blob: 59de90d5d3a362c7293e624e748cd7b6ebab73b3 [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>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070058#include <linux/prefetch.h>
Lisa Du6e543d52013-09-11 14:22:36 -070059#include <linux/mm_inline.h>
Michal Nazarewicz041d3a82011-12-29 13:09:50 +010060#include <linux/migrate.h>
Joonsoo Kime30825f2014-12-12 16:55:49 -080061#include <linux/page_ext.h>
David Rientjes949f7ec2013-04-29 15:07:48 -070062#include <linux/hugetlb.h>
Clark Williams8bd75c72013-02-07 09:47:07 -060063#include <linux/sched/rt.h>
Joonsoo Kim48c96a32014-12-12 16:56:01 -080064#include <linux/page_owner.h>
Mel Gorman0e1cc952015-06-30 14:57:27 -070065#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Jiang Liu7ee3d4e2013-07-03 15:03:41 -070067#include <asm/sections.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070069#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include "internal.h"
71
Cody P Schaferc8e251f2013-07-03 15:01:29 -070072/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
73static DEFINE_MUTEX(pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -070074#define MIN_PERCPU_PAGELIST_FRACTION (8)
Cody P Schaferc8e251f2013-07-03 15:01:29 -070075
Lee Schermerhorn72812012010-05-26 14:44:56 -070076#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
77DEFINE_PER_CPU(int, numa_node);
78EXPORT_PER_CPU_SYMBOL(numa_node);
79#endif
80
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070081#ifdef CONFIG_HAVE_MEMORYLESS_NODES
82/*
83 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
84 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
85 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
86 * defined in <linux/topology.h>.
87 */
88DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
89EXPORT_PER_CPU_SYMBOL(_numa_mem_);
Joonsoo Kimad2c8142014-10-09 15:26:13 -070090int _node_numa_mem_[MAX_NUMNODES];
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070091#endif
92
Linus Torvalds1da177e2005-04-16 15:20:36 -070093/*
Christoph Lameter13808912007-10-16 01:25:27 -070094 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 */
Christoph Lameter13808912007-10-16 01:25:27 -070096nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
97 [N_POSSIBLE] = NODE_MASK_ALL,
98 [N_ONLINE] = { { [0] = 1UL } },
99#ifndef CONFIG_NUMA
100 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
101#ifdef CONFIG_HIGHMEM
102 [N_HIGH_MEMORY] = { { [0] = 1UL } },
103#endif
Lai Jiangshan20b2f522012-12-12 13:52:00 -0800104#ifdef CONFIG_MOVABLE_NODE
105 [N_MEMORY] = { { [0] = 1UL } },
106#endif
Christoph Lameter13808912007-10-16 01:25:27 -0700107 [N_CPU] = { { [0] = 1UL } },
108#endif /* NUMA */
109};
110EXPORT_SYMBOL(node_states);
111
Jiang Liuc3d5f5f2013-07-03 15:03:14 -0700112/* Protect totalram_pages and zone->managed_pages */
113static DEFINE_SPINLOCK(managed_page_count_lock);
114
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700115unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -0700116unsigned long totalreserve_pages __read_mostly;
Pintu Kumare48322a2014-12-18 16:17:15 -0800117unsigned long totalcma_pages __read_mostly;
Johannes Weinerab8fabd2012-01-10 15:07:42 -0800118
Hugh Dickins1b76b022012-05-11 01:00:07 -0700119int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000120gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -0700122/*
123 * A cached value of the page's pageblock's migratetype, used when the page is
124 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
125 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
126 * Also the migratetype set in the page does not necessarily match the pcplist
127 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
128 * other index - this ensures that it will be put on the correct CMA freelist.
129 */
130static inline int get_pcppage_migratetype(struct page *page)
131{
132 return page->index;
133}
134
135static inline void set_pcppage_migratetype(struct page *page, int migratetype)
136{
137 page->index = migratetype;
138}
139
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800140#ifdef CONFIG_PM_SLEEP
141/*
142 * The following functions are used by the suspend/hibernate code to temporarily
143 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
144 * while devices are suspended. To avoid races with the suspend/hibernate code,
145 * they should always be called with pm_mutex held (gfp_allowed_mask also should
146 * only be modified with pm_mutex held, unless the suspend/hibernate code is
147 * guaranteed not to run in parallel with that modification).
148 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100149
150static gfp_t saved_gfp_mask;
151
152void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800153{
154 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100155 if (saved_gfp_mask) {
156 gfp_allowed_mask = saved_gfp_mask;
157 saved_gfp_mask = 0;
158 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800159}
160
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100161void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800162{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800163 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100164 WARN_ON(saved_gfp_mask);
165 saved_gfp_mask = gfp_allowed_mask;
Mel Gormand0164ad2015-11-06 16:28:21 -0800166 gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800167}
Mel Gormanf90ac392012-01-10 15:07:15 -0800168
169bool pm_suspended_storage(void)
170{
Mel Gormand0164ad2015-11-06 16:28:21 -0800171 if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
Mel Gormanf90ac392012-01-10 15:07:15 -0800172 return false;
173 return true;
174}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800175#endif /* CONFIG_PM_SLEEP */
176
Mel Gormand9c23402007-10-16 01:26:01 -0700177#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -0800178unsigned int pageblock_order __read_mostly;
Mel Gormand9c23402007-10-16 01:26:01 -0700179#endif
180
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800181static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183/*
184 * results with 256, 32 in the lowmem_reserve sysctl:
185 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
186 * 1G machine -> (16M dma, 784M normal, 224M high)
187 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
188 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
Yaowei Bai84109e12015-02-12 15:00:22 -0800189 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100190 *
191 * TBD: should special case ZONE_DMA32 machines here - in those we normally
192 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700194int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800195#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700196 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800197#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700198#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700199 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700200#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700201#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700202 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700203#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700204 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700205};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
207EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
Helge Deller15ad7cd2006-12-06 20:40:36 -0800209static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800210#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700211 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800212#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700213#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700214 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700215#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700216 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700217#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700218 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700219#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700220 "Movable",
Dan Williams033fbae2015-08-09 15:29:06 -0400221#ifdef CONFIG_ZONE_DEVICE
222 "Device",
223#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700224};
225
Vlastimil Babka60f30352016-03-15 14:56:08 -0700226char * const migratetype_names[MIGRATE_TYPES] = {
227 "Unmovable",
228 "Movable",
229 "Reclaimable",
230 "HighAtomic",
231#ifdef CONFIG_CMA
232 "CMA",
233#endif
234#ifdef CONFIG_MEMORY_ISOLATION
235 "Isolate",
236#endif
237};
238
Kirill A. Shutemovf1e61552015-11-06 16:29:50 -0800239compound_page_dtor * const compound_page_dtors[] = {
240 NULL,
241 free_compound_page,
242#ifdef CONFIG_HUGETLB_PAGE
243 free_huge_page,
244#endif
Kirill A. Shutemov9a982252016-01-15 16:54:17 -0800245#ifdef CONFIG_TRANSPARENT_HUGEPAGE
246 free_transhuge_page,
247#endif
Kirill A. Shutemovf1e61552015-11-06 16:29:50 -0800248};
249
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250int min_free_kbytes = 1024;
Han Pingtian42aa83c2014-01-23 15:53:28 -0800251int user_min_free_kbytes = -1;
Johannes Weiner795ae7a2016-03-17 14:19:14 -0700252int watermark_scale_factor = 10;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Jan Beulich2c85f512009-09-21 17:03:07 -0700254static unsigned long __meminitdata nr_kernel_pages;
255static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700256static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
Tejun Heo0ee332c2011-12-08 10:22:09 -0800258#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
259static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
260static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
261static unsigned long __initdata required_kernelcore;
262static unsigned long __initdata required_movablecore;
263static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Taku Izumi342332e2016-03-15 14:55:22 -0700264static bool mirrored_kernelcore;
Mel Gormanc7132162006-09-27 01:49:43 -0700265
Tejun Heo0ee332c2011-12-08 10:22:09 -0800266/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
267int movable_zone;
268EXPORT_SYMBOL(movable_zone);
269#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700270
Miklos Szeredi418508c2007-05-23 13:57:55 -0700271#if MAX_NUMNODES > 1
272int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700273int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700274EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700275EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700276#endif
277
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700278int page_group_by_mobility_disabled __read_mostly;
279
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700280#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
281static inline void reset_deferred_meminit(pg_data_t *pgdat)
282{
283 pgdat->first_deferred_pfn = ULONG_MAX;
284}
285
286/* Returns true if the struct page for the pfn is uninitialised */
Mel Gorman0e1cc952015-06-30 14:57:27 -0700287static inline bool __meminit early_page_uninitialised(unsigned long pfn)
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700288{
Mel Gormanae026b22015-07-17 16:23:48 -0700289 if (pfn >= NODE_DATA(early_pfn_to_nid(pfn))->first_deferred_pfn)
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700290 return true;
291
292 return false;
293}
294
Mel Gorman7e18adb2015-06-30 14:57:05 -0700295static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
296{
297 if (pfn >= NODE_DATA(nid)->first_deferred_pfn)
298 return true;
299
300 return false;
301}
302
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700303/*
304 * Returns false when the remaining initialisation should be deferred until
305 * later in the boot cycle when it can be parallelised.
306 */
307static inline bool update_defer_init(pg_data_t *pgdat,
308 unsigned long pfn, unsigned long zone_end,
309 unsigned long *nr_initialised)
310{
Li Zhang987b3092016-03-17 14:20:16 -0700311 unsigned long max_initialise;
312
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700313 /* Always populate low zones for address-contrained allocations */
314 if (zone_end < pgdat_end_pfn(pgdat))
315 return true;
Li Zhang987b3092016-03-17 14:20:16 -0700316 /*
317 * Initialise at least 2G of a node but also take into account that
318 * two large system hashes that can take up 1GB for 0.25TB/node.
319 */
320 max_initialise = max(2UL << (30 - PAGE_SHIFT),
321 (pgdat->node_spanned_pages >> 8));
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700322
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700323 (*nr_initialised)++;
Li Zhang987b3092016-03-17 14:20:16 -0700324 if ((*nr_initialised > max_initialise) &&
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700325 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
326 pgdat->first_deferred_pfn = pfn;
327 return false;
328 }
329
330 return true;
331}
332#else
333static inline void reset_deferred_meminit(pg_data_t *pgdat)
334{
335}
336
337static inline bool early_page_uninitialised(unsigned long pfn)
338{
339 return false;
340}
341
Mel Gorman7e18adb2015-06-30 14:57:05 -0700342static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
343{
344 return false;
345}
346
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700347static 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
355
Minchan Kimee6f5092012-07-31 16:43:50 -0700356void set_pageblock_migratetype(struct page *page, int migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700357{
KOSAKI Motohiro5d0f3f72013-11-12 15:08:18 -0800358 if (unlikely(page_group_by_mobility_disabled &&
359 migratetype < MIGRATE_PCPTYPES))
Mel Gorman49255c62009-06-16 15:31:58 -0700360 migratetype = MIGRATE_UNMOVABLE;
361
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700362 set_pageblock_flags_group(page, (unsigned long)migratetype,
363 PB_migrate, PB_migrate_end);
364}
365
Nick Piggin13e74442006-01-06 00:10:58 -0800366#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700367static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700369 int ret = 0;
370 unsigned seq;
371 unsigned long pfn = page_to_pfn(page);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800372 unsigned long sp, start_pfn;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700373
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700374 do {
375 seq = zone_span_seqbegin(zone);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800376 start_pfn = zone->zone_start_pfn;
377 sp = zone->spanned_pages;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800378 if (!zone_spans_pfn(zone, pfn))
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700379 ret = 1;
380 } while (zone_span_seqretry(zone, seq));
381
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800382 if (ret)
Dave Hansen613813e2014-06-04 16:07:27 -0700383 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
384 pfn, zone_to_nid(zone), zone->name,
385 start_pfn, start_pfn + sp);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800386
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700387 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700388}
389
390static int page_is_consistent(struct zone *zone, struct page *page)
391{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700392 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700393 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700395 return 0;
396
397 return 1;
398}
399/*
400 * Temporary debugging check for pages not lying within a given zone.
401 */
402static int bad_range(struct zone *zone, struct page *page)
403{
404 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700406 if (!page_is_consistent(zone, page))
407 return 1;
408
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 return 0;
410}
Nick Piggin13e74442006-01-06 00:10:58 -0800411#else
412static inline int bad_range(struct zone *zone, struct page *page)
413{
414 return 0;
415}
416#endif
417
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700418static void bad_page(struct page *page, const char *reason,
419 unsigned long bad_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800421 static unsigned long resume;
422 static unsigned long nr_shown;
423 static unsigned long nr_unshown;
424
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200425 /* Don't complain about poisoned pages */
426 if (PageHWPoison(page)) {
Mel Gorman22b751c2013-02-22 16:34:59 -0800427 page_mapcount_reset(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200428 return;
429 }
430
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800431 /*
432 * Allow a burst of 60 reports, then keep quiet for that minute;
433 * or allow a steady drip of one report per second.
434 */
435 if (nr_shown == 60) {
436 if (time_before(jiffies, resume)) {
437 nr_unshown++;
438 goto out;
439 }
440 if (nr_unshown) {
Vlastimil Babkaff8e8112016-03-15 14:56:24 -0700441 pr_alert(
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800442 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800443 nr_unshown);
444 nr_unshown = 0;
445 }
446 nr_shown = 0;
447 }
448 if (nr_shown++ == 0)
449 resume = jiffies + 60 * HZ;
450
Vlastimil Babkaff8e8112016-03-15 14:56:24 -0700451 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800452 current->comm, page_to_pfn(page));
Vlastimil Babkaff8e8112016-03-15 14:56:24 -0700453 __dump_page(page, reason);
454 bad_flags &= page->flags;
455 if (bad_flags)
456 pr_alert("bad because of flags: %#lx(%pGp)\n",
457 bad_flags, &bad_flags);
Vlastimil Babka4e462112016-03-15 14:56:21 -0700458 dump_page_owner(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700459
Dave Jones4f318882011-10-31 17:07:24 -0700460 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800462out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800463 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800464 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030465 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466}
467
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468/*
469 * Higher-order pages are called "compound pages". They are structured thusly:
470 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800471 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800473 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
474 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800476 * The first tail page's ->compound_dtor holds the offset in array of compound
477 * page destructors. See compound_page_dtors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800479 * The first tail page's ->compound_order holds the order of allocation.
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800480 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800482
Kirill A. Shutemov9a982252016-01-15 16:54:17 -0800483void free_compound_page(struct page *page)
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800484{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700485 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800486}
487
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -0800488void prep_compound_page(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489{
490 int i;
491 int nr_pages = 1 << order;
492
Kirill A. Shutemovf1e61552015-11-06 16:29:50 -0800493 set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700494 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700495 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800496 for (i = 1; i < nr_pages; i++) {
497 struct page *p = page + i;
Youquan Song58a84aa2011-12-08 14:34:18 -0800498 set_page_count(p, 0);
Kirill A. Shutemov1c290f62016-01-15 16:52:07 -0800499 p->mapping = TAIL_MAPPING;
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800500 set_compound_head(p, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 }
Kirill A. Shutemov53f92632016-01-15 16:53:42 -0800502 atomic_set(compound_mapcount_ptr(page), -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503}
504
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800505#ifdef CONFIG_DEBUG_PAGEALLOC
506unsigned int _debug_guardpage_minorder;
Christian Borntraegerea6eabb2016-03-15 14:55:30 -0700507bool _debug_pagealloc_enabled __read_mostly
508 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
Joonsoo Kim505f6d22016-03-17 14:17:56 -0700509EXPORT_SYMBOL(_debug_pagealloc_enabled);
Joonsoo Kime30825f2014-12-12 16:55:49 -0800510bool _debug_guardpage_enabled __read_mostly;
511
Joonsoo Kim031bc572014-12-12 16:55:52 -0800512static int __init early_debug_pagealloc(char *buf)
513{
514 if (!buf)
515 return -EINVAL;
516
517 if (strcmp(buf, "on") == 0)
518 _debug_pagealloc_enabled = true;
519
Christian Borntraegerea6eabb2016-03-15 14:55:30 -0700520 if (strcmp(buf, "off") == 0)
521 _debug_pagealloc_enabled = false;
522
Joonsoo Kim031bc572014-12-12 16:55:52 -0800523 return 0;
524}
525early_param("debug_pagealloc", early_debug_pagealloc);
526
Joonsoo Kime30825f2014-12-12 16:55:49 -0800527static bool need_debug_guardpage(void)
528{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800529 /* If we don't use debug_pagealloc, we don't need guard page */
530 if (!debug_pagealloc_enabled())
531 return false;
532
Joonsoo Kime30825f2014-12-12 16:55:49 -0800533 return true;
534}
535
536static void init_debug_guardpage(void)
537{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800538 if (!debug_pagealloc_enabled())
539 return;
540
Joonsoo Kime30825f2014-12-12 16:55:49 -0800541 _debug_guardpage_enabled = true;
542}
543
544struct page_ext_operations debug_guardpage_ops = {
545 .need = need_debug_guardpage,
546 .init = init_debug_guardpage,
547};
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800548
549static int __init debug_guardpage_minorder_setup(char *buf)
550{
551 unsigned long res;
552
553 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
Joe Perches11705322016-03-17 14:19:50 -0700554 pr_err("Bad debug_guardpage_minorder value\n");
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800555 return 0;
556 }
557 _debug_guardpage_minorder = res;
Joe Perches11705322016-03-17 14:19:50 -0700558 pr_info("Setting debug_guardpage_minorder to %lu\n", res);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800559 return 0;
560}
561__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
562
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800563static inline void set_page_guard(struct zone *zone, struct page *page,
564 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800565{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800566 struct page_ext *page_ext;
567
568 if (!debug_guardpage_enabled())
569 return;
570
571 page_ext = lookup_page_ext(page);
572 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
573
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800574 INIT_LIST_HEAD(&page->lru);
575 set_page_private(page, order);
576 /* Guard pages are not available for any usage */
577 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800578}
579
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800580static inline void clear_page_guard(struct zone *zone, struct page *page,
581 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800582{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800583 struct page_ext *page_ext;
584
585 if (!debug_guardpage_enabled())
586 return;
587
588 page_ext = lookup_page_ext(page);
589 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
590
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800591 set_page_private(page, 0);
592 if (!is_migrate_isolate(migratetype))
593 __mod_zone_freepage_state(zone, (1 << order), migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800594}
595#else
Joonsoo Kime30825f2014-12-12 16:55:49 -0800596struct page_ext_operations debug_guardpage_ops = { NULL, };
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800597static inline void set_page_guard(struct zone *zone, struct page *page,
598 unsigned int order, int migratetype) {}
599static inline void clear_page_guard(struct zone *zone, struct page *page,
600 unsigned int order, int migratetype) {}
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800601#endif
602
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700603static inline void set_page_order(struct page *page, unsigned int order)
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700604{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700605 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000606 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607}
608
609static inline void rmv_page_order(struct page *page)
610{
Nick Piggin676165a2006-04-10 11:21:48 +1000611 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700612 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613}
614
615/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 * This function checks whether a page is free && is the buddy
617 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800618 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000619 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700620 * (c) a page and its buddy have the same order &&
621 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 *
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700623 * For recording whether a page is in the buddy system, we set ->_mapcount
624 * PAGE_BUDDY_MAPCOUNT_VALUE.
625 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
626 * serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000627 *
628 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700630static inline int page_is_buddy(struct page *page, struct page *buddy,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700631 unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700633 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800634 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800635
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800636 if (page_is_guard(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700637 if (page_zone_id(page) != page_zone_id(buddy))
638 return 0;
639
Weijie Yang4c5018c2015-02-10 14:11:39 -0800640 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
641
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800642 return 1;
643 }
644
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700645 if (PageBuddy(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700646 /*
647 * zone check is done late to avoid uselessly
648 * calculating zone/node ids for pages that could
649 * never merge.
650 */
651 if (page_zone_id(page) != page_zone_id(buddy))
652 return 0;
653
Weijie Yang4c5018c2015-02-10 14:11:39 -0800654 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
655
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700656 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000657 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700658 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659}
660
661/*
662 * Freeing function for a buddy system allocator.
663 *
664 * The concept of a buddy system is to maintain direct-mapped table
665 * (containing bit values) for memory blocks of various "orders".
666 * The bottom level table contains the map for the smallest allocatable
667 * units of memory (here, pages), and each level above it describes
668 * pairs of units from the levels below, hence, "buddies".
669 * At a high level, all that happens here is marking the table entry
670 * at the bottom level available, and propagating the changes upward
671 * as necessary, plus some accounting needed to play nicely with other
672 * parts of the VM system.
673 * At each level, we keep a list of pages, which are heads of continuous
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700674 * free pages of length of (1 << order) and marked with _mapcount
675 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
676 * field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100678 * other. That is, if we allocate a small block, and both were
679 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100681 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100683 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 */
685
Nick Piggin48db57f2006-01-08 01:00:42 -0800686static inline void __free_one_page(struct page *page,
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700687 unsigned long pfn,
Mel Gormaned0ae212009-06-16 15:32:07 -0700688 struct zone *zone, unsigned int order,
689 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
691 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700692 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800693 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700694 struct page *buddy;
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700695 unsigned int max_order;
696
697 max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Cody P Schaferd29bb972013-02-22 16:35:25 -0800699 VM_BUG_ON(!zone_is_initialized(zone));
Kirill A. Shutemov6e9f0d52015-02-11 15:25:50 -0800700 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
Mel Gormaned0ae212009-06-16 15:32:07 -0700702 VM_BUG_ON(migratetype == -1);
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700703 if (likely(!is_migrate_isolate(migratetype)))
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800704 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Mel Gormaned0ae212009-06-16 15:32:07 -0700705
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700706 page_idx = pfn & ((1 << MAX_ORDER) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
Sasha Levin309381fea2014-01-23 15:52:54 -0800708 VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
709 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700711continue_merging:
Joonsoo Kim3c605092014-11-13 15:19:21 -0800712 while (order < max_order - 1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800713 buddy_idx = __find_buddy_index(page_idx, order);
714 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700715 if (!page_is_buddy(page, buddy, order))
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700716 goto done_merging;
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800717 /*
718 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
719 * merge with it and move up one order.
720 */
721 if (page_is_guard(buddy)) {
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800722 clear_page_guard(zone, buddy, order, migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800723 } else {
724 list_del(&buddy->lru);
725 zone->free_area[order].nr_free--;
726 rmv_page_order(buddy);
727 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800728 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 page = page + (combined_idx - page_idx);
730 page_idx = combined_idx;
731 order++;
732 }
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700733 if (max_order < MAX_ORDER) {
734 /* If we are here, it means order is >= pageblock_order.
735 * We want to prevent merge between freepages on isolate
736 * pageblock and normal pageblock. Without this, pageblock
737 * isolation could cause incorrect freepage or CMA accounting.
738 *
739 * We don't want to hit this code for the more frequent
740 * low-order merging.
741 */
742 if (unlikely(has_isolate_pageblock(zone))) {
743 int buddy_mt;
744
745 buddy_idx = __find_buddy_index(page_idx, order);
746 buddy = page + (buddy_idx - page_idx);
747 buddy_mt = get_pageblock_migratetype(buddy);
748
749 if (migratetype != buddy_mt
750 && (is_migrate_isolate(migratetype) ||
751 is_migrate_isolate(buddy_mt)))
752 goto done_merging;
753 }
754 max_order++;
755 goto continue_merging;
756 }
757
758done_merging:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700760
761 /*
762 * If this is not the largest possible page, check if the buddy
763 * of the next-highest order is free. If it is, it's possible
764 * that pages are being freed that will coalesce soon. In case,
765 * that is happening, add the free page to the tail of the list
766 * so it's less likely to be used soon and more likely to be merged
767 * as a higher order page
768 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700769 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700770 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800771 combined_idx = buddy_idx & page_idx;
772 higher_page = page + (combined_idx - page_idx);
773 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700774 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700775 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
776 list_add_tail(&page->lru,
777 &zone->free_area[order].free_list[migratetype]);
778 goto out;
779 }
780 }
781
782 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
783out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 zone->free_area[order].nr_free++;
785}
786
Nick Piggin224abf92006-01-06 00:11:11 -0800787static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700789 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800790 unsigned long bad_flags = 0;
791
Kirill A. Shutemov53f92632016-01-15 16:53:42 -0800792 if (unlikely(atomic_read(&page->_mapcount) != -1))
Dave Hansenf0b791a2014-01-23 15:52:49 -0800793 bad_reason = "nonzero mapcount";
794 if (unlikely(page->mapping != NULL))
795 bad_reason = "non-NULL mapping";
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700796 if (unlikely(page_ref_count(page) != 0))
Dave Hansenf0b791a2014-01-23 15:52:49 -0800797 bad_reason = "nonzero _count";
798 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
799 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
800 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
801 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -0800802#ifdef CONFIG_MEMCG
803 if (unlikely(page->mem_cgroup))
804 bad_reason = "page still charged to cgroup";
805#endif
Dave Hansenf0b791a2014-01-23 15:52:49 -0800806 if (unlikely(bad_reason)) {
807 bad_page(page, bad_reason, bad_flags);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800808 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800809 }
Peter Zijlstra90572892013-10-07 11:29:20 +0100810 page_cpupid_reset_last(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800811 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
812 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
813 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814}
815
816/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700817 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700819 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 *
821 * If the zone was previously in an "all pages pinned" state then look to
822 * see if this freeing clears that state.
823 *
824 * And clear the zone's pages_scanned counter, to hold off the "all pages are
825 * pinned" detection logic.
826 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700827static void free_pcppages_bulk(struct zone *zone, int count,
828 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700830 int migratetype = 0;
Mel Gormana6f9edd62009-09-21 17:03:20 -0700831 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700832 int to_free = count;
Mel Gorman0d5d8232014-08-06 16:07:16 -0700833 unsigned long nr_scanned;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700834
Nick Pigginc54ad302006-01-06 00:10:56 -0800835 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700836 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
837 if (nr_scanned)
838 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700839
Mel Gorman72853e22010-09-09 16:38:16 -0700840 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800841 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700842 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800843
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700844 /*
Mel Gormana6f9edd62009-09-21 17:03:20 -0700845 * Remove pages from lists in a round-robin fashion. A
846 * batch_free count is maintained that is incremented when an
847 * empty list is encountered. This is so more pages are freed
848 * off fuller lists instead of spinning excessively around empty
849 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700850 */
851 do {
Mel Gormana6f9edd62009-09-21 17:03:20 -0700852 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700853 if (++migratetype == MIGRATE_PCPTYPES)
854 migratetype = 0;
855 list = &pcp->lists[migratetype];
856 } while (list_empty(list));
857
Namhyung Kim1d168712011-03-22 16:32:45 -0700858 /* This is the only non-empty list. Free them all. */
859 if (batch_free == MIGRATE_PCPTYPES)
860 batch_free = to_free;
861
Mel Gormana6f9edd62009-09-21 17:03:20 -0700862 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700863 int mt; /* migratetype of the to-be-freed page */
864
Geliang Tanga16601c2016-01-14 15:20:30 -0800865 page = list_last_entry(list, struct page, lru);
Mel Gormana6f9edd62009-09-21 17:03:20 -0700866 /* must delete as __free_one_page list manipulates */
867 list_del(&page->lru);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -0700868
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -0700869 mt = get_pcppage_migratetype(page);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -0700870 /* MIGRATE_ISOLATE page should not go to pcplists */
871 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
872 /* Pageblock could have been isolated meanwhile */
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800873 if (unlikely(has_isolate_pageblock(zone)))
Joonsoo Kim51bb1a42014-11-13 15:19:14 -0800874 mt = get_pageblock_migratetype(page);
Joonsoo Kim51bb1a42014-11-13 15:19:14 -0800875
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700876 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700877 trace_mm_page_pcpu_drain(page, 0, mt);
Mel Gorman72853e22010-09-09 16:38:16 -0700878 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800880 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881}
882
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700883static void free_one_page(struct zone *zone,
884 struct page *page, unsigned long pfn,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700885 unsigned int order,
Mel Gormaned0ae212009-06-16 15:32:07 -0700886 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800887{
Mel Gorman0d5d8232014-08-06 16:07:16 -0700888 unsigned long nr_scanned;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700889 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700890 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
891 if (nr_scanned)
892 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700893
Joonsoo Kimad53f922014-11-13 15:19:11 -0800894 if (unlikely(has_isolate_pageblock(zone) ||
895 is_migrate_isolate(migratetype))) {
896 migratetype = get_pfnblock_migratetype(page, pfn);
Joonsoo Kimad53f922014-11-13 15:19:11 -0800897 }
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700898 __free_one_page(page, pfn, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700899 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800900}
901
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800902static int free_tail_pages_check(struct page *head_page, struct page *page)
903{
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800904 int ret = 1;
905
906 /*
907 * We rely page->lru.next never has bit 0 set, unless the page
908 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
909 */
910 BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
911
912 if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
913 ret = 0;
914 goto out;
915 }
Kirill A. Shutemov9a982252016-01-15 16:54:17 -0800916 switch (page - head_page) {
917 case 1:
918 /* the first tail page: ->mapping is compound_mapcount() */
Kirill A. Shutemov53f92632016-01-15 16:53:42 -0800919 if (unlikely(compound_mapcount(page))) {
920 bad_page(page, "nonzero compound_mapcount", 0);
921 goto out;
922 }
Kirill A. Shutemov9a982252016-01-15 16:54:17 -0800923 break;
924 case 2:
925 /*
926 * the second tail page: ->mapping is
927 * page_deferred_list().next -- ignore value.
