ad2015-11-06 16:28:21 -08003689 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003690 wake_all_kswapds(order, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003691
Paul Jackson9bf22292005-09-06 15:18:12 -07003692 /*
Vlastimil Babka23771232016-07-28 15:49:16 -07003693 * The adjusted alloc_flags might result in immediate success, so try
3694 * that first
3695 */
3696 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3697 if (page)
3698 goto got_pg;
3699
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003700 /*
3701 * For costly allocations, try direct compaction first, as it's likely
3702 * that we have enough base pages and don't need to reclaim. Don't try
3703 * that for allocations that are allowed to ignore watermarks, as the
3704 * ALLOC_NO_WATERMARKS attempt didn't yet happen.
3705 */
3706 if (can_direct_reclaim && order > PAGE_ALLOC_COSTLY_ORDER &&
3707 !gfp_pfmemalloc_allowed(gfp_mask)) {
3708 page = __alloc_pages_direct_compact(gfp_mask, order,
3709 alloc_flags, ac,
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003710 INIT_COMPACT_PRIORITY,
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003711 &compact_result);
3712 if (page)
3713 goto got_pg;
3714
Vlastimil Babka3eb27712016-07-28 15:49:22 -07003715 /*
3716 * Checks for costly allocations with __GFP_NORETRY, which
3717 * includes THP page fault allocations
3718 */
3719 if (gfp_mask & __GFP_NORETRY) {
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003720 /*
3721 * If compaction is deferred for high-order allocations,
3722 * it is because sync compaction recently failed. If
3723 * this is the case and the caller requested a THP
3724 * allocation, we do not want to heavily disrupt the
3725 * system, so we fail the allocation instead of entering
3726 * direct reclaim.
3727 */
3728 if (compact_result == COMPACT_DEFERRED)
3729 goto nopage;
3730
3731 /*
Vlastimil Babka3eb27712016-07-28 15:49:22 -07003732 * Looks like reclaim/compaction is worth trying, but
3733 * sync compaction could be very expensive, so keep
Vlastimil Babka25160352016-07-28 15:49:25 -07003734 * using async compaction.
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003735 */
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003736 compact_priority = INIT_COMPACT_PRIORITY;
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003737 }
3738 }
Vlastimil Babka23771232016-07-28 15:49:16 -07003739
3740retry:
3741 /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
3742 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3743 wake_all_kswapds(order, ac);
3744
3745 if (gfp_pfmemalloc_allowed(gfp_mask))
3746 alloc_flags = ALLOC_NO_WATERMARKS;
3747
3748 /*
Mel Gormane46e7b72016-06-03 14:56:01 -07003749 * Reset the zonelist iterators if memory policies can be ignored.
3750 * These allocations are high priority and system rather than user
3751 * orientated.
3752 */
Vlastimil Babka23771232016-07-28 15:49:16 -07003753 if (!(alloc_flags & ALLOC_CPUSET) || (alloc_flags & ALLOC_NO_WATERMARKS)) {
Mel Gormane46e7b72016-06-03 14:56:01 -07003754 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
3755 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
3756 ac->high_zoneidx, ac->nodemask);
3757 }
3758
Vlastimil Babka23771232016-07-28 15:49:16 -07003759 /* Attempt with potentially adjusted zonelist and alloc_flags */
Vlastimil Babka31a6c192016-07-28 15:49:13 -07003760 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003761 if (page)
3762 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763
Mel Gormand0164ad2015-11-06 16:28:21 -08003764 /* Caller is not willing to reclaim, we can't balance anything */
Michal Hocko9a67f642017-02-22 15:46:19 -08003765 if (!can_direct_reclaim)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766 goto nopage;
Michal Hocko9a67f642017-02-22 15:46:19 -08003767
3768 /* Make sure we know about allocations which stall for too long */
3769 if (time_after(jiffies, alloc_start + stall_timeout)) {
3770 warn_alloc(gfp_mask, ac->nodemask,
3771 "page allocation stalls for %ums, order:%u",
3772 jiffies_to_msecs(jiffies-alloc_start), order);
3773 stall_timeout += 10 * HZ;
David Rientjesaed0a0e2014-01-21 15:51:12 -08003774 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775
Peter Zijlstra341ce062009-06-16 15:32:02 -07003776 /* Avoid recursion of direct reclaim */
Michal Hocko9a67f642017-02-22 15:46:19 -08003777 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07003778 goto nopage;
David Rientjes8fe78042014-08-06 16:07:54 -07003779
Mel Gorman11e33f62009-06-16 15:31:57 -07003780 /* Try direct reclaim and then allocating */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003781 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3782 &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07003783 if (page)
3784 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003786 /* Try direct compaction and then allocating */
3787 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003788 compact_priority, &compact_result);
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003789 if (page)
3790 goto got_pg;
3791
Johannes Weiner90839052015-06-24 16:57:21 -07003792 /* Do not loop if specifically requested */
3793 if (gfp_mask & __GFP_NORETRY)
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003794 goto nopage;
Johannes Weiner90839052015-06-24 16:57:21 -07003795
Michal Hocko0a0337e2016-05-20 16:57:00 -07003796 /*
3797 * Do not retry costly high order allocations unless they are
3798 * __GFP_REPEAT
3799 */
3800 if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_REPEAT))
Vlastimil Babkaa8161d12016-07-28 15:49:19 -07003801 goto nopage;
Michal Hocko0a0337e2016-05-20 16:57:00 -07003802
Michal Hocko0a0337e2016-05-20 16:57:00 -07003803 if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
Vlastimil Babka423b4522016-10-07 17:00:40 -07003804 did_some_progress > 0, &no_progress_loops))
Michal Hocko0a0337e2016-05-20 16:57:00 -07003805 goto retry;
3806
Michal Hocko33c2d212016-05-20 16:57:06 -07003807 /*
3808 * It doesn't make any sense to retry for the compaction if the order-0
3809 * reclaim is not able to make any progress because the current
3810 * implementation of the compaction depends on the sufficient amount
3811 * of free memory (see __compaction_suitable)
3812 */
3813 if (did_some_progress > 0 &&
Michal Hocko86a294a2016-05-20 16:57:12 -07003814 should_compact_retry(ac, order, alloc_flags,
Vlastimil Babkaa5508cd2016-07-28 15:49:28 -07003815 compact_result, &compact_priority,
Vlastimil Babkad9436492016-10-07 17:00:31 -07003816 &compaction_retries))
Michal Hocko33c2d212016-05-20 16:57:06 -07003817 goto retry;
3818
Vlastimil Babkae47483b2017-01-24 15:18:41 -08003819 /*
3820 * It's possible we raced with cpuset update so the OOM would be
3821 * premature (see below the nopage: label for full explanation).
3822 */
3823 if (read_mems_allowed_retry(cpuset_mems_cookie))
3824 goto retry_cpuset;
3825
Johannes Weiner90839052015-06-24 16:57:21 -07003826 /* Reclaim has failed us, start killing things */
3827 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3828 if (page)
3829 goto got_pg;
3830
Michal Hocko9a67f642017-02-22 15:46:19 -08003831 /* Avoid allocations with no watermarks from looping endlessly */
3832 if (test_thread_flag(TIF_MEMDIE))
3833 goto nopage;
3834
Johannes Weiner90839052015-06-24 16:57:21 -07003835 /* Retry as long as the OOM killer is making progress */
Michal Hocko0a0337e2016-05-20 16:57:00 -07003836 if (did_some_progress) {
3837 no_progress_loops = 0;
Johannes Weiner90839052015-06-24 16:57:21 -07003838 goto retry;
Michal Hocko0a0337e2016-05-20 16:57:00 -07003839 }
Johannes Weiner90839052015-06-24 16:57:21 -07003840
Linus Torvalds1da177e2005-04-16 15:20:36 -07003841nopage:
Vlastimil Babka5ce9bfe2017-01-24 15:18:38 -08003842 /*
Vlastimil Babkae47483b2017-01-24 15:18:41 -08003843 * When updating a task's mems_allowed or mempolicy nodemask, it is
3844 * possible to race with parallel threads in such a way that our
3845 * allocation can fail while the mask is being updated. If we are about
3846 * to fail, check if the cpuset changed during allocation and if so,
3847 * retry.
Vlastimil Babka5ce9bfe2017-01-24 15:18:38 -08003848 */
3849 if (read_mems_allowed_retry(cpuset_mems_cookie))
3850 goto retry_cpuset;
3851
Michal Hocko9a67f642017-02-22 15:46:19 -08003852 /*
3853 * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
3854 * we always retry
3855 */
3856 if (gfp_mask & __GFP_NOFAIL) {
3857 /*
3858 * All existing users of the __GFP_NOFAIL are blockable, so warn
3859 * of any new users that actually require GFP_NOWAIT
3860 */
3861 if (WARN_ON_ONCE(!can_direct_reclaim))
3862 goto fail;
3863
3864 /*
3865 * PF_MEMALLOC request from this context is rather bizarre
3866 * because we cannot reclaim anything and only can loop waiting
3867 * for somebody to do a work for us
3868 */
3869 WARN_ON_ONCE(current->flags & PF_MEMALLOC);
3870
3871 /*
3872 * non failing costly orders are a hard requirement which we
3873 * are not prepared for much so let's warn about these users
3874 * so that we can identify them and convert them to something
3875 * else.
3876 */
3877 WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER);
3878
Michal Hocko6c18ba72017-02-22 15:46:25 -08003879 /*
3880 * Help non-failing allocations by giving them access to memory
3881 * reserves but do not use ALLOC_NO_WATERMARKS because this
3882 * could deplete whole memory reserves which would just make
3883 * the situation worse
3884 */
3885 page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac);
3886 if (page)
3887 goto got_pg;
3888
Michal Hocko9a67f642017-02-22 15:46:19 -08003889 cond_resched();
3890 goto retry;
3891 }
3892fail:
Michal Hockoa8e99252017-02-22 15:46:10 -08003893 warn_alloc(gfp_mask, ac->nodemask,
Michal Hocko7877cdc2016-10-07 17:01:55 -07003894 "page allocation failure: order:%u", order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003895got_pg:
Mel Gorman072bb0a2012-07-31 16:43:58 -07003896 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003897}
Mel Gorman11e33f62009-06-16 15:31:57 -07003898
Mel Gorman9cd75552017-02-24 14:56:29 -08003899static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
3900 struct zonelist *zonelist, nodemask_t *nodemask,
3901 struct alloc_context *ac, gfp_t *alloc_mask,
3902 unsigned int *alloc_flags)
3903{
3904 ac->high_zoneidx = gfp_zone(gfp_mask);
3905 ac->zonelist = zonelist;
3906 ac->nodemask = nodemask;
3907 ac->migratetype = gfpflags_to_migratetype(gfp_mask);
3908
3909 if (cpusets_enabled()) {
3910 *alloc_mask |= __GFP_HARDWALL;
Mel Gorman9cd75552017-02-24 14:56:29 -08003911 if (!ac->nodemask)
3912 ac->nodemask = &cpuset_current_mems_allowed;
Vlastimil Babka51047822017-02-24 14:56:53 -08003913 else
3914 *alloc_flags |= ALLOC_CPUSET;
Mel Gorman9cd75552017-02-24 14:56:29 -08003915 }
3916
3917 lockdep_trace_alloc(gfp_mask);
3918
3919 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
3920
3921 if (should_fail_alloc_page(gfp_mask, order))
3922 return false;
3923
Mel Gorman9cd75552017-02-24 14:56:29 -08003924 if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE)
3925 *alloc_flags |= ALLOC_CMA;
3926
3927 return true;
3928}
3929
3930/* Determine whether to spread dirty pages and what the first usable zone */
3931static inline void finalise_ac(gfp_t gfp_mask,
3932 unsigned int order, struct alloc_context *ac)
3933{
3934 /* Dirty zone balancing only done in the fast path */
3935 ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE);
3936
3937 /*
3938 * The preferred zone is used for statistics but crucially it is
3939 * also used as the starting point for the zonelist iterator. It
3940 * may get reset for allocations that ignore memory policies.
3941 */
3942 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
3943 ac->high_zoneidx, ac->nodemask);
3944}
3945
Mel Gorman11e33f62009-06-16 15:31:57 -07003946/*
3947 * This is the 'heart' of the zoned buddy allocator.
3948 */
3949struct page *
3950__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3951 struct zonelist *zonelist, nodemask_t *nodemask)
3952{
Mel Gorman5bb1b162016-05-19 17:13:50 -07003953 struct page *page;
Mel Gormane6cbd7f2016-07-28 15:46:50 -07003954 unsigned int alloc_flags = ALLOC_WMARK_LOW;
Mel Gorman83d4ca82016-05-19 17:13:56 -07003955 gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */
Mel Gorman9cd75552017-02-24 14:56:29 -08003956 struct alloc_context ac = { };
Mel Gorman682a3382016-05-19 17:13:30 -07003957
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003958 gfp_mask &= gfp_allowed_mask;
Mel Gorman9cd75552017-02-24 14:56:29 -08003959 if (!prepare_alloc_pages(gfp_mask, order, zonelist, nodemask, &ac, &alloc_mask, &alloc_flags))
Mel Gorman11e33f62009-06-16 15:31:57 -07003960 return NULL;
3961
Mel Gorman9cd75552017-02-24 14:56:29 -08003962 finalise_ac(gfp_mask, order, &ac);
Mel Gorman5bb1b162016-05-19 17:13:50 -07003963
Mel Gorman5117f452009-06-16 15:31:59 -07003964 /* First allocation attempt */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003965 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
Mel Gorman4fcb0972016-05-19 17:14:01 -07003966 if (likely(page))
3967 goto out;
Andrew Morton91fbdc02015-02-11 15:25:04 -08003968
Mel Gorman4fcb0972016-05-19 17:14:01 -07003969 /*
3970 * Runtime PM, block IO and its error handling path can deadlock
3971 * because I/O on the device might not complete.
3972 */
3973 alloc_mask = memalloc_noio_flags(gfp_mask);
3974 ac.spread_dirty_pages = false;
Mel Gorman11e33f62009-06-16 15:31:57 -07003975
Mel Gorman47415262016-05-19 17:14:44 -07003976 /*
3977 * Restore the original nodemask if it was potentially replaced with
3978 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
3979 */
Vlastimil Babkae47483b2017-01-24 15:18:41 -08003980 if (unlikely(ac.nodemask != nodemask))
Mel Gorman47415262016-05-19 17:14:44 -07003981 ac.nodemask = nodemask;
Vlastimil Babka16096c22017-01-24 15:18:35 -08003982
Mel Gorman4fcb0972016-05-19 17:14:01 -07003983 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
Xishi Qiu23f086f2015-02-11 15:25:07 -08003984
Mel Gorman4fcb0972016-05-19 17:14:01 -07003985out:
Vladimir Davydovc4159a72016-08-08 23:03:12 +03003986 if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page &&
3987 unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) {
3988 __free_pages(page, order);
3989 page = NULL;
Vladimir Davydov49491482016-07-26 15:24:24 -07003990 }
3991
Mel Gorman4fcb0972016-05-19 17:14:01 -07003992 if (kmemcheck_enabled && page)
3993 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3994
3995 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
3996
Mel Gorman11e33f62009-06-16 15:31:57 -07003997 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003998}
Mel Gormand2391712009-06-16 15:31:52 -07003999EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000
4001/*
4002 * Common helper functions.
4003 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08004004unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005{
Akinobu Mita945a1112009-09-21 17:01:47 -07004006 struct page *page;
4007
4008 /*
4009 * __get_free_pages() returns a 32-bit address, which cannot represent
4010 * a highmem page
4011 */
4012 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
4013
Linus Torvalds1da177e2005-04-16 15:20:36 -07004014 page = alloc_pages(gfp_mask, order);
4015 if (!page)
4016 return 0;
4017 return (unsigned long) page_address(page);
4018}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019EXPORT_SYMBOL(__get_free_pages);
4020
Harvey Harrison920c7a52008-02-04 22:29:26 -08004021unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004022{
Akinobu Mita945a1112009-09-21 17:01:47 -07004023 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004024}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004025EXPORT_SYMBOL(get_zeroed_page);
4026
Harvey Harrison920c7a52008-02-04 22:29:26 -08004027void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028{
Nick Pigginb5810032005-10-29 18:16:12 -07004029 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004030 if (order == 0)
Mel Gormanb745bc82014-06-04 16:10:22 -07004031 free_hot_cold_page(page, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004032 else
4033 __free_pages_ok(page, order);
4034 }
4035}
4036
4037EXPORT_SYMBOL(__free_pages);
4038
Harvey Harrison920c7a52008-02-04 22:29:26 -08004039void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004040{
4041 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07004042 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004043 __free_pages(virt_to_page((void *)addr), order);
4044 }
4045}
4046
4047EXPORT_SYMBOL(free_pages);
4048
Glauber Costa6a1a0d32012-12-18 14:22:00 -08004049/*
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07004050 * Page Fragment:
4051 * An arbitrary-length arbitrary-offset area of memory which resides
4052 * within a 0 or higher order page. Multiple fragments within that page
4053 * are individually refcounted, in the page's reference counter.
4054 *
4055 * The page_frag functions below provide a simple allocation framework for
4056 * page fragments. This is used by the network stack and network device
4057 * drivers to provide a backing region of memory for use as either an
4058 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
4059 */
Alexander Duyck2976db82017-01-10 16:58:09 -08004060static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
4061 gfp_t gfp_mask)
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07004062{
4063 struct page *page = NULL;
4064 gfp_t gfp = gfp_mask;
4065
4066#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4067 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
4068 __GFP_NOMEMALLOC;
4069 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
4070 PAGE_FRAG_CACHE_MAX_ORDER);
4071 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
4072#endif
4073 if (unlikely(!page))
4074 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
4075
4076 nc->va = page ? page_address(page) : NULL;
4077
4078 return page;
4079}
4080
Alexander Duyck2976db82017-01-10 16:58:09 -08004081void __page_frag_cache_drain(struct page *page, unsigned int count)
Alexander Duyck44fdffd2016-12-14 15:05:26 -08004082{
4083 VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
4084
4085 if (page_ref_sub_and_test(page, count)) {
Alexander Duyck2976db82017-01-10 16:58:09 -08004086 unsigned int order = compound_order(page);
4087
Alexander Duyck44fdffd2016-12-14 15:05:26 -08004088 if (order == 0)
4089 free_hot_cold_page(page, false);
4090 else
4091 __free_pages_ok(page, order);
4092 }
4093}
Alexander Duyck2976db82017-01-10 16:58:09 -08004094EXPORT_SYMBOL(__page_frag_cache_drain);
Alexander Duyck44fdffd2016-12-14 15:05:26 -08004095
Alexander Duyck8c2dd3e2017-01-10 16:58:06 -08004096void *page_frag_alloc(struct page_frag_cache *nc,
4097 unsigned int fragsz, gfp_t gfp_mask)
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07004098{
4099 unsigned int size = PAGE_SIZE;
4100 struct page *page;
4101 int offset;
4102
4103 if (unlikely(!nc->va)) {
4104refill:
Alexander Duyck2976db82017-01-10 16:58:09 -08004105 page = __page_frag_cache_refill(nc, gfp_mask);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07004106 if (!page)
4107 return NULL;
4108
4109#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4110 /* if size can vary use size else just use PAGE_SIZE */
4111 size = nc->size;
4112#endif
4113 /* Even if we own the page, we do not use atomic_set().