928 */
929 break;
930 default:
931 if (page->mapping != TAIL_MAPPING) {
932 bad_page(page, "corrupted mapping in tail page", 0);
933 goto out;
934 }
935 break;
Kirill A. Shutemov1c290f62016-01-15 16:52:07 -0800936 }
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800937 if (unlikely(!PageTail(page))) {
938 bad_page(page, "PageTail not set", 0);
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800939 goto out;
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800940 }
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800941 if (unlikely(compound_head(page) != head_page)) {
942 bad_page(page, "compound_head not consistent", 0);
943 goto out;
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800944 }
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800945 ret = 0;
946out:
Kirill A. Shutemov1c290f62016-01-15 16:52:07 -0800947 page->mapping = NULL;
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800948 clear_compound_head(page);
949 return ret;
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800950}
951
Robin Holt1e8ce832015-06-30 14:56:45 -0700952static void __meminit __init_single_page(struct page *page, unsigned long pfn,
953 unsigned long zone, int nid)
954{
Robin Holt1e8ce832015-06-30 14:56:45 -0700955 set_page_links(page, zone, nid, pfn);
Robin Holt1e8ce832015-06-30 14:56:45 -0700956 init_page_count(page);
957 page_mapcount_reset(page);
958 page_cpupid_reset_last(page);
Robin Holt1e8ce832015-06-30 14:56:45 -0700959
Robin Holt1e8ce832015-06-30 14:56:45 -0700960 INIT_LIST_HEAD(&page->lru);
961#ifdef WANT_PAGE_VIRTUAL
962 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
963 if (!is_highmem_idx(zone))
964 set_page_address(page, __va(pfn << PAGE_SHIFT));
965#endif
966}
967
968static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
969 int nid)
970{
971 return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
972}
973
Mel Gorman7e18adb2015-06-30 14:57:05 -0700974#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
975static void init_reserved_page(unsigned long pfn)
976{
977 pg_data_t *pgdat;
978 int nid, zid;
979
980 if (!early_page_uninitialised(pfn))
981 return;
982
983 nid = early_pfn_to_nid(pfn);
984 pgdat = NODE_DATA(nid);
985
986 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
987 struct zone *zone = &pgdat->node_zones[zid];
988
989 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
990 break;
991 }
992 __init_single_pfn(pfn, zid, nid);
993}
994#else
995static inline void init_reserved_page(unsigned long pfn)
996{
997}
998#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
999
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001000/*
1001 * Initialised pages do not have PageReserved set. This function is
1002 * called for each range allocated by the bootmem allocator and
1003 * marks the pages PageReserved. The remaining valid pages are later
1004 * sent to the buddy page allocator.
1005 */
Mel Gorman7e18adb2015-06-30 14:57:05 -07001006void __meminit reserve_bootmem_region(unsigned long start, unsigned long end)
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001007{
1008 unsigned long start_pfn = PFN_DOWN(start);
1009 unsigned long end_pfn = PFN_UP(end);
1010
Mel Gorman7e18adb2015-06-30 14:57:05 -07001011 for (; start_pfn < end_pfn; start_pfn++) {
1012 if (pfn_valid(start_pfn)) {
1013 struct page *page = pfn_to_page(start_pfn);
1014
1015 init_reserved_page(start_pfn);
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -08001016
1017 /* Avoid false-positive PageTail() */
1018 INIT_LIST_HEAD(&page->lru);
1019
Mel Gorman7e18adb2015-06-30 14:57:05 -07001020 SetPageReserved(page);
1021 }
1022 }
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001023}
1024
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001025static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026{
Kirill A. Shutemov81422f22015-02-11 15:25:52 -08001027 bool compound = PageCompound(page);
1028 int i, bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
Yu Zhaoab1f3062014-12-10 15:43:17 -08001030 VM_BUG_ON_PAGE(PageTail(page), page);
Kirill A. Shutemov81422f22015-02-11 15:25:52 -08001031 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
Yu Zhaoab1f3062014-12-10 15:43:17 -08001032
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001033 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001034 kmemcheck_free_shadow(page, order);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -08001035 kasan_free_pages(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001036
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -08001037 if (PageAnon(page))
1038 page->mapping = NULL;
Kirill A. Shutemov81422f22015-02-11 15:25:52 -08001039 bad += free_pages_check(page);
1040 for (i = 1; i < (1 << order); i++) {
1041 if (compound)
1042 bad += free_tail_pages_check(page, page + i);
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -08001043 bad += free_pages_check(page + i);
Kirill A. Shutemov81422f22015-02-11 15:25:52 -08001044 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -08001045 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001046 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001047
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001048 reset_page_owner(page, order);
1049
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001050 if (!PageHighMem(page)) {
Pintu Kumarb8af2942013-09-11 14:20:34 -07001051 debug_check_no_locks_freed(page_address(page),
1052 PAGE_SIZE << order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001053 debug_check_no_obj_freed(page_address(page),
1054 PAGE_SIZE << order);
1055 }
Nick Piggindafb1362006-10-11 01:21:30 -07001056 arch_free_page(page, order);
Laura Abbott8823b1d2016-03-15 14:56:27 -07001057 kernel_poison_pages(page, 1 << order, 0);
Nick Piggin48db57f2006-01-08 01:00:42 -08001058 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -07001059
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001060 return true;
1061}
1062
1063static void __free_pages_ok(struct page *page, unsigned int order)
1064{
1065 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -07001066 int migratetype;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001067 unsigned long pfn = page_to_pfn(page);
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001068
1069 if (!free_pages_prepare(page, order))
1070 return;
1071
Mel Gormancfc47a22014-06-04 16:10:19 -07001072 migratetype = get_pfnblock_migratetype(page, pfn);
Nick Pigginc54ad302006-01-06 00:10:56 -08001073 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001074 __count_vm_events(PGFREE, 1 << order);
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001075 free_one_page(page_zone(page), page, pfn, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -08001076 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077}
1078
Mel Gorman0e1cc952015-06-30 14:57:27 -07001079static void __init __free_pages_boot_core(struct page *page,
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001080 unsigned long pfn, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -08001081{
Johannes Weinerc3993072012-01-10 15:08:10 -08001082 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001083 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -08001084 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -08001085
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001086 prefetchw(p);
1087 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1088 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -08001089 __ClearPageReserved(p);
1090 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -08001091 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001092 __ClearPageReserved(p);
1093 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -08001094
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001095 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -08001096 set_page_refcounted(page);
1097 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -08001098}
1099
Mel Gorman75a592a2015-06-30 14:56:59 -07001100#if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1101 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
Mel Gorman7ace9912015-08-06 15:46:13 -07001102
Mel Gorman75a592a2015-06-30 14:56:59 -07001103static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1104
1105int __meminit early_pfn_to_nid(unsigned long pfn)
1106{
Mel Gorman7ace9912015-08-06 15:46:13 -07001107 static DEFINE_SPINLOCK(early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -07001108 int nid;
1109
Mel Gorman7ace9912015-08-06 15:46:13 -07001110 spin_lock(&early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -07001111 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
Mel Gorman7ace9912015-08-06 15:46:13 -07001112 if (nid < 0)
1113 nid = 0;
1114 spin_unlock(&early_pfn_lock);
1115
1116 return nid;
Mel Gorman75a592a2015-06-30 14:56:59 -07001117}
1118#endif
1119
1120#ifdef CONFIG_NODES_SPAN_OTHER_NODES
1121static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1122 struct mminit_pfnnid_cache *state)
1123{
1124 int nid;
1125
1126 nid = __early_pfn_to_nid(pfn, state);
1127 if (nid >= 0 && nid != node)
1128 return false;
1129 return true;
1130}
1131
1132/* Only safe to use early in boot when initialisation is single-threaded */
1133static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1134{
1135 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1136}
1137
1138#else
1139
1140static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1141{
1142 return true;
1143}
1144static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1145 struct mminit_pfnnid_cache *state)
1146{
1147 return true;
1148}
1149#endif
1150
1151
Mel Gorman0e1cc952015-06-30 14:57:27 -07001152void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001153 unsigned int order)
1154{
1155 if (early_page_uninitialised(pfn))
1156 return;
1157 return __free_pages_boot_core(page, pfn, order);
1158}
1159
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001160/*
1161 * Check that the whole (or subset of) a pageblock given by the interval of
1162 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1163 * with the migration of free compaction scanner. The scanners then need to
1164 * use only pfn_valid_within() check for arches that allow holes within
1165 * pageblocks.
1166 *
1167 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1168 *
1169 * It's possible on some configurations to have a setup like node0 node1 node0
1170 * i.e. it's possible that all pages within a zones range of pages do not
1171 * belong to a single zone. We assume that a border between node0 and node1
1172 * can occur within a single pageblock, but not a node0 node1 node0
1173 * interleaving within a single pageblock. It is therefore sufficient to check
1174 * the first and last page of a pageblock and avoid checking each individual
1175 * page in a pageblock.
1176 */
1177struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1178 unsigned long end_pfn, struct zone *zone)
1179{
1180 struct page *start_page;
1181 struct page *end_page;
1182
1183 /* end_pfn is one past the range we are checking */
1184 end_pfn--;
1185
1186 if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1187 return NULL;
1188
1189 start_page = pfn_to_page(start_pfn);
1190
1191 if (page_zone(start_page) != zone)
1192 return NULL;
1193
1194 end_page = pfn_to_page(end_pfn);
1195
1196 /* This gives a shorter code than deriving page_zone(end_page) */
1197 if (page_zone_id(start_page) != page_zone_id(end_page))
1198 return NULL;
1199
1200 return start_page;
1201}
1202
1203void set_zone_contiguous(struct zone *zone)
1204{
1205 unsigned long block_start_pfn = zone->zone_start_pfn;
1206 unsigned long block_end_pfn;
1207
1208 block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1209 for (; block_start_pfn < zone_end_pfn(zone);
1210 block_start_pfn = block_end_pfn,
1211 block_end_pfn += pageblock_nr_pages) {
1212
1213 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1214
1215 if (!__pageblock_pfn_to_page(block_start_pfn,
1216 block_end_pfn, zone))
1217 return;
1218 }
1219
1220 /* We confirm that there is no hole */
1221 zone->contiguous = true;
1222}
1223
1224void clear_zone_contiguous(struct zone *zone)
1225{
1226 zone->contiguous = false;
1227}
1228
Mel Gorman7e18adb2015-06-30 14:57:05 -07001229#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
Mel Gorman0e1cc952015-06-30 14:57:27 -07001230static void __init deferred_free_range(struct page *page,
Mel Gormana4de83d2015-06-30 14:57:16 -07001231 unsigned long pfn, int nr_pages)
1232{
1233 int i;
1234
1235 if (!page)
1236 return;
1237
1238 /* Free a large naturally-aligned chunk if possible */
1239 if (nr_pages == MAX_ORDER_NR_PAGES &&
1240 (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
Mel Gormanac5d2532015-06-30 14:57:20 -07001241 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormana4de83d2015-06-30 14:57:16 -07001242 __free_pages_boot_core(page, pfn, MAX_ORDER-1);
1243 return;
1244 }
1245
1246 for (i = 0; i < nr_pages; i++, page++, pfn++)
1247 __free_pages_boot_core(page, pfn, 0);
1248}
1249
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001250/* Completion tracking for deferred_init_memmap() threads */
1251static atomic_t pgdat_init_n_undone __initdata;
1252static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1253
1254static inline void __init pgdat_init_report_one_done(void)
1255{
1256 if (atomic_dec_and_test(&pgdat_init_n_undone))
1257 complete(&pgdat_init_all_done_comp);
1258}
Mel Gorman0e1cc952015-06-30 14:57:27 -07001259
Mel Gorman7e18adb2015-06-30 14:57:05 -07001260/* Initialise remaining memory on a node */
Mel Gorman0e1cc952015-06-30 14:57:27 -07001261static int __init deferred_init_memmap(void *data)
Mel Gorman7e18adb2015-06-30 14:57:05 -07001262{
Mel Gorman0e1cc952015-06-30 14:57:27 -07001263 pg_data_t *pgdat = data;
1264 int nid = pgdat->node_id;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001265 struct mminit_pfnnid_cache nid_init_state = { };
1266 unsigned long start = jiffies;
1267 unsigned long nr_pages = 0;
1268 unsigned long walk_start, walk_end;
1269 int i, zid;
1270 struct zone *zone;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001271 unsigned long first_init_pfn = pgdat->first_deferred_pfn;
Mel Gorman0e1cc952015-06-30 14:57:27 -07001272 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001273
Mel Gorman0e1cc952015-06-30 14:57:27 -07001274 if (first_init_pfn == ULONG_MAX) {
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001275 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001276 return 0;
1277 }
1278
1279 /* Bind memory initialisation thread to a local node if possible */
1280 if (!cpumask_empty(cpumask))
1281 set_cpus_allowed_ptr(current, cpumask);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001282
1283 /* Sanity check boundaries */
1284 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1285 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1286 pgdat->first_deferred_pfn = ULONG_MAX;
1287
1288 /* Only the highest zone is deferred so find it */
1289 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1290 zone = pgdat->node_zones + zid;
1291 if (first_init_pfn < zone_end_pfn(zone))
1292 break;
1293 }
1294
1295 for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1296 unsigned long pfn, end_pfn;
Mel Gorman54608c32015-06-30 14:57:09 -07001297 struct page *page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001298 struct page *free_base_page = NULL;
1299 unsigned long free_base_pfn = 0;
1300 int nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001301
1302 end_pfn = min(walk_end, zone_end_pfn(zone));
1303 pfn = first_init_pfn;
1304 if (pfn < walk_start)
1305 pfn = walk_start;
1306 if (pfn < zone->zone_start_pfn)
1307 pfn = zone->zone_start_pfn;
1308
1309 for (; pfn < end_pfn; pfn++) {
Mel Gorman54608c32015-06-30 14:57:09 -07001310 if (!pfn_valid_within(pfn))
Mel Gormana4de83d2015-06-30 14:57:16 -07001311 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001312
Mel Gorman54608c32015-06-30 14:57:09 -07001313 /*
1314 * Ensure pfn_valid is checked every
1315 * MAX_ORDER_NR_PAGES for memory holes
1316 */
1317 if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1318 if (!pfn_valid(pfn)) {
1319 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001320 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001321 }
1322 }
1323
1324 if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1325 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001326 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001327 }
1328
1329 /* Minimise pfn page lookups and scheduler checks */
1330 if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1331 page++;
1332 } else {
Mel Gormana4de83d2015-06-30 14:57:16 -07001333 nr_pages += nr_to_free;
1334 deferred_free_range(free_base_page,
1335 free_base_pfn, nr_to_free);
1336 free_base_page = NULL;
1337 free_base_pfn = nr_to_free = 0;
1338
Mel Gorman54608c32015-06-30 14:57:09 -07001339 page = pfn_to_page(pfn);
1340 cond_resched();
1341 }
Mel Gorman7e18adb2015-06-30 14:57:05 -07001342
1343 if (page->flags) {
1344 VM_BUG_ON(page_zone(page) != zone);
Mel Gormana4de83d2015-06-30 14:57:16 -07001345 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001346 }
1347
1348 __init_single_page(page, pfn, zid, nid);
Mel Gormana4de83d2015-06-30 14:57:16 -07001349 if (!free_base_page) {
1350 free_base_page = page;
1351 free_base_pfn = pfn;
1352 nr_to_free = 0;
1353 }
1354 nr_to_free++;
1355
1356 /* Where possible, batch up pages for a single free */
1357 continue;
1358free_range:
1359 /* Free the current block of pages to allocator */
1360 nr_pages += nr_to_free;
1361 deferred_free_range(free_base_page, free_base_pfn,
1362 nr_to_free);
1363 free_base_page = NULL;
1364 free_base_pfn = nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001365 }
Mel Gormana4de83d2015-06-30 14:57:16 -07001366
Mel Gorman7e18adb2015-06-30 14:57:05 -07001367 first_init_pfn = max(end_pfn, first_init_pfn);
1368 }
1369
1370 /* Sanity check that the next zone really is unpopulated */
1371 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1372
Mel Gorman0e1cc952015-06-30 14:57:27 -07001373 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
Mel Gorman7e18adb2015-06-30 14:57:05 -07001374 jiffies_to_msecs(jiffies - start));
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001375
1376 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001377 return 0;
1378}
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001379#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
Mel Gorman0e1cc952015-06-30 14:57:27 -07001380
1381void __init page_alloc_init_late(void)
1382{
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001383 struct zone *zone;
1384
1385#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
Mel Gorman0e1cc952015-06-30 14:57:27 -07001386 int nid;
1387
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001388 /* There will be num_node_state(N_MEMORY) threads */
1389 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
Mel Gorman0e1cc952015-06-30 14:57:27 -07001390 for_each_node_state(nid, N_MEMORY) {
Mel Gorman0e1cc952015-06-30 14:57:27 -07001391 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1392 }
1393
1394 /* Block until all are initialised */
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001395 wait_for_completion(&pgdat_init_all_done_comp);
Mel Gorman4248b0d2015-08-06 15:46:20 -07001396
1397 /* Reinit limits that are based on free pages after the kernel is up */
1398 files_maxfiles_init();
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001399#endif
1400
1401 for_each_populated_zone(zone)
1402 set_zone_contiguous(zone);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001403}
Mel Gorman7e18adb2015-06-30 14:57:05 -07001404
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001405#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +08001406/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001407void __init init_cma_reserved_pageblock(struct page *page)
1408{
1409 unsigned i = pageblock_nr_pages;
1410 struct page *p = page;
1411
1412 do {
1413 __ClearPageReserved(p);
1414 set_page_count(p, 0);
1415 } while (++p, --i);
1416
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001417 set_pageblock_migratetype(page, MIGRATE_CMA);
Michal Nazarewiczdc783272014-07-02 15:22:35 -07001418
1419 if (pageblock_order >= MAX_ORDER) {
1420 i = pageblock_nr_pages;
1421 p = page;
1422 do {
1423 set_page_refcounted(p);
1424 __free_pages(p, MAX_ORDER - 1);
1425 p += MAX_ORDER_NR_PAGES;
1426 } while (i -= MAX_ORDER_NR_PAGES);
1427 } else {
1428 set_page_refcounted(page);
1429 __free_pages(page, pageblock_order);
1430 }
1431
Jiang Liu3dcc0572013-07-03 15:03:21 -07001432 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001433}
1434#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
1436/*
1437 * The order of subdivision here is critical for the IO subsystem.
1438 * Please do not alter this order without good reasons and regression
1439 * testing. Specifically, as large blocks of memory are subdivided,
1440 * the order in which smaller blocks are delivered depends on the order
1441 * they're subdivided in this function. This is the primary factor
1442 * influencing the order in which pages are delivered to the IO
1443 * subsystem according to empirical testing, and this is also justified
1444 * by considering the behavior of a buddy system containing a single
1445 * large block of memory acted on by a series of small allocations.
1446 * This behavior is a critical factor in sglist merging's success.
1447 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +01001448 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 */
Nick Piggin085cc7d2006-01-06 00:11:01 -08001450static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001451 int low, int high, struct free_area *area,
1452 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453{
1454 unsigned long size = 1 << high;
1455
1456 while (high > low) {
1457 area--;
1458 high--;
1459 size >>= 1;
Sasha Levin309381fea2014-01-23 15:52:54 -08001460 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08001461
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001462 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
Joonsoo Kime30825f2014-12-12 16:55:49 -08001463 debug_guardpage_enabled() &&
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001464 high < debug_guardpage_minorder()) {
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08001465 /*
1466 * Mark as guard pages (or page), that will allow to
1467 * merge back to allocator when buddy will be freed.
1468 * Corresponding page table entries will not be touched,
1469 * pages will stay not present in virtual address space
1470 */
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001471 set_page_guard(zone, &page[size], high, migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08001472 continue;
1473 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001474 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 area->nr_free++;
1476 set_page_order(&page[size], high);
1477 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478}
1479
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480/*
1481 * This page is about to be returned from the page allocator
1482 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001483static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -07001485 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -08001486 unsigned long bad_flags = 0;
1487
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001488 if (unlikely(atomic_read(&page->_mapcount) != -1))
Dave Hansenf0b791a2014-01-23 15:52:49 -08001489 bad_reason = "nonzero mapcount";
1490 if (unlikely(page->mapping != NULL))
1491 bad_reason = "non-NULL mapping";
Joonsoo Kimfe896d12016-03-17 14:19:26 -07001492 if (unlikely(page_ref_count(page) != 0))
Dave Hansenf0b791a2014-01-23 15:52:49 -08001493 bad_reason = "nonzero _count";
Naoya Horiguchif4c18e62015-08-06 15:47:08 -07001494 if (unlikely(page->flags & __PG_HWPOISON)) {
1495 bad_reason = "HWPoisoned (hardware-corrupted)";
1496 bad_flags = __PG_HWPOISON;
1497 }
Dave Hansenf0b791a2014-01-23 15:52:49 -08001498 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1499 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1500 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1501 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -08001502#ifdef CONFIG_MEMCG
1503 if (unlikely(page->mem_cgroup))
1504 bad_reason = "page still charged to cgroup";
1505#endif
Dave Hansenf0b791a2014-01-23 15:52:49 -08001506 if (unlikely(bad_reason)) {
1507 bad_page(page, bad_reason, bad_flags);
Hugh Dickins689bceb2005-11-21 21:32:20 -08001508 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -08001509 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001510 return 0;
1511}
1512
Laura Abbott1414c7f2016-03-15 14:56:30 -07001513static inline bool free_pages_prezeroed(bool poisoned)
1514{
1515 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1516 page_poisoning_enabled() && poisoned;
1517}
1518
Vlastimil Babka75379192015-02-11 15:25:38 -08001519static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1520 int alloc_flags)
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001521{
1522 int i;
Laura Abbott1414c7f2016-03-15 14:56:30 -07001523 bool poisoned = true;
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001524
1525 for (i = 0; i < (1 << order); i++) {
1526 struct page *p = page + i;
1527 if (unlikely(check_new_page(p)))
1528 return 1;
Laura Abbott1414c7f2016-03-15 14:56:30 -07001529 if (poisoned)
1530 poisoned &= page_is_poisoned(p);
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001531 }
Hugh Dickins689bceb2005-11-21 21:32:20 -08001532
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001533 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -08001534 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -08001535
1536 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 kernel_map_pages(page, 1 << order, 1);
Laura Abbott8823b1d2016-03-15 14:56:27 -07001538 kernel_poison_pages(page, 1 << order, 1);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -08001539 kasan_alloc_pages(page, order);
Nick Piggin17cf4402006-03-22 00:08:41 -08001540
Laura Abbott1414c7f2016-03-15 14:56:30 -07001541 if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO))
Anisse Astierf4d28972015-06-24 16:56:36 -07001542 for (i = 0; i < (1 << order); i++)
1543 clear_highpage(page + i);
Nick Piggin17cf4402006-03-22 00:08:41 -08001544
1545 if (order && (gfp_flags & __GFP_COMP))
1546 prep_compound_page(page, order);
1547
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001548 set_page_owner(page, order, gfp_flags);
1549
Vlastimil Babka75379192015-02-11 15:25:38 -08001550 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001551 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
Vlastimil Babka75379192015-02-11 15:25:38 -08001552 * allocate the page. The expectation is that the caller is taking
1553 * steps that will free more memory. The caller should avoid the page
1554 * being used for !PFMEMALLOC purposes.
1555 */
Michal Hocko2f064f32015-08-21 14:11:51 -07001556 if (alloc_flags & ALLOC_NO_WATERMARKS)
1557 set_page_pfmemalloc(page);
1558 else
1559 clear_page_pfmemalloc(page);
Vlastimil Babka75379192015-02-11 15:25:38 -08001560
Hugh Dickins689bceb2005-11-21 21:32:20 -08001561 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562}
1563
Mel Gorman56fd56b2007-10-16 01:25:58 -07001564/*
1565 * Go through the free lists for the given migratetype and remove
1566 * the smallest available page from the freelists
1567 */
Mel Gorman728ec982009-06-16 15:32:04 -07001568static inline
1569struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -07001570 int migratetype)
1571{
1572 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -07001573 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001574 struct page *page;
1575
1576 /* Find a page of the appropriate size in the preferred list */
1577 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1578 area = &(zone->free_area[current_order]);
Geliang Tanga16601c2016-01-14 15:20:30 -08001579 page = list_first_entry_or_null(&area->free_list[migratetype],
Mel Gorman56fd56b2007-10-16 01:25:58 -07001580 struct page, lru);
Geliang Tanga16601c2016-01-14 15:20:30 -08001581 if (!page)
1582 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001583 list_del(&page->lru);
1584 rmv_page_order(page);
1585 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001586 expand(zone, page, order, current_order, area, migratetype);
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001587 set_pcppage_migratetype(page, migratetype);
Mel Gorman56fd56b2007-10-16 01:25:58 -07001588 return page;
1589 }
1590
1591 return NULL;
1592}
1593
1594
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001595/*
1596 * This array describes the order lists are fallen back to when
1597 * the free lists for the desirable migrate type are depleted
1598 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001599static int fallbacks[MIGRATE_TYPES][4] = {
Mel Gorman974a7862015-11-06 16:28:34 -08001600 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1601 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1602 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
Joonsoo Kimdc676472015-04-14 15:45:15 -07001603#ifdef CONFIG_CMA
Mel Gorman974a7862015-11-06 16:28:34 -08001604 [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001605#endif
Minchan Kim194159f2013-02-22 16:33:58 -08001606#ifdef CONFIG_MEMORY_ISOLATION
Mel Gorman974a7862015-11-06 16:28:34 -08001607 [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001608#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001609};
1610
Joonsoo Kimdc676472015-04-14 15:45:15 -07001611#ifdef CONFIG_CMA
1612static struct page *__rmqueue_cma_fallback(struct zone *zone,
1613 unsigned int order)
1614{
1615 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1616}
1617#else
1618static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1619 unsigned int order) { return NULL; }
1620#endif
1621
Mel Gormanc361be52007-10-16 01:25:51 -07001622/*
1623 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -07001624 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -07001625 * boundary. If alignment is required, use move_freepages_block()
1626 */
Minchan Kim435b4052012-10-08 16:32:16 -07001627int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -07001628 struct page *start_page, struct page *end_page,
1629 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001630{
1631 struct page *page;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08001632 unsigned int order;
Mel Gormand1003132007-10-16 01:26:00 -07001633 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -07001634
1635#ifndef CONFIG_HOLES_IN_ZONE
1636 /*
1637 * page_zone is not safe to call in this context when
1638 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1639 * anyway as we check zone boundaries in move_freepages_block().
1640 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -07001641 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -07001642 */
Mel Gorman97ee4ba2014-10-09 15:28:28 -07001643 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
Mel Gormanc361be52007-10-16 01:25:51 -07001644#endif
1645
1646 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -07001647 /* Make sure we are not inadvertently changing nodes */
Sasha Levin309381fea2014-01-23 15:52:54 -08001648 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
Adam Litke344c7902008-09-02 14:35:38 -07001649
Mel Gormanc361be52007-10-16 01:25:51 -07001650 if (!pfn_valid_within(page_to_pfn(page))) {
1651 page++;
1652 continue;
1653 }
1654
1655 if (!PageBuddy(page)) {
1656 page++;
1657 continue;
1658 }
1659
1660 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001661 list_move(&page->lru,
1662 &zone->free_area[order].free_list[migratetype]);
Mel Gormanc361be52007-10-16 01:25:51 -07001663 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001664 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001665 }
1666
Mel Gormand1003132007-10-16 01:26:00 -07001667 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001668}
1669
Minchan Kimee6f5092012-07-31 16:43:50 -07001670int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001671 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001672{
1673 unsigned long start_pfn, end_pfn;
1674 struct page *start_page, *end_page;
1675
1676 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001677 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001678 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001679 end_page = start_page + pageblock_nr_pages - 1;
1680 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001681
1682 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001683 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001684 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001685 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001686 return 0;
1687
1688 return move_freepages(zone, start_page, end_page, migratetype);
1689}
1690
Mel Gorman2f66a682009-09-21 17:02:31 -07001691static void change_pageblock_range(struct page *pageblock_page,
1692 int start_order, int migratetype)
1693{
1694 int nr_pageblocks = 1 << (start_order - pageblock_order);
1695
1696 while (nr_pageblocks--) {
1697 set_pageblock_migratetype(pageblock_page, migratetype);
1698 pageblock_page += pageblock_nr_pages;
1699 }
1700}
1701
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001702/*
Vlastimil Babka9c0415e2015-02-11 15:28:21 -08001703 * When we are falling back to another migratetype during allocation, try to
1704 * steal extra free pages from the same pageblocks to satisfy further
1705 * allocations, instead of polluting multiple pageblocks.
1706 *
1707 * If we are stealing a relatively large buddy page, it is likely there will
1708 * be more free pages in the pageblock, so try to steal them all. For
1709 * reclaimable and unmovable allocations, we steal regardless of page size,
1710 * as fragmentation caused by those allocations polluting movable pageblocks
1711 * is worse than movable allocations stealing from unmovable and reclaimable
1712 * pageblocks.
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001713 */
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001714static bool can_steal_fallback(unsigned int order, int start_mt)
1715{
1716 /*
1717 * Leaving this order check is intended, although there is
1718 * relaxed order check in next check. The reason is that
1719 * we can actually steal whole pageblock if this condition met,
1720 * but, below check doesn't guarantee it and that is just heuristic
1721 * so could be changed anytime.
1722 */
1723 if (order >= pageblock_order)
1724 return true;
1725
1726 if (order >= pageblock_order / 2 ||
1727 start_mt == MIGRATE_RECLAIMABLE ||
1728 start_mt == MIGRATE_UNMOVABLE ||
1729 page_group_by_mobility_disabled)
1730 return true;
1731
1732 return false;
1733}
1734
1735/*
1736 * This function implements actual steal behaviour. If order is large enough,
1737 * we can steal whole pageblock. If not, we first move freepages in this
1738 * pageblock and check whether half of pages are moved or not. If half of
1739 * pages are moved, we can change migratetype of pageblock and permanently
1740 * use it's pages as requested migratetype in the future.
1741 */
1742static void steal_suitable_fallback(struct zone *zone, struct page *page,
1743 int start_type)
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001744{
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08001745 unsigned int current_order = page_order(page);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001746 int pages;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001747
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001748 /* Take ownership for orders >= pageblock_order */
1749 if (current_order >= pageblock_order) {
1750 change_pageblock_range(page, current_order, start_type);
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001751 return;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001752 }
1753
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001754 pages = move_freepages_block(zone, page, start_type);
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001755
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001756 /* Claim the whole block if over half of it is free */
1757 if (pages >= (1 << (pageblock_order-1)) ||
1758 page_group_by_mobility_disabled)
1759 set_pageblock_migratetype(page, start_type);
1760}
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001761
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001762/*
1763 * Check whether there is a suitable fallback freepage with requested order.