4114 * This would break get_page_unless_zero() users.
4115 */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07004116 page_ref_add(page, size - 1);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07004117
4118 /* reset page count bias and offset to start of new frag */
Michal Hocko2f064f32015-08-21 14:11:51 -07004119 nc->pfmemalloc = page_is_pfmemalloc(page);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07004120 nc->pagecnt_bias = size;
4121 nc->offset = size;
4122 }
4123
4124 offset = nc->offset - fragsz;
4125 if (unlikely(offset < 0)) {
4126 page = virt_to_page(nc->va);
4127
Joonsoo Kimfe896d12016-03-17 14:19:26 -07004128 if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07004129 goto refill;
4130
4131#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4132 /* if size can vary use size else just use PAGE_SIZE */
4133 size = nc->size;
4134#endif
4135 /* OK, page count is 0, we can safely set it */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07004136 set_page_count(page, size);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07004137
4138 /* reset page count bias and offset to start of new frag */
4139 nc->pagecnt_bias = size;
4140 offset = size - fragsz;
4141 }
4142
4143 nc->pagecnt_bias--;
4144 nc->offset = offset;
4145
4146 return nc->va + offset;
4147}
Alexander Duyck8c2dd3e2017-01-10 16:58:06 -08004148EXPORT_SYMBOL(page_frag_alloc);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07004149
4150/*
4151 * Frees a page fragment allocated out of either a compound or order 0 page.
4152 */
Alexander Duyck8c2dd3e2017-01-10 16:58:06 -08004153void page_frag_free(void *addr)
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07004154{
4155 struct page *page = virt_to_head_page(addr);
4156
4157 if (unlikely(put_page_testzero(page)))
4158 __free_pages_ok(page, compound_order(page));
4159}
Alexander Duyck8c2dd3e2017-01-10 16:58:06 -08004160EXPORT_SYMBOL(page_frag_free);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07004161
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004162static void *make_alloc_exact(unsigned long addr, unsigned int order,
4163 size_t size)
Andi Kleenee85c2e2011-05-11 15:13:34 -07004164{
4165 if (addr) {
4166 unsigned long alloc_end = addr + (PAGE_SIZE << order);
4167 unsigned long used = addr + PAGE_ALIGN(size);
4168
4169 split_page(virt_to_page((void *)addr), order);
4170 while (used < alloc_end) {
4171 free_page(used);
4172 used += PAGE_SIZE;
4173 }
4174 }
4175 return (void *)addr;
4176}
4177
Timur Tabi2be0ffe2008-07-23 21:28:11 -07004178/**
4179 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4180 * @size: the number of bytes to allocate
4181 * @gfp_mask: GFP flags for the allocation
4182 *
4183 * This function is similar to alloc_pages(), except that it allocates the
4184 * minimum number of pages to satisfy the request. alloc_pages() can only
4185 * allocate memory in power-of-two pages.
4186 *
4187 * This function is also limited by MAX_ORDER.
4188 *
4189 * Memory allocated by this function must be released by free_pages_exact().
4190 */
4191void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
4192{
4193 unsigned int order = get_order(size);
4194 unsigned long addr;
4195
4196 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07004197 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07004198}
4199EXPORT_SYMBOL(alloc_pages_exact);
4200
4201/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07004202 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4203 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07004204 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07004205 * @size: the number of bytes to allocate
4206 * @gfp_mask: GFP flags for the allocation
4207 *
4208 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4209 * back.
Andi Kleenee85c2e2011-05-11 15:13:34 -07004210 */
Fabian Fredericke1931812014-08-06 16:04:59 -07004211void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
Andi Kleenee85c2e2011-05-11 15:13:34 -07004212{
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004213 unsigned int order = get_order(size);
Andi Kleenee85c2e2011-05-11 15:13:34 -07004214 struct page *p = alloc_pages_node(nid, gfp_mask, order);
4215 if (!p)
4216 return NULL;
4217 return make_alloc_exact((unsigned long)page_address(p), order, size);
4218}
Andi Kleenee85c2e2011-05-11 15:13:34 -07004219
4220/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07004221 * free_pages_exact - release memory allocated via alloc_pages_exact()
4222 * @virt: the value returned by alloc_pages_exact.
4223 * @size: size of allocation, same value as passed to alloc_pages_exact().
4224 *
4225 * Release the memory allocated by a previous call to alloc_pages_exact.
4226 */
4227void free_pages_exact(void *virt, size_t size)
4228{
4229 unsigned long addr = (unsigned long)virt;
4230 unsigned long end = addr + PAGE_ALIGN(size);
4231
4232 while (addr < end) {
4233 free_page(addr);
4234 addr += PAGE_SIZE;
4235 }
4236}
4237EXPORT_SYMBOL(free_pages_exact);
4238
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004239/**
4240 * nr_free_zone_pages - count number of pages beyond high watermark
4241 * @offset: The zone index of the highest zone
4242 *
4243 * nr_free_zone_pages() counts the number of counts pages which are beyond the
4244 * high watermark within all zones at or below a given zone index. For each
4245 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07004246 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004247 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004248static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004249{
Mel Gormandd1a2392008-04-28 02:12:17 -07004250 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07004251 struct zone *zone;
4252
Martin J. Blighe310fd42005-07-29 22:59:18 -07004253 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004254 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004255
Mel Gorman0e884602008-04-28 02:12:14 -07004256 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004257
Mel Gorman54a6eb52008-04-28 02:12:16 -07004258 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08004259 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07004260 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07004261 if (size > high)
4262 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004263 }
4264
4265 return sum;
4266}
4267
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004268/**
4269 * nr_free_buffer_pages - count number of pages beyond high watermark
4270 *
4271 * nr_free_buffer_pages() counts the number of pages which are beyond the high
4272 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004273 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004274unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004275{
Al Viroaf4ca452005-10-21 02:55:38 -04004276 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004277}
Meelap Shahc2f1a552007-07-17 04:04:39 -07004278EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004279
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004280/**
4281 * nr_free_pagecache_pages - count number of pages beyond high watermark
4282 *
4283 * nr_free_pagecache_pages() counts the number of pages which are beyond the
4284 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004285 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004286unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004287{
Mel Gorman2a1e2742007-07-17 04:03:12 -07004288 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004289}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07004290
4291static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004292{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08004293 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08004294 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004295}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004296
Igor Redkod02bd272016-03-17 14:19:05 -07004297long si_mem_available(void)
4298{
4299 long available;
4300 unsigned long pagecache;
4301 unsigned long wmark_low = 0;
4302 unsigned long pages[NR_LRU_LISTS];
4303 struct zone *zone;
4304 int lru;
4305
4306 for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
Mel Gorman2f95ff92016-08-11 15:32:57 -07004307 pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
Igor Redkod02bd272016-03-17 14:19:05 -07004308
4309 for_each_zone(zone)
4310 wmark_low += zone->watermark[WMARK_LOW];
4311
4312 /*
4313 * Estimate the amount of memory available for userspace allocations,
4314 * without causing swapping.
4315 */
4316 available = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
4317
4318 /*
4319 * Not all the page cache can be freed, otherwise the system will
4320 * start swapping. Assume at least half of the page cache, or the
4321 * low watermark worth of cache, needs to stay.
4322 */
4323 pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
4324 pagecache -= min(pagecache / 2, wmark_low);
4325 available += pagecache;
4326
4327 /*
4328 * Part of the reclaimable slab consists of items that are in use,
4329 * and cannot be freed. Cap this estimate at the low watermark.
4330 */
4331 available += global_page_state(NR_SLAB_RECLAIMABLE) -
4332 min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low);
4333
4334 if (available < 0)
4335 available = 0;
4336 return available;
4337}
4338EXPORT_SYMBOL_GPL(si_mem_available);
4339
Linus Torvalds1da177e2005-04-16 15:20:36 -07004340void si_meminfo(struct sysinfo *val)
4341{
4342 val->totalram = totalram_pages;
Mel Gorman11fb9982016-07-28 15:46:20 -07004343 val->sharedram = global_node_page_state(NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08004344 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004345 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004346 val->totalhigh = totalhigh_pages;
4347 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004348 val->mem_unit = PAGE_SIZE;
4349}
4350
4351EXPORT_SYMBOL(si_meminfo);
4352
4353#ifdef CONFIG_NUMA
4354void si_meminfo_node(struct sysinfo *val, int nid)
4355{
Jiang Liucdd91a72013-07-03 15:03:27 -07004356 int zone_type; /* needs to be signed */
4357 unsigned long managed_pages = 0;
Joonsoo Kimfc2bd792016-05-19 17:12:23 -07004358 unsigned long managed_highpages = 0;
4359 unsigned long free_highpages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004360 pg_data_t *pgdat = NODE_DATA(nid);
4361
Jiang Liucdd91a72013-07-03 15:03:27 -07004362 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
4363 managed_pages += pgdat->node_zones[zone_type].managed_pages;
4364 val->totalram = managed_pages;
Mel Gorman11fb9982016-07-28 15:46:20 -07004365 val->sharedram = node_page_state(pgdat, NR_SHMEM);
Mel Gorman75ef7182016-07-28 15:45:24 -07004366 val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004367#ifdef CONFIG_HIGHMEM
Joonsoo Kimfc2bd792016-05-19 17:12:23 -07004368 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
4369 struct zone *zone = &pgdat->node_zones[zone_type];
4370
4371 if (is_highmem(zone)) {
4372 managed_highpages += zone->managed_pages;
4373 free_highpages += zone_page_state(zone, NR_FREE_PAGES);
4374 }
4375 }
4376 val->totalhigh = managed_highpages;
4377 val->freehigh = free_highpages;
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004378#else
Joonsoo Kimfc2bd792016-05-19 17:12:23 -07004379 val->totalhigh = managed_highpages;
4380 val->freehigh = free_highpages;
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004381#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004382 val->mem_unit = PAGE_SIZE;
4383}
4384#endif
4385
David Rientjesddd588b2011-03-22 16:30:46 -07004386/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07004387 * Determine whether the node should be displayed or not, depending on whether
4388 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07004389 */
Michal Hocko9af744d2017-02-22 15:46:16 -08004390static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask)
David Rientjesddd588b2011-03-22 16:30:46 -07004391{
David Rientjesddd588b2011-03-22 16:30:46 -07004392 if (!(flags & SHOW_MEM_FILTER_NODES))
Michal Hocko9af744d2017-02-22 15:46:16 -08004393 return false;
David Rientjesddd588b2011-03-22 16:30:46 -07004394
Michal Hocko9af744d2017-02-22 15:46:16 -08004395 /*
4396 * no node mask - aka implicit memory numa policy. Do not bother with
4397 * the synchronization - read_mems_allowed_begin - because we do not
4398 * have to be precise here.
4399 */
4400 if (!nodemask)
4401 nodemask = &cpuset_current_mems_allowed;
4402
4403 return !node_isset(nid, *nodemask);
David Rientjesddd588b2011-03-22 16:30:46 -07004404}
4405
Linus Torvalds1da177e2005-04-16 15:20:36 -07004406#define K(x) ((x) << (PAGE_SHIFT-10))
4407
Rabin Vincent377e4f12012-12-11 16:00:24 -08004408static void show_migration_types(unsigned char type)
4409{
4410 static const char types[MIGRATE_TYPES] = {
4411 [MIGRATE_UNMOVABLE] = 'U',
Rabin Vincent377e4f12012-12-11 16:00:24 -08004412 [MIGRATE_MOVABLE] = 'M',
Vlastimil Babka475a2f92015-12-11 13:40:29 -08004413 [MIGRATE_RECLAIMABLE] = 'E',
4414 [MIGRATE_HIGHATOMIC] = 'H',
Rabin Vincent377e4f12012-12-11 16:00:24 -08004415#ifdef CONFIG_CMA
4416 [MIGRATE_CMA] = 'C',
4417#endif
Minchan Kim194159f2013-02-22 16:33:58 -08004418#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08004419 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08004420#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08004421 };
4422 char tmp[MIGRATE_TYPES + 1];
4423 char *p = tmp;
4424 int i;
4425
4426 for (i = 0; i < MIGRATE_TYPES; i++) {
4427 if (type & (1 << i))
4428 *p++ = types[i];
4429 }
4430
4431 *p = '\0';
Joe Perches1f84a182016-10-27 17:46:29 -07004432 printk(KERN_CONT "(%s) ", tmp);
Rabin Vincent377e4f12012-12-11 16:00:24 -08004433}
4434
Linus Torvalds1da177e2005-04-16 15:20:36 -07004435/*
4436 * Show free area list (used inside shift_scroll-lock stuff)
4437 * We also calculate the percentage fragmentation. We do this by counting the
4438 * memory on each free list with the exception of the first item on the list.
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004439 *
4440 * Bits in @filter:
4441 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4442 * cpuset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004443 */
Michal Hocko9af744d2017-02-22 15:46:16 -08004444void show_free_areas(unsigned int filter, nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004445{
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004446 unsigned long free_pcp = 0;
Jes Sorensenc7241912006-09-27 01:50:05 -07004447 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004448 struct zone *zone;
Mel Gorman599d0c92016-07-28 15:45:31 -07004449 pg_data_t *pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004450
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07004451 for_each_populated_zone(zone) {
Michal Hocko9af744d2017-02-22 15:46:16 -08004452 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
David Rientjesddd588b2011-03-22 16:30:46 -07004453 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004454
Konstantin Khlebnikov761b0672015-04-14 15:45:32 -07004455 for_each_online_cpu(cpu)
4456 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004457 }
4458
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07004459 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4460 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004461 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4462 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004463 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004464 " free:%lu free_pcp:%lu free_cma:%lu\n",
Mel Gorman599d0c92016-07-28 15:45:31 -07004465 global_node_page_state(NR_ACTIVE_ANON),
4466 global_node_page_state(NR_INACTIVE_ANON),
4467 global_node_page_state(NR_ISOLATED_ANON),
4468 global_node_page_state(NR_ACTIVE_FILE),
4469 global_node_page_state(NR_INACTIVE_FILE),
4470 global_node_page_state(NR_ISOLATED_FILE),
4471 global_node_page_state(NR_UNEVICTABLE),
Mel Gorman11fb9982016-07-28 15:46:20 -07004472 global_node_page_state(NR_FILE_DIRTY),
4473 global_node_page_state(NR_WRITEBACK),
4474 global_node_page_state(NR_UNSTABLE_NFS),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07004475 global_page_state(NR_SLAB_RECLAIMABLE),
4476 global_page_state(NR_SLAB_UNRECLAIMABLE),
Mel Gorman50658e22016-07-28 15:46:14 -07004477 global_node_page_state(NR_FILE_MAPPED),
Mel Gorman11fb9982016-07-28 15:46:20 -07004478 global_node_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08004479 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004480 global_page_state(NR_BOUNCE),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004481 global_page_state(NR_FREE_PAGES),
4482 free_pcp,
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004483 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004484
Mel Gorman599d0c92016-07-28 15:45:31 -07004485 for_each_online_pgdat(pgdat) {
Michal Hocko9af744d2017-02-22 15:46:16 -08004486 if (show_mem_node_skip(filter, pgdat->node_id, nodemask))
Michal Hockoc02e50b2017-02-22 15:46:07 -08004487 continue;
4488
Mel Gorman599d0c92016-07-28 15:45:31 -07004489 printk("Node %d"
4490 " active_anon:%lukB"
4491 " inactive_anon:%lukB"
4492 " active_file:%lukB"
4493 " inactive_file:%lukB"
4494 " unevictable:%lukB"
4495 " isolated(anon):%lukB"
4496 " isolated(file):%lukB"
Mel Gorman50658e22016-07-28 15:46:14 -07004497 " mapped:%lukB"
Mel Gorman11fb9982016-07-28 15:46:20 -07004498 " dirty:%lukB"
4499 " writeback:%lukB"
4500 " shmem:%lukB"
4501#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4502 " shmem_thp: %lukB"
4503 " shmem_pmdmapped: %lukB"
4504 " anon_thp: %lukB"
4505#endif
4506 " writeback_tmp:%lukB"
4507 " unstable:%lukB"
Minchan Kim33e077b2016-07-28 15:47:14 -07004508 " pages_scanned:%lu"
Mel Gorman599d0c92016-07-28 15:45:31 -07004509 " all_unreclaimable? %s"
4510 "\n",
4511 pgdat->node_id,
4512 K(node_page_state(pgdat, NR_ACTIVE_ANON)),
4513 K(node_page_state(pgdat, NR_INACTIVE_ANON)),
4514 K(node_page_state(pgdat, NR_ACTIVE_FILE)),
4515 K(node_page_state(pgdat, NR_INACTIVE_FILE)),
4516 K(node_page_state(pgdat, NR_UNEVICTABLE)),
4517 K(node_page_state(pgdat, NR_ISOLATED_ANON)),
4518 K(node_page_state(pgdat, NR_ISOLATED_FILE)),
Mel Gorman50658e22016-07-28 15:46:14 -07004519 K(node_page_state(pgdat, NR_FILE_MAPPED)),
Mel Gorman11fb9982016-07-28 15:46:20 -07004520 K(node_page_state(pgdat, NR_FILE_DIRTY)),
4521 K(node_page_state(pgdat, NR_WRITEBACK)),
4522#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4523 K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR),
4524 K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)
4525 * HPAGE_PMD_NR),
4526 K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR),
4527#endif
4528 K(node_page_state(pgdat, NR_SHMEM)),
4529 K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
4530 K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
Minchan Kim33e077b2016-07-28 15:47:14 -07004531 node_page_state(pgdat, NR_PAGES_SCANNED),
Mel Gorman599d0c92016-07-28 15:45:31 -07004532 !pgdat_reclaimable(pgdat) ? "yes" : "no");
4533 }
4534
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07004535 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004536 int i;
4537
Michal Hocko9af744d2017-02-22 15:46:16 -08004538 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
David Rientjesddd588b2011-03-22 16:30:46 -07004539 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004540
4541 free_pcp = 0;
4542 for_each_online_cpu(cpu)
4543 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
4544
Linus Torvalds1da177e2005-04-16 15:20:36 -07004545 show_node(zone);
Joe Perches1f84a182016-10-27 17:46:29 -07004546 printk(KERN_CONT
4547 "%s"
Linus Torvalds1da177e2005-04-16 15:20:36 -07004548 " free:%lukB"
4549 " min:%lukB"
4550 " low:%lukB"
4551 " high:%lukB"
Minchan Kim71c799f2016-07-28 15:47:26 -07004552 " active_anon:%lukB"
4553 " inactive_anon:%lukB"
4554 " active_file:%lukB"
4555 " inactive_file:%lukB"
4556 " unevictable:%lukB"
Mel Gorman5a1c84b2016-07-28 15:47:31 -07004557 " writepending:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07004558 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08004559 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004560 " mlocked:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004561 " slab_reclaimable:%lukB"
4562 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07004563 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004564 " pagetables:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004565 " bounce:%lukB"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004566 " free_pcp:%lukB"
4567 " local_pcp:%ukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004568 " free_cma:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07004569 "\n",
4570 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08004571 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07004572 K(min_wmark_pages(zone)),
4573 K(low_wmark_pages(zone)),
4574 K(high_wmark_pages(zone)),
Minchan Kim71c799f2016-07-28 15:47:26 -07004575 K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
4576 K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
4577 K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
4578 K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
4579 K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
Mel Gorman5a1c84b2016-07-28 15:47:31 -07004580 K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07004581 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08004582 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004583 K(zone_page_state(zone, NR_MLOCK)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004584 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
4585 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
Andy Lutomirskid30dd8b2016-07-28 15:48:14 -07004586 zone_page_state(zone, NR_KERNEL_STACK_KB),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004587 K(zone_page_state(zone, NR_PAGETABLE)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004588 K(zone_page_state(zone, NR_BOUNCE)),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004589 K(free_pcp),
4590 K(this_cpu_read(zone->pageset->pcp.count)),
Minchan Kim33e077b2016-07-28 15:47:14 -07004591 K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004592 printk("lowmem_reserve[]:");
4593 for (i = 0; i < MAX_NR_ZONES; i++)
Joe Perches1f84a182016-10-27 17:46:29 -07004594 printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
4595 printk(KERN_CONT "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004596 }
4597
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07004598 for_each_populated_zone(zone) {
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004599 unsigned int order;
4600 unsigned long nr[MAX_ORDER], flags, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08004601 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004602
Michal Hocko9af744d2017-02-22 15:46:16 -08004603 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
David Rientjesddd588b2011-03-22 16:30:46 -07004604 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004605 show_node(zone);
Joe Perches1f84a182016-10-27 17:46:29 -07004606 printk(KERN_CONT "%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004607
4608 spin_lock_irqsave(&zone->lock, flags);
4609 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08004610 struct free_area *area = &zone->free_area[order];
4611 int type;
4612
4613 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07004614 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08004615
4616 types[order] = 0;
4617 for (type = 0; type < MIGRATE_TYPES; type++) {
4618 if (!list_empty(&area->free_list[type]))
4619 types[order] |= 1 << type;
4620 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004621 }
4622 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08004623 for (order = 0; order < MAX_ORDER; order++) {
Joe Perches1f84a182016-10-27 17:46:29 -07004624 printk(KERN_CONT "%lu*%lukB ",
4625 nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08004626 if (nr[order])
4627 show_migration_types(types[order]);
4628 }
Joe Perches1f84a182016-10-27 17:46:29 -07004629 printk(KERN_CONT "= %lukB\n", K(total));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004630 }
4631
David Rientjes949f7ec2013-04-29 15:07:48 -07004632 hugetlb_show_meminfo();
4633
Mel Gorman11fb9982016-07-28 15:46:20 -07004634 printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
Larry Woodmane6f36022008-02-04 22:29:30 -08004635
Linus Torvalds1da177e2005-04-16 15:20:36 -07004636 show_swap_cache_info();
4637}
4638
Mel Gorman19770b32008-04-28 02:12:18 -07004639static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
4640{
4641 zoneref->zone = zone;
4642 zoneref->zone_idx = zone_idx(zone);
4643}
4644
Linus Torvalds1da177e2005-04-16 15:20:36 -07004645/*
4646 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08004647 *
4648 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004649 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004650static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004651 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004652{
Christoph Lameter1a932052006-01-06 00:11:16 -08004653 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004654 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08004655
4656 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004657 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08004658 zone = pgdat->node_zones + zone_type;
Mel Gorman6aa303d2016-09-01 16:14:55 -07004659 if (managed_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07004660 zoneref_set_zone(zone,
4661 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08004662 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004663 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004664 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004665
Christoph Lameter070f8032006-01-06 00:11:19 -08004666 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004667}
4668
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004669
4670/*
4671 * zonelist_order:
4672 * 0 = automatic detection of better ordering.