1764 * If only_stealable is true, this function returns fallback_mt only if
1765 * we can steal other freepages all together. This would help to reduce
1766 * fragmentation due to mixed migratetype pages in one pageblock.
1767 */
1768int find_suitable_fallback(struct free_area *area, unsigned int order,
1769 int migratetype, bool only_stealable, bool *can_steal)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001770{
1771 int i;
1772 int fallback_mt;
1773
1774 if (area->nr_free == 0)
1775 return -1;
1776
1777 *can_steal = false;
1778 for (i = 0;; i++) {
1779 fallback_mt = fallbacks[migratetype][i];
Mel Gorman974a7862015-11-06 16:28:34 -08001780 if (fallback_mt == MIGRATE_TYPES)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001781 break;
1782
1783 if (list_empty(&area->free_list[fallback_mt]))
1784 continue;
1785
1786 if (can_steal_fallback(order, migratetype))
1787 *can_steal = true;
1788
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001789 if (!only_stealable)
1790 return fallback_mt;
1791
1792 if (*can_steal)
1793 return fallback_mt;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001794 }
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001795
1796 return -1;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001797}
1798
Mel Gorman0aaa29a2015-11-06 16:28:37 -08001799/*
1800 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1801 * there are no empty page blocks that contain a page with a suitable order
1802 */
1803static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
1804 unsigned int alloc_order)
1805{
1806 int mt;
1807 unsigned long max_managed, flags;
1808
1809 /*
1810 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
1811 * Check is race-prone but harmless.
1812 */
1813 max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
1814 if (zone->nr_reserved_highatomic >= max_managed)
1815 return;
1816
1817 spin_lock_irqsave(&zone->lock, flags);
1818
1819 /* Recheck the nr_reserved_highatomic limit under the lock */
1820 if (zone->nr_reserved_highatomic >= max_managed)
1821 goto out_unlock;
1822
1823 /* Yoink! */
1824 mt = get_pageblock_migratetype(page);
1825 if (mt != MIGRATE_HIGHATOMIC &&
1826 !is_migrate_isolate(mt) && !is_migrate_cma(mt)) {
1827 zone->nr_reserved_highatomic += pageblock_nr_pages;
1828 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
1829 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC);
1830 }
1831
1832out_unlock:
1833 spin_unlock_irqrestore(&zone->lock, flags);
1834}
1835
1836/*
1837 * Used when an allocation is about to fail under memory pressure. This
1838 * potentially hurts the reliability of high-order allocations when under
1839 * intense memory pressure but failed atomic allocations should be easier
1840 * to recover from than an OOM.
1841 */
1842static void unreserve_highatomic_pageblock(const struct alloc_context *ac)
1843{
1844 struct zonelist *zonelist = ac->zonelist;
1845 unsigned long flags;
1846 struct zoneref *z;
1847 struct zone *zone;
1848 struct page *page;
1849 int order;
1850
1851 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
1852 ac->nodemask) {
1853 /* Preserve at least one pageblock */
1854 if (zone->nr_reserved_highatomic <= pageblock_nr_pages)
1855 continue;
1856
1857 spin_lock_irqsave(&zone->lock, flags);
1858 for (order = 0; order < MAX_ORDER; order++) {
1859 struct free_area *area = &(zone->free_area[order]);
1860
Geliang Tanga16601c2016-01-14 15:20:30 -08001861 page = list_first_entry_or_null(
1862 &area->free_list[MIGRATE_HIGHATOMIC],
1863 struct page, lru);
1864 if (!page)
Mel Gorman0aaa29a2015-11-06 16:28:37 -08001865 continue;
1866
Mel Gorman0aaa29a2015-11-06 16:28:37 -08001867 /*
1868 * It should never happen but changes to locking could
1869 * inadvertently allow a per-cpu drain to add pages
1870 * to MIGRATE_HIGHATOMIC while unreserving so be safe
1871 * and watch for underflows.
1872 */
1873 zone->nr_reserved_highatomic -= min(pageblock_nr_pages,
1874 zone->nr_reserved_highatomic);
1875
1876 /*
1877 * Convert to ac->migratetype and avoid the normal
1878 * pageblock stealing heuristics. Minimally, the caller
1879 * is doing the work and needs the pages. More
1880 * importantly, if the block was always converted to
1881 * MIGRATE_UNMOVABLE or another type then the number
1882 * of pageblocks that cannot be completely freed
1883 * may increase.
1884 */
1885 set_pageblock_migratetype(page, ac->migratetype);
1886 move_freepages_block(zone, page, ac->migratetype);
1887 spin_unlock_irqrestore(&zone->lock, flags);
1888 return;
1889 }
1890 spin_unlock_irqrestore(&zone->lock, flags);
1891 }
1892}
1893
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001894/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001895static inline struct page *
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001896__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001897{
Pintu Kumarb8af2942013-09-11 14:20:34 -07001898 struct free_area *area;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001899 unsigned int current_order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001900 struct page *page;
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001901 int fallback_mt;
1902 bool can_steal;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001903
1904 /* Find the largest possible block of pages in the other list */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001905 for (current_order = MAX_ORDER-1;
1906 current_order >= order && current_order <= MAX_ORDER-1;
1907 --current_order) {
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001908 area = &(zone->free_area[current_order]);
1909 fallback_mt = find_suitable_fallback(area, current_order,
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001910 start_migratetype, false, &can_steal);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001911 if (fallback_mt == -1)
1912 continue;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001913
Geliang Tanga16601c2016-01-14 15:20:30 -08001914 page = list_first_entry(&area->free_list[fallback_mt],
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001915 struct page, lru);
1916 if (can_steal)
1917 steal_suitable_fallback(zone, page, start_migratetype);
Mel Gormane0104872007-10-16 01:25:53 -07001918
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001919 /* Remove the page from the freelists */
1920 area->nr_free--;
1921 list_del(&page->lru);
1922 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001923
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001924 expand(zone, page, order, current_order, area,
1925 start_migratetype);
1926 /*
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001927 * The pcppage_migratetype may differ from pageblock's
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001928 * migratetype depending on the decisions in
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001929 * find_suitable_fallback(). This is OK as long as it does not
1930 * differ for MIGRATE_CMA pageblocks. Those can be used as
1931 * fallback only via special __rmqueue_cma_fallback() function
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001932 */
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001933 set_pcppage_migratetype(page, start_migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001934
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001935 trace_mm_page_alloc_extfrag(page, order, current_order,
1936 start_migratetype, fallback_mt);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001937
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001938 return page;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001939 }
1940
Mel Gorman728ec982009-06-16 15:32:04 -07001941 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001942}
1943
Mel Gorman56fd56b2007-10-16 01:25:58 -07001944/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 * Do the hard work of removing an element from the buddy allocator.
1946 * Call me with the zone->lock already held.
1947 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001948static struct page *__rmqueue(struct zone *zone, unsigned int order,
Mel Gorman6ac02062016-01-14 15:20:28 -08001949 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 struct page *page;
1952
Mel Gorman56fd56b2007-10-16 01:25:58 -07001953 page = __rmqueue_smallest(zone, order, migratetype);
Mel Gorman974a7862015-11-06 16:28:34 -08001954 if (unlikely(!page)) {
Joonsoo Kimdc676472015-04-14 15:45:15 -07001955 if (migratetype == MIGRATE_MOVABLE)
1956 page = __rmqueue_cma_fallback(zone, order);
1957
1958 if (!page)
1959 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gorman728ec982009-06-16 15:32:04 -07001960 }
1961
Mel Gorman0d3d0622009-09-21 17:02:44 -07001962 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001963 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964}
1965
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001966/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 * Obtain a specified number of elements from the buddy allocator, all under
1968 * a single hold of the lock, for efficiency. Add them to the supplied list.
1969 * Returns the number of new pages which were placed at *list.
1970 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001971static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001972 unsigned long count, struct list_head *list,
Mel Gormanb745bc82014-06-04 16:10:22 -07001973 int migratetype, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974{
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001975 int i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001976
Nick Pigginc54ad302006-01-06 00:10:56 -08001977 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 for (i = 0; i < count; ++i) {
Mel Gorman6ac02062016-01-14 15:20:28 -08001979 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001980 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001982
1983 /*
1984 * Split buddy pages returned by expand() are received here
1985 * in physical page order. The page is added to the callers and
1986 * list and the list head then moves forward. From the callers
1987 * perspective, the linked list is ordered by page number in
1988 * some conditions. This is useful for IO devices that can
1989 * merge IO requests if the physical pages are ordered
1990 * properly.
1991 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001992 if (likely(!cold))
Mel Gormane084b2d2009-07-29 15:02:04 -07001993 list_add(&page->lru, list);
1994 else
1995 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001996 list = &page->lru;
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001997 if (is_migrate_cma(get_pcppage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001998 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1999 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 }
Mel Gormanf2260e62009-06-16 15:32:13 -07002001 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08002002 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08002003 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004}
2005
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002006#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08002007/*
Christoph Lameter4037d452007-05-09 02:35:14 -07002008 * Called from the vmstat counter updater to drain pagesets of this
2009 * currently executing processor on remote nodes after they have
2010 * expired.
2011 *
Christoph Lameter879336c2006-03-22 00:09:08 -08002012 * Note that this function must be called with the thread pinned to
2013 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08002014 */
Christoph Lameter4037d452007-05-09 02:35:14 -07002015void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002016{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002017 unsigned long flags;
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07002018 int to_drain, batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002019
Christoph Lameter4037d452007-05-09 02:35:14 -07002020 local_irq_save(flags);
Jason Low4db0c3c2015-04-15 16:14:08 -07002021 batch = READ_ONCE(pcp->batch);
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07002022 to_drain = min(pcp->count, batch);
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07002023 if (to_drain > 0) {
2024 free_pcppages_bulk(zone, to_drain, pcp);
2025 pcp->count -= to_drain;
2026 }
Christoph Lameter4037d452007-05-09 02:35:14 -07002027 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002028}
2029#endif
2030
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002031/*
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002032 * Drain pcplists of the indicated processor and zone.
2033 *
2034 * The processor must either be the current processor and the
2035 * thread pinned to the current processor or a processor that
2036 * is not online.
2037 */
2038static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2039{
2040 unsigned long flags;
2041 struct per_cpu_pageset *pset;
2042 struct per_cpu_pages *pcp;
2043
2044 local_irq_save(flags);
2045 pset = per_cpu_ptr(zone->pageset, cpu);
2046
2047 pcp = &pset->pcp;
2048 if (pcp->count) {
2049 free_pcppages_bulk(zone, pcp->count, pcp);
2050 pcp->count = 0;
2051 }
2052 local_irq_restore(flags);
2053}
2054
2055/*
2056 * Drain pcplists of all zones on the indicated processor.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002057 *
2058 * The processor must either be the current processor and the
2059 * thread pinned to the current processor or a processor that
2060 * is not online.
2061 */
2062static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063{
2064 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002066 for_each_populated_zone(zone) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002067 drain_pages_zone(cpu, zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 }
2069}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002071/*
2072 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002073 *
2074 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2075 * the single zone's pages.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002076 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002077void drain_local_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002078{
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002079 int cpu = smp_processor_id();
2080
2081 if (zone)
2082 drain_pages_zone(cpu, zone);
2083 else
2084 drain_pages(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002085}
2086
2087/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002088 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2089 *
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002090 * When zone parameter is non-NULL, spill just the single zone's pages.
2091 *
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002092 * Note that this code is protected against sending an IPI to an offline
2093 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2094 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2095 * nothing keeps CPUs from showing up after we populated the cpumask and
2096 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002097 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002098void drain_all_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002099{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002100 int cpu;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002101
2102 /*
2103 * Allocate in the BSS so we wont require allocation in
2104 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2105 */
2106 static cpumask_t cpus_with_pcps;
2107
2108 /*
2109 * We don't care about racing with CPU hotplug event
2110 * as offline notification will cause the notified
2111 * cpu to drain that CPU pcps and on_each_cpu_mask
2112 * disables preemption as part of its processing
2113 */
2114 for_each_online_cpu(cpu) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002115 struct per_cpu_pageset *pcp;
2116 struct zone *z;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002117 bool has_pcps = false;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002118
2119 if (zone) {
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002120 pcp = per_cpu_ptr(zone->pageset, cpu);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002121 if (pcp->pcp.count)
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002122 has_pcps = true;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002123 } else {
2124 for_each_populated_zone(z) {
2125 pcp = per_cpu_ptr(z->pageset, cpu);
2126 if (pcp->pcp.count) {
2127 has_pcps = true;
2128 break;
2129 }
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002130 }
2131 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002132
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002133 if (has_pcps)
2134 cpumask_set_cpu(cpu, &cpus_with_pcps);
2135 else
2136 cpumask_clear_cpu(cpu, &cpus_with_pcps);
2137 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002138 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
2139 zone, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002140}
2141
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02002142#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143
2144void mark_free_pages(struct zone *zone)
2145{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002146 unsigned long pfn, max_zone_pfn;
2147 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002148 unsigned int order, t;
Geliang Tang86760a22016-01-14 15:20:33 -08002149 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150
Xishi Qiu8080fc02013-09-11 14:21:45 -07002151 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 return;
2153
2154 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002155
Cody P Schafer108bcc92013-02-22 16:35:23 -08002156 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002157 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2158 if (pfn_valid(pfn)) {
Geliang Tang86760a22016-01-14 15:20:33 -08002159 page = pfn_to_page(pfn);
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07002160 if (!swsusp_page_is_forbidden(page))
2161 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002162 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002164 for_each_migratetype_order(order, t) {
Geliang Tang86760a22016-01-14 15:20:33 -08002165 list_for_each_entry(page,
2166 &zone->free_area[order].free_list[t], lru) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002167 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168
Geliang Tang86760a22016-01-14 15:20:33 -08002169 pfn = page_to_pfn(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002170 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07002171 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002172 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002173 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 spin_unlock_irqrestore(&zone->lock, flags);
2175}
Mel Gormane2c55dc2007-10-16 01:25:50 -07002176#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177
2178/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 * Free a 0-order page
Mel Gormanb745bc82014-06-04 16:10:22 -07002180 * cold == true ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002182void free_hot_cold_page(struct page *page, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183{
2184 struct zone *zone = page_zone(page);
2185 struct per_cpu_pages *pcp;
2186 unsigned long flags;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002187 unsigned long pfn = page_to_pfn(page);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002188 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07002190 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08002191 return;
2192
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002193 migratetype = get_pfnblock_migratetype(page, pfn);
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002194 set_pcppage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07002196 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07002197
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002198 /*
2199 * We only track unmovable, reclaimable and movable on pcp lists.
2200 * Free ISOLATE pages back to the allocator because they are being
2201 * offlined but treat RESERVE as movable pages so we can get those
2202 * areas back if necessary. Otherwise, we may have to free
2203 * excessively into the page allocator
2204 */
2205 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08002206 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002207 free_one_page(zone, page, pfn, 0, migratetype);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002208 goto out;
2209 }
2210 migratetype = MIGRATE_MOVABLE;
2211 }
2212
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002213 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Mel Gormanb745bc82014-06-04 16:10:22 -07002214 if (!cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002215 list_add(&page->lru, &pcp->lists[migratetype]);
Mel Gormanb745bc82014-06-04 16:10:22 -07002216 else
2217 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08002219 if (pcp->count >= pcp->high) {
Jason Low4db0c3c2015-04-15 16:14:08 -07002220 unsigned long batch = READ_ONCE(pcp->batch);
Cody P Schafer998d39c2013-07-03 15:01:32 -07002221 free_pcppages_bulk(zone, batch, pcp);
2222 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08002223 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002224
2225out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227}
2228
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002229/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002230 * Free a list of 0-order pages
2231 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002232void free_hot_cold_page_list(struct list_head *list, bool cold)
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002233{
2234 struct page *page, *next;
2235
2236 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08002237 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002238 free_hot_cold_page(page, cold);
2239 }
2240}
2241
2242/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002243 * split_page takes a non-compound higher-order page, and splits it into
2244 * n (1<<order) sub-pages: page[0..n]
2245 * Each sub-page must be freed individually.
2246 *
2247 * Note: this is probably too low level an operation for use in drivers.
2248 * Please consult with lkml before using this in your driver.
2249 */
2250void split_page(struct page *page, unsigned int order)
2251{
2252 int i;
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002253 gfp_t gfp_mask;
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002254
Sasha Levin309381fea2014-01-23 15:52:54 -08002255 VM_BUG_ON_PAGE(PageCompound(page), page);
2256 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002257
2258#ifdef CONFIG_KMEMCHECK
2259 /*
2260 * Split shadow pages too, because free(page[0]) would
2261 * otherwise free the whole shadow.
2262 */
2263 if (kmemcheck_page_is_tracked(page))
2264 split_page(virt_to_page(page[0].shadow), order);
2265#endif
2266
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002267 gfp_mask = get_page_owner_gfp(page);
2268 set_page_owner(page, 0, gfp_mask);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08002269 for (i = 1; i < (1 << order); i++) {
Nick Piggin7835e982006-03-22 00:08:40 -08002270 set_page_refcounted(page + i);
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002271 set_page_owner(page + i, 0, gfp_mask);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08002272 }
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002273}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07002274EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002275
Joonsoo Kim3c605092014-11-13 15:19:21 -08002276int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07002277{
Mel Gorman748446b2010-05-24 14:32:27 -07002278 unsigned long watermark;
2279 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002280 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07002281
2282 BUG_ON(!PageBuddy(page));
2283
2284 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002285 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07002286
Minchan Kim194159f2013-02-22 16:33:58 -08002287 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002288 /* Obey watermarks as if the page was being allocated */
2289 watermark = low_wmark_pages(zone) + (1 << order);
2290 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2291 return 0;
2292
Mel Gorman8fb74b92013-01-11 14:32:16 -08002293 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002294 }
Mel Gorman748446b2010-05-24 14:32:27 -07002295
2296 /* Remove page from free list */
2297 list_del(&page->lru);
2298 zone->free_area[order].nr_free--;
2299 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002300
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002301 set_page_owner(page, order, __GFP_MOVABLE);
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002302
Mel Gorman8fb74b92013-01-11 14:32:16 -08002303 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07002304 if (order >= pageblock_order - 1) {
2305 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002306 for (; page < endpage; page += pageblock_nr_pages) {
2307 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08002308 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002309 set_pageblock_migratetype(page,
2310 MIGRATE_MOVABLE);
2311 }
Mel Gorman748446b2010-05-24 14:32:27 -07002312 }
2313
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002314
Mel Gorman8fb74b92013-01-11 14:32:16 -08002315 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002316}
2317
2318/*
2319 * Similar to split_page except the page is already free. As this is only
2320 * being used for migration, the migratetype of the block also changes.
2321 * As this is called with interrupts disabled, the caller is responsible
2322 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2323 * are enabled.
2324 *
2325 * Note: this is probably too low level an operation for use in drivers.
2326 * Please consult with lkml before using this in your driver.
2327 */
2328int split_free_page(struct page *page)
2329{
2330 unsigned int order;
2331 int nr_pages;
2332
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002333 order = page_order(page);
2334
Mel Gorman8fb74b92013-01-11 14:32:16 -08002335 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002336 if (!nr_pages)
2337 return 0;
2338
2339 /* Split into individual pages */
2340 set_page_refcounted(page);
2341 split_page(page, order);
2342 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07002343}
2344
2345/*
Vlastimil Babka75379192015-02-11 15:25:38 -08002346 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07002348static inline
2349struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002350 struct zone *zone, unsigned int order,
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002351 gfp_t gfp_flags, int alloc_flags, int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352{
2353 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08002354 struct page *page;
Mel Gormanb745bc82014-06-04 16:10:22 -07002355 bool cold = ((gfp_flags & __GFP_COLD) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356
Nick Piggin48db57f2006-01-08 01:00:42 -08002357 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002359 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002362 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2363 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002364 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07002365 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002366 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07002367 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002368 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07002369 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07002370 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07002371
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002372 if (cold)
Geliang Tanga16601c2016-01-14 15:20:30 -08002373 page = list_last_entry(list, struct page, lru);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002374 else
Geliang Tanga16601c2016-01-14 15:20:30 -08002375 page = list_first_entry(list, struct page, lru);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002376
Mel Gormanb92a6ed2007-10-16 01:25:50 -07002377 list_del(&page->lru);
2378 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002379 } else {
Michal Hocko0f352e52016-03-17 14:19:32 -07002380 /*
2381 * We most definitely don't want callers attempting to
2382 * allocate greater than order-1 page units with __GFP_NOFAIL.
2383 */
2384 WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 spin_lock_irqsave(&zone->lock, flags);
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002386
2387 page = NULL;
2388 if (alloc_flags & ALLOC_HARDER) {
2389 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
2390 if (page)
2391 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2392 }
2393 if (!page)
Mel Gorman6ac02062016-01-14 15:20:28 -08002394 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08002395 spin_unlock(&zone->lock);
2396 if (!page)
2397 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07002398 __mod_zone_freepage_state(zone, -(1 << order),
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002399 get_pcppage_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 }
2401
Johannes Weiner3a025762014-04-07 15:37:48 -07002402 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Johannes Weinerabe5f972014-10-02 16:21:10 -07002403 if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
Johannes Weiner57054652014-10-09 15:28:17 -07002404 !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
2405 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Johannes Weiner27329362014-03-03 15:38:41 -08002406
Christoph Lameterf8891e52006-06-30 01:55:45 -07002407 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07002408 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002409 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410
Sasha Levin309381fea2014-01-23 15:52:54 -08002411 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08002413
2414failed:
2415 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002416 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417}
2418
Akinobu Mita933e3122006-12-08 02:39:45 -08002419#ifdef CONFIG_FAIL_PAGE_ALLOC
2420
Akinobu Mitab2588c42011-07-26 16:09:03 -07002421static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08002422 struct fault_attr attr;
2423
Viresh Kumar621a5f72015-09-26 15:04:07 -07002424 bool ignore_gfp_highmem;
Mel Gorman71baba42015-11-06 16:28:28 -08002425 bool ignore_gfp_reclaim;
Akinobu Mita54114992007-07-15 23:40:23 -07002426 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08002427} fail_page_alloc = {
2428 .attr = FAULT_ATTR_INITIALIZER,
Mel Gorman71baba42015-11-06 16:28:28 -08002429 .ignore_gfp_reclaim = true,
Viresh Kumar621a5f72015-09-26 15:04:07 -07002430 .ignore_gfp_highmem = true,
Akinobu Mita54114992007-07-15 23:40:23 -07002431 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08002432};
2433
2434static int __init setup_fail_page_alloc(char *str)
2435{
2436 return setup_fault_attr(&fail_page_alloc.attr, str);
2437}
2438__setup("fail_page_alloc=", setup_fail_page_alloc);
2439
Gavin Shandeaf3862012-07-31 16:41:51 -07002440static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002441{
Akinobu Mita54114992007-07-15 23:40:23 -07002442 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07002443 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002444 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07002445 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002446 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002447 return false;
Mel Gorman71baba42015-11-06 16:28:28 -08002448 if (fail_page_alloc.ignore_gfp_reclaim &&
2449 (gfp_mask & __GFP_DIRECT_RECLAIM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002450 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002451
2452 return should_fail(&fail_page_alloc.attr, 1 << order);
2453}
2454
2455#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2456
2457static int __init fail_page_alloc_debugfs(void)
2458{
Al Virof4ae40a2011-07-24 04:33:43 -04002459 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08002460 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08002461
Akinobu Mitadd48c082011-08-03 16:21:01 -07002462 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2463 &fail_page_alloc.attr);
2464 if (IS_ERR(dir))
2465 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002466
Akinobu Mitab2588c42011-07-26 16:09:03 -07002467 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
Mel Gorman71baba42015-11-06 16:28:28 -08002468 &fail_page_alloc.ignore_gfp_reclaim))
Akinobu Mitab2588c42011-07-26 16:09:03 -07002469 goto fail;
2470 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2471 &fail_page_alloc.ignore_gfp_highmem))
2472 goto fail;
2473 if (!debugfs_create_u32("min-order", mode, dir,
2474 &fail_page_alloc.min_order))
2475 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08002476
Akinobu Mitab2588c42011-07-26 16:09:03 -07002477 return 0;
2478fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07002479 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002480
Akinobu Mitab2588c42011-07-26 16:09:03 -07002481 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08002482}
2483
2484late_initcall(fail_page_alloc_debugfs);
2485
2486#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2487
2488#else /* CONFIG_FAIL_PAGE_ALLOC */
2489
Gavin Shandeaf3862012-07-31 16:41:51 -07002490static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002491{
Gavin Shandeaf3862012-07-31 16:41:51 -07002492 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002493}
2494
2495#endif /* CONFIG_FAIL_PAGE_ALLOC */
2496
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497/*
Mel Gorman97a16fc2015-11-06 16:28:40 -08002498 * Return true if free base pages are above 'mark'. For high-order checks it
2499 * will return true of the order-0 watermark is reached and there is at least
2500 * one free page of a suitable size. Checking now avoids taking the zone lock
2501 * to check in the allocation paths if no pages are free.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002503static bool __zone_watermark_ok(struct zone *z, unsigned int order,
2504 unsigned long mark, int classzone_idx, int alloc_flags,
2505 long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506{
Christoph Lameterd23ad422007-02-10 01:43:02 -08002507 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 int o;
Mel Gorman97a16fc2015-11-06 16:28:40 -08002509 const int alloc_harder = (alloc_flags & ALLOC_HARDER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002511 /* free_pages may go negative - that's OK */
Michal Hockodf0a6da2012-01-10 15:08:02 -08002512 free_pages -= (1 << order) - 1;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002513
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002514 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 min -= min / 2;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002516
2517 /*
2518 * If the caller does not have rights to ALLOC_HARDER then subtract
2519 * the high-atomic reserves. This will over-estimate the size of the
2520 * atomic reserve but it avoids a search.
2521 */
Mel Gorman97a16fc2015-11-06 16:28:40 -08002522 if (likely(!alloc_harder))
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002523 free_pages -= z->nr_reserved_highatomic;
2524 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 min -= min / 4;
Mel Gormane2b19192015-11-06 16:28:09 -08002526
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002527#ifdef CONFIG_CMA
2528 /* If allocation can't use CMA areas don't use free CMA pages */
2529 if (!(alloc_flags & ALLOC_CMA))
Mel Gorman97a16fc2015-11-06 16:28:40 -08002530 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002531#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002532
Mel Gorman97a16fc2015-11-06 16:28:40 -08002533 /*
2534 * Check watermarks for an order-0 allocation request. If these
2535 * are not met, then a high-order request also cannot go ahead
2536 * even if a suitable page happened to be free.
2537 */
2538 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08002539 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540
Mel Gorman97a16fc2015-11-06 16:28:40 -08002541 /* If this is an order-0 request then the watermark is fine */
2542 if (!order)
2543 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544
Mel Gorman97a16fc2015-11-06 16:28:40 -08002545 /* For a high-order request, check at least one suitable page is free */
2546 for (o = order; o < MAX_ORDER; o++) {
2547 struct free_area *area = &z->free_area[o];
2548 int mt;
2549
2550 if (!area->nr_free)
2551 continue;
2552
2553 if (alloc_harder)
2554 return true;
2555
2556 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
2557 if (!list_empty(&area->free_list[mt]))
2558 return true;
2559 }
2560
2561#ifdef CONFIG_CMA
2562 if ((alloc_flags & ALLOC_CMA) &&
2563 !list_empty(&area->free_list[MIGRATE_CMA])) {
2564 return true;
2565 }
2566#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 }
Mel Gorman97a16fc2015-11-06 16:28:40 -08002568 return false;
Mel Gorman88f5acf2011-01-13 15:45:41 -08002569}
2570
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002571bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002572 int classzone_idx, int alloc_flags)
2573{
2574 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2575 zone_page_state(z, NR_FREE_PAGES));
2576}
2577
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002578bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
Mel Gormane2b19192015-11-06 16:28:09 -08002579 unsigned long mark, int classzone_idx)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002580{
2581 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2582
2583 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2584 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2585
Mel Gormane2b19192015-11-06 16:28:09 -08002586 return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002587 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588}
2589
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002590#ifdef CONFIG_NUMA
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002591static bool zone_local(struct zone *local_zone, struct zone *zone)
2592{
Johannes Weinerfff40682013-12-20 14:54:12 +00002593 return local_zone->node == zone->node;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002594}
2595
David Rientjes957f8222012-10-08 16:33:24 -07002596static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2597{
Mel Gorman5f7a75a2014-06-04 16:07:15 -07002598 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2599 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07002600}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002601#else /* CONFIG_NUMA */
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002602static bool zone_local(struct zone *local_zone, struct zone *zone)
2603{
2604 return true;
2605}
2606
David Rientjes957f8222012-10-08 16:33:24 -07002607static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2608{
2609 return true;
2610}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002611#endif /* CONFIG_NUMA */
2612
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002613static void reset_alloc_batches(struct zone *preferred_zone)
2614{
2615 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2616
2617 do {
2618 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2619 high_wmark_pages(zone) - low_wmark_pages(zone) -
2620 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner57054652014-10-09 15:28:17 -07002621 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002622 } while (zone++ != preferred_zone);
2623}
2624
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002625/*
Paul Jackson0798e512006-12-06 20:31:38 -08002626 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002627 * a page.
2628 */
2629static struct page *
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002630get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2631 const struct alloc_context *ac)
Martin Hicks753ee722005-06-21 17:14:41 -07002632{
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002633 struct zonelist *zonelist = ac->zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07002634 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002635 struct page *page = NULL;
Mel Gorman5117f452009-06-16 15:31:59 -07002636 struct zone *zone;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002637 int nr_fair_skipped = 0;
2638 bool zonelist_rescan;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002639
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002640zonelist_scan:
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002641 zonelist_rescan = false;
2642
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002643 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002644 * Scan zonelist, looking for a zone with enough free.
Vladimir Davydov344736f2014-10-20 15:50:30 +04002645 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002646 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002647 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2648 ac->nodemask) {
Johannes Weinere085dbc2013-09-11 14:20:46 -07002649 unsigned long mark;
2650
Mel Gorman664eedd2014-06-04 16:10:08 -07002651 if (cpusets_enabled() &&
2652 (alloc_flags & ALLOC_CPUSET) &&
Vladimir Davydov344736f2014-10-20 15:50:30 +04002653 !cpuset_zone_allowed(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07002654 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08002655 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002656 * Distribute pages in proportion to the individual
2657 * zone size to ensure fair page aging. The zone a
2658 * page was allocated in should have no effect on the
2659 * time the page has in memory before being reclaimed.
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002660 */
Johannes Weiner3a025762014-04-07 15:37:48 -07002661 if (alloc_flags & ALLOC_FAIR) {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002662 if (!zone_local(ac->preferred_zone, zone))
Mel Gormanf7b5d642014-08-06 16:07:20 -07002663 break;
Johannes Weiner57054652014-10-09 15:28:17 -07002664 if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002665 nr_fair_skipped++;
Johannes Weiner3a025762014-04-07 15:37:48 -07002666 continue;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002667 }
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002668 }
2669 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08002670 * When allocating a page cache page for writing, we
2671 * want to get it from a zone that is within its dirty
2672 * limit, such that no single zone holds more than its
2673 * proportional share of globally allowed dirty pages.
2674 * The dirty limits take into account the zone's
2675 * lowmem reserves and high watermark so that kswapd
2676 * should be able to balance it without having to
2677 * write pages from its LRU list.
2678 *
2679 * This may look like it could increase pressure on
2680 * lower zones by failing allocations in higher zones
2681 * before they are full. But the pages that do spill
2682 * over are limited as the lower zones are protected
2683 * by this very same mechanism. It should not become
2684 * a practical burden to them.
2685 *
2686 * XXX: For now, allow allocations to potentially
2687 * exceed the per-zone dirty limit in the slowpath
Mel Gormanc9ab0c42015-11-06 16:28:12 -08002688 * (spread_dirty_pages unset) before going into reclaim,
Johannes Weinera756cf52012-01-10 15:07:49 -08002689 * which is important when on a NUMA setup the allowed
2690 * zones are together not big enough to reach the
2691 * global limit. The proper fix for these situations
2692 * will require awareness of zones in the
2693 * dirty-throttling and the flusher threads.
2694 */
Mel Gormanc9ab0c42015-11-06 16:28:12 -08002695 if (ac->spread_dirty_pages && !zone_dirty_ok(zone))
Mel Gorman800a1e72014-06-04 16:10:06 -07002696 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002697
Johannes Weinere085dbc2013-09-11 14:20:46 -07002698 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2699 if (!zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002700 ac->classzone_idx, alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07002701 int ret;
2702
Mel Gorman5dab2912014-06-04 16:10:14 -07002703 /* Checked here to keep the fast path fast */
2704 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2705 if (alloc_flags & ALLOC_NO_WATERMARKS)
2706 goto try_this_zone;
2707
David Rientjes957f8222012-10-08 16:33:24 -07002708 if (zone_reclaim_mode == 0 ||
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002709 !zone_allows_reclaim(ac->preferred_zone, zone))
Mel Gormancd38b112011-07-25 17:12:29 -07002710 continue;
2711
Mel Gormanfa5e0842009-06-16 15:33:22 -07002712 ret = zone_reclaim(zone, gfp_mask, order);
2713 switch (ret) {
2714 case ZONE_RECLAIM_NOSCAN:
2715 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07002716 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002717 case ZONE_RECLAIM_FULL:
2718 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002719 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002720 default:
2721 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002722 if (zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002723 ac->classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002724 goto try_this_zone;
2725
Mel Gormanfed27192013-04-29 15:07:57 -07002726 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002727 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002728 }
2729
Mel Gormanfa5e0842009-06-16 15:33:22 -07002730try_this_zone:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002731 page = buffered_rmqueue(ac->preferred_zone, zone, order,
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002732 gfp_mask, alloc_flags, ac->migratetype);
Vlastimil Babka75379192015-02-11 15:25:38 -08002733 if (page) {
2734 if (prep_new_page(page, order, gfp_mask, alloc_flags))
2735 goto try_this_zone;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002736
2737 /*
2738 * If this is a high-order atomic allocation then check
2739 * if the pageblock should be reserved for the future
2740 */
2741 if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
2742 reserve_highatomic_pageblock(page, zone, order);
2743
Vlastimil Babka75379192015-02-11 15:25:38 -08002744 return page;
2745 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002746 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002747
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002748 /*
2749 * The first pass makes sure allocations are spread fairly within the
2750 * local node. However, the local node might have free pages left
2751 * after the fairness batches are exhausted, and remote zones haven't
2752 * even been considered yet. Try once more without fairness, and
2753 * include remote zones now, before entering the slowpath and waking
2754 * kswapd: prefer spilling to a remote zone over swapping locally.
2755 */
2756 if (alloc_flags & ALLOC_FAIR) {
2757 alloc_flags &= ~ALLOC_FAIR;
2758 if (nr_fair_skipped) {
2759 zonelist_rescan = true;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002760 reset_alloc_batches(ac->preferred_zone);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002761 }
2762 if (nr_online_nodes > 1)
2763 zonelist_rescan = true;
2764 }
2765
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002766 if (zonelist_rescan)
2767 goto zonelist_scan;
2768
2769 return NULL;
Martin Hicks753ee722005-06-21 17:14:41 -07002770}
2771
David Rientjes29423e772011-03-22 16:30:47 -07002772/*
2773 * Large machines with many possible nodes should not always dump per-node
2774 * meminfo in irq context.
2775 */
2776static inline bool should_suppress_show_mem(void)
2777{
2778 bool ret = false;
2779
2780#if NODES_SHIFT > 8
2781 ret = in_interrupt();
2782#endif
2783 return ret;
2784}
2785
Dave Hansena238ab52011-05-24 17:12:16 -07002786static DEFINE_RATELIMIT_STATE(nopage_rs,
2787 DEFAULT_RATELIMIT_INTERVAL,
2788 DEFAULT_RATELIMIT_BURST);
2789
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08002790void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...)
Dave Hansena238ab52011-05-24 17:12:16 -07002791{
Dave Hansena238ab52011-05-24 17:12:16 -07002792 unsigned int filter = SHOW_MEM_FILTER_NODES;
2793
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08002794 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2795 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002796 return;
2797
2798 /*
2799 * This documents exceptions given to allocations in certain
2800 * contexts that are allowed to allocate outside current's set
2801 * of allowed nodes.
2802 */
2803 if (!(gfp_mask & __GFP_NOMEMALLOC))
2804 if (test_thread_flag(TIF_MEMDIE) ||
2805 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2806 filter &= ~SHOW_MEM_FILTER_NODES;
Mel Gormand0164ad2015-11-06 16:28:21 -08002807 if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
Dave Hansena238ab52011-05-24 17:12:16 -07002808 filter &= ~SHOW_MEM_FILTER_NODES;
2809
2810 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002811 struct va_format vaf;
2812 va_list args;
2813
Dave Hansena238ab52011-05-24 17:12:16 -07002814 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002815
2816 vaf.fmt = fmt;
2817 vaf.va = &args;
2818
2819 pr_warn("%pV", &vaf);
2820
Dave Hansena238ab52011-05-24 17:12:16 -07002821 va_end(args);
2822 }
2823
Vlastimil Babkac5c990e2016-03-15 14:56:02 -07002824 pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n",
2825 current->comm, order, gfp_mask, &gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002826 dump_stack();
2827 if (!should_suppress_show_mem())
2828 show_mem(filter);
2829}
2830
Mel Gorman11e33f62009-06-16 15:31:57 -07002831static inline struct page *
2832__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002833 const struct alloc_context *ac, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07002834{
David Rientjes6e0fc462015-09-08 15:00:36 -07002835 struct oom_control oc = {
2836 .zonelist = ac->zonelist,
2837 .nodemask = ac->nodemask,
2838 .gfp_mask = gfp_mask,
2839 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07002840 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842
Johannes Weiner9879de72015-01-26 12:58:32 -08002843 *did_some_progress = 0;
2844
Johannes Weiner9879de72015-01-26 12:58:32 -08002845 /*
Johannes Weinerdc564012015-06-24 16:57:19 -07002846 * Acquire the oom lock. If that fails, somebody else is
2847 * making progress for us.
Johannes Weiner9879de72015-01-26 12:58:32 -08002848 */
Johannes Weinerdc564012015-06-24 16:57:19 -07002849 if (!mutex_trylock(&oom_lock)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002850 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002851 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 return NULL;
2853 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002854
Mel Gorman11e33f62009-06-16 15:31:57 -07002855 /*
2856 * Go through the zonelist yet one more time, keep very high watermark
2857 * here, this is only to catch a parallel oom killing, we must fail if
2858 * we're still under heavy pressure.
2859 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002860 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2861 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002862 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002863 goto out;
2864
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002865 if (!(gfp_mask & __GFP_NOFAIL)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002866 /* Coredumps can quickly deplete all memory reserves */
2867 if (current->flags & PF_DUMPCORE)
2868 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002869 /* The OOM killer will not help higher order allocs */
2870 if (order > PAGE_ALLOC_COSTLY_ORDER)
2871 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002872 /* The OOM killer does not needlessly kill tasks for lowmem */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002873 if (ac->high_zoneidx < ZONE_NORMAL)
David Rientjes03668b32010-08-09 17:18:54 -07002874 goto out;
Johannes Weiner90839052015-06-24 16:57:21 -07002875 /* The OOM killer does not compensate for IO-less reclaim */
Johannes Weinercc873172015-02-27 15:52:09 -08002876 if (!(gfp_mask & __GFP_FS)) {
2877 /*
2878 * XXX: Page reclaim didn't yield anything,
2879 * and the OOM killer can't be invoked, but
Johannes Weiner90839052015-06-24 16:57:21 -07002880 * keep looping as per tradition.
Tetsuo Handa0a687aa2016-03-17 14:20:48 -07002881 *
2882 * But do not keep looping if oom_killer_disable()
2883 * was already called, for the system is trying to
2884 * enter a quiescent state during suspend.
Johannes Weinercc873172015-02-27 15:52:09 -08002885 */
Tetsuo Handa0a687aa2016-03-17 14:20:48 -07002886 *did_some_progress = !oom_killer_disabled;
Johannes Weiner9879de72015-01-26 12:58:32 -08002887 goto out;
Johannes Weinercc873172015-02-27 15:52:09 -08002888 }
Johannes Weiner90839052015-06-24 16:57:21 -07002889 if (pm_suspended_storage())
2890 goto out;
David Rientjes4167e9b2015-04-14 15:46:55 -07002891 /* The OOM killer may not free memory on a specific node */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002892 if (gfp_mask & __GFP_THISNODE)
2893 goto out;
2894 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002895 /* Exhausted what can be done so it's blamo time */
Michal Hocko5020e282016-01-14 15:20:36 -08002896 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
Michal Hockoc32b3cb2015-02-11 15:26:24 -08002897 *did_some_progress = 1;
Michal Hocko5020e282016-01-14 15:20:36 -08002898
2899 if (gfp_mask & __GFP_NOFAIL) {
2900 page = get_page_from_freelist(gfp_mask, order,
2901 ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac);
2902 /*
2903 * fallback to ignore cpuset restriction if our nodes
2904 * are depleted
2905 */
2906 if (!page)
2907 page = get_page_from_freelist(gfp_mask, order,
2908 ALLOC_NO_WATERMARKS, ac);
2909 }
2910 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002911out:
Johannes Weinerdc564012015-06-24 16:57:19 -07002912 mutex_unlock(&oom_lock);
Mel Gorman11e33f62009-06-16 15:31:57 -07002913 return page;
2914}
2915
Mel Gorman56de7262010-05-24 14:32:30 -07002916#ifdef CONFIG_COMPACTION
2917/* Try memory compaction for high-order allocations before reclaim */
2918static struct page *
2919__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002920 int alloc_flags, const struct alloc_context *ac,
2921 enum migrate_mode mode, int *contended_compaction,
2922 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002923{
Vlastimil Babka53853e22014-10-09 15:27:02 -07002924 unsigned long compact_result;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002925 struct page *page;
Vlastimil Babka53853e22014-10-09 15:27:02 -07002926
Mel Gorman66199712012-01-12 17:19:41 -08002927 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002928 return NULL;
2929
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002930 current->flags |= PF_MEMALLOC;
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08002931 compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2932 mode, contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002933 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002934
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002935 switch (compact_result) {
2936 case COMPACT_DEFERRED:
Vlastimil Babka53853e22014-10-09 15:27:02 -07002937 *deferred_compaction = true;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002938 /* fall-through */
2939 case COMPACT_SKIPPED:
2940 return NULL;
2941 default:
2942 break;
Mel Gorman56de7262010-05-24 14:32:30 -07002943 }
2944
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002945 /*
2946 * At least in one zone compaction wasn't deferred or skipped, so let's
2947 * count a compaction stall
2948 */
2949 count_vm_event(COMPACTSTALL);
2950
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002951 page = get_page_from_freelist(gfp_mask, order,
2952 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002953
2954 if (page) {
2955 struct zone *zone = page_zone(page);
2956
2957 zone->compact_blockskip_flush = false;
2958 compaction_defer_reset(zone, order, true);
2959 count_vm_event(COMPACTSUCCESS);
2960 return page;
2961 }
2962
2963 /*
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002964 * It's bad if compaction run occurs and fails. The most likely reason
2965 * is that pages exist, but not enough to satisfy watermarks.
2966 */
2967 count_vm_event(COMPACTFAIL);
2968
2969 cond_resched();
2970
Mel Gorman56de7262010-05-24 14:32:30 -07002971 return NULL;
2972}
2973#else
2974static inline struct page *
2975__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002976 int alloc_flags, const struct alloc_context *ac,
2977 enum migrate_mode mode, int *contended_compaction,
2978 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002979{
2980 return NULL;
2981}
2982#endif /* CONFIG_COMPACTION */
2983
Marek Szyprowskibba90712012-01-25 12:09:52 +01002984/* Perform direct synchronous page reclaim */
2985static int
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002986__perform_reclaim(gfp_t gfp_mask, unsigned int order,
2987 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002988{
Mel Gorman11e33f62009-06-16 15:31:57 -07002989 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002990 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002991
2992 cond_resched();
2993
2994 /* We now go into synchronous reclaim */
2995 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002996 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002997 lockdep_set_current_reclaim_state(gfp_mask);
2998 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002999 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07003000
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003001 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
3002 ac->nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07003003
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003004 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07003005 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003006 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07003007
3008 cond_resched();
3009
Marek Szyprowskibba90712012-01-25 12:09:52 +01003010 return progress;
3011}
3012
3013/* The really slow allocator path where we enter direct reclaim */
3014static inline struct page *
3015__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003016 int alloc_flags, const struct alloc_context *ac,
3017 unsigned long *did_some_progress)
Marek Szyprowskibba90712012-01-25 12:09:52 +01003018{
3019 struct page *page = NULL;
3020 bool drained = false;
3021
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003022 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003023 if (unlikely(!(*did_some_progress)))
3024 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07003025
Mel Gorman9ee493c2010-09-09 16:38:18 -07003026retry:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003027 page = get_page_from_freelist(gfp_mask, order,
3028 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003029
3030 /*
3031 * If an allocation failed after direct reclaim, it could be because
Mel Gorman0aaa29a2015-11-06 16:28:37 -08003032 * pages are pinned on the per-cpu lists or in high alloc reserves.
3033 * Shrink them them and try again
Mel Gorman9ee493c2010-09-09 16:38:18 -07003034 */
3035 if (!page && !drained) {
Mel Gorman0aaa29a2015-11-06 16:28:37 -08003036 unreserve_highatomic_pageblock(ac);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08003037 drain_all_pages(NULL);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003038 drained = true;
3039 goto retry;
3040 }
3041
Mel Gorman11e33f62009-06-16 15:31:57 -07003042 return page;
3043}
3044
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003045static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003046{
3047 struct zoneref *z;
3048 struct zone *zone;
3049
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003050 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
3051 ac->high_zoneidx, ac->nodemask)
3052 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
Mel Gorman11e33f62009-06-16 15:31:57 -07003053}
3054
Peter Zijlstra341ce062009-06-16 15:32:02 -07003055static inline int
3056gfp_to_alloc_flags(gfp_t gfp_mask)
3057{
Peter Zijlstra341ce062009-06-16 15:32:02 -07003058 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
Peter Zijlstra341ce062009-06-16 15:32:02 -07003059
Mel Gormana56f57f2009-06-16 15:32:02 -07003060 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07003061 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07003062
Peter Zijlstra341ce062009-06-16 15:32:02 -07003063 /*
3064 * The caller may dip into page reserves a bit more if the caller
3065 * cannot run direct reclaim, or if the caller has realtime scheduling
3066 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
Mel Gormand0164ad2015-11-06 16:28:21 -08003067 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
Peter Zijlstra341ce062009-06-16 15:32:02 -07003068 */
Namhyung Kime6223a32010-10-26 14:21:59 -07003069 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07003070
Mel Gormand0164ad2015-11-06 16:28:21 -08003071 if (gfp_mask & __GFP_ATOMIC) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003072 /*
David Rientjesb104a352014-07-30 16:08:24 -07003073 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3074 * if it can't schedule.
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003075 */
David Rientjesb104a352014-07-30 16:08:24 -07003076 if (!(gfp_mask & __GFP_NOMEMALLOC))
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003077 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07003078 /*
David Rientjesb104a352014-07-30 16:08:24 -07003079 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
Vladimir Davydov344736f2014-10-20 15:50:30 +04003080 * comment for __cpuset_node_allowed().
Peter Zijlstra341ce062009-06-16 15:32:02 -07003081 */
3082 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003083 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07003084 alloc_flags |= ALLOC_HARDER;
3085
Mel Gormanb37f1dd2012-07-31 16:44:03 -07003086 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
3087 if (gfp_mask & __GFP_MEMALLOC)
3088 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07003089 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
3090 alloc_flags |= ALLOC_NO_WATERMARKS;
3091 else if (!in_interrupt() &&
3092 ((current->flags & PF_MEMALLOC) ||
3093 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07003094 alloc_flags |= ALLOC_NO_WATERMARKS;
3095 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07003096#ifdef CONFIG_CMA
David Rientjes43e7a342014-10-09 15:27:25 -07003097 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07003098 alloc_flags |= ALLOC_CMA;
3099#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07003100 return alloc_flags;
3101}
3102
Mel Gorman072bb0a2012-07-31 16:43:58 -07003103bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
3104{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07003105 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07003106}
3107
Mel Gormand0164ad2015-11-06 16:28:21 -08003108static inline bool is_thp_gfp_mask(gfp_t gfp_mask)
3109{
3110 return (gfp_mask & (GFP_TRANSHUGE | __GFP_KSWAPD_RECLAIM)) == GFP_TRANSHUGE;
3111}
3112
Mel Gorman11e33f62009-06-16 15:31:57 -07003113static inline struct page *
3114__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003115 struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003116{
Mel Gormand0164ad2015-11-06 16:28:21 -08003117 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
Mel Gorman11e33f62009-06-16 15:31:57 -07003118 struct page *page = NULL;
3119 int alloc_flags;
3120 unsigned long pages_reclaimed = 0;
3121 unsigned long did_some_progress;
David Rientjese0b9dae2014-06-04 16:08:28 -07003122 enum migrate_mode migration_mode = MIGRATE_ASYNC;
Mel Gorman66199712012-01-12 17:19:41 -08003123 bool deferred_compaction = false;
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003124 int contended_compaction = COMPACT_CONTENDED_NONE;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003125
Christoph Lameter952f3b52006-12-06 20:33:26 -08003126 /*
Mel Gorman72807a72009-06-16 15:32:18 -07003127 * In the slowpath, we sanity check order to avoid ever trying to
3128 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3129 * be using allocators in order of preference for an area that is
3130 * too large.
3131 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07003132 if (order >= MAX_ORDER) {
3133 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07003134 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07003135 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003136
Christoph Lameter952f3b52006-12-06 20:33:26 -08003137 /*
Mel Gormand0164ad2015-11-06 16:28:21 -08003138 * We also sanity check to catch abuse of atomic reserves being used by
3139 * callers that are not in atomic context.
3140 */
3141 if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
3142 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
3143 gfp_mask &= ~__GFP_ATOMIC;
3144
Johannes Weiner9879de72015-01-26 12:58:32 -08003145retry:
Mel Gormand0164ad2015-11-06 16:28:21 -08003146 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003147 wake_all_kswapds(order, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003148
Paul Jackson9bf22292005-09-06 15:18:12 -07003149 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003150 * OK, we're below the kswapd watermark and have kicked background
3151 * reclaim. Now things get more complex, so set up alloc_flags according
3152 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07003153 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07003154 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155
David Rientjesf33261d2011-01-25 15:07:20 -08003156 /*
3157 * Find the true preferred zone if the allocation is unconstrained by
3158 * cpusets.
3159 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003160 if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
Mel Gormand8846372014-06-04 16:10:33 -07003161 struct zoneref *preferred_zoneref;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003162 preferred_zoneref = first_zones_zonelist(ac->zonelist,
3163 ac->high_zoneidx, NULL, &ac->preferred_zone);
3164 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gormand8846372014-06-04 16:10:33 -07003165 }
David Rientjesf33261d2011-01-25 15:07:20 -08003166
Peter Zijlstra341ce062009-06-16 15:32:02 -07003167 /* This is the last chance, in general, before the goto nopage. */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003168 page = get_page_from_freelist(gfp_mask, order,
3169 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003170 if (page)
3171 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172
Mel Gorman11e33f62009-06-16 15:31:57 -07003173 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07003174 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07003175 /*
3176 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3177 * the allocation is high priority and these type of
3178 * allocations are system rather than user orientated
3179 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003180 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
Michal Hocko33d53102016-01-14 15:19:05 -08003181 page = get_page_from_freelist(gfp_mask, order,
3182 ALLOC_NO_WATERMARKS, ac);
3183 if (page)
3184 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185 }
3186
Mel Gormand0164ad2015-11-06 16:28:21 -08003187 /* Caller is not willing to reclaim, we can't balance anything */
3188 if (!can_direct_reclaim) {
David Rientjesaed0a0e2014-01-21 15:51:12 -08003189 /*
Michal Hocko33d53102016-01-14 15:19:05 -08003190 * All existing users of the __GFP_NOFAIL are blockable, so warn
3191 * of any new users that actually allow this type of allocation
3192 * to fail.
David Rientjesaed0a0e2014-01-21 15:51:12 -08003193 */
3194 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195 goto nopage;
David Rientjesaed0a0e2014-01-21 15:51:12 -08003196 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197
Peter Zijlstra341ce062009-06-16 15:32:02 -07003198 /* Avoid recursion of direct reclaim */
Michal Hocko33d53102016-01-14 15:19:05 -08003199 if (current->flags & PF_MEMALLOC) {
3200 /*
3201 * __GFP_NOFAIL request from this context is rather bizarre
3202 * because we cannot reclaim anything and only can loop waiting
3203 * for somebody to do a work for us.
3204 */
3205 if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3206 cond_resched();
3207 goto retry;
3208 }
Peter Zijlstra341ce062009-06-16 15:32:02 -07003209 goto nopage;
Michal Hocko33d53102016-01-14 15:19:05 -08003210 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211
David Rientjes6583bb62009-07-29 15:02:06 -07003212 /* Avoid allocations with no watermarks from looping endlessly */
3213 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3214 goto nopage;
3215
Mel Gorman77f1fe62011-01-13 15:45:57 -08003216 /*
3217 * Try direct compaction. The first pass is asynchronous. Subsequent
3218 * attempts after direct reclaim are synchronous
3219 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003220 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3221 migration_mode,
3222 &contended_compaction,
Vlastimil Babka53853e22014-10-09 15:27:02 -07003223 &deferred_compaction);
Mel Gorman56de7262010-05-24 14:32:30 -07003224 if (page)
3225 goto got_pg;
David Rientjes75f30862014-06-04 16:08:30 -07003226
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003227 /* Checks for THP-specific high-order allocations */
Mel Gormand0164ad2015-11-06 16:28:21 -08003228 if (is_thp_gfp_mask(gfp_mask)) {
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003229 /*
3230 * If compaction is deferred for high-order allocations, it is
3231 * because sync compaction recently failed. If this is the case
3232 * and the caller requested a THP allocation, we do not want
3233 * to heavily disrupt the system, so we fail the allocation
3234 * instead of entering direct reclaim.
3235 */
3236 if (deferred_compaction)
3237 goto nopage;
3238
3239 /*
3240 * In all zones where compaction was attempted (and not
3241 * deferred or skipped), lock contention has been detected.
3242 * For THP allocation we do not want to disrupt the others
3243 * so we fallback to base pages instead.
3244 */
3245 if (contended_compaction == COMPACT_CONTENDED_LOCK)
3246 goto nopage;
3247
3248 /*
3249 * If compaction was aborted due to need_resched(), we do not
3250 * want to further increase allocation latency, unless it is
3251 * khugepaged trying to collapse.
3252 */
3253 if (contended_compaction == COMPACT_CONTENDED_SCHED
3254 && !(current->flags & PF_KTHREAD))
3255 goto nopage;
3256 }
Mel Gorman66199712012-01-12 17:19:41 -08003257
David Rientjes8fe78042014-08-06 16:07:54 -07003258 /*
3259 * It can become very expensive to allocate transparent hugepages at
3260 * fault, so use asynchronous memory compaction for THP unless it is
3261 * khugepaged trying to collapse.
3262 */
Mel Gormand0164ad2015-11-06 16:28:21 -08003263 if (!is_thp_gfp_mask(gfp_mask) || (current->flags & PF_KTHREAD))
David Rientjes8fe78042014-08-06 16:07:54 -07003264 migration_mode = MIGRATE_SYNC_LIGHT;
3265
Mel Gorman11e33f62009-06-16 15:31:57 -07003266 /* Try direct reclaim and then allocating */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003267 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3268 &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07003269 if (page)
3270 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271
Johannes Weiner90839052015-06-24 16:57:21 -07003272 /* Do not loop if specifically requested */
3273 if (gfp_mask & __GFP_NORETRY)
3274 goto noretry;
3275
3276 /* Keep reclaiming pages as long as there is reasonable progress */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07003277 pages_reclaimed += did_some_progress;
Johannes Weiner90839052015-06-24 16:57:21 -07003278 if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) ||
3279 ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) {
Mel Gorman11e33f62009-06-16 15:31:57 -07003280 /* Wait for some write requests to complete then retry */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003281 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
Johannes Weiner9879de72015-01-26 12:58:32 -08003282 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283 }
3284
Johannes Weiner90839052015-06-24 16:57:21 -07003285 /* Reclaim has failed us, start killing things */
3286 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3287 if (page)
3288 goto got_pg;
3289
3290 /* Retry as long as the OOM killer is making progress */
3291 if (did_some_progress)
3292 goto retry;
3293
3294noretry:
3295 /*
3296 * High-order allocations do not necessarily loop after
3297 * direct reclaim and reclaim/compaction depends on compaction
3298 * being called after reclaim so call directly if necessary
3299 */
3300 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags,
3301 ac, migration_mode,
3302 &contended_compaction,
3303 &deferred_compaction);
3304 if (page)
3305 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07003307 warn_alloc_failed(gfp_mask, order, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308got_pg:
Mel Gorman072bb0a2012-07-31 16:43:58 -07003309 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310}
Mel Gorman11e33f62009-06-16 15:31:57 -07003311
3312/*
3313 * This is the 'heart' of the zoned buddy allocator.
3314 */
3315struct page *
3316__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3317 struct zonelist *zonelist, nodemask_t *nodemask)
3318{
Mel Gormand8846372014-06-04 16:10:33 -07003319 struct zoneref *preferred_zoneref;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003320 struct page *page = NULL;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003321 unsigned int cpuset_mems_cookie;
Johannes Weiner3a025762014-04-07 15:37:48 -07003322 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
Andrew Morton91fbdc02015-02-11 15:25:04 -08003323 gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003324 struct alloc_context ac = {
3325 .high_zoneidx = gfp_zone(gfp_mask),
3326 .nodemask = nodemask,
3327 .migratetype = gfpflags_to_migratetype(gfp_mask),
3328 };
Mel Gorman11e33f62009-06-16 15:31:57 -07003329
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003330 gfp_mask &= gfp_allowed_mask;
3331
Mel Gorman11e33f62009-06-16 15:31:57 -07003332 lockdep_trace_alloc(gfp_mask);
3333
Mel Gormand0164ad2015-11-06 16:28:21 -08003334 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
Mel Gorman11e33f62009-06-16 15:31:57 -07003335
3336 if (should_fail_alloc_page(gfp_mask, order))
3337 return NULL;
3338
3339 /*
3340 * Check the zones suitable for the gfp_mask contain at least one
3341 * valid zone. It's possible to have an empty zonelist as a result
David Rientjes4167e9b2015-04-14 15:46:55 -07003342 * of __GFP_THISNODE and a memoryless node
Mel Gorman11e33f62009-06-16 15:31:57 -07003343 */
3344 if (unlikely(!zonelist->_zonerefs->zone))
3345 return NULL;
3346
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003347 if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
Vlastimil Babka21bb9bd2014-10-09 15:26:51 -07003348 alloc_flags |= ALLOC_CMA;
3349
Mel Gormancc9a6c82012-03-21 16:34:11 -07003350retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07003351 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003352
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003353 /* We set it here, as __alloc_pages_slowpath might have changed it */
3354 ac.zonelist = zonelist;
Mel Gormanc9ab0c42015-11-06 16:28:12 -08003355
3356 /* Dirty zone balancing only done in the fast path */
3357 ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE);
3358
Mel Gorman5117f452009-06-16 15:31:59 -07003359 /* The preferred zone is used for statistics later */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003360 preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
3361 ac.nodemask ? : &cpuset_current_mems_allowed,
3362 &ac.preferred_zone);
3363 if (!ac.preferred_zone)
Mel Gormancc9a6c82012-03-21 16:34:11 -07003364 goto out;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003365 ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gorman5117f452009-06-16 15:31:59 -07003366
3367 /* First allocation attempt */
Andrew Morton91fbdc02015-02-11 15:25:04 -08003368 alloc_mask = gfp_mask|__GFP_HARDWALL;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003369 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08003370 if (unlikely(!page)) {
3371 /*
3372 * Runtime PM, block IO and its error handling path
3373 * can deadlock because I/O on the device might not
3374 * complete.