4673 * 1 = order by ([node] distance, -zonetype)
4674 * 2 = order by (-zonetype, [node] distance)
4675 *
4676 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4677 * the same zonelist. So only NUMA can configure this param.
4678 */
4679#define ZONELIST_ORDER_DEFAULT 0
4680#define ZONELIST_ORDER_NODE 1
4681#define ZONELIST_ORDER_ZONE 2
4682
4683/* zonelist order in the kernel.
4684 * set_zonelist_order() will set this to NODE or ZONE.
4685 */
4686static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
4687static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
4688
4689
Linus Torvalds1da177e2005-04-16 15:20:36 -07004690#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004691/* The value user specified ....changed by config */
4692static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4693/* string for sysctl */
4694#define NUMA_ZONELIST_ORDER_LEN 16
4695char numa_zonelist_order[16] = "default";
4696
4697/*
4698 * interface for configure zonelist ordering.
4699 * command line option "numa_zonelist_order"
4700 * = "[dD]efault - default, automatic configuration.
4701 * = "[nN]ode - order by node locality, then by zone within node
4702 * = "[zZ]one - order by zone, then by locality within zone
4703 */
4704
4705static int __parse_numa_zonelist_order(char *s)
4706{
4707 if (*s == 'd' || *s == 'D') {
4708 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4709 } else if (*s == 'n' || *s == 'N') {
4710 user_zonelist_order = ZONELIST_ORDER_NODE;
4711 } else if (*s == 'z' || *s == 'Z') {
4712 user_zonelist_order = ZONELIST_ORDER_ZONE;
4713 } else {
Joe Perches11705322016-03-17 14:19:50 -07004714 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004715 return -EINVAL;
4716 }
4717 return 0;
4718}
4719
4720static __init int setup_numa_zonelist_order(char *s)
4721{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08004722 int ret;
4723
4724 if (!s)
4725 return 0;
4726
4727 ret = __parse_numa_zonelist_order(s);
4728 if (ret == 0)
4729 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
4730
4731 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004732}
4733early_param("numa_zonelist_order", setup_numa_zonelist_order);
4734
4735/*
4736 * sysctl handler for numa_zonelist_order
4737 */
Joe Perchescccad5b2014-06-06 14:38:09 -07004738int numa_zonelist_order_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004739 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004740 loff_t *ppos)
4741{
4742 char saved_string[NUMA_ZONELIST_ORDER_LEN];
4743 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01004744 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004745
Andi Kleen443c6f12009-12-23 21:00:47 +01004746 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07004747 if (write) {
4748 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
4749 ret = -EINVAL;
4750 goto out;
4751 }
4752 strcpy(saved_string, (char *)table->data);
4753 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004754 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004755 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01004756 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004757 if (write) {
4758 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07004759
4760 ret = __parse_numa_zonelist_order((char *)table->data);
4761 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004762 /*
4763 * bogus value. restore saved string
4764 */
Chen Gangdacbde02013-07-03 15:02:35 -07004765 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004766 NUMA_ZONELIST_ORDER_LEN);
4767 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004768 } else if (oldval != user_zonelist_order) {
4769 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07004770 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004771 mutex_unlock(&zonelists_mutex);
4772 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004773 }
Andi Kleen443c6f12009-12-23 21:00:47 +01004774out:
4775 mutex_unlock(&zl_order_mutex);
4776 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004777}
4778
4779
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004780#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004781static int node_load[MAX_NUMNODES];
4782
Linus Torvalds1da177e2005-04-16 15:20:36 -07004783/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07004784 * 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 -07004785 * @node: node whose fallback list we're appending
4786 * @used_node_mask: nodemask_t of already used nodes
4787 *
4788 * We use a number of factors to determine which is the next node that should
4789 * appear on a given node's fallback list. The node should not have appeared
4790 * already in @node's fallback list, and it should be the next closest node
4791 * according to the distance array (which contains arbitrary distance values
4792 * from each node to each node in the system), and should also prefer nodes
4793 * with no CPUs, since presumably they'll have very little allocation pressure
4794 * on them otherwise.
4795 * It returns -1 if no node is found.
4796 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004797static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004798{
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01004799 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004800 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08004801 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10304802 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004803
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01004804 /* Use the local node if we haven't already */
4805 if (!node_isset(node, *used_node_mask)) {
4806 node_set(node, *used_node_mask);
4807 return node;
4808 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004809
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004810 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004811
4812 /* Don't want a node to appear more than once */
4813 if (node_isset(n, *used_node_mask))
4814 continue;
4815
Linus Torvalds1da177e2005-04-16 15:20:36 -07004816 /* Use the distance array to find the distance */
4817 val = node_distance(node, n);
4818
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01004819 /* Penalize nodes under us ("prefer the next node") */
4820 val += (n < node);
4821
Linus Torvalds1da177e2005-04-16 15:20:36 -07004822 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10304823 tmp = cpumask_of_node(n);
4824 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004825 val += PENALTY_FOR_NODE_WITH_CPUS;
4826
4827 /* Slight preference for less loaded node */
4828 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4829 val += node_load[n];
4830
4831 if (val < min_val) {
4832 min_val = val;
4833 best_node = n;
4834 }
4835 }
4836
4837 if (best_node >= 0)
4838 node_set(best_node, *used_node_mask);
4839
4840 return best_node;
4841}
4842
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004843
4844/*
4845 * Build zonelists ordered by node and zones within node.
4846 * This results in maximum locality--normal zone overflows into local
4847 * DMA zone, if any--but risks exhausting DMA zone.
4848 */
4849static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004850{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004851 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004852 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004853
Aneesh Kumar K.Vc9634cf2016-10-07 16:59:12 -07004854 zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK];
Mel Gormandd1a2392008-04-28 02:12:17 -07004855 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07004856 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004857 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07004858 zonelist->_zonerefs[j].zone = NULL;
4859 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004860}
4861
4862/*
Christoph Lameter523b9452007-10-16 01:25:37 -07004863 * Build gfp_thisnode zonelists
4864 */
4865static void build_thisnode_zonelists(pg_data_t *pgdat)
4866{
Christoph Lameter523b9452007-10-16 01:25:37 -07004867 int j;
4868 struct zonelist *zonelist;
4869
Aneesh Kumar K.Vc9634cf2016-10-07 16:59:12 -07004870 zonelist = &pgdat->node_zonelists[ZONELIST_NOFALLBACK];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004871 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07004872 zonelist->_zonerefs[j].zone = NULL;
4873 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07004874}
4875
4876/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004877 * Build zonelists ordered by zone and nodes within zones.
4878 * This results in conserving DMA zone[s] until all Normal memory is
4879 * exhausted, but results in overflowing to remote node while memory
4880 * may still exist in local DMA zone.
4881 */
4882static int node_order[MAX_NUMNODES];
4883
4884static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4885{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004886 int pos, j, node;
4887 int zone_type; /* needs to be signed */
4888 struct zone *z;
4889 struct zonelist *zonelist;
4890
Aneesh Kumar K.Vc9634cf2016-10-07 16:59:12 -07004891 zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK];
Mel Gorman54a6eb52008-04-28 02:12:16 -07004892 pos = 0;
4893 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4894 for (j = 0; j < nr_nodes; j++) {
4895 node = node_order[j];
4896 z = &NODE_DATA(node)->node_zones[zone_type];
Mel Gorman6aa303d2016-09-01 16:14:55 -07004897 if (managed_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07004898 zoneref_set_zone(z,
4899 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004900 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004901 }
4902 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004903 }
Mel Gormandd1a2392008-04-28 02:12:17 -07004904 zonelist->_zonerefs[pos].zone = NULL;
4905 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004906}
4907
Mel Gorman31939132014-10-09 15:28:30 -07004908#if defined(CONFIG_64BIT)
4909/*
4910 * Devices that require DMA32/DMA are relatively rare and do not justify a
4911 * penalty to every machine in case the specialised case applies. Default
4912 * to Node-ordering on 64-bit NUMA machines
4913 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004914static int default_zonelist_order(void)
4915{
Mel Gorman31939132014-10-09 15:28:30 -07004916 return ZONELIST_ORDER_NODE;
4917}
4918#else
4919/*
4920 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4921 * by the kernel. If processes running on node 0 deplete the low memory zone
4922 * then reclaim will occur more frequency increasing stalls and potentially
4923 * be easier to OOM if a large percentage of the zone is under writeback or
4924 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4925 * Hence, default to zone ordering on 32-bit.
4926 */
4927static int default_zonelist_order(void)
4928{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004929 return ZONELIST_ORDER_ZONE;
4930}
Mel Gorman31939132014-10-09 15:28:30 -07004931#endif /* CONFIG_64BIT */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004932
4933static void set_zonelist_order(void)
4934{
4935 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4936 current_zonelist_order = default_zonelist_order();
4937 else
4938 current_zonelist_order = user_zonelist_order;
4939}
4940
4941static void build_zonelists(pg_data_t *pgdat)
4942{
Yaowei Baic00eb152016-01-14 15:19:00 -08004943 int i, node, load;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004944 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004945 int local_node, prev_node;
4946 struct zonelist *zonelist;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004947 unsigned int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004948
4949 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07004950 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004951 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07004952 zonelist->_zonerefs[0].zone = NULL;
4953 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004954 }
4955
4956 /* NUMA-aware ordering of nodes */
4957 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004958 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004959 prev_node = local_node;
4960 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004961
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004962 memset(node_order, 0, sizeof(node_order));
Yaowei Baic00eb152016-01-14 15:19:00 -08004963 i = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004964
Linus Torvalds1da177e2005-04-16 15:20:36 -07004965 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4966 /*
4967 * We don't want to pressure a particular node.
4968 * So adding penalty to the first node in same
4969 * distance group to make it round-robin.
4970 */
David Rientjes957f8222012-10-08 16:33:24 -07004971 if (node_distance(local_node, node) !=
4972 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004973 node_load[node] = load;
4974
Linus Torvalds1da177e2005-04-16 15:20:36 -07004975 prev_node = node;
4976 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004977 if (order == ZONELIST_ORDER_NODE)
4978 build_zonelists_in_node_order(pgdat, node);
4979 else
Yaowei Baic00eb152016-01-14 15:19:00 -08004980 node_order[i++] = node; /* remember order */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004981 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004982
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004983 if (order == ZONELIST_ORDER_ZONE) {
4984 /* calculate node order -- i.e., DMA last! */
Yaowei Baic00eb152016-01-14 15:19:00 -08004985 build_zonelists_in_zone_order(pgdat, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004986 }
Christoph Lameter523b9452007-10-16 01:25:37 -07004987
4988 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004989}
4990
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004991#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4992/*
4993 * Return node id of node used for "local" allocations.
4994 * I.e., first node id of first zone in arg node's generic zonelist.
4995 * Used for initializing percpu 'numa_mem', which is used primarily
4996 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4997 */
4998int local_memory_node(int node)
4999{
Mel Gormanc33d6c02016-05-19 17:14:10 -07005000 struct zoneref *z;
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07005001
Mel Gormanc33d6c02016-05-19 17:14:10 -07005002 z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07005003 gfp_zone(GFP_KERNEL),
Mel Gormanc33d6c02016-05-19 17:14:10 -07005004 NULL);
5005 return z->zone->node;
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07005006}
5007#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07005008
Joonsoo Kim6423aa82016-08-10 16:27:49 -07005009static void setup_min_unmapped_ratio(void);
5010static void setup_min_slab_ratio(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005011#else /* CONFIG_NUMA */
5012
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07005013static void set_zonelist_order(void)
5014{
5015 current_zonelist_order = ZONELIST_ORDER_ZONE;
5016}
5017
5018static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005019{
Christoph Lameter19655d32006-09-25 23:31:19 -07005020 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07005021 enum zone_type j;
5022 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005023
5024 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005025
Aneesh Kumar K.Vc9634cf2016-10-07 16:59:12 -07005026 zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07005027 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005028
Mel Gorman54a6eb52008-04-28 02:12:16 -07005029 /*
5030 * Now we build the zonelist so that it contains the zones
5031 * of all the other nodes.
5032 * We don't want to pressure a particular node, so when
5033 * building the zones for node N, we make sure that the
5034 * zones coming right after the local ones are those from
5035 * node N+1 (modulo N)
5036 */
5037 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
5038 if (!node_online(node))
5039 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07005040 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005041 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07005042 for (node = 0; node < local_node; node++) {
5043 if (!node_online(node))
5044 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07005045 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07005046 }
5047
Mel Gormandd1a2392008-04-28 02:12:17 -07005048 zonelist->_zonerefs[j].zone = NULL;
5049 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005050}
5051
5052#endif /* CONFIG_NUMA */
5053
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005054/*
5055 * Boot pageset table. One per cpu which is going to be used for all
5056 * zones and all nodes. The parameters will be set in such a way
5057 * that an item put on a list will immediately be handed over to
5058 * the buddy list. This is safe since pageset manipulation is done
5059 * with interrupts disabled.
5060 *
5061 * The boot_pagesets must be kept even after bootup is complete for
5062 * unused processors and/or zones. They do play a role for bootstrapping
5063 * hotplugged processors.
5064 *
5065 * zoneinfo_show() and maybe other functions do
5066 * not check if the processor is online before following the pageset pointer.
5067 * Other parts of the kernel may not check if the zone is available.
5068 */
5069static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
5070static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07005071static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005072
Haicheng Li4eaf3f62010-05-24 14:32:52 -07005073/*
5074 * Global mutex to protect against size modification of zonelists
5075 * as well as to serialize pageset setup for the new populated zone.