3375 */
Andrew Morton91fbdc02015-02-11 15:25:04 -08003376 alloc_mask = memalloc_noio_flags(gfp_mask);
Mel Gormanc9ab0c42015-11-06 16:28:12 -08003377 ac.spread_dirty_pages = false;
Andrew Morton91fbdc02015-02-11 15:25:04 -08003378
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003379 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08003380 }
Mel Gorman11e33f62009-06-16 15:31:57 -07003381
Xishi Qiu23f086f2015-02-11 15:25:07 -08003382 if (kmemcheck_enabled && page)
3383 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3384
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003385 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07003386
3387out:
3388 /*
3389 * When updating a task's mems_allowed, it is possible to race with
3390 * parallel threads in such a way that an allocation can fail while
3391 * the mask is being updated. If a page allocation is about to fail,
3392 * check if the cpuset changed during allocation and if so, retry.
3393 */
Mel Gormand26914d2014-04-03 14:47:24 -07003394 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07003395 goto retry_cpuset;
3396
Mel Gorman11e33f62009-06-16 15:31:57 -07003397 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398}
Mel Gormand2391712009-06-16 15:31:52 -07003399EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400
3401/*
3402 * Common helper functions.
3403 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08003404unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405{
Akinobu Mita945a1112009-09-21 17:01:47 -07003406 struct page *page;
3407
3408 /*
3409 * __get_free_pages() returns a 32-bit address, which cannot represent
3410 * a highmem page
3411 */
3412 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3413
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414 page = alloc_pages(gfp_mask, order);
3415 if (!page)
3416 return 0;
3417 return (unsigned long) page_address(page);
3418}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419EXPORT_SYMBOL(__get_free_pages);
3420
Harvey Harrison920c7a52008-02-04 22:29:26 -08003421unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422{
Akinobu Mita945a1112009-09-21 17:01:47 -07003423 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003425EXPORT_SYMBOL(get_zeroed_page);
3426
Harvey Harrison920c7a52008-02-04 22:29:26 -08003427void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428{
Nick Pigginb5810032005-10-29 18:16:12 -07003429 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430 if (order == 0)
Mel Gormanb745bc82014-06-04 16:10:22 -07003431 free_hot_cold_page(page, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432 else
3433 __free_pages_ok(page, order);
3434 }
3435}
3436
3437EXPORT_SYMBOL(__free_pages);
3438
Harvey Harrison920c7a52008-02-04 22:29:26 -08003439void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440{
3441 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07003442 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003443 __free_pages(virt_to_page((void *)addr), order);
3444 }
3445}
3446
3447EXPORT_SYMBOL(free_pages);
3448
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003449/*
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003450 * Page Fragment:
3451 * An arbitrary-length arbitrary-offset area of memory which resides
3452 * within a 0 or higher order page. Multiple fragments within that page
3453 * are individually refcounted, in the page's reference counter.
3454 *
3455 * The page_frag functions below provide a simple allocation framework for
3456 * page fragments. This is used by the network stack and network device
3457 * drivers to provide a backing region of memory for use as either an
3458 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3459 */
3460static struct page *__page_frag_refill(struct page_frag_cache *nc,
3461 gfp_t gfp_mask)
3462{
3463 struct page *page = NULL;
3464 gfp_t gfp = gfp_mask;
3465
3466#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3467 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3468 __GFP_NOMEMALLOC;
3469 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3470 PAGE_FRAG_CACHE_MAX_ORDER);
3471 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3472#endif
3473 if (unlikely(!page))
3474 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3475
3476 nc->va = page ? page_address(page) : NULL;
3477
3478 return page;
3479}
3480
3481void *__alloc_page_frag(struct page_frag_cache *nc,
3482 unsigned int fragsz, gfp_t gfp_mask)
3483{
3484 unsigned int size = PAGE_SIZE;
3485 struct page *page;
3486 int offset;
3487
3488 if (unlikely(!nc->va)) {
3489refill:
3490 page = __page_frag_refill(nc, gfp_mask);
3491 if (!page)
3492 return NULL;
3493
3494#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3495 /* if size can vary use size else just use PAGE_SIZE */
3496 size = nc->size;
3497#endif
3498 /* Even if we own the page, we do not use atomic_set().
3499 * This would break get_page_unless_zero() users.
3500 */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07003501 page_ref_add(page, size - 1);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003502
3503 /* reset page count bias and offset to start of new frag */
Michal Hocko2f064f32015-08-21 14:11:51 -07003504 nc->pfmemalloc = page_is_pfmemalloc(page);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003505 nc->pagecnt_bias = size;
3506 nc->offset = size;
3507 }
3508
3509 offset = nc->offset - fragsz;
3510 if (unlikely(offset < 0)) {
3511 page = virt_to_page(nc->va);
3512
Joonsoo Kimfe896d12016-03-17 14:19:26 -07003513 if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003514 goto refill;
3515
3516#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3517 /* if size can vary use size else just use PAGE_SIZE */
3518 size = nc->size;
3519#endif
3520 /* OK, page count is 0, we can safely set it */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07003521 set_page_count(page, size);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003522
3523 /* reset page count bias and offset to start of new frag */
3524 nc->pagecnt_bias = size;
3525 offset = size - fragsz;
3526 }
3527
3528 nc->pagecnt_bias--;
3529 nc->offset = offset;
3530
3531 return nc->va + offset;
3532}
3533EXPORT_SYMBOL(__alloc_page_frag);
3534
3535/*
3536 * Frees a page fragment allocated out of either a compound or order 0 page.
3537 */
3538void __free_page_frag(void *addr)
3539{
3540 struct page *page = virt_to_head_page(addr);
3541
3542 if (unlikely(put_page_testzero(page)))
3543 __free_pages_ok(page, compound_order(page));
3544}
3545EXPORT_SYMBOL(__free_page_frag);
3546
3547/*
Vladimir Davydov52383432014-06-04 16:06:39 -07003548 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
Vladimir Davydova9bb7e62016-01-14 15:18:12 -08003549 * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is
3550 * equivalent to alloc_pages.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003551 *
Vladimir Davydov52383432014-06-04 16:06:39 -07003552 * It should be used when the caller would like to use kmalloc, but since the
3553 * allocation is large, it has to fall back to the page allocator.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003554 */
Vladimir Davydov52383432014-06-04 16:06:39 -07003555struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
3556{
3557 struct page *page;
Vladimir Davydov52383432014-06-04 16:06:39 -07003558
Vladimir Davydov52383432014-06-04 16:06:39 -07003559 page = alloc_pages(gfp_mask, order);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08003560 if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
3561 __free_pages(page, order);
3562 page = NULL;
3563 }
Vladimir Davydov52383432014-06-04 16:06:39 -07003564 return page;
3565}
3566
3567struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
3568{
3569 struct page *page;
Vladimir Davydov52383432014-06-04 16:06:39 -07003570
Vladimir Davydov52383432014-06-04 16:06:39 -07003571 page = alloc_pages_node(nid, gfp_mask, order);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08003572 if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
3573 __free_pages(page, order);
3574 page = NULL;
3575 }
Vladimir Davydov52383432014-06-04 16:06:39 -07003576 return page;
3577}
3578
3579/*
3580 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3581 * alloc_kmem_pages.
3582 */
3583void __free_kmem_pages(struct page *page, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003584{
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08003585 memcg_kmem_uncharge(page, order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003586 __free_pages(page, order);
3587}
3588
Vladimir Davydov52383432014-06-04 16:06:39 -07003589void free_kmem_pages(unsigned long addr, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003590{
3591 if (addr != 0) {
3592 VM_BUG_ON(!virt_addr_valid((void *)addr));
Vladimir Davydov52383432014-06-04 16:06:39 -07003593 __free_kmem_pages(virt_to_page((void *)addr), order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003594 }
3595}
3596
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08003597static void *make_alloc_exact(unsigned long addr, unsigned int order,
3598 size_t size)
Andi Kleenee85c2e2011-05-11 15:13:34 -07003599{
3600 if (addr) {
3601 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3602 unsigned long used = addr + PAGE_ALIGN(size);
3603
3604 split_page(virt_to_page((void *)addr), order);
3605 while (used < alloc_end) {
3606 free_page(used);
3607 used += PAGE_SIZE;
3608 }
3609 }
3610 return (void *)addr;
3611}
3612
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003613/**
3614 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3615 * @size: the number of bytes to allocate
3616 * @gfp_mask: GFP flags for the allocation
3617 *
3618 * This function is similar to alloc_pages(), except that it allocates the
3619 * minimum number of pages to satisfy the request. alloc_pages() can only
3620 * allocate memory in power-of-two pages.
3621 *
3622 * This function is also limited by MAX_ORDER.
3623 *
3624 * Memory allocated by this function must be released by free_pages_exact().
3625 */
3626void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3627{
3628 unsigned int order = get_order(size);
3629 unsigned long addr;
3630
3631 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07003632 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003633}
3634EXPORT_SYMBOL(alloc_pages_exact);
3635
3636/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07003637 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3638 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07003639 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07003640 * @size: the number of bytes to allocate
3641 * @gfp_mask: GFP flags for the allocation
3642 *
3643 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3644 * back.
Andi Kleenee85c2e2011-05-11 15:13:34 -07003645 */
Fabian Fredericke1931812014-08-06 16:04:59 -07003646void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
Andi Kleenee85c2e2011-05-11 15:13:34 -07003647{
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08003648 unsigned int order = get_order(size);
Andi Kleenee85c2e2011-05-11 15:13:34 -07003649 struct page *p = alloc_pages_node(nid, gfp_mask, order);
3650 if (!p)
3651 return NULL;
3652 return make_alloc_exact((unsigned long)page_address(p), order, size);
3653}
Andi Kleenee85c2e2011-05-11 15:13:34 -07003654
3655/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003656 * free_pages_exact - release memory allocated via alloc_pages_exact()
3657 * @virt: the value returned by alloc_pages_exact.
3658 * @size: size of allocation, same value as passed to alloc_pages_exact().
3659 *
3660 * Release the memory allocated by a previous call to alloc_pages_exact.
3661 */
3662void free_pages_exact(void *virt, size_t size)
3663{
3664 unsigned long addr = (unsigned long)virt;
3665 unsigned long end = addr + PAGE_ALIGN(size);
3666
3667 while (addr < end) {
3668 free_page(addr);
3669 addr += PAGE_SIZE;
3670 }
3671}
3672EXPORT_SYMBOL(free_pages_exact);
3673
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003674/**
3675 * nr_free_zone_pages - count number of pages beyond high watermark
3676 * @offset: The zone index of the highest zone
3677 *
3678 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3679 * high watermark within all zones at or below a given zone index. For each
3680 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07003681 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003682 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003683static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003684{
Mel Gormandd1a2392008-04-28 02:12:17 -07003685 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003686 struct zone *zone;
3687
Martin J. Blighe310fd42005-07-29 22:59:18 -07003688 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003689 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003690
Mel Gorman0e884602008-04-28 02:12:14 -07003691 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692
Mel Gorman54a6eb52008-04-28 02:12:16 -07003693 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08003694 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003695 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07003696 if (size > high)
3697 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003698 }
3699
3700 return sum;
3701}
3702
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003703/**
3704 * nr_free_buffer_pages - count number of pages beyond high watermark
3705 *
3706 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3707 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003708 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003709unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003710{
Al Viroaf4ca452005-10-21 02:55:38 -04003711 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712}
Meelap Shahc2f1a552007-07-17 04:04:39 -07003713EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003714
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003715/**
3716 * nr_free_pagecache_pages - count number of pages beyond high watermark
3717 *
3718 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3719 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003720 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003721unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722{
Mel Gorman2a1e2742007-07-17 04:03:12 -07003723 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003724}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07003725
3726static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003727{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08003728 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08003729 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003730}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003731
Igor Redkod02bd272016-03-17 14:19:05 -07003732long si_mem_available(void)
3733{
3734 long available;
3735 unsigned long pagecache;
3736 unsigned long wmark_low = 0;
3737 unsigned long pages[NR_LRU_LISTS];
3738 struct zone *zone;
3739 int lru;
3740
3741 for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
3742 pages[lru] = global_page_state(NR_LRU_BASE + lru);
3743
3744 for_each_zone(zone)
3745 wmark_low += zone->watermark[WMARK_LOW];
3746
3747 /*
3748 * Estimate the amount of memory available for userspace allocations,
3749 * without causing swapping.
3750 */
3751 available = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
3752
3753 /*
3754 * Not all the page cache can be freed, otherwise the system will
3755 * start swapping. Assume at least half of the page cache, or the
3756 * low watermark worth of cache, needs to stay.
3757 */
3758 pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
3759 pagecache -= min(pagecache / 2, wmark_low);
3760 available += pagecache;
3761
3762 /*
3763 * Part of the reclaimable slab consists of items that are in use,
3764 * and cannot be freed. Cap this estimate at the low watermark.
3765 */
3766 available += global_page_state(NR_SLAB_RECLAIMABLE) -
3767 min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low);
3768
3769 if (available < 0)
3770 available = 0;
3771 return available;
3772}
3773EXPORT_SYMBOL_GPL(si_mem_available);
3774
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775void si_meminfo(struct sysinfo *val)
3776{
3777 val->totalram = totalram_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003778 val->sharedram = global_page_state(NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003779 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781 val->totalhigh = totalhigh_pages;
3782 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783 val->mem_unit = PAGE_SIZE;
3784}
3785
3786EXPORT_SYMBOL(si_meminfo);
3787
3788#ifdef CONFIG_NUMA
3789void si_meminfo_node(struct sysinfo *val, int nid)
3790{
Jiang Liucdd91a72013-07-03 15:03:27 -07003791 int zone_type; /* needs to be signed */
3792 unsigned long managed_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793 pg_data_t *pgdat = NODE_DATA(nid);
3794
Jiang Liucdd91a72013-07-03 15:03:27 -07003795 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3796 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3797 val->totalram = managed_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003798 val->sharedram = node_page_state(nid, NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003799 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003800#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08003801 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08003802 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3803 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003804#else
3805 val->totalhigh = 0;
3806 val->freehigh = 0;
3807#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003808 val->mem_unit = PAGE_SIZE;
3809}
3810#endif
3811
David Rientjesddd588b2011-03-22 16:30:46 -07003812/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07003813 * Determine whether the node should be displayed or not, depending on whether
3814 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07003815 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003816bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07003817{
3818 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003819 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07003820
3821 if (!(flags & SHOW_MEM_FILTER_NODES))
3822 goto out;
3823
Mel Gormancc9a6c82012-03-21 16:34:11 -07003824 do {
Mel Gormand26914d2014-04-03 14:47:24 -07003825 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003826 ret = !node_isset(nid, cpuset_current_mems_allowed);
Mel Gormand26914d2014-04-03 14:47:24 -07003827 } while (read_mems_allowed_retry(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07003828out:
3829 return ret;
3830}
3831
Linus Torvalds1da177e2005-04-16 15:20:36 -07003832#define K(x) ((x) << (PAGE_SHIFT-10))
3833
Rabin Vincent377e4f12012-12-11 16:00:24 -08003834static void show_migration_types(unsigned char type)
3835{
3836 static const char types[MIGRATE_TYPES] = {
3837 [MIGRATE_UNMOVABLE] = 'U',
Rabin Vincent377e4f12012-12-11 16:00:24 -08003838 [MIGRATE_MOVABLE] = 'M',
Vlastimil Babka475a2f92015-12-11 13:40:29 -08003839 [MIGRATE_RECLAIMABLE] = 'E',
3840 [MIGRATE_HIGHATOMIC] = 'H',
Rabin Vincent377e4f12012-12-11 16:00:24 -08003841#ifdef CONFIG_CMA
3842 [MIGRATE_CMA] = 'C',
3843#endif
Minchan Kim194159f2013-02-22 16:33:58 -08003844#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08003845 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08003846#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08003847 };
3848 char tmp[MIGRATE_TYPES + 1];
3849 char *p = tmp;
3850 int i;
3851
3852 for (i = 0; i < MIGRATE_TYPES; i++) {
3853 if (type & (1 << i))
3854 *p++ = types[i];
3855 }
3856
3857 *p = '\0';
3858 printk("(%s) ", tmp);
3859}
3860
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861/*
3862 * Show free area list (used inside shift_scroll-lock stuff)
3863 * We also calculate the percentage fragmentation. We do this by counting the
3864 * memory on each free list with the exception of the first item on the list.
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003865 *
3866 * Bits in @filter:
3867 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3868 * cpuset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003869 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003870void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003871{
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003872 unsigned long free_pcp = 0;
Jes Sorensenc7241912006-09-27 01:50:05 -07003873 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874 struct zone *zone;
3875
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003876 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07003877 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003878 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003879
Konstantin Khlebnikov761b0672015-04-14 15:45:32 -07003880 for_each_online_cpu(cpu)
3881 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882 }
3883
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003884 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3885 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003886 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3887 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003888 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003889 " free:%lu free_pcp:%lu free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003890 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003891 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003892 global_page_state(NR_ISOLATED_ANON),
3893 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003894 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003895 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003896 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003897 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003898 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003899 global_page_state(NR_UNSTABLE_NFS),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003900 global_page_state(NR_SLAB_RECLAIMABLE),
3901 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003902 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003903 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003904 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003905 global_page_state(NR_BOUNCE),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003906 global_page_state(NR_FREE_PAGES),
3907 free_pcp,
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003908 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003909
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003910 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911 int i;
3912
David Rientjes7bf02ea2011-05-24 17:11:16 -07003913 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003914 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003915
3916 free_pcp = 0;
3917 for_each_online_cpu(cpu)
3918 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3919
Linus Torvalds1da177e2005-04-16 15:20:36 -07003920 show_node(zone);
3921 printk("%s"
3922 " free:%lukB"
3923 " min:%lukB"
3924 " low:%lukB"
3925 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003926 " active_anon:%lukB"
3927 " inactive_anon:%lukB"
3928 " active_file:%lukB"
3929 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003930 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003931 " isolated(anon):%lukB"
3932 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003933 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003934 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003935 " mlocked:%lukB"
3936 " dirty:%lukB"
3937 " writeback:%lukB"
3938 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003939 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003940 " slab_reclaimable:%lukB"
3941 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003942 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003943 " pagetables:%lukB"
3944 " unstable:%lukB"
3945 " bounce:%lukB"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003946 " free_pcp:%lukB"
3947 " local_pcp:%ukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003948 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003949 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003950 " pages_scanned:%lu"
3951 " all_unreclaimable? %s"
3952 "\n",
3953 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003954 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003955 K(min_wmark_pages(zone)),
3956 K(low_wmark_pages(zone)),
3957 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003958 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3959 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3960 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3961 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003962 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003963 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3964 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003966 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003967 K(zone_page_state(zone, NR_MLOCK)),
3968 K(zone_page_state(zone, NR_FILE_DIRTY)),
3969 K(zone_page_state(zone, NR_WRITEBACK)),
3970 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003971 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003972 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3973 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003974 zone_page_state(zone, NR_KERNEL_STACK) *
3975 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003976 K(zone_page_state(zone, NR_PAGETABLE)),
3977 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3978 K(zone_page_state(zone, NR_BOUNCE)),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003979 K(free_pcp),
3980 K(this_cpu_read(zone->pageset->pcp.count)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003981 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003982 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Mel Gorman0d5d8232014-08-06 16:07:16 -07003983 K(zone_page_state(zone, NR_PAGES_SCANNED)),
Lisa Du6e543d52013-09-11 14:22:36 -07003984 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003985 );
3986 printk("lowmem_reserve[]:");
3987 for (i = 0; i < MAX_NR_ZONES; i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07003988 printk(" %ld", zone->lowmem_reserve[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989 printk("\n");
3990 }
3991
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003992 for_each_populated_zone(zone) {
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08003993 unsigned int order;
3994 unsigned long nr[MAX_ORDER], flags, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003995 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003996
David Rientjes7bf02ea2011-05-24 17:11:16 -07003997 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003998 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003999 show_node(zone);
4000 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001
4002 spin_lock_irqsave(&zone->lock, flags);
4003 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08004004 struct free_area *area = &zone->free_area[order];
4005 int type;
4006
4007 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07004008 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08004009
4010 types[order] = 0;
4011 for (type = 0; type < MIGRATE_TYPES; type++) {
4012 if (!list_empty(&area->free_list[type]))
4013 types[order] |= 1 << type;
4014 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015 }
4016 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08004017 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07004018 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08004019 if (nr[order])
4020 show_migration_types(types[order]);
4021 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004022 printk("= %lukB\n", K(total));
4023 }
4024
David Rientjes949f7ec2013-04-29 15:07:48 -07004025 hugetlb_show_meminfo();
4026
Larry Woodmane6f36022008-02-04 22:29:30 -08004027 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
4028
Linus Torvalds1da177e2005-04-16 15:20:36 -07004029 show_swap_cache_info();
4030}
4031
Mel Gorman19770b32008-04-28 02:12:18 -07004032static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
4033{
4034 zoneref->zone = zone;
4035 zoneref->zone_idx = zone_idx(zone);
4036}
4037
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038/*
4039 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08004040 *
4041 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004043static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004044 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045{
Christoph Lameter1a932052006-01-06 00:11:16 -08004046 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004047 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08004048
4049 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004050 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08004051 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08004052 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07004053 zoneref_set_zone(zone,
4054 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08004055 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004056 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004057 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004058
Christoph Lameter070f8032006-01-06 00:11:19 -08004059 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060}
4061
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004062
4063/*
4064 * zonelist_order:
4065 * 0 = automatic detection of better ordering.
4066 * 1 = order by ([node] distance, -zonetype)
4067 * 2 = order by (-zonetype, [node] distance)
4068 *
4069 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4070 * the same zonelist. So only NUMA can configure this param.
4071 */
4072#define ZONELIST_ORDER_DEFAULT 0
4073#define ZONELIST_ORDER_NODE 1
4074#define ZONELIST_ORDER_ZONE 2
4075
4076/* zonelist order in the kernel.
4077 * set_zonelist_order() will set this to NODE or ZONE.
4078 */
4079static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
4080static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
4081
4082
Linus Torvalds1da177e2005-04-16 15:20:36 -07004083#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004084/* The value user specified ....changed by config */
4085static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4086/* string for sysctl */
4087#define NUMA_ZONELIST_ORDER_LEN 16
4088char numa_zonelist_order[16] = "default";
4089
4090/*
4091 * interface for configure zonelist ordering.
4092 * command line option "numa_zonelist_order"
4093 * = "[dD]efault - default, automatic configuration.
4094 * = "[nN]ode - order by node locality, then by zone within node
4095 * = "[zZ]one - order by zone, then by locality within zone
4096 */
4097
4098static int __parse_numa_zonelist_order(char *s)
4099{
4100 if (*s == 'd' || *s == 'D') {
4101 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4102 } else if (*s == 'n' || *s == 'N') {
4103 user_zonelist_order = ZONELIST_ORDER_NODE;
4104 } else if (*s == 'z' || *s == 'Z') {
4105 user_zonelist_order = ZONELIST_ORDER_ZONE;
4106 } else {
Joe Perches11705322016-03-17 14:19:50 -07004107 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004108 return -EINVAL;
4109 }
4110 return 0;
4111}
4112
4113static __init int setup_numa_zonelist_order(char *s)
4114{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08004115 int ret;
4116
4117 if (!s)
4118 return 0;
4119
4120 ret = __parse_numa_zonelist_order(s);
4121 if (ret == 0)
4122 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
4123
4124 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004125}
4126early_param("numa_zonelist_order", setup_numa_zonelist_order);
4127
4128/*
4129 * sysctl handler for numa_zonelist_order
4130 */
Joe Perchescccad5b2014-06-06 14:38:09 -07004131int numa_zonelist_order_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004132 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004133 loff_t *ppos)
4134{
4135 char saved_string[NUMA_ZONELIST_ORDER_LEN];
4136 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01004137 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004138
Andi Kleen443c6f12009-12-23 21:00:47 +01004139 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07004140 if (write) {
4141 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
4142 ret = -EINVAL;
4143 goto out;
4144 }
4145 strcpy(saved_string, (char *)table->data);
4146 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004147 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004148 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01004149 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004150 if (write) {
4151 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07004152
4153 ret = __parse_numa_zonelist_order((char *)table->data);
4154 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004155 /*
4156 * bogus value. restore saved string
4157 */
Chen Gangdacbde02013-07-03 15:02:35 -07004158 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004159 NUMA_ZONELIST_ORDER_LEN);
4160 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004161 } else if (oldval != user_zonelist_order) {
4162 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07004163 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004164 mutex_unlock(&zonelists_mutex);
4165 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004166 }
Andi Kleen443c6f12009-12-23 21:00:47 +01004167out:
4168 mutex_unlock(&zl_order_mutex);
4169 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004170}
4171
4172
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004173#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004174static int node_load[MAX_NUMNODES];
4175
Linus Torvalds1da177e2005-04-16 15:20:36 -07004176/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07004177 * 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 -07004178 * @node: node whose fallback list we're appending
4179 * @used_node_mask: nodemask_t of already used nodes
4180 *
4181 * We use a number of factors to determine which is the next node that should
4182 * appear on a given node's fallback list. The node should not have appeared
4183 * already in @node's fallback list, and it should be the next closest node
4184 * according to the distance array (which contains arbitrary distance values
4185 * from each node to each node in the system), and should also prefer nodes
4186 * with no CPUs, since presumably they'll have very little allocation pressure
4187 * on them otherwise.
4188 * It returns -1 if no node is found.
4189 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004190static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004191{
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01004192 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004193 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08004194 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10304195 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004196
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01004197 /* Use the local node if we haven't already */
4198 if (!node_isset(node, *used_node_mask)) {
4199 node_set(node, *used_node_mask);
4200 return node;
4201 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004202
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004203 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004204
4205 /* Don't want a node to appear more than once */
4206 if (node_isset(n, *used_node_mask))
4207 continue;
4208
Linus Torvalds1da177e2005-04-16 15:20:36 -07004209 /* Use the distance array to find the distance */
4210 val = node_distance(node, n);
4211
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01004212 /* Penalize nodes under us ("prefer the next node") */
4213 val += (n < node);
4214
Linus Torvalds1da177e2005-04-16 15:20:36 -07004215 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10304216 tmp = cpumask_of_node(n);
4217 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004218 val += PENALTY_FOR_NODE_WITH_CPUS;
4219
4220 /* Slight preference for less loaded node */
4221 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4222 val += node_load[n];
4223
4224 if (val < min_val) {
4225 min_val = val;
4226 best_node = n;
4227 }
4228 }
4229
4230 if (best_node >= 0)
4231 node_set(best_node, *used_node_mask);
4232
4233 return best_node;
4234}
4235
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004236
4237/*
4238 * Build zonelists ordered by node and zones within node.
4239 * This results in maximum locality--normal zone overflows into local
4240 * DMA zone, if any--but risks exhausting DMA zone.
4241 */
4242static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004243{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004244 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004245 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004246
Mel Gorman54a6eb52008-04-28 02:12:16 -07004247 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07004248 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07004249 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004250 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07004251 zonelist->_zonerefs[j].zone = NULL;
4252 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004253}
4254
4255/*
Christoph Lameter523b9452007-10-16 01:25:37 -07004256 * Build gfp_thisnode zonelists
4257 */
4258static void build_thisnode_zonelists(pg_data_t *pgdat)
4259{
Christoph Lameter523b9452007-10-16 01:25:37 -07004260 int j;
4261 struct zonelist *zonelist;
4262
Mel Gorman54a6eb52008-04-28 02:12:16 -07004263 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004264 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07004265 zonelist->_zonerefs[j].zone = NULL;
4266 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07004267}
4268
4269/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004270 * Build zonelists ordered by zone and nodes within zones.
4271 * This results in conserving DMA zone[s] until all Normal memory is
4272 * exhausted, but results in overflowing to remote node while memory
4273 * may still exist in local DMA zone.
4274 */
4275static int node_order[MAX_NUMNODES];
4276
4277static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4278{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004279 int pos, j, node;
4280 int zone_type; /* needs to be signed */
4281 struct zone *z;
4282 struct zonelist *zonelist;
4283
Mel Gorman54a6eb52008-04-28 02:12:16 -07004284 zonelist = &pgdat->node_zonelists[0];
4285 pos = 0;
4286 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4287 for (j = 0; j < nr_nodes; j++) {
4288 node = node_order[j];
4289 z = &NODE_DATA(node)->node_zones[zone_type];
4290 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07004291 zoneref_set_zone(z,
4292 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004293 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004294 }
4295 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004296 }
Mel Gormandd1a2392008-04-28 02:12:17 -07004297 zonelist->_zonerefs[pos].zone = NULL;
4298 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004299}
4300
Mel Gorman31939132014-10-09 15:28:30 -07004301#if defined(CONFIG_64BIT)
4302/*
4303 * Devices that require DMA32/DMA are relatively rare and do not justify a
4304 * penalty to every machine in case the specialised case applies. Default
4305 * to Node-ordering on 64-bit NUMA machines
4306 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004307static int default_zonelist_order(void)
4308{
Mel Gorman31939132014-10-09 15:28:30 -07004309 return ZONELIST_ORDER_NODE;
4310}
4311#else
4312/*
4313 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4314 * by the kernel. If processes running on node 0 deplete the low memory zone
4315 * then reclaim will occur more frequency increasing stalls and potentially
4316 * be easier to OOM if a large percentage of the zone is under writeback or
4317 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4318 * Hence, default to zone ordering on 32-bit.
4319 */
4320static int default_zonelist_order(void)
4321{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004322 return ZONELIST_ORDER_ZONE;
4323}
Mel Gorman31939132014-10-09 15:28:30 -07004324#endif /* CONFIG_64BIT */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004325
4326static void set_zonelist_order(void)
4327{
4328 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4329 current_zonelist_order = default_zonelist_order();
4330 else
4331 current_zonelist_order = user_zonelist_order;
4332}
4333
4334static void build_zonelists(pg_data_t *pgdat)
4335{
Yaowei Baic00eb152016-01-14 15:19:00 -08004336 int i, node, load;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004337 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004338 int local_node, prev_node;
4339 struct zonelist *zonelist;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004340 unsigned int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004341
4342 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07004343 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004344 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07004345 zonelist->_zonerefs[0].zone = NULL;
4346 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004347 }
4348
4349 /* NUMA-aware ordering of nodes */
4350 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004351 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004352 prev_node = local_node;
4353 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004354
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004355 memset(node_order, 0, sizeof(node_order));
Yaowei Baic00eb152016-01-14 15:19:00 -08004356 i = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004357
Linus Torvalds1da177e2005-04-16 15:20:36 -07004358 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4359 /*
4360 * We don't want to pressure a particular node.
4361 * So adding penalty to the first node in same
4362 * distance group to make it round-robin.
4363 */
David Rientjes957f8222012-10-08 16:33:24 -07004364 if (node_distance(local_node, node) !=
4365 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004366 node_load[node] = load;
4367
Linus Torvalds1da177e2005-04-16 15:20:36 -07004368 prev_node = node;
4369 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004370 if (order == ZONELIST_ORDER_NODE)
4371 build_zonelists_in_node_order(pgdat, node);
4372 else
Yaowei Baic00eb152016-01-14 15:19:00 -08004373 node_order[i++] = node; /* remember order */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004374 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004375
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004376 if (order == ZONELIST_ORDER_ZONE) {
4377 /* calculate node order -- i.e., DMA last! */
Yaowei Baic00eb152016-01-14 15:19:00 -08004378 build_zonelists_in_zone_order(pgdat, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004379 }
Christoph Lameter523b9452007-10-16 01:25:37 -07004380
4381 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004382}
4383
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004384#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4385/*
4386 * Return node id of node used for "local" allocations.
4387 * I.e., first node id of first zone in arg node's generic zonelist.
4388 * Used for initializing percpu 'numa_mem', which is used primarily
4389 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4390 */
4391int local_memory_node(int node)
4392{
4393 struct zone *zone;
4394
4395 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
4396 gfp_zone(GFP_KERNEL),
4397 NULL,
4398 &zone);
4399 return zone->node;
4400}
4401#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004402
Linus Torvalds1da177e2005-04-16 15:20:36 -07004403#else /* CONFIG_NUMA */
4404
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004405static void set_zonelist_order(void)
4406{
4407 current_zonelist_order = ZONELIST_ORDER_ZONE;
4408}
4409
4410static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004411{
Christoph Lameter19655d32006-09-25 23:31:19 -07004412 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07004413 enum zone_type j;
4414 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004415
4416 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004417
Mel Gorman54a6eb52008-04-28 02:12:16 -07004418 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004419 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004420
Mel Gorman54a6eb52008-04-28 02:12:16 -07004421 /*
4422 * Now we build the zonelist so that it contains the zones
4423 * of all the other nodes.
4424 * We don't want to pressure a particular node, so when
4425 * building the zones for node N, we make sure that the
4426 * zones coming right after the local ones are those from
4427 * node N+1 (modulo N)
4428 */
4429 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4430 if (!node_online(node))
4431 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004432 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004433 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07004434 for (node = 0; node < local_node; node++) {
4435 if (!node_online(node))
4436 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004437 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004438 }
4439
Mel Gormandd1a2392008-04-28 02:12:17 -07004440 zonelist->_zonerefs[j].zone = NULL;
4441 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004442}
4443
4444#endif /* CONFIG_NUMA */
4445
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004446/*
4447 * Boot pageset table. One per cpu which is going to be used for all
4448 * zones and all nodes. The parameters will be set in such a way
4449 * that an item put on a list will immediately be handed over to
4450 * the buddy list. This is safe since pageset manipulation is done
4451 * with interrupts disabled.
4452 *
4453 * The boot_pagesets must be kept even after bootup is complete for
4454 * unused processors and/or zones. They do play a role for bootstrapping
4455 * hotplugged processors.
4456 *
4457 * zoneinfo_show() and maybe other functions do
4458 * not check if the processor is online before following the pageset pointer.
4459 * Other parts of the kernel may not check if the zone is available.
4460 */
4461static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4462static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07004463static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004464
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004465/*
4466 * Global mutex to protect against size modification of zonelists
4467 * as well as to serialize pageset setup for the new populated zone.
4468 */
4469DEFINE_MUTEX(zonelists_mutex);
4470
Rusty Russell9b1a4d32008-07-28 12:16:30 -05004471/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07004472static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004473{
Yasunori Goto68113782006-06-23 02:03:11 -07004474 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004475 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07004476 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004477
Bo Liu7f9cfb32009-08-18 14:11:19 -07004478#ifdef CONFIG_NUMA
4479 memset(node_load, 0, sizeof(node_load));
4480#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07004481
4482 if (self && !node_online(self->node_id)) {
4483 build_zonelists(self);
Jiang Liu9adb62a2012-07-31 16:43:28 -07004484 }
4485
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004486 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07004487 pg_data_t *pgdat = NODE_DATA(nid);
4488
4489 build_zonelists(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004490 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004491
4492 /*
4493 * Initialize the boot_pagesets that are going to be used
4494 * for bootstrapping processors. The real pagesets for
4495 * each zone will be allocated later when the per cpu
4496 * allocator is available.
4497 *
4498 * boot_pagesets are used also for bootstrapping offline
4499 * cpus if the system is already booted because the pagesets
4500 * are needed to initialize allocators on a specific cpu too.
4501 * F.e. the percpu allocator needs the page allocator which
4502 * needs the percpu allocator in order to allocate its pagesets
4503 * (a chicken-egg dilemma).
4504 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004505 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004506 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4507
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004508#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4509 /*
4510 * We now know the "local memory node" for each node--
4511 * i.e., the node of the first zone in the generic zonelist.
4512 * Set up numa_mem percpu variable for on-line cpus. During
4513 * boot, only the boot cpu should be on-line; we'll init the
4514 * secondary cpus' numa_mem as they come on-line. During
4515 * node/memory hotplug, we'll fixup all on-line cpus.
4516 */
4517 if (cpu_online(cpu))
4518 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4519#endif
4520 }
4521
Yasunori Goto68113782006-06-23 02:03:11 -07004522 return 0;
4523}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004524
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004525static noinline void __init
4526build_all_zonelists_init(void)
4527{
4528 __build_all_zonelists(NULL);
4529 mminit_verify_zonelist();
4530 cpuset_init_current_mems_allowed();
4531}
4532
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004533/*
4534 * Called with zonelists_mutex held always
4535 * unless system_state == SYSTEM_BOOTING.
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004536 *
4537 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4538 * [we're only called with non-NULL zone through __meminit paths] and
4539 * (2) call of __init annotated helper build_all_zonelists_init
4540 * [protected by SYSTEM_BOOTING].
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004541 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07004542void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07004543{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004544 set_zonelist_order();
4545
Yasunori Goto68113782006-06-23 02:03:11 -07004546 if (system_state == SYSTEM_BOOTING) {
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004547 build_all_zonelists_init();
Yasunori Goto68113782006-06-23 02:03:11 -07004548 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08004549#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07004550 if (zone)
4551 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08004552#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07004553 /* we have to stop all cpus to guarantee there is no user
4554 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07004555 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07004556 /* cpuset refresh routine should be here */
4557 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07004558 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004559 /*
4560 * Disable grouping by mobility if the number of pages in the
4561 * system is too low to allow the mechanism to work. It would be
4562 * more accurate, but expensive to check per-zone. This check is
4563 * made on memory-hotadd so a system can start with mobility
4564 * disabled and enable it later
4565 */
Mel Gormand9c23402007-10-16 01:26:01 -07004566 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004567 page_group_by_mobility_disabled = 1;
4568 else
4569 page_group_by_mobility_disabled = 0;
4570
Joe Perches756a0252016-03-17 14:19:47 -07004571 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
4572 nr_online_nodes,
4573 zonelist_order_name[current_zonelist_order],
4574 page_group_by_mobility_disabled ? "off" : "on",
4575 vm_total_pages);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004576#ifdef CONFIG_NUMA
Anton Blanchardf88dfff2014-12-10 15:42:53 -08004577 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004578#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004579}
4580
4581/*
4582 * Helper functions to size the waitqueue hash table.
4583 * Essentially these want to choose hash table sizes sufficiently
4584 * large so that collisions trying to wait on pages are rare.
4585 * But in fact, the number of active page waitqueues on typical
4586 * systems is ridiculously low, less than 200. So this is even
4587 * conservative, even though it seems large.
4588 *
4589 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4590 * waitqueues, i.e. the size of the waitq table given the number of pages.
4591 */
4592#define PAGES_PER_WAITQUEUE 256
4593
Yasunori Gotocca448f2006-06-23 02:03:10 -07004594#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07004595static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004596{
4597 unsigned long size = 1;
4598
4599 pages /= PAGES_PER_WAITQUEUE;
4600
4601 while (size < pages)
4602 size <<= 1;
4603
4604 /*
4605 * Once we have dozens or even hundreds of threads sleeping
4606 * on IO we've got bigger problems than wait queue collision.
4607 * Limit the size of the wait table to a reasonable size.
4608 */
4609 size = min(size, 4096UL);
4610
4611 return max(size, 4UL);
4612}
Yasunori Gotocca448f2006-06-23 02:03:10 -07004613#else
4614/*
4615 * A zone's size might be changed by hot-add, so it is not possible to determine
4616 * a suitable size for its wait_table. So we use the maximum size now.
4617 *
4618 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4619 *
4620 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4621 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4622 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4623 *
4624 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4625 * or more by the traditional way. (See above). It equals:
4626 *
4627 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4628 * ia64(16K page size) : = ( 8G + 4M)byte.
4629 * powerpc (64K page size) : = (32G +16M)byte.
4630 */
4631static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4632{
4633 return 4096UL;
4634}
4635#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004636
4637/*
4638 * This is an integer logarithm so that shifts can be used later
4639 * to extract the more random high bits from the multiplicative
4640 * hash function before the remainder is taken.
4641 */
4642static inline unsigned long wait_table_bits(unsigned long size)
4643{
4644 return ffz(~size);
4645}
4646
Mel Gorman56fd56b2007-10-16 01:25:58 -07004647/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004648 * Initially all pages are reserved - free ones are freed
4649 * up by free_all_bootmem() once the early boot process is
4650 * done. Non-atomic initialization, single-pass.
4651 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004652void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004653 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004654{
Dan Williams4b94ffd2016-01-15 16:56:22 -08004655 struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn));
Andy Whitcroft29751f62005-06-23 00:08:00 -07004656 unsigned long end_pfn = start_pfn + size;
Dan Williams4b94ffd2016-01-15 16:56:22 -08004657 pg_data_t *pgdat = NODE_DATA(nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -07004658 unsigned long pfn;
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004659 unsigned long nr_initialised = 0;
Taku Izumi342332e2016-03-15 14:55:22 -07004660#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4661 struct memblock_region *r = NULL, *tmp;
4662#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004663
Hugh Dickins22b31ee2009-01-06 14:40:09 -08004664 if (highest_memmap_pfn < end_pfn - 1)
4665 highest_memmap_pfn = end_pfn - 1;
4666
Dan Williams4b94ffd2016-01-15 16:56:22 -08004667 /*
4668 * Honor reservation requested by the driver for this ZONE_DEVICE
4669 * memory
4670 */
4671 if (altmap && start_pfn == altmap->base_pfn)
4672 start_pfn += altmap->reserve;
4673
Greg Ungerercbe8dd42006-01-12 01:05:24 -08004674 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08004675 /*
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07004676 * There can be holes in boot-time mem_map[]s handed to this
4677 * function. They do not exist on hotplugged memory.
Dave Hansena2f3aa022007-01-10 23:15:30 -08004678 */
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07004679 if (context != MEMMAP_EARLY)
4680 goto not_early;
4681
4682 if (!early_pfn_valid(pfn))
4683 continue;
4684 if (!early_pfn_in_nid(pfn, nid))
4685 continue;
4686 if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised))
4687 break;
Taku Izumi342332e2016-03-15 14:55:22 -07004688
4689#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07004690 /*
4691 * If not mirrored_kernelcore and ZONE_MOVABLE exists, range
4692 * from zone_movable_pfn[nid] to end of each node should be
4693 * ZONE_MOVABLE not ZONE_NORMAL. skip it.
4694 */
4695 if (!mirrored_kernelcore && zone_movable_pfn[nid])
4696 if (zone == ZONE_NORMAL && pfn >= zone_movable_pfn[nid])
4697 continue;
Taku Izumi342332e2016-03-15 14:55:22 -07004698
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07004699 /*
4700 * Check given memblock attribute by firmware which can affect
4701 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
4702 * mirrored, it's an overlapped memmap init. skip it.
4703 */
4704 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
4705 if (!r || pfn >= memblock_region_memory_end_pfn(r)) {
4706 for_each_memblock(memory, tmp)
4707 if (pfn < memblock_region_memory_end_pfn(tmp))
4708 break;
4709 r = tmp;
Taku Izumi342332e2016-03-15 14:55:22 -07004710 }
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07004711 if (pfn >= memblock_region_memory_base_pfn(r) &&
4712 memblock_is_mirror(r)) {
4713 /* already initialized as NORMAL */
4714 pfn = memblock_region_memory_end_pfn(r);
4715 continue;
4716 }
Dave Hansena2f3aa022007-01-10 23:15:30 -08004717 }
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07004718#endif
Mel Gormanac5d2532015-06-30 14:57:20 -07004719
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07004720not_early:
Mel Gormanac5d2532015-06-30 14:57:20 -07004721 /*
4722 * Mark the block movable so that blocks are reserved for
4723 * movable at startup. This will force kernel allocations
4724 * to reserve their blocks rather than leaking throughout
4725 * the address space during boot when many long-lived
Mel Gorman974a7862015-11-06 16:28:34 -08004726 * kernel allocations are made.
Mel Gormanac5d2532015-06-30 14:57:20 -07004727 *
4728 * bitmap is created for zone's valid pfn range. but memmap
4729 * can be created for invalid pages (for alignment)
4730 * check here not to call set_pageblock_migratetype() against
4731 * pfn out of zone.
4732 */
4733 if (!(pfn & (pageblock_nr_pages - 1))) {
4734 struct page *page = pfn_to_page(pfn);
4735
4736 __init_single_page(page, pfn, zone, nid);
4737 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4738 } else {
4739 __init_single_pfn(pfn, zone, nid);
4740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004741 }
4742}
4743
Andi Kleen1e548de2008-02-04 22:29:26 -08004744static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004745{
Mel Gorman7aeb09f2014-06-04 16:10:21 -07004746 unsigned int order, t;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004747 for_each_migratetype_order(order, t) {
4748 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004749 zone->free_area[order].nr_free = 0;
4750 }
4751}
4752
4753#ifndef __HAVE_ARCH_MEMMAP_INIT
4754#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08004755 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004756#endif
4757
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004758static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004759{
David Howells3a6be872009-05-06 16:03:03 -07004760#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004761 int batch;
4762
4763 /*
4764 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07004765 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004766 *
4767 * OK, so we don't know how big the cache is. So guess.
4768 */
Jiang Liub40da042013-02-22 16:33:52 -08004769 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07004770 if (batch * PAGE_SIZE > 512 * 1024)
4771 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004772 batch /= 4; /* We effectively *= 4 below */
4773 if (batch < 1)
4774 batch = 1;
4775
4776 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004777 * Clamp the batch to a 2^n - 1 value. Having a power
4778 * of 2 value was found to be more likely to have
4779 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004780 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004781 * For example if 2 tasks are alternately allocating
4782 * batches of pages, one task can end up with a lot
4783 * of pages of one half of the possible page colors
4784 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004785 */
David Howells91552032009-05-06 16:03:02 -07004786 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004787
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004788 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004789
4790#else
4791 /* The deferral and batching of frees should be suppressed under NOMMU
4792 * conditions.
4793 *
4794 * The problem is that NOMMU needs to be able to allocate large chunks
4795 * of contiguous memory as there's no hardware page translation to
4796 * assemble apparent contiguous memory from discontiguous pages.
4797 *
4798 * Queueing large contiguous runs of pages for batching, however,
4799 * causes the pages to actually be freed in smaller chunks. As there
4800 * can be a significant delay between the individual batches being
4801 * recycled, this leads to the once large chunks of space being
4802 * fragmented and becoming unavailable for high-order allocations.
4803 */
4804 return 0;
4805#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004806}
4807
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004808/*
4809 * pcp->high and pcp->batch values are related and dependent on one another:
4810 * ->batch must never be higher then ->high.
4811 * The following function updates them in a safe manner without read side
4812 * locking.
4813 *
4814 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4815 * those fields changing asynchronously (acording the the above rule).
4816 *
4817 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4818 * outside of boot time (or some other assurance that no concurrent updaters
4819 * exist).
4820 */
4821static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4822 unsigned long batch)
4823{
4824 /* start with a fail safe value for batch */
4825 pcp->batch = 1;
4826 smp_wmb();
4827
4828 /* Update high, then batch, in order */
4829 pcp->high = high;
4830 smp_wmb();
4831
4832 pcp->batch = batch;
4833}
4834
Cody P Schafer36640332013-07-03 15:01:40 -07004835/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07004836static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4837{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004838 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07004839}
4840
Cody P Schafer88c90db2013-07-03 15:01:35 -07004841static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004842{
4843 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004844 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004845
Magnus Damm1c6fe942005-10-26 01:58:59 -07004846 memset(p, 0, sizeof(*p));
4847
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004848 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004849 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004850 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4851 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004852}
4853
Cody P Schafer88c90db2013-07-03 15:01:35 -07004854static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4855{
4856 pageset_init(p);
4857 pageset_set_batch(p, batch);
4858}
4859
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004860/*
Cody P Schafer36640332013-07-03 15:01:40 -07004861 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004862 * to the value high for the pageset p.
4863 */
Cody P Schafer36640332013-07-03 15:01:40 -07004864static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004865 unsigned long high)
4866{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004867 unsigned long batch = max(1UL, high / 4);
4868 if ((high / 4) > (PAGE_SHIFT * 8))
4869 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004870
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004871 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004872}
4873
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004874static void pageset_set_high_and_batch(struct zone *zone,
4875 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004876{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004877 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07004878 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004879 (zone->managed_pages /
4880 percpu_pagelist_fraction));
4881 else
4882 pageset_set_batch(pcp, zone_batchsize(zone));
4883}
4884
Cody P Schafer169f6c12013-07-03 15:01:41 -07004885static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4886{
4887 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4888
4889 pageset_init(pcp);
4890 pageset_set_high_and_batch(zone, pcp);
4891}
4892
Jiang Liu4ed7e022012-07-31 16:43:35 -07004893static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004894{
4895 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07004896 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004897 for_each_possible_cpu(cpu)
4898 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07004899}
4900
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004901/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004902 * Allocate per cpu pagesets and initialize them.
4903 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004904 */
Al Viro78d99552005-12-15 09:18:25 +00004905void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004906{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004907 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004908
Wu Fengguang319774e2010-05-24 14:32:49 -07004909 for_each_populated_zone(zone)
4910 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004911}
4912
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004913static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004914int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004915{
4916 int i;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004917 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004918
4919 /*
4920 * The per-page waitqueue mechanism uses hashed waitqueues
4921 * per zone.
4922 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004923 zone->wait_table_hash_nr_entries =
4924 wait_table_hash_nr_entries(zone_size_pages);
4925 zone->wait_table_bits =
4926 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004927 alloc_size = zone->wait_table_hash_nr_entries
4928 * sizeof(wait_queue_head_t);
4929
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004930 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004931 zone->wait_table = (wait_queue_head_t *)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004932 memblock_virt_alloc_node_nopanic(
4933 alloc_size, zone->zone_pgdat->node_id);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004934 } else {
4935 /*
4936 * This case means that a zone whose size was 0 gets new memory
4937 * via memory hot-add.
4938 * But it may be the case that a new node was hot-added. In
4939 * this case vmalloc() will not be able to use this new node's
4940 * memory - this wait_table must be initialized to use this new
4941 * node itself as well.
4942 * To use this new node's memory, further consideration will be
4943 * necessary.
4944 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004945 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004946 }
4947 if (!zone->wait_table)
4948 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004949
Pintu Kumarb8af2942013-09-11 14:20:34 -07004950 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004951 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004952
4953 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004954}
4955
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004956static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004957{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004958 /*
4959 * per cpu subsystem is not up at this point. The following code
4960 * relies on the ability of the linker to provide the
4961 * offset of a (static) per cpu variable into the per cpu area.
4962 */
4963 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004964
Xishi Qiub38a8722013-11-12 15:07:20 -08004965 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004966 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4967 zone->name, zone->present_pages,
4968 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004969}
4970
Jiang Liu4ed7e022012-07-31 16:43:35 -07004971int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004972 unsigned long zone_start_pfn,
Yaowei Baib171e402015-11-05 18:47:06 -08004973 unsigned long size)
Dave Hansened8ece22005-10-29 18:16:50 -07004974{
4975 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004976 int ret;
4977 ret = zone_wait_table_init(zone, size);
4978 if (ret)
4979 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004980 pgdat->nr_zones = zone_idx(zone) + 1;
4981
Dave Hansened8ece22005-10-29 18:16:50 -07004982 zone->zone_start_pfn = zone_start_pfn;
4983
Mel Gorman708614e2008-07-23 21:26:51 -07004984 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4985 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4986 pgdat->node_id,
4987 (unsigned long)zone_idx(zone),
4988 zone_start_pfn, (zone_start_pfn + size));
4989
Andi Kleen1e548de2008-02-04 22:29:26 -08004990 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004991
4992 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004993}
4994
Tejun Heo0ee332c2011-12-08 10:22:09 -08004995#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004996#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
Mel Gorman8a942fd2015-06-30 14:56:55 -07004997
Mel Gormanc7132162006-09-27 01:49:43 -07004998/*
4999 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
Mel Gormanc7132162006-09-27 01:49:43 -07005000 */
Mel Gorman8a942fd2015-06-30 14:56:55 -07005001int __meminit __early_pfn_to_nid(unsigned long pfn,
5002 struct mminit_pfnnid_cache *state)
Mel Gormanc7132162006-09-27 01:49:43 -07005003{
Tejun Heoc13291a2011-07-12 10:46:30 +02005004 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07005005 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07005006
Mel Gorman8a942fd2015-06-30 14:56:55 -07005007 if (state->last_start <= pfn && pfn < state->last_end)
5008 return state->last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005009
Yinghai Lue76b63f2013-09-11 14:22:17 -07005010 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
5011 if (nid != -1) {
Mel Gorman8a942fd2015-06-30 14:56:55 -07005012 state->last_start = start_pfn;
5013 state->last_end = end_pfn;
5014 state->last_nid = nid;
Yinghai Lue76b63f2013-09-11 14:22:17 -07005015 }
5016
5017 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005018}
5019#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5020
Mel Gormanc7132162006-09-27 01:49:43 -07005021/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08005022 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005023 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08005024 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07005025 *
Zhang Zhen7d018172014-06-04 16:10:53 -07005026 * If an architecture guarantees that all ranges registered contain no holes
5027 * and may be freed, this this function may be used instead of calling
5028 * memblock_free_early_nid() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07005029 */
Tejun Heoc13291a2011-07-12 10:46:30 +02005030void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07005031{
Tejun Heoc13291a2011-07-12 10:46:30 +02005032 unsigned long start_pfn, end_pfn;
5033 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005034
Tejun Heoc13291a2011-07-12 10:46:30 +02005035 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
5036 start_pfn = min(start_pfn, max_low_pfn);
5037 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005038
Tejun Heoc13291a2011-07-12 10:46:30 +02005039 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005040 memblock_free_early_nid(PFN_PHYS(start_pfn),
5041 (end_pfn - start_pfn) << PAGE_SHIFT,
5042 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005043 }
5044}
5045
5046/**
5047 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005048 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07005049 *
Zhang Zhen7d018172014-06-04 16:10:53 -07005050 * If an architecture guarantees that all ranges registered contain no holes and may
5051 * be freed, this function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07005052 */
5053void __init sparse_memory_present_with_active_regions(int nid)
5054{
Tejun Heoc13291a2011-07-12 10:46:30 +02005055 unsigned long start_pfn, end_pfn;
5056 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005057
Tejun Heoc13291a2011-07-12 10:46:30 +02005058 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
5059 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005060}
5061
5062/**
5063 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005064 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5065 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5066 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07005067 *
5068 * It returns the start and end page frame of a node based on information
Zhang Zhen7d018172014-06-04 16:10:53 -07005069 * provided by memblock_set_node(). If called for a node
Mel Gormanc7132162006-09-27 01:49:43 -07005070 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005071 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07005072 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005073void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005074 unsigned long *start_pfn, unsigned long *end_pfn)
5075{
Tejun Heoc13291a2011-07-12 10:46:30 +02005076 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005077 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02005078
Mel Gormanc7132162006-09-27 01:49:43 -07005079 *start_pfn = -1UL;
5080 *end_pfn = 0;
5081
Tejun Heoc13291a2011-07-12 10:46:30 +02005082 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5083 *start_pfn = min(*start_pfn, this_start_pfn);
5084 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005085 }
5086
Christoph Lameter633c0662007-10-16 01:25:37 -07005087 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07005088 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005089}
5090
5091/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07005092 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5093 * assumption is made that zones within a node are ordered in monotonic
5094 * increasing memory addresses so that the "highest" populated zone is used
5095 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005096static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005097{
5098 int zone_index;
5099 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5100 if (zone_index == ZONE_MOVABLE)
5101 continue;
5102
5103 if (arch_zone_highest_possible_pfn[zone_index] >
5104 arch_zone_lowest_possible_pfn[zone_index])
5105 break;
5106 }
5107
5108 VM_BUG_ON(zone_index == -1);
5109 movable_zone = zone_index;
5110}
5111
5112/*
5113 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03005114 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005115 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5116 * in each node depending on the size of each node and how evenly kernelcore
5117 * is distributed. This helper function adjusts the zone ranges
5118 * provided by the architecture for a given node by using the end of the
5119 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5120 * zones within a node are in order of monotonic increases memory addresses
5121 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005122static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005123 unsigned long zone_type,
5124 unsigned long node_start_pfn,
5125 unsigned long node_end_pfn,
5126 unsigned long *zone_start_pfn,
5127 unsigned long *zone_end_pfn)
5128{
5129 /* Only adjust if ZONE_MOVABLE is on this node */
5130 if (zone_movable_pfn[nid]) {
5131 /* Size ZONE_MOVABLE */
5132 if (zone_type == ZONE_MOVABLE) {
5133 *zone_start_pfn = zone_movable_pfn[nid];
5134 *zone_end_pfn = min(node_end_pfn,
5135 arch_zone_highest_possible_pfn[movable_zone]);
5136
Mel Gorman2a1e2742007-07-17 04:03:12 -07005137 /* Check if this whole range is within ZONE_MOVABLE */
5138 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5139 *zone_start_pfn = *zone_end_pfn;
5140 }
5141}
5142
5143/*
Mel Gormanc7132162006-09-27 01:49:43 -07005144 * Return the number of pages a zone spans in a node, including holes
5145 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5146 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005147static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005148 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005149 unsigned long node_start_pfn,
5150 unsigned long node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005151 unsigned long *zone_start_pfn,
5152 unsigned long *zone_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005153 unsigned long *ignored)
5154{
Xishi Qiub5685e92015-09-08 15:04:16 -07005155 /* When hotadd a new node from cpu_up(), the node should be empty */
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005156 if (!node_start_pfn && !node_end_pfn)
5157 return 0;
5158
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005159 /* Get the start and end of the zone */
Taku Izumid91749c2016-03-15 14:55:18 -07005160 *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5161 *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07005162 adjust_zone_range_for_zone_movable(nid, zone_type,
5163 node_start_pfn, node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005164 zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005165
5166 /* Check that this node has pages within the zone's required range */
Taku Izumid91749c2016-03-15 14:55:18 -07005167 if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07005168 return 0;
5169
5170 /* Move the zone boundaries inside the node if necessary */
Taku Izumid91749c2016-03-15 14:55:18 -07005171 *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
5172 *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005173
5174 /* Return the spanned pages */
Taku Izumid91749c2016-03-15 14:55:18 -07005175 return *zone_end_pfn - *zone_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005176}
5177
5178/*
5179 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005180 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07005181 */
Yinghai Lu32996252009-12-15 17:59:02 -08005182unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005183 unsigned long range_start_pfn,
5184 unsigned long range_end_pfn)
5185{
Tejun Heo96e907d2011-07-12 10:46:29 +02005186 unsigned long nr_absent = range_end_pfn - range_start_pfn;
5187 unsigned long start_pfn, end_pfn;
5188 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07005189
Tejun Heo96e907d2011-07-12 10:46:29 +02005190 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5191 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5192 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5193 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005194 }
Tejun Heo96e907d2011-07-12 10:46:29 +02005195 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07005196}
5197
5198/**
5199 * absent_pages_in_range - Return number of page frames in holes within a range
5200 * @start_pfn: The start PFN to start searching for holes
5201 * @end_pfn: The end PFN to stop searching for holes
5202 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005203 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07005204 */
5205unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5206 unsigned long end_pfn)
5207{
5208 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5209}
5210
5211/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005212static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005213 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005214 unsigned long node_start_pfn,
5215 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005216 unsigned long *ignored)
5217{
Tejun Heo96e907d2011-07-12 10:46:29 +02005218 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5219 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07005220 unsigned long zone_start_pfn, zone_end_pfn;
Taku Izumi342332e2016-03-15 14:55:22 -07005221 unsigned long nr_absent;
Mel Gorman9c7cd682006-09-27 01:49:58 -07005222
Xishi Qiub5685e92015-09-08 15:04:16 -07005223 /* When hotadd a new node from cpu_up(), the node should be empty */
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005224 if (!node_start_pfn && !node_end_pfn)
5225 return 0;
5226
Tejun Heo96e907d2011-07-12 10:46:29 +02005227 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5228 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07005229
Mel Gorman2a1e2742007-07-17 04:03:12 -07005230 adjust_zone_range_for_zone_movable(nid, zone_type,
5231 node_start_pfn, node_end_pfn,
5232 &zone_start_pfn, &zone_end_pfn);
Taku Izumi342332e2016-03-15 14:55:22 -07005233 nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5234
5235 /*
5236 * ZONE_MOVABLE handling.