5076 */
5077DEFINE_MUTEX(zonelists_mutex);
5078
Rusty Russell9b1a4d32008-07-28 12:16:30 -05005079/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07005080static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005081{
Yasunori Goto68113782006-06-23 02:03:11 -07005082 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005083 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07005084 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08005085
Bo Liu7f9cfb32009-08-18 14:11:19 -07005086#ifdef CONFIG_NUMA
5087 memset(node_load, 0, sizeof(node_load));
5088#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07005089
5090 if (self && !node_online(self->node_id)) {
5091 build_zonelists(self);
Jiang Liu9adb62a2012-07-31 16:43:28 -07005092 }
5093
Paul Jackson9276b1bc2006-12-06 20:31:48 -08005094 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07005095 pg_data_t *pgdat = NODE_DATA(nid);
5096
5097 build_zonelists(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08005098 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005099
5100 /*
5101 * Initialize the boot_pagesets that are going to be used
5102 * for bootstrapping processors. The real pagesets for
5103 * each zone will be allocated later when the per cpu
5104 * allocator is available.
5105 *
5106 * boot_pagesets are used also for bootstrapping offline
5107 * cpus if the system is already booted because the pagesets
5108 * are needed to initialize allocators on a specific cpu too.
5109 * F.e. the percpu allocator needs the page allocator which
5110 * needs the percpu allocator in order to allocate its pagesets
5111 * (a chicken-egg dilemma).
5112 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07005113 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005114 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
5115
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07005116#ifdef CONFIG_HAVE_MEMORYLESS_NODES
5117 /*
5118 * We now know the "local memory node" for each node--
5119 * i.e., the node of the first zone in the generic zonelist.
5120 * Set up numa_mem percpu variable for on-line cpus. During
5121 * boot, only the boot cpu should be on-line; we'll init the
5122 * secondary cpus' numa_mem as they come on-line. During
5123 * node/memory hotplug, we'll fixup all on-line cpus.
5124 */
5125 if (cpu_online(cpu))
5126 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
5127#endif
5128 }
5129
Yasunori Goto68113782006-06-23 02:03:11 -07005130 return 0;
5131}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005132
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08005133static noinline void __init
5134build_all_zonelists_init(void)
5135{
5136 __build_all_zonelists(NULL);
5137 mminit_verify_zonelist();
5138 cpuset_init_current_mems_allowed();
5139}
5140
Haicheng Li4eaf3f62010-05-24 14:32:52 -07005141/*
5142 * Called with zonelists_mutex held always
5143 * unless system_state == SYSTEM_BOOTING.
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08005144 *
5145 * __ref due to (1) call of __meminit annotated setup_zone_pageset
5146 * [we're only called with non-NULL zone through __meminit paths] and
5147 * (2) call of __init annotated helper build_all_zonelists_init
5148 * [protected by SYSTEM_BOOTING].
Haicheng Li4eaf3f62010-05-24 14:32:52 -07005149 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07005150void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07005151{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07005152 set_zonelist_order();
5153
Yasunori Goto68113782006-06-23 02:03:11 -07005154 if (system_state == SYSTEM_BOOTING) {
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08005155 build_all_zonelists_init();
Yasunori Goto68113782006-06-23 02:03:11 -07005156 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08005157#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07005158 if (zone)
5159 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08005160#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07005161 /* we have to stop all cpus to guarantee there is no user
5162 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07005163 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07005164 /* cpuset refresh routine should be here */
5165 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07005166 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07005167 /*
5168 * Disable grouping by mobility if the number of pages in the
5169 * system is too low to allow the mechanism to work. It would be
5170 * more accurate, but expensive to check per-zone. This check is
5171 * made on memory-hotadd so a system can start with mobility
5172 * disabled and enable it later
5173 */
Mel Gormand9c23402007-10-16 01:26:01 -07005174 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07005175 page_group_by_mobility_disabled = 1;
5176 else
5177 page_group_by_mobility_disabled = 0;
5178
Joe Perches756a0252016-03-17 14:19:47 -07005179 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
5180 nr_online_nodes,
5181 zonelist_order_name[current_zonelist_order],
5182 page_group_by_mobility_disabled ? "off" : "on",
5183 vm_total_pages);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07005184#ifdef CONFIG_NUMA
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005185 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07005186#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005187}
5188
5189/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005190 * Initially all pages are reserved - free ones are freed
5191 * up by free_all_bootmem() once the early boot process is
5192 * done. Non-atomic initialization, single-pass.
5193 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01005194void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08005195 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005196{
Dan Williams4b94ffd2016-01-15 16:56:22 -08005197 struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn));
Andy Whitcroft29751f62005-06-23 00:08:00 -07005198 unsigned long end_pfn = start_pfn + size;
Dan Williams4b94ffd2016-01-15 16:56:22 -08005199 pg_data_t *pgdat = NODE_DATA(nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -07005200 unsigned long pfn;
Mel Gorman3a80a7f2015-06-30 14:57:02 -07005201 unsigned long nr_initialised = 0;
Taku Izumi342332e2016-03-15 14:55:22 -07005202#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5203 struct memblock_region *r = NULL, *tmp;
5204#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005205
Hugh Dickins22b31ee2009-01-06 14:40:09 -08005206 if (highest_memmap_pfn < end_pfn - 1)
5207 highest_memmap_pfn = end_pfn - 1;
5208
Dan Williams4b94ffd2016-01-15 16:56:22 -08005209 /*
5210 * Honor reservation requested by the driver for this ZONE_DEVICE
5211 * memory
5212 */
5213 if (altmap && start_pfn == altmap->base_pfn)
5214 start_pfn += altmap->reserve;
5215
Greg Ungerercbe8dd42006-01-12 01:05:24 -08005216 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08005217 /*
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005218 * There can be holes in boot-time mem_map[]s handed to this
5219 * function. They do not exist on hotplugged memory.
Dave Hansena2f3aa022007-01-10 23:15:30 -08005220 */
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005221 if (context != MEMMAP_EARLY)
5222 goto not_early;
5223
Paul Burtonb92df1d2017-02-22 15:44:53 -08005224 if (!early_pfn_valid(pfn)) {
5225#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5226 /*
5227 * Skip to the pfn preceding the next valid one (or
5228 * end_pfn), such that we hit a valid pfn (or end_pfn)
5229 * on our next iteration of the loop.
5230 */
5231 pfn = memblock_next_valid_pfn(pfn, end_pfn) - 1;
5232#endif
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005233 continue;
Paul Burtonb92df1d2017-02-22 15:44:53 -08005234 }
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005235 if (!early_pfn_in_nid(pfn, nid))
5236 continue;
5237 if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised))
5238 break;
Taku Izumi342332e2016-03-15 14:55:22 -07005239
5240#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005241 /*
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005242 * Check given memblock attribute by firmware which can affect
5243 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
5244 * mirrored, it's an overlapped memmap init. skip it.
5245 */
5246 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
5247 if (!r || pfn >= memblock_region_memory_end_pfn(r)) {
5248 for_each_memblock(memory, tmp)
5249 if (pfn < memblock_region_memory_end_pfn(tmp))
5250 break;
5251 r = tmp;
Taku Izumi342332e2016-03-15 14:55:22 -07005252 }
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005253 if (pfn >= memblock_region_memory_base_pfn(r) &&
5254 memblock_is_mirror(r)) {
5255 /* already initialized as NORMAL */
5256 pfn = memblock_region_memory_end_pfn(r);
5257 continue;
5258 }
Dave Hansena2f3aa022007-01-10 23:15:30 -08005259 }
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005260#endif
Mel Gormanac5d2532015-06-30 14:57:20 -07005261
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005262not_early:
Mel Gormanac5d2532015-06-30 14:57:20 -07005263 /*
5264 * Mark the block movable so that blocks are reserved for
5265 * movable at startup. This will force kernel allocations
5266 * to reserve their blocks rather than leaking throughout
5267 * the address space during boot when many long-lived
Mel Gorman974a7862015-11-06 16:28:34 -08005268 * kernel allocations are made.
Mel Gormanac5d2532015-06-30 14:57:20 -07005269 *
5270 * bitmap is created for zone's valid pfn range. but memmap
5271 * can be created for invalid pages (for alignment)
5272 * check here not to call set_pageblock_migratetype() against
5273 * pfn out of zone.
5274 */
5275 if (!(pfn & (pageblock_nr_pages - 1))) {
5276 struct page *page = pfn_to_page(pfn);
5277
5278 __init_single_page(page, pfn, zone, nid);
5279 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5280 } else {
5281 __init_single_pfn(pfn, zone, nid);
5282 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005283 }
5284}
5285
Andi Kleen1e548de2008-02-04 22:29:26 -08005286static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005287{
Mel Gorman7aeb09f2014-06-04 16:10:21 -07005288 unsigned int order, t;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07005289 for_each_migratetype_order(order, t) {
5290 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005291 zone->free_area[order].nr_free = 0;
5292 }
5293}
5294
5295#ifndef __HAVE_ARCH_MEMMAP_INIT
5296#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08005297 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005298#endif
5299
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005300static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005301{
David Howells3a6be872009-05-06 16:03:03 -07005302#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005303 int batch;
5304
5305 /*
5306 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07005307 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005308 *
5309 * OK, so we don't know how big the cache is. So guess.
5310 */
Jiang Liub40da042013-02-22 16:33:52 -08005311 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07005312 if (batch * PAGE_SIZE > 512 * 1024)
5313 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005314 batch /= 4; /* We effectively *= 4 below */
5315 if (batch < 1)
5316 batch = 1;
5317
5318 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11005319 * Clamp the batch to a 2^n - 1 value. Having a power
5320 * of 2 value was found to be more likely to have
5321 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005322 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11005323 * For example if 2 tasks are alternately allocating
5324 * batches of pages, one task can end up with a lot
5325 * of pages of one half of the possible page colors
5326 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005327 */
David Howells91552032009-05-06 16:03:02 -07005328 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07005329
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005330 return batch;
David Howells3a6be872009-05-06 16:03:03 -07005331
5332#else
5333 /* The deferral and batching of frees should be suppressed under NOMMU
5334 * conditions.
5335 *
5336 * The problem is that NOMMU needs to be able to allocate large chunks
5337 * of contiguous memory as there's no hardware page translation to
5338 * assemble apparent contiguous memory from discontiguous pages.
5339 *
5340 * Queueing large contiguous runs of pages for batching, however,
5341 * causes the pages to actually be freed in smaller chunks. As there
5342 * can be a significant delay between the individual batches being
5343 * recycled, this leads to the once large chunks of space being
5344 * fragmented and becoming unavailable for high-order allocations.
5345 */
5346 return 0;
5347#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005348}
5349
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005350/*
5351 * pcp->high and pcp->batch values are related and dependent on one another:
5352 * ->batch must never be higher then ->high.
5353 * The following function updates them in a safe manner without read side
5354 * locking.
5355 *
5356 * Any new users of pcp->batch and pcp->high should ensure they can cope with
5357 * those fields changing asynchronously (acording the the above rule).
5358 *
5359 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5360 * outside of boot time (or some other assurance that no concurrent updaters
5361 * exist).
5362 */
5363static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
5364 unsigned long batch)
5365{
5366 /* start with a fail safe value for batch */
5367 pcp->batch = 1;
5368 smp_wmb();
5369
5370 /* Update high, then batch, in order */
5371 pcp->high = high;
5372 smp_wmb();
5373
5374 pcp->batch = batch;
5375}
5376
Cody P Schafer36640332013-07-03 15:01:40 -07005377/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07005378static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
5379{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005380 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07005381}
5382
Cody P Schafer88c90db2013-07-03 15:01:35 -07005383static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07005384{
5385 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07005386 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07005387
Magnus Damm1c6fe942005-10-26 01:58:59 -07005388 memset(p, 0, sizeof(*p));
5389
Christoph Lameter3dfa5722008-02-04 22:29:19 -08005390 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07005391 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07005392 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
5393 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07005394}
5395
Cody P Schafer88c90db2013-07-03 15:01:35 -07005396static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
5397{
5398 pageset_init(p);
5399 pageset_set_batch(p, batch);
5400}
5401
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005402/*
Cody P Schafer36640332013-07-03 15:01:40 -07005403 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005404 * to the value high for the pageset p.
5405 */
Cody P Schafer36640332013-07-03 15:01:40 -07005406static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005407 unsigned long high)
5408{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005409 unsigned long batch = max(1UL, high / 4);
5410 if ((high / 4) > (PAGE_SHIFT * 8))
5411 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005412
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005413 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005414}
5415
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005416static void pageset_set_high_and_batch(struct zone *zone,
5417 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005418{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005419 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07005420 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005421 (zone->managed_pages /
5422 percpu_pagelist_fraction));
5423 else
5424 pageset_set_batch(pcp, zone_batchsize(zone));
5425}
5426
Cody P Schafer169f6c12013-07-03 15:01:41 -07005427static void __meminit zone_pageset_init(struct zone *zone, int cpu)
5428{
5429 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
5430
5431 pageset_init(pcp);
5432 pageset_set_high_and_batch(zone, pcp);
5433}
5434
Jiang Liu4ed7e022012-07-31 16:43:35 -07005435static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07005436{
5437 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07005438 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005439 for_each_possible_cpu(cpu)
5440 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07005441}
5442
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005443/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005444 * Allocate per cpu pagesets and initialize them.
5445 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07005446 */
Al Viro78d99552005-12-15 09:18:25 +00005447void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005448{
Mel Gormanb4911ea2016-08-04 15:31:49 -07005449 struct pglist_data *pgdat;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005450 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005451
Wu Fengguang319774e2010-05-24 14:32:49 -07005452 for_each_populated_zone(zone)
5453 setup_zone_pageset(zone);
Mel Gormanb4911ea2016-08-04 15:31:49 -07005454
5455 for_each_online_pgdat(pgdat)
5456 pgdat->per_cpu_nodestats =
5457 alloc_percpu(struct per_cpu_nodestat);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005458}
5459
Matt Tolentinoc09b4242006-01-17 07:03:44 +01005460static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07005461{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005462 /*
5463 * per cpu subsystem is not up at this point. The following code
5464 * relies on the ability of the linker to provide the
5465 * offset of a (static) per cpu variable into the per cpu area.