5237 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5238 * and vice versa.
5239 */
5240 if (zone_movable_pfn[nid]) {
5241 if (mirrored_kernelcore) {
5242 unsigned long start_pfn, end_pfn;
5243 struct memblock_region *r;
5244
5245 for_each_memblock(memory, r) {
5246 start_pfn = clamp(memblock_region_memory_base_pfn(r),
5247 zone_start_pfn, zone_end_pfn);
5248 end_pfn = clamp(memblock_region_memory_end_pfn(r),
5249 zone_start_pfn, zone_end_pfn);
5250
5251 if (zone_type == ZONE_MOVABLE &&
5252 memblock_is_mirror(r))
5253 nr_absent += end_pfn - start_pfn;
5254
5255 if (zone_type == ZONE_NORMAL &&
5256 !memblock_is_mirror(r))
5257 nr_absent += end_pfn - start_pfn;
5258 }
5259 } else {
5260 if (zone_type == ZONE_NORMAL)
5261 nr_absent += node_end_pfn - zone_movable_pfn[nid];
5262 }
5263 }
5264
5265 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07005266}
Mel Gorman0e0b8642006-09-27 01:49:56 -07005267
Tejun Heo0ee332c2011-12-08 10:22:09 -08005268#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005269static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005270 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005271 unsigned long node_start_pfn,
5272 unsigned long node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005273 unsigned long *zone_start_pfn,
5274 unsigned long *zone_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005275 unsigned long *zones_size)
5276{
Taku Izumid91749c2016-03-15 14:55:18 -07005277 unsigned int zone;
5278
5279 *zone_start_pfn = node_start_pfn;
5280 for (zone = 0; zone < zone_type; zone++)
5281 *zone_start_pfn += zones_size[zone];
5282
5283 *zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
5284
Mel Gormanc7132162006-09-27 01:49:43 -07005285 return zones_size[zone_type];
5286}
5287
Paul Mundt6ea6e682007-07-15 23:38:20 -07005288static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005289 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005290 unsigned long node_start_pfn,
5291 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005292 unsigned long *zholes_size)
5293{
5294 if (!zholes_size)
5295 return 0;
5296
5297 return zholes_size[zone_type];
5298}
Yinghai Lu20e69262013-03-01 14:51:27 -08005299
Tejun Heo0ee332c2011-12-08 10:22:09 -08005300#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005301
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005302static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005303 unsigned long node_start_pfn,
5304 unsigned long node_end_pfn,
5305 unsigned long *zones_size,
5306 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07005307{
Gu Zhengfebd5942015-06-24 16:57:02 -07005308 unsigned long realtotalpages = 0, totalpages = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005309 enum zone_type i;
5310
Gu Zhengfebd5942015-06-24 16:57:02 -07005311 for (i = 0; i < MAX_NR_ZONES; i++) {
5312 struct zone *zone = pgdat->node_zones + i;
Taku Izumid91749c2016-03-15 14:55:18 -07005313 unsigned long zone_start_pfn, zone_end_pfn;
Gu Zhengfebd5942015-06-24 16:57:02 -07005314 unsigned long size, real_size;
Mel Gormanc7132162006-09-27 01:49:43 -07005315
Gu Zhengfebd5942015-06-24 16:57:02 -07005316 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5317 node_start_pfn,
5318 node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005319 &zone_start_pfn,
5320 &zone_end_pfn,
Gu Zhengfebd5942015-06-24 16:57:02 -07005321 zones_size);
5322 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005323 node_start_pfn, node_end_pfn,
5324 zholes_size);
Taku Izumid91749c2016-03-15 14:55:18 -07005325 if (size)
5326 zone->zone_start_pfn = zone_start_pfn;
5327 else
5328 zone->zone_start_pfn = 0;
Gu Zhengfebd5942015-06-24 16:57:02 -07005329 zone->spanned_pages = size;
5330 zone->present_pages = real_size;
5331
5332 totalpages += size;
5333 realtotalpages += real_size;
5334 }
5335
5336 pgdat->node_spanned_pages = totalpages;
Mel Gormanc7132162006-09-27 01:49:43 -07005337 pgdat->node_present_pages = realtotalpages;
5338 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5339 realtotalpages);
5340}
5341
Mel Gorman835c1342007-10-16 01:25:47 -07005342#ifndef CONFIG_SPARSEMEM
5343/*
5344 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07005345 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5346 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07005347 * round what is now in bits to nearest long in bits, then return it in
5348 * bytes.
5349 */
Linus Torvalds7c455122013-02-18 09:58:02 -08005350static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005351{
5352 unsigned long usemapsize;
5353
Linus Torvalds7c455122013-02-18 09:58:02 -08005354 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005355 usemapsize = roundup(zonesize, pageblock_nr_pages);
5356 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07005357 usemapsize *= NR_PAGEBLOCK_BITS;
5358 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5359
5360 return usemapsize / 8;
5361}
5362
5363static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08005364 struct zone *zone,
5365 unsigned long zone_start_pfn,
5366 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005367{
Linus Torvalds7c455122013-02-18 09:58:02 -08005368 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07005369 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08005370 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005371 zone->pageblock_flags =
5372 memblock_virt_alloc_node_nopanic(usemapsize,
5373 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07005374}
5375#else
Linus Torvalds7c455122013-02-18 09:58:02 -08005376static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5377 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07005378#endif /* CONFIG_SPARSEMEM */
5379
Mel Gormand9c23402007-10-16 01:26:01 -07005380#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08005381
Mel Gormand9c23402007-10-16 01:26:01 -07005382/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07005383void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07005384{
Andrew Morton955c1cd2012-05-29 15:06:31 -07005385 unsigned int order;
5386
Mel Gormand9c23402007-10-16 01:26:01 -07005387 /* Check that pageblock_nr_pages has not already been setup */
5388 if (pageblock_order)
5389 return;
5390
Andrew Morton955c1cd2012-05-29 15:06:31 -07005391 if (HPAGE_SHIFT > PAGE_SHIFT)
5392 order = HUGETLB_PAGE_ORDER;
5393 else
5394 order = MAX_ORDER - 1;
5395
Mel Gormand9c23402007-10-16 01:26:01 -07005396 /*
5397 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07005398 * This value may be variable depending on boot parameters on IA64 and
5399 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07005400 */
5401 pageblock_order = order;
5402}
5403#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5404
Mel Gormanba72cb82007-11-28 16:21:13 -08005405/*
5406 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07005407 * is unused as pageblock_order is set at compile-time. See
5408 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5409 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08005410 */
Chen Gang15ca2202013-09-11 14:20:27 -07005411void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08005412{
Mel Gormanba72cb82007-11-28 16:21:13 -08005413}
Mel Gormand9c23402007-10-16 01:26:01 -07005414
5415#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5416
Jiang Liu01cefae2012-12-12 13:52:19 -08005417static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5418 unsigned long present_pages)
5419{
5420 unsigned long pages = spanned_pages;
5421
5422 /*
5423 * Provide a more accurate estimation if there are holes within
5424 * the zone and SPARSEMEM is in use. If there are holes within the
5425 * zone, each populated memory region may cost us one or two extra
5426 * memmap pages due to alignment because memmap pages for each
5427 * populated regions may not naturally algined on page boundary.
5428 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5429 */
5430 if (spanned_pages > present_pages + (present_pages >> 4) &&
5431 IS_ENABLED(CONFIG_SPARSEMEM))
5432 pages = present_pages;
5433
5434 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5435}
5436
Linus Torvalds1da177e2005-04-16 15:20:36 -07005437/*
5438 * Set up the zone data structures:
5439 * - mark all pages reserved
5440 * - mark all memory queues empty
5441 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07005442 *
5443 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005444 */
Wei Yang7f3eb552015-09-08 14:59:50 -07005445static void __paginginit free_area_init_core(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005446{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07005447 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07005448 int nid = pgdat->node_id;
Yasunori Goto718127c2006-06-23 02:03:10 -07005449 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005450
Dave Hansen208d54e2005-10-29 18:16:52 -07005451 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01005452#ifdef CONFIG_NUMA_BALANCING
5453 spin_lock_init(&pgdat->numabalancing_migrate_lock);
5454 pgdat->numabalancing_migrate_nr_pages = 0;
5455 pgdat->numabalancing_migrate_next_window = jiffies;
5456#endif
Kirill A. Shutemova3d0a9182016-02-02 16:57:08 -08005457#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5458 spin_lock_init(&pgdat->split_queue_lock);
5459 INIT_LIST_HEAD(&pgdat->split_queue);
5460 pgdat->split_queue_len = 0;
5461#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005462 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07005463 init_waitqueue_head(&pgdat->pfmemalloc_wait);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07005464#ifdef CONFIG_COMPACTION
5465 init_waitqueue_head(&pgdat->kcompactd_wait);
5466#endif
Joonsoo Kimeefa864b2014-12-12 16:55:46 -08005467 pgdat_page_ext_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01005468
Linus Torvalds1da177e2005-04-16 15:20:36 -07005469 for (j = 0; j < MAX_NR_ZONES; j++) {
5470 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08005471 unsigned long size, realsize, freesize, memmap_pages;
Taku Izumid91749c2016-03-15 14:55:18 -07005472 unsigned long zone_start_pfn = zone->zone_start_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005473
Gu Zhengfebd5942015-06-24 16:57:02 -07005474 size = zone->spanned_pages;
5475 realsize = freesize = zone->present_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005476
Mel Gorman0e0b8642006-09-27 01:49:56 -07005477 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08005478 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07005479 * is used by this zone for memmap. This affects the watermark
5480 * and per-cpu initialisations
5481 */
Jiang Liu01cefae2012-12-12 13:52:19 -08005482 memmap_pages = calc_memmap_size(size, realsize);
Zhong Hongboba914f42014-12-12 16:56:21 -08005483 if (!is_highmem_idx(j)) {
5484 if (freesize >= memmap_pages) {
5485 freesize -= memmap_pages;
5486 if (memmap_pages)
5487 printk(KERN_DEBUG
5488 " %s zone: %lu pages used for memmap\n",
5489 zone_names[j], memmap_pages);
5490 } else
Joe Perches11705322016-03-17 14:19:50 -07005491 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
Zhong Hongboba914f42014-12-12 16:56:21 -08005492 zone_names[j], memmap_pages, freesize);
5493 }
Mel Gorman0e0b8642006-09-27 01:49:56 -07005494
Christoph Lameter62672762007-02-10 01:43:07 -08005495 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08005496 if (j == 0 && freesize > dma_reserve) {
5497 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07005498 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08005499 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07005500 }
5501
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07005502 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08005503 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08005504 /* Charge for highmem memmap if there are enough kernel pages */
5505 else if (nr_kernel_pages > memmap_pages * 2)
5506 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08005507 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005508
Jiang Liu9feedc92012-12-12 13:52:12 -08005509 /*
5510 * Set an approximate value for lowmem here, it will be adjusted
5511 * when the bootmem allocator frees pages into the buddy system.
5512 * And all highmem pages will be managed by the buddy system.
5513 */
5514 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07005515#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07005516 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08005517 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07005518 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08005519 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07005520#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005521 zone->name = zone_names[j];
5522 spin_lock_init(&zone->lock);
5523 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07005524 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005525 zone->zone_pgdat = pgdat;
Dave Hansened8ece22005-10-29 18:16:50 -07005526 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005527
5528 /* For bootup, initialized properly in watermark setup */
5529 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
5530
Hugh Dickinsbea8c152012-11-16 14:14:54 -08005531 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005532 if (!size)
5533 continue;
5534
Andrew Morton955c1cd2012-05-29 15:06:31 -07005535 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08005536 setup_usemap(pgdat, zone, zone_start_pfn, size);
Yaowei Baib171e402015-11-05 18:47:06 -08005537 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
Yasunori Goto718127c2006-06-23 02:03:10 -07005538 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07005539 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005540 }
5541}
5542
Sam Ravnborg577a32f2007-05-17 23:29:25 +02005543static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005544{
Tony Luckb0aeba72015-11-10 10:09:47 -08005545 unsigned long __maybe_unused start = 0;
Laura Abbotta1c34a32015-11-05 18:48:46 -08005546 unsigned long __maybe_unused offset = 0;
5547
Linus Torvalds1da177e2005-04-16 15:20:36 -07005548 /* Skip empty nodes */
5549 if (!pgdat->node_spanned_pages)
5550 return;
5551
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005552#ifdef CONFIG_FLAT_NODE_MEM_MAP
Tony Luckb0aeba72015-11-10 10:09:47 -08005553 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
5554 offset = pgdat->node_start_pfn - start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005555 /* ia64 gets its own node_mem_map, before this, without bootmem */
5556 if (!pgdat->node_mem_map) {
Tony Luckb0aeba72015-11-10 10:09:47 -08005557 unsigned long size, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005558 struct page *map;
5559
Bob Piccoe984bb42006-05-20 15:00:31 -07005560 /*
5561 * The zone's endpoints aren't required to be MAX_ORDER
5562 * aligned but the node_mem_map endpoints must be in order
5563 * for the buddy allocator to function correctly.
5564 */
Cody P Schafer108bcc92013-02-22 16:35:23 -08005565 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07005566 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5567 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07005568 map = alloc_remap(pgdat->node_id, size);
5569 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005570 map = memblock_virt_alloc_node_nopanic(size,
5571 pgdat->node_id);
Laura Abbotta1c34a32015-11-05 18:48:46 -08005572 pgdat->node_mem_map = map + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005573 }
Roman Zippel12d810c2007-05-31 00:40:54 -07005574#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07005575 /*
5576 * With no DISCONTIG, the global mem_map is just set as node 0's
5577 */
Mel Gormanc7132162006-09-27 01:49:43 -07005578 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005579 mem_map = NODE_DATA(0)->node_mem_map;
Laura Abbotta1c34a32015-11-05 18:48:46 -08005580#if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
Mel Gormanc7132162006-09-27 01:49:43 -07005581 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Laura Abbotta1c34a32015-11-05 18:48:46 -08005582 mem_map -= offset;
Tejun Heo0ee332c2011-12-08 10:22:09 -08005583#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005584 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005585#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005586#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005587}
5588
Johannes Weiner9109fb72008-07-23 21:27:20 -07005589void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5590 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005591{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005592 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005593 unsigned long start_pfn = 0;
5594 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07005595
Minchan Kim88fdf752012-07-31 16:46:14 -07005596 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07005597 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07005598
Mel Gorman3a80a7f2015-06-30 14:57:02 -07005599 reset_deferred_meminit(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005600 pgdat->node_id = nid;
5601 pgdat->node_start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005602#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5603 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
Juergen Gross8d29e182015-02-11 15:26:01 -08005604 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
Zhen Lei4ada0c52015-09-08 15:04:19 -07005605 (u64)start_pfn << PAGE_SHIFT,
5606 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
Taku Izumid91749c2016-03-15 14:55:18 -07005607#else
5608 start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005609#endif
5610 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5611 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005612
5613 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07005614#ifdef CONFIG_FLAT_NODE_MEM_MAP
5615 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5616 nid, (unsigned long)pgdat,
5617 (unsigned long)pgdat->node_mem_map);
5618#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005619
Wei Yang7f3eb552015-09-08 14:59:50 -07005620 free_area_init_core(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005621}
5622
Tejun Heo0ee332c2011-12-08 10:22:09 -08005623#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07005624
5625#if MAX_NUMNODES > 1
5626/*
5627 * Figure out the number of possible node ids.
5628 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07005629void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07005630{
Wei Yang904a9552015-09-08 14:59:48 -07005631 unsigned int highest;
Miklos Szeredi418508c2007-05-23 13:57:55 -07005632
Wei Yang904a9552015-09-08 14:59:48 -07005633 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
Miklos Szeredi418508c2007-05-23 13:57:55 -07005634 nr_node_ids = highest + 1;
5635}
Miklos Szeredi418508c2007-05-23 13:57:55 -07005636#endif
5637
Mel Gormanc7132162006-09-27 01:49:43 -07005638/**
Tejun Heo1e019792011-07-12 09:45:34 +02005639 * node_map_pfn_alignment - determine the maximum internode alignment
5640 *
5641 * This function should be called after node map is populated and sorted.
5642 * It calculates the maximum power of two alignment which can distinguish
5643 * all the nodes.
5644 *
5645 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5646 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5647 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5648 * shifted, 1GiB is enough and this function will indicate so.
5649 *
5650 * This is used to test whether pfn -> nid mapping of the chosen memory
5651 * model has fine enough granularity to avoid incorrect mapping for the
5652 * populated node map.
5653 *
5654 * Returns the determined alignment in pfn's. 0 if there is no alignment
5655 * requirement (single node).
5656 */
5657unsigned long __init node_map_pfn_alignment(void)
5658{
5659 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005660 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02005661 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02005662 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02005663
Tejun Heoc13291a2011-07-12 10:46:30 +02005664 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02005665 if (!start || last_nid < 0 || last_nid == nid) {
5666 last_nid = nid;
5667 last_end = end;
5668 continue;
5669 }
5670
5671 /*
5672 * Start with a mask granular enough to pin-point to the
5673 * start pfn and tick off bits one-by-one until it becomes
5674 * too coarse to separate the current node from the last.
5675 */
5676 mask = ~((1 << __ffs(start)) - 1);
5677 while (mask && last_end <= (start & (mask << 1)))
5678 mask <<= 1;
5679
5680 /* accumulate all internode masks */
5681 accl_mask |= mask;
5682 }
5683
5684 /* convert mask to number of pages */
5685 return ~accl_mask + 1;
5686}
5687
Mel Gormana6af2bc2007-02-10 01:42:57 -08005688/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005689static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07005690{
Mel Gormana6af2bc2007-02-10 01:42:57 -08005691 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02005692 unsigned long start_pfn;
5693 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00005694
Tejun Heoc13291a2011-07-12 10:46:30 +02005695 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5696 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005697
Mel Gormana6af2bc2007-02-10 01:42:57 -08005698 if (min_pfn == ULONG_MAX) {
Joe Perches11705322016-03-17 14:19:50 -07005699 pr_warn("Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08005700 return 0;
5701 }
5702
5703 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005704}
5705
5706/**
5707 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5708 *
5709 * It returns the minimum PFN based on information provided via
Zhang Zhen7d018172014-06-04 16:10:53 -07005710 * memblock_set_node().
Mel Gormanc7132162006-09-27 01:49:43 -07005711 */
5712unsigned long __init find_min_pfn_with_active_regions(void)
5713{
5714 return find_min_pfn_for_node(MAX_NUMNODES);
5715}
5716
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005717/*
5718 * early_calculate_totalpages()
5719 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005720 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005721 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07005722static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07005723{
Mel Gorman7e63efe2007-07-17 04:03:15 -07005724 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005725 unsigned long start_pfn, end_pfn;
5726 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005727
Tejun Heoc13291a2011-07-12 10:46:30 +02005728 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5729 unsigned long pages = end_pfn - start_pfn;
5730
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005731 totalpages += pages;
5732 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005733 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005734 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07005735 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005736}
5737
Mel Gorman2a1e2742007-07-17 04:03:12 -07005738/*
5739 * Find the PFN the Movable zone begins in each node. Kernel memory
5740 * is spread evenly between nodes as long as the nodes have enough
5741 * memory. When they don't, some nodes will have more kernelcore than
5742 * others
5743 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07005744static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005745{
5746 int i, nid;
5747 unsigned long usable_startpfn;
5748 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07005749 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005750 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005751 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005752 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07005753 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005754
5755 /* Need to find movable_zone earlier when movable_node is specified. */
5756 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07005757
Mel Gorman7e63efe2007-07-17 04:03:15 -07005758 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005759 * If movable_node is specified, ignore kernelcore and movablecore
5760 * options.
5761 */
5762 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07005763 for_each_memblock(memory, r) {
5764 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005765 continue;
5766
Emil Medve136199f2014-04-07 15:37:52 -07005767 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005768
Emil Medve136199f2014-04-07 15:37:52 -07005769 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005770 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5771 min(usable_startpfn, zone_movable_pfn[nid]) :
5772 usable_startpfn;
5773 }
5774
5775 goto out2;
5776 }
5777
5778 /*
Taku Izumi342332e2016-03-15 14:55:22 -07005779 * If kernelcore=mirror is specified, ignore movablecore option
5780 */
5781 if (mirrored_kernelcore) {
5782 bool mem_below_4gb_not_mirrored = false;
5783
5784 for_each_memblock(memory, r) {
5785 if (memblock_is_mirror(r))
5786 continue;
5787
5788 nid = r->nid;
5789
5790 usable_startpfn = memblock_region_memory_base_pfn(r);
5791
5792 if (usable_startpfn < 0x100000) {
5793 mem_below_4gb_not_mirrored = true;
5794 continue;
5795 }
5796
5797 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5798 min(usable_startpfn, zone_movable_pfn[nid]) :
5799 usable_startpfn;
5800 }
5801
5802 if (mem_below_4gb_not_mirrored)
5803 pr_warn("This configuration results in unmirrored kernel memory.");
5804
5805 goto out2;
5806 }
5807
5808 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005809 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07005810 * kernelcore that corresponds so that memory usable for
5811 * any allocation type is evenly spread. If both kernelcore
5812 * and movablecore are specified, then the value of kernelcore
5813 * will be used for required_kernelcore if it's greater than
5814 * what movablecore would have allowed.
5815 */
5816 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07005817 unsigned long corepages;
5818
5819 /*
5820 * Round-up so that ZONE_MOVABLE is at least as large as what
5821 * was requested by the user
5822 */
5823 required_movablecore =
5824 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
Xishi Qiu9fd745d2015-11-05 18:48:11 -08005825 required_movablecore = min(totalpages, required_movablecore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005826 corepages = totalpages - required_movablecore;
5827
5828 required_kernelcore = max(required_kernelcore, corepages);
5829 }
5830
Xishi Qiubde304b2015-11-05 18:48:56 -08005831 /*
5832 * If kernelcore was not specified or kernelcore size is larger
5833 * than totalpages, there is no ZONE_MOVABLE.
5834 */
5835 if (!required_kernelcore || required_kernelcore >= totalpages)
Yinghai Lu66918dc2009-06-30 11:41:37 -07005836 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005837
5838 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005839 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5840
5841restart:
5842 /* Spread kernelcore memory as evenly as possible throughout nodes */
5843 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005844 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02005845 unsigned long start_pfn, end_pfn;
5846
Mel Gorman2a1e2742007-07-17 04:03:12 -07005847 /*
5848 * Recalculate kernelcore_node if the division per node
5849 * now exceeds what is necessary to satisfy the requested
5850 * amount of memory for the kernel
5851 */
5852 if (required_kernelcore < kernelcore_node)
5853 kernelcore_node = required_kernelcore / usable_nodes;
5854
5855 /*
5856 * As the map is walked, we track how much memory is usable
5857 * by the kernel using kernelcore_remaining. When it is
5858 * 0, the rest of the node is usable by ZONE_MOVABLE
5859 */
5860 kernelcore_remaining = kernelcore_node;
5861
5862 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02005863 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005864 unsigned long size_pages;
5865
Tejun Heoc13291a2011-07-12 10:46:30 +02005866 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005867 if (start_pfn >= end_pfn)
5868 continue;
5869
5870 /* Account for what is only usable for kernelcore */
5871 if (start_pfn < usable_startpfn) {
5872 unsigned long kernel_pages;
5873 kernel_pages = min(end_pfn, usable_startpfn)
5874 - start_pfn;
5875
5876 kernelcore_remaining -= min(kernel_pages,
5877 kernelcore_remaining);
5878 required_kernelcore -= min(kernel_pages,
5879 required_kernelcore);
5880
5881 /* Continue if range is now fully accounted */
5882 if (end_pfn <= usable_startpfn) {
5883
5884 /*
5885 * Push zone_movable_pfn to the end so
5886 * that if we have to rebalance
5887 * kernelcore across nodes, we will
5888 * not double account here
5889 */
5890 zone_movable_pfn[nid] = end_pfn;
5891 continue;
5892 }
5893 start_pfn = usable_startpfn;
5894 }
5895
5896 /*
5897 * The usable PFN range for ZONE_MOVABLE is from
5898 * start_pfn->end_pfn. Calculate size_pages as the
5899 * number of pages used as kernelcore
5900 */
5901 size_pages = end_pfn - start_pfn;
5902 if (size_pages > kernelcore_remaining)
5903 size_pages = kernelcore_remaining;
5904 zone_movable_pfn[nid] = start_pfn + size_pages;
5905
5906 /*
5907 * Some kernelcore has been met, update counts and
5908 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07005909 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005910 */
5911 required_kernelcore -= min(required_kernelcore,
5912 size_pages);
5913 kernelcore_remaining -= size_pages;
5914 if (!kernelcore_remaining)
5915 break;
5916 }
5917 }
5918
5919 /*
5920 * If there is still required_kernelcore, we do another pass with one
5921 * less node in the count. This will push zone_movable_pfn[nid] further
5922 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07005923 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005924 */
5925 usable_nodes--;
5926 if (usable_nodes && required_kernelcore > usable_nodes)
5927 goto restart;
5928
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005929out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07005930 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5931 for (nid = 0; nid < MAX_NUMNODES; nid++)
5932 zone_movable_pfn[nid] =
5933 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005934
Yinghai Lu20e69262013-03-01 14:51:27 -08005935out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005936 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005937 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005938}
5939
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005940/* Any regular or high memory on that node ? */
5941static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005942{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005943 enum zone_type zone_type;
5944
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005945 if (N_MEMORY == N_NORMAL_MEMORY)
5946 return;
5947
5948 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005949 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08005950 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005951 node_set_state(nid, N_HIGH_MEMORY);
5952 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5953 zone_type <= ZONE_NORMAL)
5954 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005955 break;
5956 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005957 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005958}
5959
Mel Gormanc7132162006-09-27 01:49:43 -07005960/**
5961 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005962 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005963 *
5964 * This will call free_area_init_node() for each active node in the system.
Zhang Zhen7d018172014-06-04 16:10:53 -07005965 * Using the page ranges provided by memblock_set_node(), the size of each
Mel Gormanc7132162006-09-27 01:49:43 -07005966 * zone in each node and their holes is calculated. If the maximum PFN
5967 * between two adjacent zones match, it is assumed that the zone is empty.
5968 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5969 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5970 * starts where the previous one ended. For example, ZONE_DMA32 starts
5971 * at arch_max_dma_pfn.
5972 */
5973void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5974{
Tejun Heoc13291a2011-07-12 10:46:30 +02005975 unsigned long start_pfn, end_pfn;
5976 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005977
Mel Gormanc7132162006-09-27 01:49:43 -07005978 /* Record where the zone boundaries are */
5979 memset(arch_zone_lowest_possible_pfn, 0,
5980 sizeof(arch_zone_lowest_possible_pfn));
5981 memset(arch_zone_highest_possible_pfn, 0,
5982 sizeof(arch_zone_highest_possible_pfn));
5983 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5984 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5985 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005986 if (i == ZONE_MOVABLE)
5987 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005988 arch_zone_lowest_possible_pfn[i] =
5989 arch_zone_highest_possible_pfn[i-1];
5990 arch_zone_highest_possible_pfn[i] =
5991 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5992 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005993 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5994 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5995
5996 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5997 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005998 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005999
Mel Gormanc7132162006-09-27 01:49:43 -07006000 /* Print out the zone ranges */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006001 pr_info("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07006002 for (i = 0; i < MAX_NR_ZONES; i++) {
6003 if (i == ZONE_MOVABLE)
6004 continue;
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006005 pr_info(" %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08006006 if (arch_zone_lowest_possible_pfn[i] ==
6007 arch_zone_highest_possible_pfn[i])
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006008 pr_cont("empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08006009 else
Juergen Gross8d29e182015-02-11 15:26:01 -08006010 pr_cont("[mem %#018Lx-%#018Lx]\n",
6011 (u64)arch_zone_lowest_possible_pfn[i]
6012 << PAGE_SHIFT,
6013 ((u64)arch_zone_highest_possible_pfn[i]
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07006014 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07006015 }
6016
6017 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006018 pr_info("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07006019 for (i = 0; i < MAX_NUMNODES; i++) {
6020 if (zone_movable_pfn[i])
Juergen Gross8d29e182015-02-11 15:26:01 -08006021 pr_info(" Node %d: %#018Lx\n", i,
6022 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07006023 }
Mel Gormanc7132162006-09-27 01:49:43 -07006024
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07006025 /* Print out the early node map */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006026 pr_info("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02006027 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Juergen Gross8d29e182015-02-11 15:26:01 -08006028 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
6029 (u64)start_pfn << PAGE_SHIFT,
6030 ((u64)end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07006031
6032 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07006033 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08006034 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07006035 for_each_online_node(nid) {
6036 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07006037 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07006038 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006039
6040 /* Any memory on that node */
6041 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006042 node_set_state(nid, N_MEMORY);
6043 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07006044 }
6045}
Mel Gorman2a1e2742007-07-17 04:03:12 -07006046
Mel Gorman7e63efe2007-07-17 04:03:15 -07006047static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07006048{
6049 unsigned long long coremem;
6050 if (!p)
6051 return -EINVAL;
6052
6053 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07006054 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07006055
Mel Gorman7e63efe2007-07-17 04:03:15 -07006056 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07006057 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
6058
6059 return 0;
6060}
Mel Gormaned7ed362007-07-17 04:03:14 -07006061
Mel Gorman7e63efe2007-07-17 04:03:15 -07006062/*
6063 * kernelcore=size sets the amount of memory for use for allocations that
6064 * cannot be reclaimed or migrated.