5466 */
5467 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07005468
Xishi Qiub38a8722013-11-12 15:07:20 -08005469 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005470 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
5471 zone->name, zone->present_pages,
5472 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07005473}
5474
Jiang Liu4ed7e022012-07-31 16:43:35 -07005475int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07005476 unsigned long zone_start_pfn,
Yaowei Baib171e402015-11-05 18:47:06 -08005477 unsigned long size)
Dave Hansened8ece22005-10-29 18:16:50 -07005478{
5479 struct pglist_data *pgdat = zone->zone_pgdat;
Linus Torvalds9dcb8b62016-10-26 10:15:30 -07005480
Dave Hansened8ece22005-10-29 18:16:50 -07005481 pgdat->nr_zones = zone_idx(zone) + 1;
5482
Dave Hansened8ece22005-10-29 18:16:50 -07005483 zone->zone_start_pfn = zone_start_pfn;
5484
Mel Gorman708614e2008-07-23 21:26:51 -07005485 mminit_dprintk(MMINIT_TRACE, "memmap_init",
5486 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5487 pgdat->node_id,
5488 (unsigned long)zone_idx(zone),
5489 zone_start_pfn, (zone_start_pfn + size));
5490
Andi Kleen1e548de2008-02-04 22:29:26 -08005491 zone_init_free_lists(zone);
Linus Torvalds9dcb8b62016-10-26 10:15:30 -07005492 zone->initialized = 1;
Yasunori Goto718127c2006-06-23 02:03:10 -07005493
5494 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07005495}
5496
Tejun Heo0ee332c2011-12-08 10:22:09 -08005497#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07005498#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
Mel Gorman8a942fd2015-06-30 14:56:55 -07005499
Mel Gormanc7132162006-09-27 01:49:43 -07005500/*
5501 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
Mel Gormanc7132162006-09-27 01:49:43 -07005502 */
Mel Gorman8a942fd2015-06-30 14:56:55 -07005503int __meminit __early_pfn_to_nid(unsigned long pfn,
5504 struct mminit_pfnnid_cache *state)
Mel Gormanc7132162006-09-27 01:49:43 -07005505{
Tejun Heoc13291a2011-07-12 10:46:30 +02005506 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07005507 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07005508
Mel Gorman8a942fd2015-06-30 14:56:55 -07005509 if (state->last_start <= pfn && pfn < state->last_end)
5510 return state->last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005511
Yinghai Lue76b63f2013-09-11 14:22:17 -07005512 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
5513 if (nid != -1) {
Mel Gorman8a942fd2015-06-30 14:56:55 -07005514 state->last_start = start_pfn;
5515 state->last_end = end_pfn;
5516 state->last_nid = nid;
Yinghai Lue76b63f2013-09-11 14:22:17 -07005517 }
5518
5519 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005520}
5521#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5522
Mel Gormanc7132162006-09-27 01:49:43 -07005523/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08005524 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005525 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08005526 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07005527 *
Zhang Zhen7d018172014-06-04 16:10:53 -07005528 * If an architecture guarantees that all ranges registered contain no holes
5529 * and may be freed, this this function may be used instead of calling
5530 * memblock_free_early_nid() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07005531 */
Tejun Heoc13291a2011-07-12 10:46:30 +02005532void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07005533{
Tejun Heoc13291a2011-07-12 10:46:30 +02005534 unsigned long start_pfn, end_pfn;
5535 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005536
Tejun Heoc13291a2011-07-12 10:46:30 +02005537 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
5538 start_pfn = min(start_pfn, max_low_pfn);
5539 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005540
Tejun Heoc13291a2011-07-12 10:46:30 +02005541 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005542 memblock_free_early_nid(PFN_PHYS(start_pfn),
5543 (end_pfn - start_pfn) << PAGE_SHIFT,
5544 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005545 }
5546}
5547
5548/**
5549 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005550 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07005551 *
Zhang Zhen7d018172014-06-04 16:10:53 -07005552 * If an architecture guarantees that all ranges registered contain no holes and may
5553 * be freed, this function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07005554 */
5555void __init sparse_memory_present_with_active_regions(int nid)
5556{
Tejun Heoc13291a2011-07-12 10:46:30 +02005557 unsigned long start_pfn, end_pfn;
5558 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005559
Tejun Heoc13291a2011-07-12 10:46:30 +02005560 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
5561 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005562}
5563
5564/**
5565 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005566 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5567 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5568 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07005569 *
5570 * It returns the start and end page frame of a node based on information
Zhang Zhen7d018172014-06-04 16:10:53 -07005571 * provided by memblock_set_node(). If called for a node
Mel Gormanc7132162006-09-27 01:49:43 -07005572 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005573 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07005574 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005575void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005576 unsigned long *start_pfn, unsigned long *end_pfn)
5577{
Tejun Heoc13291a2011-07-12 10:46:30 +02005578 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005579 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02005580
Mel Gormanc7132162006-09-27 01:49:43 -07005581 *start_pfn = -1UL;
5582 *end_pfn = 0;
5583
Tejun Heoc13291a2011-07-12 10:46:30 +02005584 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5585 *start_pfn = min(*start_pfn, this_start_pfn);
5586 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005587 }
5588
Christoph Lameter633c0662007-10-16 01:25:37 -07005589 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07005590 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005591}
5592
5593/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07005594 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5595 * assumption is made that zones within a node are ordered in monotonic
5596 * increasing memory addresses so that the "highest" populated zone is used
5597 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005598static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005599{
5600 int zone_index;
5601 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5602 if (zone_index == ZONE_MOVABLE)
5603 continue;
5604
5605 if (arch_zone_highest_possible_pfn[zone_index] >
5606 arch_zone_lowest_possible_pfn[zone_index])
5607 break;
5608 }
5609
5610 VM_BUG_ON(zone_index == -1);
5611 movable_zone = zone_index;
5612}
5613
5614/*
5615 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03005616 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005617 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5618 * in each node depending on the size of each node and how evenly kernelcore
5619 * is distributed. This helper function adjusts the zone ranges
5620 * provided by the architecture for a given node by using the end of the
5621 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5622 * zones within a node are in order of monotonic increases memory addresses
5623 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005624static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005625 unsigned long zone_type,
5626 unsigned long node_start_pfn,
5627 unsigned long node_end_pfn,
5628 unsigned long *zone_start_pfn,
5629 unsigned long *zone_end_pfn)
5630{
5631 /* Only adjust if ZONE_MOVABLE is on this node */
5632 if (zone_movable_pfn[nid]) {
5633 /* Size ZONE_MOVABLE */
5634 if (zone_type == ZONE_MOVABLE) {
5635 *zone_start_pfn = zone_movable_pfn[nid];
5636 *zone_end_pfn = min(node_end_pfn,
5637 arch_zone_highest_possible_pfn[movable_zone]);
5638
Xishi Qiue506b992016-10-07 16:58:06 -07005639 /* Adjust for ZONE_MOVABLE starting within this range */
5640 } else if (!mirrored_kernelcore &&
5641 *zone_start_pfn < zone_movable_pfn[nid] &&
5642 *zone_end_pfn > zone_movable_pfn[nid]) {
5643 *zone_end_pfn = zone_movable_pfn[nid];
5644
Mel Gorman2a1e2742007-07-17 04:03:12 -07005645 /* Check if this whole range is within ZONE_MOVABLE */
5646 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5647 *zone_start_pfn = *zone_end_pfn;
5648 }
5649}
5650
5651/*
Mel Gormanc7132162006-09-27 01:49:43 -07005652 * Return the number of pages a zone spans in a node, including holes
5653 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5654 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005655static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005656 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005657 unsigned long node_start_pfn,
5658 unsigned long node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005659 unsigned long *zone_start_pfn,
5660 unsigned long *zone_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005661 unsigned long *ignored)
5662{
Xishi Qiub5685e92015-09-08 15:04:16 -07005663 /* When hotadd a new node from cpu_up(), the node should be empty */
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005664 if (!node_start_pfn && !node_end_pfn)
5665 return 0;
5666
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005667 /* Get the start and end of the zone */
Taku Izumid91749c2016-03-15 14:55:18 -07005668 *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5669 *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07005670 adjust_zone_range_for_zone_movable(nid, zone_type,
5671 node_start_pfn, node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005672 zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005673
5674 /* Check that this node has pages within the zone's required range */
Taku Izumid91749c2016-03-15 14:55:18 -07005675 if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07005676 return 0;
5677
5678 /* Move the zone boundaries inside the node if necessary */
Taku Izumid91749c2016-03-15 14:55:18 -07005679 *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
5680 *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005681
5682 /* Return the spanned pages */
Taku Izumid91749c2016-03-15 14:55:18 -07005683 return *zone_end_pfn - *zone_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005684}
5685
5686/*
5687 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005688 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07005689 */
Yinghai Lu32996252009-12-15 17:59:02 -08005690unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005691 unsigned long range_start_pfn,
5692 unsigned long range_end_pfn)
5693{
Tejun Heo96e907d2011-07-12 10:46:29 +02005694 unsigned long nr_absent = range_end_pfn - range_start_pfn;
5695 unsigned long start_pfn, end_pfn;
5696 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07005697
Tejun Heo96e907d2011-07-12 10:46:29 +02005698 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5699 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5700 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5701 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005702 }
Tejun Heo96e907d2011-07-12 10:46:29 +02005703 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07005704}
5705
5706/**
5707 * absent_pages_in_range - Return number of page frames in holes within a range
5708 * @start_pfn: The start PFN to start searching for holes
5709 * @end_pfn: The end PFN to stop searching for holes
5710 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005711 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07005712 */
5713unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5714 unsigned long end_pfn)
5715{
5716 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5717}
5718
5719/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005720static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005721 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005722 unsigned long node_start_pfn,
5723 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005724 unsigned long *ignored)
5725{
Tejun Heo96e907d2011-07-12 10:46:29 +02005726 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5727 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07005728 unsigned long zone_start_pfn, zone_end_pfn;
Taku Izumi342332e2016-03-15 14:55:22 -07005729 unsigned long nr_absent;
Mel Gorman9c7cd682006-09-27 01:49:58 -07005730
Xishi Qiub5685e92015-09-08 15:04:16 -07005731 /* When hotadd a new node from cpu_up(), the node should be empty */
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005732 if (!node_start_pfn && !node_end_pfn)
5733 return 0;
5734
Tejun Heo96e907d2011-07-12 10:46:29 +02005735 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5736 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07005737
Mel Gorman2a1e2742007-07-17 04:03:12 -07005738 adjust_zone_range_for_zone_movable(nid, zone_type,
5739 node_start_pfn, node_end_pfn,
5740 &zone_start_pfn, &zone_end_pfn);
Taku Izumi342332e2016-03-15 14:55:22 -07005741 nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5742
5743 /*
5744 * ZONE_MOVABLE handling.
5745 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5746 * and vice versa.
5747 */
Xishi Qiue506b992016-10-07 16:58:06 -07005748 if (mirrored_kernelcore && zone_movable_pfn[nid]) {
5749 unsigned long start_pfn, end_pfn;
5750 struct memblock_region *r;
Taku Izumi342332e2016-03-15 14:55:22 -07005751
Xishi Qiue506b992016-10-07 16:58:06 -07005752 for_each_memblock(memory, r) {
5753 start_pfn = clamp(memblock_region_memory_base_pfn(r),
5754 zone_start_pfn, zone_end_pfn);
5755 end_pfn = clamp(memblock_region_memory_end_pfn(r),
5756 zone_start_pfn, zone_end_pfn);
Taku Izumi342332e2016-03-15 14:55:22 -07005757
Xishi Qiue506b992016-10-07 16:58:06 -07005758 if (zone_type == ZONE_MOVABLE &&
5759 memblock_is_mirror(r))
5760 nr_absent += end_pfn - start_pfn;
Taku Izumi342332e2016-03-15 14:55:22 -07005761
Xishi Qiue506b992016-10-07 16:58:06 -07005762 if (zone_type == ZONE_NORMAL &&
5763 !memblock_is_mirror(r))
5764 nr_absent += end_pfn - start_pfn;
Taku Izumi342332e2016-03-15 14:55:22 -07005765 }
5766 }
5767
5768 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07005769}
Mel Gorman0e0b8642006-09-27 01:49:56 -07005770
Tejun Heo0ee332c2011-12-08 10:22:09 -08005771#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005772static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005773 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005774 unsigned long node_start_pfn,
5775 unsigned long node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005776 unsigned long *zone_start_pfn,
5777 unsigned long *zone_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005778 unsigned long *zones_size)
5779{
Taku Izumid91749c2016-03-15 14:55:18 -07005780 unsigned int zone;
5781
5782 *zone_start_pfn = node_start_pfn;
5783 for (zone = 0; zone < zone_type; zone++)
5784 *zone_start_pfn += zones_size[zone];
5785
5786 *zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
5787
Mel Gormanc7132162006-09-27 01:49:43 -07005788 return zones_size[zone_type];
5789}
5790
Paul Mundt6ea6e682007-07-15 23:38:20 -07005791static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005792 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005793 unsigned long node_start_pfn,
5794 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005795 unsigned long *zholes_size)
5796{
5797 if (!zholes_size)
5798 return 0;
5799
5800 return zholes_size[zone_type];
5801}
Yinghai Lu20e69262013-03-01 14:51:27 -08005802
Tejun Heo0ee332c2011-12-08 10:22:09 -08005803#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005804
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005805static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005806 unsigned long node_start_pfn,
5807 unsigned long node_end_pfn,
5808 unsigned long *zones_size,
5809 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07005810{
Gu Zhengfebd5942015-06-24 16:57:02 -07005811 unsigned long realtotalpages = 0, totalpages = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005812 enum zone_type i;
5813
Gu Zhengfebd5942015-06-24 16:57:02 -07005814 for (i = 0; i < MAX_NR_ZONES; i++) {
5815 struct zone *zone = pgdat->node_zones + i;
Taku Izumid91749c2016-03-15 14:55:18 -07005816 unsigned long zone_start_pfn, zone_end_pfn;
Gu Zhengfebd5942015-06-24 16:57:02 -07005817 unsigned long size, real_size;
Mel Gormanc7132162006-09-27 01:49:43 -07005818
Gu Zhengfebd5942015-06-24 16:57:02 -07005819 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5820 node_start_pfn,
5821 node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005822 &zone_start_pfn,
5823 &zone_end_pfn,
Gu Zhengfebd5942015-06-24 16:57:02 -07005824 zones_size);
5825 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005826 node_start_pfn, node_end_pfn,
5827 zholes_size);
Taku Izumid91749c2016-03-15 14:55:18 -07005828 if (size)
5829 zone->zone_start_pfn = zone_start_pfn;
5830 else
5831 zone->zone_start_pfn = 0;
Gu Zhengfebd5942015-06-24 16:57:02 -07005832 zone->spanned_pages = size;
5833 zone->present_pages = real_size;
5834
5835 totalpages += size;
5836 realtotalpages += real_size;
5837 }
5838
5839 pgdat->node_spanned_pages = totalpages;
Mel Gormanc7132162006-09-27 01:49:43 -07005840 pgdat->node_present_pages = realtotalpages;
5841 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5842 realtotalpages);
5843}
5844
Mel Gorman835c1342007-10-16 01:25:47 -07005845#ifndef CONFIG_SPARSEMEM
5846/*
5847 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07005848 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5849 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07005850 * round what is now in bits to nearest long in bits, then return it in
5851 * bytes.
5852 */
Linus Torvalds7c455122013-02-18 09:58:02 -08005853static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005854{
5855 unsigned long usemapsize;
5856
Linus Torvalds7c455122013-02-18 09:58:02 -08005857 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005858 usemapsize = roundup(zonesize, pageblock_nr_pages);
5859 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07005860 usemapsize *= NR_PAGEBLOCK_BITS;
5861 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5862
5863 return usemapsize / 8;
5864}
5865
5866static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08005867 struct zone *zone,
5868 unsigned long zone_start_pfn,
5869 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005870{
Linus Torvalds7c455122013-02-18 09:58:02 -08005871 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07005872 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08005873 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005874 zone->pageblock_flags =
5875 memblock_virt_alloc_node_nopanic(usemapsize,
5876 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07005877}
5878#else
Linus Torvalds7c455122013-02-18 09:58:02 -08005879static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5880 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07005881#endif /* CONFIG_SPARSEMEM */
5882
Mel Gormand9c23402007-10-16 01:26:01 -07005883#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08005884
Mel Gormand9c23402007-10-16 01:26:01 -07005885/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07005886void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07005887{
Andrew Morton955c1cd2012-05-29 15:06:31 -07005888 unsigned int order;
5889
Mel Gormand9c23402007-10-16 01:26:01 -07005890 /* Check that pageblock_nr_pages has not already been setup */
5891 if (pageblock_order)
5892 return;
5893
Andrew Morton955c1cd2012-05-29 15:06:31 -07005894 if (HPAGE_SHIFT > PAGE_SHIFT)
5895 order = HUGETLB_PAGE_ORDER;
5896 else
5897 order = MAX_ORDER - 1;
5898
Mel Gormand9c23402007-10-16 01:26:01 -07005899 /*
5900 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07005901 * This value may be variable depending on boot parameters on IA64 and
5902 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07005903 */
5904 pageblock_order = order;
5905}
5906#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5907
Mel Gormanba72cb82007-11-28 16:21:13 -08005908/*
5909 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07005910 * is unused as pageblock_order is set at compile-time. See
5911 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5912 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08005913 */
Chen Gang15ca2202013-09-11 14:20:27 -07005914void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08005915{
Mel Gormanba72cb82007-11-28 16:21:13 -08005916}
Mel Gormand9c23402007-10-16 01:26:01 -07005917
5918#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5919
Jiang Liu01cefae2012-12-12 13:52:19 -08005920static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5921 unsigned long present_pages)
5922{
5923 unsigned long pages = spanned_pages;
5924
5925 /*
5926 * Provide a more accurate estimation if there are holes within
5927 * the zone and SPARSEMEM is in use. If there are holes within the
5928 * zone, each populated memory region may cost us one or two extra
5929 * memmap pages due to alignment because memmap pages for each
Masahiro Yamada89d790a2017-02-27 14:29:01 -08005930 * populated regions may not be naturally aligned on page boundary.
Jiang Liu01cefae2012-12-12 13:52:19 -08005931 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5932 */
5933 if (spanned_pages > present_pages + (present_pages >> 4) &&
5934 IS_ENABLED(CONFIG_SPARSEMEM))
5935 pages = present_pages;
5936
5937 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5938}
5939
Linus Torvalds1da177e2005-04-16 15:20:36 -07005940/*
5941 * Set up the zone data structures:
5942 * - mark all pages reserved
5943 * - mark all memory queues empty
5944 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07005945 *
5946 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005947 */
Wei Yang7f3eb552015-09-08 14:59:50 -07005948static void __paginginit free_area_init_core(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005949{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07005950 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07005951 int nid = pgdat->node_id;
Yasunori Goto718127c2006-06-23 02:03:10 -07005952 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005953
Dave Hansen208d54e2005-10-29 18:16:52 -07005954 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01005955#ifdef CONFIG_NUMA_BALANCING
5956 spin_lock_init(&pgdat->numabalancing_migrate_lock);
5957 pgdat->numabalancing_migrate_nr_pages = 0;
5958 pgdat->numabalancing_migrate_next_window = jiffies;
5959#endif
Kirill A. Shutemova3d0a9182016-02-02 16:57:08 -08005960#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5961 spin_lock_init(&pgdat->split_queue_lock);
5962 INIT_LIST_HEAD(&pgdat->split_queue);
5963 pgdat->split_queue_len = 0;
5964#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005965 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07005966 init_waitqueue_head(&pgdat->pfmemalloc_wait);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07005967#ifdef CONFIG_COMPACTION
5968 init_waitqueue_head(&pgdat->kcompactd_wait);
5969#endif
Joonsoo Kimeefa864b2014-12-12 16:55:46 -08005970 pgdat_page_ext_init(pgdat);
Mel Gormana52633d2016-07-28 15:45:28 -07005971 spin_lock_init(&pgdat->lru_lock);
Mel Gormana9dd0a82016-07-28 15:46:02 -07005972 lruvec_init(node_lruvec(pgdat));
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01005973
Linus Torvalds1da177e2005-04-16 15:20:36 -07005974 for (j = 0; j < MAX_NR_ZONES; j++) {
5975 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08005976 unsigned long size, realsize, freesize, memmap_pages;
Taku Izumid91749c2016-03-15 14:55:18 -07005977 unsigned long zone_start_pfn = zone->zone_start_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005978
Gu Zhengfebd5942015-06-24 16:57:02 -07005979 size = zone->spanned_pages;
5980 realsize = freesize = zone->present_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005981
Mel Gorman0e0b8642006-09-27 01:49:56 -07005982 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08005983 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07005984 * is used by this zone for memmap. This affects the watermark
5985 * and per-cpu initialisations
5986 */
Jiang Liu01cefae2012-12-12 13:52:19 -08005987 memmap_pages = calc_memmap_size(size, realsize);
Zhong Hongboba914f42014-12-12 16:56:21 -08005988 if (!is_highmem_idx(j)) {
5989 if (freesize >= memmap_pages) {
5990 freesize -= memmap_pages;
5991 if (memmap_pages)
5992 printk(KERN_DEBUG
5993 " %s zone: %lu pages used for memmap\n",
5994 zone_names[j], memmap_pages);
5995 } else
Joe Perches11705322016-03-17 14:19:50 -07005996 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
Zhong Hongboba914f42014-12-12 16:56:21 -08005997 zone_names[j], memmap_pages, freesize);
5998 }
Mel Gorman0e0b8642006-09-27 01:49:56 -07005999
Christoph Lameter62672762007-02-10 01:43:07 -08006000 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08006001 if (j == 0 && freesize > dma_reserve) {
6002 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07006003 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08006004 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07006005 }
6006
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07006007 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08006008 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08006009 /* Charge for highmem memmap if there are enough kernel pages */
6010 else if (nr_kernel_pages > memmap_pages * 2)
6011 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08006012 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006013
Jiang Liu9feedc92012-12-12 13:52:12 -08006014 /*
6015 * Set an approximate value for lowmem here, it will be adjusted
6016 * when the bootmem allocator frees pages into the buddy system.
6017 * And all highmem pages will be managed by the buddy system.
6018 */
6019 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07006020#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07006021 zone->node = nid;
Christoph Lameter96146342006-07-03 00:24:13 -07006022#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07006023 zone->name = zone_names[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -07006024 zone->zone_pgdat = pgdat;
Mel Gormana52633d2016-07-28 15:45:28 -07006025 spin_lock_init(&zone->lock);
6026 zone_seqlock_init(zone);
Dave Hansened8ece22005-10-29 18:16:50 -07006027 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006028
Linus Torvalds1da177e2005-04-16 15:20:36 -07006029 if (!size)
6030 continue;
6031
Andrew Morton955c1cd2012-05-29 15:06:31 -07006032 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08006033 setup_usemap(pgdat, zone, zone_start_pfn, size);
Yaowei Baib171e402015-11-05 18:47:06 -08006034 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
Yasunori Goto718127c2006-06-23 02:03:10 -07006035 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07006036 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006037 }
6038}
6039
Fabian Frederickbd721ea2016-08-02 14:03:33 -07006040static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006041{
Tony Luckb0aeba72015-11-10 10:09:47 -08006042 unsigned long __maybe_unused start = 0;
Laura Abbotta1c34a32015-11-05 18:48:46 -08006043 unsigned long __maybe_unused offset = 0;
6044
Linus Torvalds1da177e2005-04-16 15:20:36 -07006045 /* Skip empty nodes */
6046 if (!pgdat->node_spanned_pages)
6047 return;
6048
Andy Whitcroftd41dee32005-06-23 00:07:54 -07006049#ifdef CONFIG_FLAT_NODE_MEM_MAP
Tony Luckb0aeba72015-11-10 10:09:47 -08006050 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
6051 offset = pgdat->node_start_pfn - start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006052 /* ia64 gets its own node_mem_map, before this, without bootmem */
6053 if (!pgdat->node_mem_map) {
Tony Luckb0aeba72015-11-10 10:09:47 -08006054 unsigned long size, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07006055 struct page *map;
6056
Bob Piccoe984bb42006-05-20 15:00:31 -07006057 /*
6058 * The zone's endpoints aren't required to be MAX_ORDER
6059 * aligned but the node_mem_map endpoints must be in order
6060 * for the buddy allocator to function correctly.
6061 */
Cody P Schafer108bcc92013-02-22 16:35:23 -08006062 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07006063 end = ALIGN(end, MAX_ORDER_NR_PAGES);
6064 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07006065 map = alloc_remap(pgdat->node_id, size);
6066 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08006067 map = memblock_virt_alloc_node_nopanic(size,
6068 pgdat->node_id);
Laura Abbotta1c34a32015-11-05 18:48:46 -08006069 pgdat->node_mem_map = map + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006070 }
Roman Zippel12d810c2007-05-31 00:40:54 -07006071#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07006072 /*
6073 * With no DISCONTIG, the global mem_map is just set as node 0's
6074 */
Mel Gormanc7132162006-09-27 01:49:43 -07006075 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006076 mem_map = NODE_DATA(0)->node_mem_map;
Laura Abbotta1c34a32015-11-05 18:48:46 -08006077#if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
Mel Gormanc7132162006-09-27 01:49:43 -07006078 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Laura Abbotta1c34a32015-11-05 18:48:46 -08006079 mem_map -= offset;
Tejun Heo0ee332c2011-12-08 10:22:09 -08006080#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07006081 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006082#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07006083#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006084}
6085
Johannes Weiner9109fb72008-07-23 21:27:20 -07006086void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
6087 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006088{
Johannes Weiner9109fb72008-07-23 21:27:20 -07006089 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07006090 unsigned long start_pfn = 0;
6091 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07006092
Minchan Kim88fdf752012-07-31 16:46:14 -07006093 /* pg_data_t should be reset to zero when it's allocated */
Mel Gorman38087d92016-07-28 15:45:49 -07006094 WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07006095
Mel Gorman3a80a7f2015-06-30 14:57:02 -07006096 reset_deferred_meminit(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006097 pgdat->node_id = nid;
6098 pgdat->node_start_pfn = node_start_pfn;
Mel Gorman75ef7182016-07-28 15:45:24 -07006099 pgdat->per_cpu_nodestats = NULL;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07006100#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6101 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
Juergen Gross8d29e182015-02-11 15:26:01 -08006102 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
Zhen Lei4ada0c52015-09-08 15:04:19 -07006103 (u64)start_pfn << PAGE_SHIFT,
6104 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
Taku Izumid91749c2016-03-15 14:55:18 -07006105#else
6106 start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07006107#endif
6108 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
6109 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006110
6111 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07006112#ifdef CONFIG_FLAT_NODE_MEM_MAP
6113 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
6114 nid, (unsigned long)pgdat,
6115 (unsigned long)pgdat->node_mem_map);
6116#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07006117
Wei Yang7f3eb552015-09-08 14:59:50 -07006118 free_area_init_core(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006119}
6120
Tejun Heo0ee332c2011-12-08 10:22:09 -08006121#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07006122
6123#if MAX_NUMNODES > 1
6124/*
6125 * Figure out the number of possible node ids.
6126 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07006127void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07006128{
Wei Yang904a9552015-09-08 14:59:48 -07006129 unsigned int highest;
Miklos Szeredi418508c2007-05-23 13:57:55 -07006130
Wei Yang904a9552015-09-08 14:59:48 -07006131 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
Miklos Szeredi418508c2007-05-23 13:57:55 -07006132 nr_node_ids = highest + 1;
6133}
Miklos Szeredi418508c2007-05-23 13:57:55 -07006134#endif
6135
Mel Gormanc7132162006-09-27 01:49:43 -07006136/**
Tejun Heo1e019792011-07-12 09:45:34 +02006137 * node_map_pfn_alignment - determine the maximum internode alignment
6138 *
6139 * This function should be called after node map is populated and sorted.
6140 * It calculates the maximum power of two alignment which can distinguish
6141 * all the nodes.
6142 *
6143 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6144 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
6145 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
6146 * shifted, 1GiB is enough and this function will indicate so.
6147 *
6148 * This is used to test whether pfn -> nid mapping of the chosen memory
6149 * model has fine enough granularity to avoid incorrect mapping for the
6150 * populated node map.
6151 *
6152 * Returns the determined alignment in pfn's. 0 if there is no alignment
6153 * requirement (single node).
6154 */
6155unsigned long __init node_map_pfn_alignment(void)
6156{
6157 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02006158 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02006159 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02006160 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02006161
Tejun Heoc13291a2011-07-12 10:46:30 +02006162 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02006163 if (!start || last_nid < 0 || last_nid == nid) {
6164 last_nid = nid;
6165 last_end = end;
6166 continue;
6167 }
6168
6169 /*
6170 * Start with a mask granular enough to pin-point to the
6171 * start pfn and tick off bits one-by-one until it becomes
6172 * too coarse to separate the current node from the last.
6173 */
6174 mask = ~((1 << __ffs(start)) - 1);
6175 while (mask && last_end <= (start & (mask << 1)))
6176 mask <<= 1;
6177
6178 /* accumulate all internode masks */
6179 accl_mask |= mask;
6180 }
6181
6182 /* convert mask to number of pages */
6183 return ~accl_mask + 1;
6184}
6185
Mel Gormana6af2bc2007-02-10 01:42:57 -08006186/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07006187static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07006188{
Mel Gormana6af2bc2007-02-10 01:42:57 -08006189 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02006190 unsigned long start_pfn;
6191 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00006192
Tejun Heoc13291a2011-07-12 10:46:30 +02006193 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
6194 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07006195
Mel Gormana6af2bc2007-02-10 01:42:57 -08006196 if (min_pfn == ULONG_MAX) {
Joe Perches11705322016-03-17 14:19:50 -07006197 pr_warn("Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08006198 return 0;
6199 }
6200
6201 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07006202}
6203
6204/**
6205 * find_min_pfn_with_active_regions - Find the minimum PFN registered
6206 *
6207 * It returns the minimum PFN based on information provided via
Zhang Zhen7d018172014-06-04 16:10:53 -07006208 * memblock_set_node().
Mel Gormanc7132162006-09-27 01:49:43 -07006209 */
6210unsigned long __init find_min_pfn_with_active_regions(void)
6211{
6212 return find_min_pfn_for_node(MAX_NUMNODES);
6213}
6214
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006215/*
6216 * early_calculate_totalpages()
6217 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006218 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006219 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07006220static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07006221{
Mel Gorman7e63efe2007-07-17 04:03:15 -07006222 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02006223 unsigned long start_pfn, end_pfn;
6224 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07006225
Tejun Heoc13291a2011-07-12 10:46:30 +02006226 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
6227 unsigned long pages = end_pfn - start_pfn;
6228
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006229 totalpages += pages;
6230 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006231 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006232 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07006233 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07006234}
6235
Mel Gorman2a1e2742007-07-17 04:03:12 -07006236/*
6237 * Find the PFN the Movable zone begins in each node. Kernel memory
6238 * is spread evenly between nodes as long as the nodes have enough
6239 * memory. When they don't, some nodes will have more kernelcore than
6240 * others
6241 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07006242static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07006243{
6244 int i, nid;
6245 unsigned long usable_startpfn;
6246 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07006247 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006248 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006249 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006250 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07006251 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006252
6253 /* Need to find movable_zone earlier when movable_node is specified. */
6254 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07006255
Mel Gorman7e63efe2007-07-17 04:03:15 -07006256 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006257 * If movable_node is specified, ignore kernelcore and movablecore
6258 * options.
6259 */
6260 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07006261 for_each_memblock(memory, r) {
6262 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006263 continue;
6264
Emil Medve136199f2014-04-07 15:37:52 -07006265 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006266
Emil Medve136199f2014-04-07 15:37:52 -07006267 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006268 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6269 min(usable_startpfn, zone_movable_pfn[nid]) :
6270 usable_startpfn;
6271 }
6272
6273 goto out2;
6274 }
6275
6276 /*
Taku Izumi342332e2016-03-15 14:55:22 -07006277 * If kernelcore=mirror is specified, ignore movablecore option
6278 */
6279 if (mirrored_kernelcore) {
6280 bool mem_below_4gb_not_mirrored = false;
6281
6282 for_each_memblock(memory, r) {
6283 if (memblock_is_mirror(r))
6284 continue;
6285
6286 nid = r->nid;
6287
6288 usable_startpfn = memblock_region_memory_base_pfn(r);
6289
6290 if (usable_startpfn < 0x100000) {
6291 mem_below_4gb_not_mirrored = true;
6292 continue;
6293 }
6294
6295 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6296 min(usable_startpfn, zone_movable_pfn[nid]) :
6297 usable_startpfn;
6298 }
6299
6300 if (mem_below_4gb_not_mirrored)
6301 pr_warn("This configuration results in unmirrored kernel memory.");
6302
6303 goto out2;
6304 }
6305
6306 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006307 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07006308 * kernelcore that corresponds so that memory usable for
6309 * any allocation type is evenly spread. If both kernelcore
6310 * and movablecore are specified, then the value of kernelcore
6311 * will be used for required_kernelcore if it's greater than
6312 * what movablecore would have allowed.
6313 */
6314 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07006315 unsigned long corepages;
6316
6317 /*
6318 * Round-up so that ZONE_MOVABLE is at least as large as what
6319 * was requested by the user
6320 */
6321 required_movablecore =
6322 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
Xishi Qiu9fd745d2015-11-05 18:48:11 -08006323 required_movablecore = min(totalpages, required_movablecore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07006324 corepages = totalpages - required_movablecore;
6325
6326 required_kernelcore = max(required_kernelcore, corepages);
6327 }
6328
Xishi Qiubde304b2015-11-05 18:48:56 -08006329 /*
6330 * If kernelcore was not specified or kernelcore size is larger
6331 * than totalpages, there is no ZONE_MOVABLE.
6332 */
6333 if (!required_kernelcore || required_kernelcore >= totalpages)
Yinghai Lu66918dc2009-06-30 11:41:37 -07006334 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07006335
6336 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07006337 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
6338
6339restart:
6340 /* Spread kernelcore memory as evenly as possible throughout nodes */
6341 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006342 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02006343 unsigned long start_pfn, end_pfn;
6344
Mel Gorman2a1e2742007-07-17 04:03:12 -07006345 /*
6346 * Recalculate kernelcore_node if the division per node
6347 * now exceeds what is necessary to satisfy the requested
6348 * amount of memory for the kernel
6349 */
6350 if (required_kernelcore < kernelcore_node)
6351 kernelcore_node = required_kernelcore / usable_nodes;
6352
6353 /*
6354 * As the map is walked, we track how much memory is usable
6355 * by the kernel using kernelcore_remaining. When it is
6356 * 0, the rest of the node is usable by ZONE_MOVABLE
6357 */
6358 kernelcore_remaining = kernelcore_node;
6359
6360 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02006361 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07006362 unsigned long size_pages;
6363
Tejun Heoc13291a2011-07-12 10:46:30 +02006364 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07006365 if (start_pfn >= end_pfn)
6366 continue;
6367
6368 /* Account for what is only usable for kernelcore */
6369 if (start_pfn < usable_startpfn) {
6370 unsigned long kernel_pages;
6371 kernel_pages = min(end_pfn, usable_startpfn)
6372 - start_pfn;
6373
6374 kernelcore_remaining -= min(kernel_pages,
6375 kernelcore_remaining);
6376 required_kernelcore -= min(kernel_pages,
6377 required_kernelcore);
6378
6379 /* Continue if range is now fully accounted */
6380 if (end_pfn <= usable_startpfn) {
6381
6382 /*
6383 * Push zone_movable_pfn to the end so
6384 * that if we have to rebalance
6385 * kernelcore across nodes, we will
6386 * not double account here
6387 */
6388 zone_movable_pfn[nid] = end_pfn;
6389 continue;
6390 }
6391 start_pfn = usable_startpfn;
6392 }
6393
6394 /*
6395 * The usable PFN range for ZONE_MOVABLE is from
6396 * start_pfn->end_pfn. Calculate size_pages as the
6397 * number of pages used as kernelcore
6398 */
6399 size_pages = end_pfn - start_pfn;
6400 if (size_pages > kernelcore_remaining)
6401 size_pages = kernelcore_remaining;
6402 zone_movable_pfn[nid] = start_pfn + size_pages;
6403
6404 /*
6405 * Some kernelcore has been met, update counts and
6406 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07006407 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07006408 */
6409 required_kernelcore -= min(required_kernelcore,
6410 size_pages);
6411 kernelcore_remaining -= size_pages;
6412 if (!kernelcore_remaining)
6413 break;
6414 }
6415 }
6416
6417 /*
6418 * If there is still required_kernelcore, we do another pass with one
6419 * less node in the count. This will push zone_movable_pfn[nid] further
6420 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07006421 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07006422 */
6423 usable_nodes--;
6424 if (usable_nodes && required_kernelcore > usable_nodes)
6425 goto restart;
6426
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006427out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07006428 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6429 for (nid = 0; nid < MAX_NUMNODES; nid++)
6430 zone_movable_pfn[nid] =
6431 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07006432
Yinghai Lu20e69262013-03-01 14:51:27 -08006433out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07006434 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006435 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07006436}
6437
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006438/* Any regular or high memory on that node ? */
6439static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006440{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006441 enum zone_type zone_type;
6442
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006443 if (N_MEMORY == N_NORMAL_MEMORY)
6444 return;
6445
6446 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006447 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08006448 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006449 node_set_state(nid, N_HIGH_MEMORY);
6450 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
6451 zone_type <= ZONE_NORMAL)
6452 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08006453 break;
6454 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006455 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006456}
6457
Mel Gormanc7132162006-09-27 01:49:43 -07006458/**
6459 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006460 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07006461 *
6462 * This will call free_area_init_node() for each active node in the system.
Zhang Zhen7d018172014-06-04 16:10:53 -07006463 * Using the page ranges provided by memblock_set_node(), the size of each
Mel Gormanc7132162006-09-27 01:49:43 -07006464 * zone in each node and their holes is calculated. If the maximum PFN
6465 * between two adjacent zones match, it is assumed that the zone is empty.
6466 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6467 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6468 * starts where the previous one ended. For example, ZONE_DMA32 starts
6469 * at arch_max_dma_pfn.
6470 */
6471void __init free_area_init_nodes(unsigned long *max_zone_pfn)
6472{
Tejun Heoc13291a2011-07-12 10:46:30 +02006473 unsigned long start_pfn, end_pfn;
6474 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08006475
Mel Gormanc7132162006-09-27 01:49:43 -07006476 /* Record where the zone boundaries are */
6477 memset(arch_zone_lowest_possible_pfn, 0,
6478 sizeof(arch_zone_lowest_possible_pfn));
6479 memset(arch_zone_highest_possible_pfn, 0,
6480 sizeof(arch_zone_highest_possible_pfn));
Oliver O'Halloran90cae1f2016-07-26 15:22:17 -07006481
6482 start_pfn = find_min_pfn_with_active_regions();
6483
6484 for (i = 0; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07006485 if (i == ZONE_MOVABLE)
6486 continue;
Oliver O'Halloran90cae1f2016-07-26 15:22:17 -07006487
6488 end_pfn = max(max_zone_pfn[i], start_pfn);
6489 arch_zone_lowest_possible_pfn[i] = start_pfn;
6490 arch_zone_highest_possible_pfn[i] = end_pfn;
6491
6492 start_pfn = end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07006493 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07006494
6495 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6496 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07006497 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07006498
Mel Gormanc7132162006-09-27 01:49:43 -07006499 /* Print out the zone ranges */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006500 pr_info("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07006501 for (i = 0; i < MAX_NR_ZONES; i++) {
6502 if (i == ZONE_MOVABLE)
6503 continue;
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006504 pr_info(" %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08006505 if (arch_zone_lowest_possible_pfn[i] ==
6506 arch_zone_highest_possible_pfn[i])
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006507 pr_cont("empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08006508 else
Juergen Gross8d29e182015-02-11 15:26:01 -08006509 pr_cont("[mem %#018Lx-%#018Lx]\n",
6510 (u64)arch_zone_lowest_possible_pfn[i]
6511 << PAGE_SHIFT,
6512 ((u64)arch_zone_highest_possible_pfn[i]
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07006513 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07006514 }
6515
6516 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006517 pr_info("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07006518 for (i = 0; i < MAX_NUMNODES; i++) {
6519 if (zone_movable_pfn[i])
Juergen Gross8d29e182015-02-11 15:26:01 -08006520 pr_info(" Node %d: %#018Lx\n", i,
6521 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07006522 }
Mel Gormanc7132162006-09-27 01:49:43 -07006523
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07006524 /* Print out the early node map */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006525 pr_info("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02006526 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Juergen Gross8d29e182015-02-11 15:26:01 -08006527 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
6528 (u64)start_pfn << PAGE_SHIFT,
6529 ((u64)end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07006530
6531 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07006532 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08006533 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07006534 for_each_online_node(nid) {
6535 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07006536 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07006537 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006538
6539 /* Any memory on that node */
6540 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006541 node_set_state(nid, N_MEMORY);
6542 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07006543 }
6544}
Mel Gorman2a1e2742007-07-17 04:03:12 -07006545
Mel Gorman7e63efe2007-07-17 04:03:15 -07006546static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07006547{
6548 unsigned long long coremem;
6549 if (!p)
6550 return -EINVAL;
6551
6552 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07006553 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07006554
Mel Gorman7e63efe2007-07-17 04:03:15 -07006555 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07006556 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
6557
6558 return 0;
6559}
Mel Gormaned7ed362007-07-17 04:03:14 -07006560
Mel Gorman7e63efe2007-07-17 04:03:15 -07006561/*
6562 * kernelcore=size sets the amount of memory for use for allocations that
6563 * cannot be reclaimed or migrated.
6564 */
6565static int __init cmdline_parse_kernelcore(char *p)
6566{
Taku Izumi342332e2016-03-15 14:55:22 -07006567 /* parse kernelcore=mirror */
6568 if (parse_option_str(p, "mirror")) {
6569 mirrored_kernelcore = true;
6570 return 0;
6571 }
6572
Mel Gorman7e63efe2007-07-17 04:03:15 -07006573 return cmdline_parse_core(p, &required_kernelcore);
6574}
6575
6576/*
6577 * movablecore=size sets the amount of memory for use for allocations that
6578 * can be reclaimed or migrated.
6579 */
6580static int __init cmdline_parse_movablecore(char *p)
6581{
6582 return cmdline_parse_core(p, &required_movablecore);
6583}
6584
Mel Gormaned7ed362007-07-17 04:03:14 -07006585early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07006586early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07006587
Tejun Heo0ee332c2011-12-08 10:22:09 -08006588#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07006589
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006590void adjust_managed_page_count(struct page *page, long count)
6591{
6592 spin_lock(&managed_page_count_lock);
6593 page_zone(page)->managed_pages += count;
6594 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07006595#ifdef CONFIG_HIGHMEM
6596 if (PageHighMem(page))
6597 totalhigh_pages += count;
6598#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006599 spin_unlock(&managed_page_count_lock);
6600}
Jiang Liu3dcc0572013-07-03 15:03:21 -07006601EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006602
Jiang Liu11199692013-07-03 15:02:48 -07006603unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07006604{
Jiang Liu11199692013-07-03 15:02:48 -07006605 void *pos;
6606 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07006607
Jiang Liu11199692013-07-03 15:02:48 -07006608 start = (void *)PAGE_ALIGN((unsigned long)start);
6609 end = (void *)((unsigned long)end & PAGE_MASK);
6610 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07006611 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07006612 memset(pos, poison, PAGE_SIZE);
6613 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07006614 }
6615
6616 if (pages && s)
Josh Poimboeufadb1fe92016-10-25 09:51:14 -05006617 pr_info("Freeing %s memory: %ldK\n",
6618 s, pages << (PAGE_SHIFT - 10));
Jiang Liu69afade2013-04-29 15:06:21 -07006619
6620 return pages;
6621}
Jiang Liu11199692013-07-03 15:02:48 -07006622EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07006623
Jiang Liucfa11e02013-04-29 15:07:00 -07006624#ifdef CONFIG_HIGHMEM
6625void free_highmem_page(struct page *page)
6626{
6627 __free_reserved_page(page);
6628 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07006629 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07006630 totalhigh_pages++;
6631}
6632#endif
6633
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006634
6635void __init mem_init_print_info(const char *str)
6636{
6637 unsigned long physpages, codesize, datasize, rosize, bss_size;
6638 unsigned long init_code_size, init_data_size;
6639
6640 physpages = get_num_physpages();
6641 codesize = _etext - _stext;
6642 datasize = _edata - _sdata;
6643 rosize = __end_rodata - __start_rodata;
6644 bss_size = __bss_stop - __bss_start;
6645 init_data_size = __init_end - __init_begin;
6646 init_code_size = _einittext - _sinittext;
6647
6648 /*
6649 * Detect special cases and adjust section sizes accordingly:
6650 * 1) .init.* may be embedded into .data sections
6651 * 2) .init.text.* may be out of [__init_begin, __init_end],
6652 * please refer to arch/tile/kernel/vmlinux.lds.S.
6653 * 3) .rodata.* may be embedded into .text or .data sections.
6654 */
6655#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07006656 do { \
6657 if (start <= pos && pos < end && size > adj) \
6658 size -= adj; \
6659 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006660
6661 adj_init_size(__init_begin, __init_end, init_data_size,
6662 _sinittext, init_code_size);
6663 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
6664 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
6665 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
6666 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
6667
6668#undef adj_init_size
6669
Joe Perches756a0252016-03-17 14:19:47 -07006670 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 -07006671#ifdef CONFIG_HIGHMEM
Joe Perches756a0252016-03-17 14:19:47 -07006672 ", %luK highmem"
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006673#endif
Joe Perches756a0252016-03-17 14:19:47 -07006674 "%s%s)\n",
6675 nr_free_pages() << (PAGE_SHIFT - 10),
6676 physpages << (PAGE_SHIFT - 10),
6677 codesize >> 10, datasize >> 10, rosize >> 10,
6678 (init_data_size + init_code_size) >> 10, bss_size >> 10,
6679 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10),
6680 totalcma_pages << (PAGE_SHIFT - 10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006681#ifdef CONFIG_HIGHMEM
Joe Perches756a0252016-03-17 14:19:47 -07006682 totalhigh_pages << (PAGE_SHIFT - 10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006683#endif
Joe Perches756a0252016-03-17 14:19:47 -07006684 str ? ", " : "", str ? str : "");
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006685}
6686
Mel Gorman0e0b8642006-09-27 01:49:56 -07006687/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006688 * set_dma_reserve - set the specified number of pages reserved in the first zone
6689 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07006690 *
Yaowei Bai013110a2015-09-08 15:04:10 -07006691 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
Mel Gorman0e0b8642006-09-27 01:49:56 -07006692 * In the DMA zone, a significant percentage may be consumed by kernel image
6693 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006694 * function may optionally be used to account for unfreeable pages in the
6695 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6696 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07006697 */
6698void __init set_dma_reserve(unsigned long new_dma_reserve)
6699{
6700 dma_reserve = new_dma_reserve;
6701}
6702
Linus Torvalds1da177e2005-04-16 15:20:36 -07006703void __init free_area_init(unsigned long *zones_size)
6704{
Johannes Weiner9109fb72008-07-23 21:27:20 -07006705 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006706 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
6707}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006708
Sebastian Andrzej Siewior005fd4b2016-11-03 15:50:02 +01006709static int page_alloc_cpu_dead(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006710{
Linus Torvalds1da177e2005-04-16 15:20:36 -07006711
Sebastian Andrzej Siewior005fd4b2016-11-03 15:50:02 +01006712 lru_add_drain_cpu(cpu);
6713 drain_pages(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006714
Sebastian Andrzej Siewior005fd4b2016-11-03 15:50:02 +01006715 /*
6716 * Spill the event counters of the dead processor
6717 * into the current processors event counters.
6718 * This artificially elevates the count of the current
6719 * processor.
6720 */
6721 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006722
Sebastian Andrzej Siewior005fd4b2016-11-03 15:50:02 +01006723 /*
6724 * Zero the differential counters of the dead processor
6725 * so that the vm statistics are consistent.
6726 *
6727 * This is only okay since the processor is dead and cannot
6728 * race with what we are doing.
6729 */
6730 cpu_vm_stats_fold(cpu);
6731 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006732}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006733
6734void __init page_alloc_init(void)
6735{
Sebastian Andrzej Siewior005fd4b2016-11-03 15:50:02 +01006736 int ret;
6737
6738 ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD,
6739 "mm/page_alloc:dead", NULL,
6740 page_alloc_cpu_dead);
6741 WARN_ON(ret < 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006742}
6743
6744/*
Yaowei Bai34b10062015-09-08 15:04:13 -07006745 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006746 * or min_free_kbytes changes.
6747 */
6748static void calculate_totalreserve_pages(void)
6749{
6750 struct pglist_data *pgdat;
6751 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006752 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006753
6754 for_each_online_pgdat(pgdat) {
Mel Gorman281e3722016-07-28 15:46:11 -07006755
6756 pgdat->totalreserve_pages = 0;
6757
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006758 for (i = 0; i < MAX_NR_ZONES; i++) {
6759 struct zone *zone = pgdat->node_zones + i;
Mel Gorman3484b2d2014-08-06 16:07:14 -07006760 long max = 0;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006761
6762 /* Find valid and maximum lowmem_reserve in the zone */
6763 for (j = i; j < MAX_NR_ZONES; j++) {
6764 if (zone->lowmem_reserve[j] > max)
6765 max = zone->lowmem_reserve[j];
6766 }
6767
Mel Gorman41858962009-06-16 15:32:12 -07006768 /* we treat the high watermark as reserved pages. */
6769 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006770
Jiang Liub40da042013-02-22 16:33:52 -08006771 if (max > zone->managed_pages)
6772 max = zone->managed_pages;
Johannes Weinera8d01432016-01-14 15:20:15 -08006773
Mel Gorman281e3722016-07-28 15:46:11 -07006774 pgdat->totalreserve_pages += max;
Johannes Weinera8d01432016-01-14 15:20:15 -08006775
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006776 reserve_pages += max;
6777 }
6778 }
6779 totalreserve_pages = reserve_pages;
6780}
6781
6782/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07006783 * setup_per_zone_lowmem_reserve - called whenever
Yaowei Bai34b10062015-09-08 15:04:13 -07006784 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
Linus Torvalds1da177e2005-04-16 15:20:36 -07006785 * has a correct pages reserved value, so an adequate number of
6786 * pages are left in the zone after a successful __alloc_pages().
6787 */
6788static void setup_per_zone_lowmem_reserve(void)
6789{
6790 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006791 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006792
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08006793 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006794 for (j = 0; j < MAX_NR_ZONES; j++) {
6795 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08006796 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006797
6798 zone->lowmem_reserve[j] = 0;
6799
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006800 idx = j;
6801 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006802 struct zone *lower_zone;
6803
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006804 idx--;
6805
Linus Torvalds1da177e2005-04-16 15:20:36 -07006806 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6807 sysctl_lowmem_reserve_ratio[idx] = 1;
6808
6809 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08006810 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07006811 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08006812 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006813 }
6814 }
6815 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006816
6817 /* update totalreserve_pages */
6818 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006819}
6820
Mel Gormancfd3da12011-04-25 21:36:42 +00006821static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006822{
6823 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6824 unsigned long lowmem_pages = 0;
6825 struct zone *zone;
6826 unsigned long flags;
6827
6828 /* Calculate total number of !ZONE_HIGHMEM pages */
6829 for_each_zone(zone) {
6830 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08006831 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006832 }
6833
6834 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07006835 u64 tmp;
6836
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006837 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08006838 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07006839 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006840 if (is_highmem(zone)) {
6841 /*
Nick Piggin669ed172005-11-13 16:06:45 -08006842 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6843 * need highmem pages, so cap pages_min to a small
6844 * value here.
6845 *
Mel Gorman41858962009-06-16 15:32:12 -07006846 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Yaowei Bai42ff2702015-04-14 15:47:14 -07006847 * deltas control asynch page reclaim, and so should
Nick Piggin669ed172005-11-13 16:06:45 -08006848 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006849 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08006850 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006851
Jiang Liub40da042013-02-22 16:33:52 -08006852 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08006853 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07006854 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006855 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08006856 /*
6857 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07006858 * proportionate to the zone's size.
6859 */
Mel Gorman41858962009-06-16 15:32:12 -07006860 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006861 }
6862
Johannes Weiner795ae7a2016-03-17 14:19:14 -07006863 /*
6864 * Set the kswapd watermarks distance according to the
6865 * scale factor in proportion to available memory, but
6866 * ensure a minimum size on small systems.
6867 */
6868 tmp = max_t(u64, tmp >> 2,
6869 mult_frac(zone->managed_pages,
6870 watermark_scale_factor, 10000));
6871
6872 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp;
6873 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006874
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006875 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006876 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006877
6878 /* update totalreserve_pages */
6879 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006880}
6881
Mel Gormancfd3da12011-04-25 21:36:42 +00006882/**
6883 * setup_per_zone_wmarks - called when min_free_kbytes changes
6884 * or when memory is hot-{added|removed}
6885 *
6886 * Ensures that the watermark[min,low,high] values for each zone are set
6887 * correctly with respect to min_free_kbytes.
6888 */
6889void setup_per_zone_wmarks(void)
6890{
6891 mutex_lock(&zonelists_mutex);
6892 __setup_per_zone_wmarks();
6893 mutex_unlock(&zonelists_mutex);
6894}
6895
Randy Dunlap55a44622009-09-21 17:01:20 -07006896/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07006897 * Initialise min_free_kbytes.
6898 *
6899 * For small machines we want it small (128k min). For large machines
6900 * we want it large (64MB max). But it is not linear, because network
6901 * bandwidth does not increase linearly with machine size. We use
6902 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006903 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07006904 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6905 *
6906 * which yields
6907 *
6908 * 16MB: 512k
6909 * 32MB: 724k
6910 * 64MB: 1024k
6911 * 128MB: 1448k
6912 * 256MB: 2048k
6913 * 512MB: 2896k
6914 * 1024MB: 4096k
6915 * 2048MB: 5792k
6916 * 4096MB: 8192k
6917 * 8192MB: 11584k
6918 * 16384MB: 16384k
6919 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07006920int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006921{
6922 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07006923 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006924
6925 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07006926 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006927
Michal Hocko5f127332013-07-08 16:00:40 -07006928 if (new_min_free_kbytes > user_min_free_kbytes) {
6929 min_free_kbytes = new_min_free_kbytes;
6930 if (min_free_kbytes < 128)
6931 min_free_kbytes = 128;
6932 if (min_free_kbytes > 65536)
6933 min_free_kbytes = 65536;
6934 } else {
6935 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6936 new_min_free_kbytes, user_min_free_kbytes);
6937 }
Minchan Kimbc75d332009-06-16 15:32:48 -07006938 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07006939 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006940 setup_per_zone_lowmem_reserve();
Joonsoo Kim6423aa82016-08-10 16:27:49 -07006941
6942#ifdef CONFIG_NUMA
6943 setup_min_unmapped_ratio();
6944 setup_min_slab_ratio();
6945#endif
6946
Linus Torvalds1da177e2005-04-16 15:20:36 -07006947 return 0;
6948}
Jason Baronbc22af72016-05-05 16:22:12 -07006949core_initcall(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006950
6951/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07006952 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07006953 * that we can call two helper functions whenever min_free_kbytes
6954 * changes.
6955 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006956int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006957 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006958{
Han Pingtianda8c7572014-01-23 15:53:17 -08006959 int rc;
6960
6961 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6962 if (rc)
6963 return rc;
6964
Michal Hocko5f127332013-07-08 16:00:40 -07006965 if (write) {
6966 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07006967 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07006968 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006969 return 0;
6970}
6971
Johannes Weiner795ae7a2016-03-17 14:19:14 -07006972int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
6973 void __user *buffer, size_t *length, loff_t *ppos)
6974{
6975 int rc;
6976
6977 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6978 if (rc)
6979 return rc;
6980
6981 if (write)
6982 setup_per_zone_wmarks();
6983
6984 return 0;
6985}
6986
Christoph Lameter96146342006-07-03 00:24:13 -07006987#ifdef CONFIG_NUMA
Joonsoo Kim6423aa82016-08-10 16:27:49 -07006988static void setup_min_unmapped_ratio(void)
Christoph Lameter96146342006-07-03 00:24:13 -07006989{
Joonsoo Kim6423aa82016-08-10 16:27:49 -07006990 pg_data_t *pgdat;
Christoph Lameter96146342006-07-03 00:24:13 -07006991 struct zone *zone;
Christoph Lameter96146342006-07-03 00:24:13 -07006992
Mel Gormana5f5f912016-07-28 15:46:32 -07006993 for_each_online_pgdat(pgdat)
Joonsoo Kim81cbcbc2016-08-10 16:27:46 -07006994 pgdat->min_unmapped_pages = 0;
Mel Gormana5f5f912016-07-28 15:46:32 -07006995
Christoph Lameter96146342006-07-03 00:24:13 -07006996 for_each_zone(zone)
Mel Gormana5f5f912016-07-28 15:46:32 -07006997 zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07006998 sysctl_min_unmapped_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07006999}
Christoph Lameter0ff38492006-09-25 23:31:52 -07007000
Joonsoo Kim6423aa82016-08-10 16:27:49 -07007001
7002int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07007003 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07007004{
Christoph Lameter0ff38492006-09-25 23:31:52 -07007005 int rc;
7006
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07007007 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07007008 if (rc)
7009 return rc;
7010
Joonsoo Kim6423aa82016-08-10 16:27:49 -07007011 setup_min_unmapped_ratio();
7012
7013 return 0;
7014}
7015
7016static void setup_min_slab_ratio(void)
7017{
7018 pg_data_t *pgdat;
7019 struct zone *zone;
7020
Mel Gormana5f5f912016-07-28 15:46:32 -07007021 for_each_online_pgdat(pgdat)
7022 pgdat->min_slab_pages = 0;
7023
Christoph Lameter0ff38492006-09-25 23:31:52 -07007024 for_each_zone(zone)
Mel Gormana5f5f912016-07-28 15:46:32 -07007025 zone->zone_pgdat->min_slab_pages += (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07007026 sysctl_min_slab_ratio) / 100;
Joonsoo Kim6423aa82016-08-10 16:27:49 -07007027}
7028
7029int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
7030 void __user *buffer, size_t *length, loff_t *ppos)
7031{
7032 int rc;
7033
7034 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7035 if (rc)
7036 return rc;
7037
7038 setup_min_slab_ratio();
7039
Christoph Lameter0ff38492006-09-25 23:31:52 -07007040 return 0;
7041}
Christoph Lameter96146342006-07-03 00:24:13 -07007042#endif
7043
Linus Torvalds1da177e2005-04-16 15:20:36 -07007044/*
7045 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
7046 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
7047 * whenever sysctl_lowmem_reserve_ratio changes.
7048 *
7049 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07007050 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07007051 * if in function of the boot time zone sizes.
7052 */
Joe Perchescccad5b2014-06-06 14:38:09 -07007053int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07007054 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007055{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07007056 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007057 setup_per_zone_lowmem_reserve();
7058 return 0;
7059}
7060
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08007061/*
7062 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07007063 * cpu. It is the fraction of total pages in each zone that a hot per cpu
7064 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08007065 */
Joe Perchescccad5b2014-06-06 14:38:09 -07007066int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07007067 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08007068{
7069 struct zone *zone;
David Rientjes7cd2b0a2014-06-23 13:22:04 -07007070 int old_percpu_pagelist_fraction;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08007071 int ret;
7072
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007073 mutex_lock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07007074 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
7075
7076 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
7077 if (!write || ret < 0)
7078 goto out;
7079
7080 /* Sanity checking to avoid pcp imbalance */
7081 if (percpu_pagelist_fraction &&
7082 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
7083 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
7084 ret = -EINVAL;
7085 goto out;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08007086 }
David Rientjes7cd2b0a2014-06-23 13:22:04 -07007087
7088 /* No change? */
7089 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
7090 goto out;
7091
7092 for_each_populated_zone(zone) {
7093 unsigned int cpu;
7094
7095 for_each_possible_cpu(cpu)
7096 pageset_set_high_and_batch(zone,
7097 per_cpu_ptr(zone->pageset, cpu));
7098 }
7099out:
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007100 mutex_unlock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07007101 return ret;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08007102}
7103
Rasmus Villemoesa9919c72015-06-24 16:56:28 -07007104#ifdef CONFIG_NUMA
David S. Millerf034b5d2006-08-24 03:08:07 -07007105int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007106
Linus Torvalds1da177e2005-04-16 15:20:36 -07007107static int __init set_hashdist(char *str)
7108{
7109 if (!str)
7110 return 0;
7111 hashdist = simple_strtoul(str, &str, 0);
7112 return 1;
7113}
7114__setup("hashdist=", set_hashdist);
7115#endif
7116
Srikar Dronamrajuf6f34b42016-10-07 16:59:15 -07007117#ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
7118/*
7119 * Returns the number of pages that arch has reserved but
7120 * is not known to alloc_large_system_hash().
7121 */
7122static unsigned long __init arch_reserved_kernel_pages(void)
7123{
7124 return 0;
7125}
7126#endif
7127
Linus Torvalds1da177e2005-04-16 15:20:36 -07007128/*
7129 * allocate a large system hash table from bootmem
7130 * - it is assumed that the hash table must contain an exact power-of-2
7131 * quantity of entries
7132 * - limit is the number of hash buckets, not the total allocation size
7133 */
7134void *__init alloc_large_system_hash(const char *tablename,
7135 unsigned long bucketsize,
7136 unsigned long numentries,
7137 int scale,
7138 int flags,
7139 unsigned int *_hash_shift,
7140 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00007141 unsigned long low_limit,
7142 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007143{
Tim Bird31fe62b2012-05-23 13:33:35 +00007144 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007145 unsigned long log2qty, size;
7146 void *table = NULL;
7147
7148 /* allow the kernel cmdline to have a say */
7149 if (!numentries) {
7150 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08007151 numentries = nr_kernel_pages;
Srikar Dronamrajuf6f34b42016-10-07 16:59:15 -07007152 numentries -= arch_reserved_kernel_pages();
Jerry Zhoua7e83312013-09-11 14:20:26 -07007153
7154 /* It isn't necessary when PAGE_SIZE >= 1MB */
7155 if (PAGE_SHIFT < 20)
7156 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007157
7158 /* limit to 1 bucket per 2^scale bytes of low memory */
7159 if (scale > PAGE_SHIFT)
7160 numentries >>= (scale - PAGE_SHIFT);
7161 else
7162 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08007163
7164 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07007165 if (unlikely(flags & HASH_SMALL)) {
7166 /* Makes no sense without HASH_EARLY */
7167 WARN_ON(!(flags & HASH_EARLY));
7168 if (!(numentries >> *_hash_shift)) {
7169 numentries = 1UL << *_hash_shift;
7170 BUG_ON(!numentries);
7171 }
7172 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08007173 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007174 }
John Hawkes6e692ed2006-03-25 03:08:02 -08007175 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007176
7177 /* limit allocation size to 1/16 total memory by default */
7178 if (max == 0) {
7179 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
7180 do_div(max, bucketsize);
7181 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08007182 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007183
Tim Bird31fe62b2012-05-23 13:33:35 +00007184 if (numentries < low_limit)
7185 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007186 if (numentries > max)
7187 numentries = max;
7188
David Howellsf0d1b0b2006-12-08 02:37:49 -08007189 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007190
7191 do {
7192 size = bucketsize << log2qty;
7193 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08007194 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007195 else if (hashdist)
7196 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
7197 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07007198 /*
7199 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07007200 * some pages at the end of hash table which
7201 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07007202 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01007203 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07007204 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01007205 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
7206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007207 }
7208 } while (!table && size > PAGE_SIZE && --log2qty);
7209
7210 if (!table)
7211 panic("Failed to allocate %s hash table\n", tablename);
7212
Joe Perches11705322016-03-17 14:19:50 -07007213 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7214 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007215
7216 if (_hash_shift)
7217 *_hash_shift = log2qty;
7218 if (_hash_mask)
7219 *_hash_mask = (1 << log2qty) - 1;
7220
7221 return table;
7222}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08007223
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07007224/*
Minchan Kim80934512012-07-31 16:43:01 -07007225 * This function checks whether pageblock includes unmovable pages or not.
7226 * If @count is not zero, it is okay to include less @count unmovable pages
7227 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07007228 * PageLRU check without isolation or lru_lock could race so that
Yisheng Xie0efadf42017-02-24 14:57:39 -08007229 * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
7230 * check without lock_page also may miss some movable non-lru pages at
7231 * race condition. So you can't expect this function should be exact.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07007232 */
Wen Congyangb023f462012-12-11 16:00:45 -08007233bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
7234 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007235{
7236 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01007237 int mt;
7238
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007239 /*
7240 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07007241 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007242 */
7243 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07007244 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01007245 mt = get_pageblock_migratetype(page);
7246 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07007247 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007248
7249 pfn = page_to_pfn(page);
7250 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
7251 unsigned long check = pfn + iter;
7252
Namhyung Kim29723fc2011-02-25 14:44:25 -08007253 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007254 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08007255
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007256 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07007257
7258 /*
7259 * Hugepages are not in LRU lists, but they're movable.
7260 * We need not scan over tail pages bacause we don't
7261 * handle each tail page individually in migration.
7262 */
7263 if (PageHuge(page)) {
7264 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
7265 continue;
7266 }
7267
Minchan Kim97d255c2012-07-31 16:42:59 -07007268 /*
7269 * We can't use page_count without pin a page
7270 * because another CPU can free compound page.
7271 * This check already skips compound tails of THP
Joonsoo Kim0139aa72016-05-19 17:10:49 -07007272 * because their page->_refcount is zero at all time.
Minchan Kim97d255c2012-07-31 16:42:59 -07007273 */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07007274 if (!page_ref_count(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007275 if (PageBuddy(page))
7276 iter += (1 << page_order(page)) - 1;
7277 continue;
7278 }
Minchan Kim97d255c2012-07-31 16:42:59 -07007279
Wen Congyangb023f462012-12-11 16:00:45 -08007280 /*
7281 * The HWPoisoned page may be not in buddy system, and
7282 * page_count() is not 0.
7283 */
7284 if (skip_hwpoisoned_pages && PageHWPoison(page))
7285 continue;
7286
Yisheng Xie0efadf42017-02-24 14:57:39 -08007287 if (__PageMovable(page))
7288 continue;
7289
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007290 if (!PageLRU(page))
7291 found++;
7292 /*
Johannes Weiner6b4f7792014-12-12 16:56:13 -08007293 * If there are RECLAIMABLE pages, we need to check
7294 * it. But now, memory offline itself doesn't call
7295 * shrink_node_slabs() and it still to be fixed.
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007296 */
7297 /*
7298 * If the page is not RAM, page_count()should be 0.
7299 * we don't need more check. This is an _used_ not-movable page.
7300 *
7301 * The problematic thing here is PG_reserved pages. PG_reserved
7302 * is set to both of a memory hole page and a _used_ kernel
7303 * page at boot.
7304 */
7305 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07007306 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007307 }
Minchan Kim80934512012-07-31 16:43:01 -07007308 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007309}
7310
7311bool is_pageblock_removable_nolock(struct page *page)
7312{
Michal Hocko656a0702012-01-20 14:33:58 -08007313 struct zone *zone;
7314 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08007315
7316 /*
7317 * We have to be careful here because we are iterating over memory
7318 * sections which are not zone aware so we might end up outside of
7319 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08007320 * We have to take care about the node as well. If the node is offline
7321 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08007322 */
Michal Hocko656a0702012-01-20 14:33:58 -08007323 if (!node_online(page_to_nid(page)))
7324 return false;
7325
7326 zone = page_zone(page);
7327 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08007328 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08007329 return false;
7330
Wen Congyangb023f462012-12-11 16:00:45 -08007331 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07007332}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007333
Vlastimil Babka080fe202016-02-05 15:36:41 -08007334#if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007335
7336static unsigned long pfn_max_align_down(unsigned long pfn)
7337{
7338 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
7339 pageblock_nr_pages) - 1);
7340}
7341
7342static unsigned long pfn_max_align_up(unsigned long pfn)
7343{
7344 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
7345 pageblock_nr_pages));
7346}
7347
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007348/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007349static int __alloc_contig_migrate_range(struct compact_control *cc,
7350 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007351{
7352 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07007353 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007354 unsigned long pfn = start;
7355 unsigned int tries = 0;
7356 int ret = 0;
7357
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08007358 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007359
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007360 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007361 if (fatal_signal_pending(current)) {
7362 ret = -EINTR;
7363 break;
7364 }
7365
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007366 if (list_empty(&cc->migratepages)) {
7367 cc->nr_migratepages = 0;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07007368 pfn = isolate_migratepages_range(cc, pfn, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007369 if (!pfn) {
7370 ret = -EINTR;
7371 break;
7372 }
7373 tries = 0;
7374 } else if (++tries == 5) {
7375 ret = ret < 0 ? ret : -EBUSY;
7376 break;
7377 }
7378
Minchan Kimbeb51ea2012-10-08 16:33:51 -07007379 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
7380 &cc->migratepages);
7381 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07007382
Hugh Dickins9c620e22013-02-22 16:35:14 -08007383 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07007384 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007385 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08007386 if (ret < 0) {
7387 putback_movable_pages(&cc->migratepages);
7388 return ret;
7389 }
7390 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007391}
7392
7393/**
7394 * alloc_contig_range() -- tries to allocate given range of pages
7395 * @start: start PFN to allocate
7396 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02007397 * @migratetype: migratetype of the underlaying pageblocks (either
7398 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
7399 * in range must have the same migratetype and it must
7400 * be either of the two.
Lucas Stachca96b622017-02-24 14:58:37 -08007401 * @gfp_mask: GFP mask to use during compaction
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007402 *
7403 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7404 * aligned, however it's the caller's responsibility to guarantee that
7405 * we are the only thread that changes migrate type of pageblocks the
7406 * pages fall in.
7407 *
7408 * The PFN range must belong to a single zone.
7409 *
7410 * Returns zero on success or negative error code. On success all
7411 * pages which PFN is in [start, end) are allocated for the caller and
7412 * need to be freed with free_contig_range().
7413 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02007414int alloc_contig_range(unsigned long start, unsigned long end,
Lucas Stachca96b622017-02-24 14:58:37 -08007415 unsigned migratetype, gfp_t gfp_mask)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007416{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007417 unsigned long outer_start, outer_end;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08007418 unsigned int order;
7419 int ret = 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007420
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007421 struct compact_control cc = {
7422 .nr_migratepages = 0,
7423 .order = -1,
7424 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07007425 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007426 .ignore_skip_hint = true,
Lucas Stachca96b622017-02-24 14:58:37 -08007427 .gfp_mask = memalloc_noio_flags(gfp_mask),
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007428 };
7429 INIT_LIST_HEAD(&cc.migratepages);
7430
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007431 /*
7432 * What we do here is we mark all pageblocks in range as
7433 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7434 * have different sizes, and due to the way page allocator
7435 * work, we align the range to biggest of the two pages so
7436 * that page allocator won't try to merge buddies from
7437 * different pageblocks and change MIGRATE_ISOLATE to some
7438 * other migration type.
7439 *
7440 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7441 * migrate the pages from an unaligned range (ie. pages that
7442 * we are interested in). This will put all the pages in
7443 * range back to page allocator as MIGRATE_ISOLATE.
7444 *
7445 * When this is done, we take the pages in range from page
7446 * allocator removing them from the buddy system. This way
7447 * page allocator will never consider using them.
7448 *
7449 * This lets us mark the pageblocks back as
7450 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7451 * aligned range but not in the unaligned, original range are
7452 * put back to page allocator so that buddy can use them.
7453 */
7454
7455 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08007456 pfn_max_align_up(end), migratetype,
7457 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007458 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07007459 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007460
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007461 /*
7462 * In case of -EBUSY, we'd like to know which page causes problem.
7463 * So, just fall through. We will check it in test_pages_isolated().
7464 */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007465 ret = __alloc_contig_migrate_range(&cc, start, end);
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007466 if (ret && ret != -EBUSY)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007467 goto done;
7468
7469 /*
7470 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7471 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7472 * more, all pages in [start, end) are free in page allocator.
7473 * What we are going to do is to allocate all pages from
7474 * [start, end) (that is remove them from page allocator).
7475 *
7476 * The only problem is that pages at the beginning and at the
7477 * end of interesting range may be not aligned with pages that
7478 * page allocator holds, ie. they can be part of higher order
7479 * pages. Because of this, we reserve the bigger range and
7480 * once this is done free the pages we are not interested in.
7481 *
7482 * We don't have to hold zone->lock here because the pages are
7483 * isolated thus they won't get removed from buddy.
7484 */
7485
7486 lru_add_drain_all();
Vlastimil Babka510f5502014-12-10 15:43:07 -08007487 drain_all_pages(cc.zone);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007488
7489 order = 0;
7490 outer_start = start;
7491 while (!PageBuddy(pfn_to_page(outer_start))) {
7492 if (++order >= MAX_ORDER) {
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007493 outer_start = start;
7494 break;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007495 }
7496 outer_start &= ~0UL << order;
7497 }
7498
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007499 if (outer_start != start) {
7500 order = page_order(pfn_to_page(outer_start));
7501
7502 /*
7503 * outer_start page could be small order buddy page and
7504 * it doesn't include start page. Adjust outer_start
7505 * in this case to report failed page properly
7506 * on tracepoint in test_pages_isolated()
7507 */
7508 if (outer_start + (1UL << order) <= start)
7509 outer_start = start;
7510 }
7511
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007512 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08007513 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewiczdae803e2014-11-13 15:19:27 -08007514 pr_info("%s: [%lx, %lx) PFNs busy\n",
7515 __func__, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007516 ret = -EBUSY;
7517 goto done;
7518 }
7519
Marek Szyprowski49f223a2012-01-25 12:49:24 +01007520 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007521 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007522 if (!outer_end) {
7523 ret = -EBUSY;
7524 goto done;
7525 }
7526
7527 /* Free head and tail (if any) */
7528 if (start != outer_start)
7529 free_contig_range(outer_start, start - outer_start);
7530 if (end != outer_end)
7531 free_contig_range(end, outer_end - end);
7532
7533done:
7534 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02007535 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007536 return ret;
7537}
7538
7539void free_contig_range(unsigned long pfn, unsigned nr_pages)
7540{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08007541 unsigned int count = 0;
7542
7543 for (; nr_pages--; pfn++) {
7544 struct page *page = pfn_to_page(pfn);
7545
7546 count += page_count(page) != 1;
7547 __free_page(page);
7548 }
7549 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007550}
7551#endif
7552
Jiang Liu4ed7e022012-07-31 16:43:35 -07007553#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07007554/*
7555 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7556 * page high values need to be recalulated.
7557 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07007558void __meminit zone_pcp_update(struct zone *zone)
7559{
Cody P Schafer0a647f32013-07-03 15:01:33 -07007560 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007561 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07007562 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07007563 pageset_set_high_and_batch(zone,
7564 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007565 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07007566}
7567#endif
7568
Jiang Liu340175b2012-07-31 16:43:32 -07007569void zone_pcp_reset(struct zone *zone)
7570{
7571 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07007572 int cpu;
7573 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07007574
7575 /* avoid races with drain_pages() */
7576 local_irq_save(flags);
7577 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07007578 for_each_online_cpu(cpu) {
7579 pset = per_cpu_ptr(zone->pageset, cpu);
7580 drain_zonestat(zone, pset);
7581 }
Jiang Liu340175b2012-07-31 16:43:32 -07007582 free_percpu(zone->pageset);
7583 zone->pageset = &boot_pageset;
7584 }
7585 local_irq_restore(flags);
7586}
7587
Wen Congyang6dcd73d2012-12-11 16:01:01 -08007588#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007589/*
Joonsoo Kimb9eb6312016-05-19 17:12:06 -07007590 * All pages in the range must be in a single zone and isolated
7591 * before calling this.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007592 */
7593void
7594__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
7595{
7596 struct page *page;
7597 struct zone *zone;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07007598 unsigned int order, i;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007599 unsigned long pfn;
7600 unsigned long flags;
7601 /* find the first valid pfn */
7602 for (pfn = start_pfn; pfn < end_pfn; pfn++)
7603 if (pfn_valid(pfn))
7604 break;
7605 if (pfn == end_pfn)
7606 return;
7607 zone = page_zone(pfn_to_page(pfn));
7608 spin_lock_irqsave(&zone->lock, flags);
7609 pfn = start_pfn;
7610 while (pfn < end_pfn) {
7611 if (!pfn_valid(pfn)) {
7612 pfn++;
7613 continue;
7614 }
7615 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08007616 /*
7617 * The HWPoisoned page may be not in buddy system, and
7618 * page_count() is not 0.
7619 */
7620 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
7621 pfn++;
7622 SetPageReserved(page);
7623 continue;
7624 }
7625
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007626 BUG_ON(page_count(page));
7627 BUG_ON(!PageBuddy(page));
7628 order = page_order(page);
7629#ifdef CONFIG_DEBUG_VM
Joe Perches11705322016-03-17 14:19:50 -07007630 pr_info("remove from free list %lx %d %lx\n",
7631 pfn, 1 << order, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007632#endif
7633 list_del(&page->lru);
7634 rmv_page_order(page);
7635 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007636 for (i = 0; i < (1 << order); i++)
7637 SetPageReserved((page+i));
7638 pfn += (1 << order);
7639 }
7640 spin_unlock_irqrestore(&zone->lock, flags);
7641}
7642#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007643
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007644bool is_free_buddy_page(struct page *page)
7645{
7646 struct zone *zone = page_zone(page);
7647 unsigned long pfn = page_to_pfn(page);
7648 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07007649 unsigned int order;
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007650
7651 spin_lock_irqsave(&zone->lock, flags);
7652 for (order = 0; order < MAX_ORDER; order++) {
7653 struct page *page_head = page - (pfn & ((1 << order) - 1));
7654
7655 if (PageBuddy(page_head) && page_order(page_head) >= order)
7656 break;
7657 }
7658 spin_unlock_irqrestore(&zone->lock, flags);
7659
7660 return order < MAX_ORDER;
7661}