6065 */
6066static int __init cmdline_parse_kernelcore(char *p)
6067{
Taku Izumi342332e2016-03-15 14:55:22 -07006068 /* parse kernelcore=mirror */
6069 if (parse_option_str(p, "mirror")) {
6070 mirrored_kernelcore = true;
6071 return 0;
6072 }
6073
Mel Gorman7e63efe2007-07-17 04:03:15 -07006074 return cmdline_parse_core(p, &required_kernelcore);
6075}
6076
6077/*
6078 * movablecore=size sets the amount of memory for use for allocations that
6079 * can be reclaimed or migrated.
6080 */
6081static int __init cmdline_parse_movablecore(char *p)
6082{
6083 return cmdline_parse_core(p, &required_movablecore);
6084}
6085
Mel Gormaned7ed362007-07-17 04:03:14 -07006086early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07006087early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07006088
Tejun Heo0ee332c2011-12-08 10:22:09 -08006089#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07006090
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006091void adjust_managed_page_count(struct page *page, long count)
6092{
6093 spin_lock(&managed_page_count_lock);
6094 page_zone(page)->managed_pages += count;
6095 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07006096#ifdef CONFIG_HIGHMEM
6097 if (PageHighMem(page))
6098 totalhigh_pages += count;
6099#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006100 spin_unlock(&managed_page_count_lock);
6101}
Jiang Liu3dcc0572013-07-03 15:03:21 -07006102EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006103
Jiang Liu11199692013-07-03 15:02:48 -07006104unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07006105{
Jiang Liu11199692013-07-03 15:02:48 -07006106 void *pos;
6107 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07006108
Jiang Liu11199692013-07-03 15:02:48 -07006109 start = (void *)PAGE_ALIGN((unsigned long)start);
6110 end = (void *)((unsigned long)end & PAGE_MASK);
6111 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07006112 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07006113 memset(pos, poison, PAGE_SIZE);
6114 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07006115 }
6116
6117 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07006118 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07006119 s, pages << (PAGE_SHIFT - 10), start, end);
6120
6121 return pages;
6122}
Jiang Liu11199692013-07-03 15:02:48 -07006123EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07006124
Jiang Liucfa11e02013-04-29 15:07:00 -07006125#ifdef CONFIG_HIGHMEM
6126void free_highmem_page(struct page *page)
6127{
6128 __free_reserved_page(page);
6129 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07006130 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07006131 totalhigh_pages++;
6132}
6133#endif
6134
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006135
6136void __init mem_init_print_info(const char *str)
6137{
6138 unsigned long physpages, codesize, datasize, rosize, bss_size;
6139 unsigned long init_code_size, init_data_size;
6140
6141 physpages = get_num_physpages();
6142 codesize = _etext - _stext;
6143 datasize = _edata - _sdata;
6144 rosize = __end_rodata - __start_rodata;
6145 bss_size = __bss_stop - __bss_start;
6146 init_data_size = __init_end - __init_begin;
6147 init_code_size = _einittext - _sinittext;
6148
6149 /*
6150 * Detect special cases and adjust section sizes accordingly:
6151 * 1) .init.* may be embedded into .data sections
6152 * 2) .init.text.* may be out of [__init_begin, __init_end],
6153 * please refer to arch/tile/kernel/vmlinux.lds.S.
6154 * 3) .rodata.* may be embedded into .text or .data sections.
6155 */
6156#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07006157 do { \
6158 if (start <= pos && pos < end && size > adj) \
6159 size -= adj; \
6160 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006161
6162 adj_init_size(__init_begin, __init_end, init_data_size,
6163 _sinittext, init_code_size);
6164 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
6165 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
6166 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
6167 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
6168
6169#undef adj_init_size
6170
Joe Perches756a0252016-03-17 14:19:47 -07006171 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 -07006172#ifdef CONFIG_HIGHMEM
Joe Perches756a0252016-03-17 14:19:47 -07006173 ", %luK highmem"
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006174#endif
Joe Perches756a0252016-03-17 14:19:47 -07006175 "%s%s)\n",
6176 nr_free_pages() << (PAGE_SHIFT - 10),
6177 physpages << (PAGE_SHIFT - 10),
6178 codesize >> 10, datasize >> 10, rosize >> 10,
6179 (init_data_size + init_code_size) >> 10, bss_size >> 10,
6180 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10),
6181 totalcma_pages << (PAGE_SHIFT - 10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006182#ifdef CONFIG_HIGHMEM
Joe Perches756a0252016-03-17 14:19:47 -07006183 totalhigh_pages << (PAGE_SHIFT - 10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006184#endif
Joe Perches756a0252016-03-17 14:19:47 -07006185 str ? ", " : "", str ? str : "");
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006186}
6187
Mel Gorman0e0b8642006-09-27 01:49:56 -07006188/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006189 * set_dma_reserve - set the specified number of pages reserved in the first zone
6190 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07006191 *
Yaowei Bai013110a2015-09-08 15:04:10 -07006192 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
Mel Gorman0e0b8642006-09-27 01:49:56 -07006193 * In the DMA zone, a significant percentage may be consumed by kernel image
6194 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006195 * function may optionally be used to account for unfreeable pages in the
6196 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6197 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07006198 */
6199void __init set_dma_reserve(unsigned long new_dma_reserve)
6200{
6201 dma_reserve = new_dma_reserve;
6202}
6203
Linus Torvalds1da177e2005-04-16 15:20:36 -07006204void __init free_area_init(unsigned long *zones_size)
6205{
Johannes Weiner9109fb72008-07-23 21:27:20 -07006206 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006207 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
6208}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006209
Linus Torvalds1da177e2005-04-16 15:20:36 -07006210static int page_alloc_cpu_notify(struct notifier_block *self,
6211 unsigned long action, void *hcpu)
6212{
6213 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006214
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07006215 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07006216 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006217 drain_pages(cpu);
6218
6219 /*
6220 * Spill the event counters of the dead processor
6221 * into the current processors event counters.
6222 * This artificially elevates the count of the current
6223 * processor.
6224 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07006225 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006226
6227 /*
6228 * Zero the differential counters of the dead processor
6229 * so that the vm statistics are consistent.
6230 *
6231 * This is only okay since the processor is dead and cannot
6232 * race with what we are doing.
6233 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07006234 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006235 }
6236 return NOTIFY_OK;
6237}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006238
6239void __init page_alloc_init(void)
6240{
6241 hotcpu_notifier(page_alloc_cpu_notify, 0);
6242}
6243
6244/*
Yaowei Bai34b10062015-09-08 15:04:13 -07006245 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006246 * or min_free_kbytes changes.
6247 */
6248static void calculate_totalreserve_pages(void)
6249{
6250 struct pglist_data *pgdat;
6251 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006252 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006253
6254 for_each_online_pgdat(pgdat) {
6255 for (i = 0; i < MAX_NR_ZONES; i++) {
6256 struct zone *zone = pgdat->node_zones + i;
Mel Gorman3484b2d2014-08-06 16:07:14 -07006257 long max = 0;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006258
6259 /* Find valid and maximum lowmem_reserve in the zone */
6260 for (j = i; j < MAX_NR_ZONES; j++) {
6261 if (zone->lowmem_reserve[j] > max)
6262 max = zone->lowmem_reserve[j];
6263 }
6264
Mel Gorman41858962009-06-16 15:32:12 -07006265 /* we treat the high watermark as reserved pages. */
6266 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006267
Jiang Liub40da042013-02-22 16:33:52 -08006268 if (max > zone->managed_pages)
6269 max = zone->managed_pages;
Johannes Weinera8d01432016-01-14 15:20:15 -08006270
6271 zone->totalreserve_pages = max;
6272
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006273 reserve_pages += max;
6274 }
6275 }
6276 totalreserve_pages = reserve_pages;
6277}
6278
6279/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07006280 * setup_per_zone_lowmem_reserve - called whenever
Yaowei Bai34b10062015-09-08 15:04:13 -07006281 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
Linus Torvalds1da177e2005-04-16 15:20:36 -07006282 * has a correct pages reserved value, so an adequate number of
6283 * pages are left in the zone after a successful __alloc_pages().
6284 */
6285static void setup_per_zone_lowmem_reserve(void)
6286{
6287 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006288 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006289
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08006290 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006291 for (j = 0; j < MAX_NR_ZONES; j++) {
6292 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08006293 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006294
6295 zone->lowmem_reserve[j] = 0;
6296
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006297 idx = j;
6298 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006299 struct zone *lower_zone;
6300
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006301 idx--;
6302
Linus Torvalds1da177e2005-04-16 15:20:36 -07006303 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6304 sysctl_lowmem_reserve_ratio[idx] = 1;
6305
6306 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08006307 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07006308 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08006309 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006310 }
6311 }
6312 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006313
6314 /* update totalreserve_pages */
6315 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006316}
6317
Mel Gormancfd3da12011-04-25 21:36:42 +00006318static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006319{
6320 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6321 unsigned long lowmem_pages = 0;
6322 struct zone *zone;
6323 unsigned long flags;
6324
6325 /* Calculate total number of !ZONE_HIGHMEM pages */
6326 for_each_zone(zone) {
6327 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08006328 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006329 }
6330
6331 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07006332 u64 tmp;
6333
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006334 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08006335 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07006336 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006337 if (is_highmem(zone)) {
6338 /*
Nick Piggin669ed172005-11-13 16:06:45 -08006339 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6340 * need highmem pages, so cap pages_min to a small
6341 * value here.
6342 *
Mel Gorman41858962009-06-16 15:32:12 -07006343 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Yaowei Bai42ff2702015-04-14 15:47:14 -07006344 * deltas control asynch page reclaim, and so should
Nick Piggin669ed172005-11-13 16:06:45 -08006345 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006346 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08006347 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006348
Jiang Liub40da042013-02-22 16:33:52 -08006349 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08006350 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07006351 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006352 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08006353 /*
6354 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07006355 * proportionate to the zone's size.
6356 */
Mel Gorman41858962009-06-16 15:32:12 -07006357 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006358 }
6359
Johannes Weiner795ae7a2016-03-17 14:19:14 -07006360 /*
6361 * Set the kswapd watermarks distance according to the
6362 * scale factor in proportion to available memory, but
6363 * ensure a minimum size on small systems.
6364 */
6365 tmp = max_t(u64, tmp >> 2,
6366 mult_frac(zone->managed_pages,
6367 watermark_scale_factor, 10000));
6368
6369 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp;
6370 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006371
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006372 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
Johannes Weinerabe5f972014-10-02 16:21:10 -07006373 high_wmark_pages(zone) - low_wmark_pages(zone) -
6374 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006375
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006376 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006377 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006378
6379 /* update totalreserve_pages */
6380 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006381}
6382
Mel Gormancfd3da12011-04-25 21:36:42 +00006383/**
6384 * setup_per_zone_wmarks - called when min_free_kbytes changes
6385 * or when memory is hot-{added|removed}
6386 *
6387 * Ensures that the watermark[min,low,high] values for each zone are set
6388 * correctly with respect to min_free_kbytes.
6389 */
6390void setup_per_zone_wmarks(void)
6391{
6392 mutex_lock(&zonelists_mutex);
6393 __setup_per_zone_wmarks();
6394 mutex_unlock(&zonelists_mutex);
6395}
6396
Randy Dunlap55a44622009-09-21 17:01:20 -07006397/*
Rik van Riel556adec2008-10-18 20:26:34 -07006398 * The inactive anon list should be small enough that the VM never has to
6399 * do too much work, but large enough that each inactive page has a chance
6400 * to be referenced again before it is swapped out.
6401 *
6402 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6403 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6404 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6405 * the anonymous pages are kept on the inactive list.
6406 *
6407 * total target max
6408 * memory ratio inactive anon
6409 * -------------------------------------
6410 * 10MB 1 5MB
6411 * 100MB 1 50MB
6412 * 1GB 3 250MB
6413 * 10GB 10 0.9GB
6414 * 100GB 31 3GB
6415 * 1TB 101 10GB
6416 * 10TB 320 32GB
6417 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07006418static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07006419{
6420 unsigned int gb, ratio;
6421
6422 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08006423 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07006424 if (gb)
6425 ratio = int_sqrt(10 * gb);
6426 else
6427 ratio = 1;
6428
6429 zone->inactive_ratio = ratio;
6430}
6431
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07006432static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07006433{
6434 struct zone *zone;
6435
Minchan Kim96cb4df2009-06-16 15:32:49 -07006436 for_each_zone(zone)
6437 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07006438}
6439
Linus Torvalds1da177e2005-04-16 15:20:36 -07006440/*
6441 * Initialise min_free_kbytes.
6442 *
6443 * For small machines we want it small (128k min). For large machines
6444 * we want it large (64MB max). But it is not linear, because network
6445 * bandwidth does not increase linearly with machine size. We use
6446 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006447 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07006448 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6449 *
6450 * which yields
6451 *
6452 * 16MB: 512k
6453 * 32MB: 724k
6454 * 64MB: 1024k
6455 * 128MB: 1448k
6456 * 256MB: 2048k
6457 * 512MB: 2896k
6458 * 1024MB: 4096k
6459 * 2048MB: 5792k
6460 * 4096MB: 8192k
6461 * 8192MB: 11584k
6462 * 16384MB: 16384k
6463 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07006464int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006465{
6466 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07006467 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006468
6469 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07006470 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006471
Michal Hocko5f127332013-07-08 16:00:40 -07006472 if (new_min_free_kbytes > user_min_free_kbytes) {
6473 min_free_kbytes = new_min_free_kbytes;
6474 if (min_free_kbytes < 128)
6475 min_free_kbytes = 128;
6476 if (min_free_kbytes > 65536)
6477 min_free_kbytes = 65536;
6478 } else {
6479 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6480 new_min_free_kbytes, user_min_free_kbytes);
6481 }
Minchan Kimbc75d332009-06-16 15:32:48 -07006482 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07006483 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006484 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07006485 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006486 return 0;
6487}
Minchan Kimbc75d332009-06-16 15:32:48 -07006488module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006489
6490/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07006491 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07006492 * that we can call two helper functions whenever min_free_kbytes
6493 * changes.
6494 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006495int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006496 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006497{
Han Pingtianda8c7572014-01-23 15:53:17 -08006498 int rc;
6499
6500 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6501 if (rc)
6502 return rc;
6503
Michal Hocko5f127332013-07-08 16:00:40 -07006504 if (write) {
6505 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07006506 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07006507 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006508 return 0;
6509}
6510
Johannes Weiner795ae7a2016-03-17 14:19:14 -07006511int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
6512 void __user *buffer, size_t *length, loff_t *ppos)
6513{
6514 int rc;
6515
6516 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6517 if (rc)
6518 return rc;
6519
6520 if (write)
6521 setup_per_zone_wmarks();
6522
6523 return 0;
6524}
6525
Christoph Lameter96146342006-07-03 00:24:13 -07006526#ifdef CONFIG_NUMA
Joe Perchescccad5b2014-06-06 14:38:09 -07006527int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006528 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07006529{
6530 struct zone *zone;
6531 int rc;
6532
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006533 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07006534 if (rc)
6535 return rc;
6536
6537 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08006538 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07006539 sysctl_min_unmapped_ratio) / 100;
6540 return 0;
6541}
Christoph Lameter0ff38492006-09-25 23:31:52 -07006542
Joe Perchescccad5b2014-06-06 14:38:09 -07006543int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006544 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07006545{
6546 struct zone *zone;
6547 int rc;
6548
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006549 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07006550 if (rc)
6551 return rc;
6552
6553 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08006554 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07006555 sysctl_min_slab_ratio) / 100;
6556 return 0;
6557}
Christoph Lameter96146342006-07-03 00:24:13 -07006558#endif
6559
Linus Torvalds1da177e2005-04-16 15:20:36 -07006560/*
6561 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6562 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6563 * whenever sysctl_lowmem_reserve_ratio changes.
6564 *
6565 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07006566 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07006567 * if in function of the boot time zone sizes.
6568 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006569int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006570 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006571{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006572 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006573 setup_per_zone_lowmem_reserve();
6574 return 0;
6575}
6576
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006577/*
6578 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07006579 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6580 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006581 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006582int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006583 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006584{
6585 struct zone *zone;
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006586 int old_percpu_pagelist_fraction;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006587 int ret;
6588
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006589 mutex_lock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006590 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6591
6592 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
6593 if (!write || ret < 0)
6594 goto out;
6595
6596 /* Sanity checking to avoid pcp imbalance */
6597 if (percpu_pagelist_fraction &&
6598 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6599 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6600 ret = -EINVAL;
6601 goto out;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006602 }
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006603
6604 /* No change? */
6605 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6606 goto out;
6607
6608 for_each_populated_zone(zone) {
6609 unsigned int cpu;
6610
6611 for_each_possible_cpu(cpu)
6612 pageset_set_high_and_batch(zone,
6613 per_cpu_ptr(zone->pageset, cpu));
6614 }
6615out:
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006616 mutex_unlock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006617 return ret;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006618}
6619
Rasmus Villemoesa9919c72015-06-24 16:56:28 -07006620#ifdef CONFIG_NUMA
David S. Millerf034b5d2006-08-24 03:08:07 -07006621int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006622
Linus Torvalds1da177e2005-04-16 15:20:36 -07006623static int __init set_hashdist(char *str)
6624{
6625 if (!str)
6626 return 0;
6627 hashdist = simple_strtoul(str, &str, 0);
6628 return 1;
6629}
6630__setup("hashdist=", set_hashdist);
6631#endif
6632
6633/*
6634 * allocate a large system hash table from bootmem
6635 * - it is assumed that the hash table must contain an exact power-of-2
6636 * quantity of entries
6637 * - limit is the number of hash buckets, not the total allocation size
6638 */
6639void *__init alloc_large_system_hash(const char *tablename,
6640 unsigned long bucketsize,
6641 unsigned long numentries,
6642 int scale,
6643 int flags,
6644 unsigned int *_hash_shift,
6645 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00006646 unsigned long low_limit,
6647 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006648{
Tim Bird31fe62b2012-05-23 13:33:35 +00006649 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006650 unsigned long log2qty, size;
6651 void *table = NULL;
6652
6653 /* allow the kernel cmdline to have a say */
6654 if (!numentries) {
6655 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08006656 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07006657
6658 /* It isn't necessary when PAGE_SIZE >= 1MB */
6659 if (PAGE_SHIFT < 20)
6660 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006661
6662 /* limit to 1 bucket per 2^scale bytes of low memory */
6663 if (scale > PAGE_SHIFT)
6664 numentries >>= (scale - PAGE_SHIFT);
6665 else
6666 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08006667
6668 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07006669 if (unlikely(flags & HASH_SMALL)) {
6670 /* Makes no sense without HASH_EARLY */
6671 WARN_ON(!(flags & HASH_EARLY));
6672 if (!(numentries >> *_hash_shift)) {
6673 numentries = 1UL << *_hash_shift;
6674 BUG_ON(!numentries);
6675 }
6676 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08006677 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006678 }
John Hawkes6e692ed2006-03-25 03:08:02 -08006679 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006680
6681 /* limit allocation size to 1/16 total memory by default */
6682 if (max == 0) {
6683 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6684 do_div(max, bucketsize);
6685 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08006686 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006687
Tim Bird31fe62b2012-05-23 13:33:35 +00006688 if (numentries < low_limit)
6689 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006690 if (numentries > max)
6691 numentries = max;
6692
David Howellsf0d1b0b2006-12-08 02:37:49 -08006693 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006694
6695 do {
6696 size = bucketsize << log2qty;
6697 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08006698 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006699 else if (hashdist)
6700 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6701 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07006702 /*
6703 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07006704 * some pages at the end of hash table which
6705 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07006706 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006707 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07006708 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006709 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6710 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006711 }
6712 } while (!table && size > PAGE_SIZE && --log2qty);
6713
6714 if (!table)
6715 panic("Failed to allocate %s hash table\n", tablename);
6716
Joe Perches11705322016-03-17 14:19:50 -07006717 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
6718 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006719
6720 if (_hash_shift)
6721 *_hash_shift = log2qty;
6722 if (_hash_mask)
6723 *_hash_mask = (1 << log2qty) - 1;
6724
6725 return table;
6726}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08006727
Mel Gorman835c1342007-10-16 01:25:47 -07006728/* Return a pointer to the bitmap storing bits affecting a block of pages */
6729static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6730 unsigned long pfn)
6731{
6732#ifdef CONFIG_SPARSEMEM
6733 return __pfn_to_section(pfn)->pageblock_flags;
6734#else
6735 return zone->pageblock_flags;
6736#endif /* CONFIG_SPARSEMEM */
6737}
Andrew Morton6220ec72006-10-19 23:29:05 -07006738
Mel Gorman835c1342007-10-16 01:25:47 -07006739static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6740{
6741#ifdef CONFIG_SPARSEMEM
6742 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07006743 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006744#else
Laura Abbottc060f942013-01-11 14:31:51 -08006745 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07006746 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006747#endif /* CONFIG_SPARSEMEM */
6748}
6749
6750/**
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006751 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
Mel Gorman835c1342007-10-16 01:25:47 -07006752 * @page: The page within the block of interest
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006753 * @pfn: The target page frame number
6754 * @end_bitidx: The last bit of interest to retrieve
6755 * @mask: mask of bits that the caller is interested in
6756 *
6757 * Return: pageblock_bits flags
Mel Gorman835c1342007-10-16 01:25:47 -07006758 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006759unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006760 unsigned long end_bitidx,
6761 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006762{
6763 struct zone *zone;
6764 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006765 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006766 unsigned long word;
Mel Gorman835c1342007-10-16 01:25:47 -07006767
6768 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006769 bitmap = get_pageblock_bitmap(zone, pfn);
6770 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006771 word_bitidx = bitidx / BITS_PER_LONG;
6772 bitidx &= (BITS_PER_LONG-1);
Mel Gorman835c1342007-10-16 01:25:47 -07006773
Mel Gormane58469b2014-06-04 16:10:16 -07006774 word = bitmap[word_bitidx];
6775 bitidx += end_bitidx;
6776 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
Mel Gorman835c1342007-10-16 01:25:47 -07006777}
6778
6779/**
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006780 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
Mel Gorman835c1342007-10-16 01:25:47 -07006781 * @page: The page within the block of interest
Mel Gorman835c1342007-10-16 01:25:47 -07006782 * @flags: The flags to set
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006783 * @pfn: The target page frame number
6784 * @end_bitidx: The last bit of interest
6785 * @mask: mask of bits that the caller is interested in
Mel Gorman835c1342007-10-16 01:25:47 -07006786 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006787void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6788 unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006789 unsigned long end_bitidx,
6790 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006791{
6792 struct zone *zone;
6793 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006794 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006795 unsigned long old_word, word;
6796
6797 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
Mel Gorman835c1342007-10-16 01:25:47 -07006798
6799 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006800 bitmap = get_pageblock_bitmap(zone, pfn);
6801 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006802 word_bitidx = bitidx / BITS_PER_LONG;
6803 bitidx &= (BITS_PER_LONG-1);
6804
Sasha Levin309381fea2014-01-23 15:52:54 -08006805 VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
Mel Gorman835c1342007-10-16 01:25:47 -07006806
Mel Gormane58469b2014-06-04 16:10:16 -07006807 bitidx += end_bitidx;
6808 mask <<= (BITS_PER_LONG - bitidx - 1);
6809 flags <<= (BITS_PER_LONG - bitidx - 1);
6810
Jason Low4db0c3c2015-04-15 16:14:08 -07006811 word = READ_ONCE(bitmap[word_bitidx]);
Mel Gormane58469b2014-06-04 16:10:16 -07006812 for (;;) {
6813 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6814 if (word == old_word)
6815 break;
6816 word = old_word;
6817 }
Mel Gorman835c1342007-10-16 01:25:47 -07006818}
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07006819
6820/*
Minchan Kim80934512012-07-31 16:43:01 -07006821 * This function checks whether pageblock includes unmovable pages or not.
6822 * If @count is not zero, it is okay to include less @count unmovable pages
6823 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006824 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07006825 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6826 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07006827 */
Wen Congyangb023f462012-12-11 16:00:45 -08006828bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6829 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006830{
6831 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006832 int mt;
6833
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006834 /*
6835 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07006836 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006837 */
6838 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07006839 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006840 mt = get_pageblock_migratetype(page);
6841 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07006842 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006843
6844 pfn = page_to_pfn(page);
6845 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6846 unsigned long check = pfn + iter;
6847
Namhyung Kim29723fc2011-02-25 14:44:25 -08006848 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006849 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08006850
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006851 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07006852
6853 /*
6854 * Hugepages are not in LRU lists, but they're movable.
6855 * We need not scan over tail pages bacause we don't
6856 * handle each tail page individually in migration.
6857 */
6858 if (PageHuge(page)) {
6859 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6860 continue;
6861 }
6862
Minchan Kim97d255c2012-07-31 16:42:59 -07006863 /*
6864 * We can't use page_count without pin a page
6865 * because another CPU can free compound page.
6866 * This check already skips compound tails of THP
6867 * because their page->_count is zero at all time.
6868 */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07006869 if (!page_ref_count(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006870 if (PageBuddy(page))
6871 iter += (1 << page_order(page)) - 1;
6872 continue;
6873 }
Minchan Kim97d255c2012-07-31 16:42:59 -07006874
Wen Congyangb023f462012-12-11 16:00:45 -08006875 /*
6876 * The HWPoisoned page may be not in buddy system, and
6877 * page_count() is not 0.
6878 */
6879 if (skip_hwpoisoned_pages && PageHWPoison(page))
6880 continue;
6881
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006882 if (!PageLRU(page))
6883 found++;
6884 /*
Johannes Weiner6b4f7792014-12-12 16:56:13 -08006885 * If there are RECLAIMABLE pages, we need to check
6886 * it. But now, memory offline itself doesn't call
6887 * shrink_node_slabs() and it still to be fixed.
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006888 */
6889 /*
6890 * If the page is not RAM, page_count()should be 0.
6891 * we don't need more check. This is an _used_ not-movable page.
6892 *
6893 * The problematic thing here is PG_reserved pages. PG_reserved
6894 * is set to both of a memory hole page and a _used_ kernel
6895 * page at boot.
6896 */
6897 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07006898 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006899 }
Minchan Kim80934512012-07-31 16:43:01 -07006900 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006901}
6902
6903bool is_pageblock_removable_nolock(struct page *page)
6904{
Michal Hocko656a0702012-01-20 14:33:58 -08006905 struct zone *zone;
6906 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08006907
6908 /*
6909 * We have to be careful here because we are iterating over memory
6910 * sections which are not zone aware so we might end up outside of
6911 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08006912 * We have to take care about the node as well. If the node is offline
6913 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08006914 */
Michal Hocko656a0702012-01-20 14:33:58 -08006915 if (!node_online(page_to_nid(page)))
6916 return false;
6917
6918 zone = page_zone(page);
6919 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08006920 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08006921 return false;
6922
Wen Congyangb023f462012-12-11 16:00:45 -08006923 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07006924}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006925
Vlastimil Babka080fe202016-02-05 15:36:41 -08006926#if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006927
6928static unsigned long pfn_max_align_down(unsigned long pfn)
6929{
6930 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6931 pageblock_nr_pages) - 1);
6932}
6933
6934static unsigned long pfn_max_align_up(unsigned long pfn)
6935{
6936 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6937 pageblock_nr_pages));
6938}
6939
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006940/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006941static int __alloc_contig_migrate_range(struct compact_control *cc,
6942 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006943{
6944 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006945 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006946 unsigned long pfn = start;
6947 unsigned int tries = 0;
6948 int ret = 0;
6949
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08006950 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006951
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006952 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006953 if (fatal_signal_pending(current)) {
6954 ret = -EINTR;
6955 break;
6956 }
6957
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006958 if (list_empty(&cc->migratepages)) {
6959 cc->nr_migratepages = 0;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07006960 pfn = isolate_migratepages_range(cc, pfn, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006961 if (!pfn) {
6962 ret = -EINTR;
6963 break;
6964 }
6965 tries = 0;
6966 } else if (++tries == 5) {
6967 ret = ret < 0 ? ret : -EBUSY;
6968 break;
6969 }
6970
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006971 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6972 &cc->migratepages);
6973 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07006974
Hugh Dickins9c620e22013-02-22 16:35:14 -08006975 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07006976 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006977 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08006978 if (ret < 0) {
6979 putback_movable_pages(&cc->migratepages);
6980 return ret;
6981 }
6982 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006983}
6984
6985/**
6986 * alloc_contig_range() -- tries to allocate given range of pages
6987 * @start: start PFN to allocate
6988 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006989 * @migratetype: migratetype of the underlaying pageblocks (either
6990 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6991 * in range must have the same migratetype and it must
6992 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006993 *
6994 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6995 * aligned, however it's the caller's responsibility to guarantee that
6996 * we are the only thread that changes migrate type of pageblocks the
6997 * pages fall in.
6998 *
6999 * The PFN range must belong to a single zone.
7000 *
7001 * Returns zero on success or negative error code. On success all
7002 * pages which PFN is in [start, end) are allocated for the caller and
7003 * need to be freed with free_contig_range().
7004 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02007005int alloc_contig_range(unsigned long start, unsigned long end,
7006 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007007{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007008 unsigned long outer_start, outer_end;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08007009 unsigned int order;
7010 int ret = 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007011
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007012 struct compact_control cc = {
7013 .nr_migratepages = 0,
7014 .order = -1,
7015 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07007016 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007017 .ignore_skip_hint = true,
7018 };
7019 INIT_LIST_HEAD(&cc.migratepages);
7020
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007021 /*
7022 * What we do here is we mark all pageblocks in range as
7023 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7024 * have different sizes, and due to the way page allocator
7025 * work, we align the range to biggest of the two pages so
7026 * that page allocator won't try to merge buddies from
7027 * different pageblocks and change MIGRATE_ISOLATE to some
7028 * other migration type.
7029 *
7030 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7031 * migrate the pages from an unaligned range (ie. pages that
7032 * we are interested in). This will put all the pages in
7033 * range back to page allocator as MIGRATE_ISOLATE.
7034 *
7035 * When this is done, we take the pages in range from page
7036 * allocator removing them from the buddy system. This way
7037 * page allocator will never consider using them.
7038 *
7039 * This lets us mark the pageblocks back as
7040 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7041 * aligned range but not in the unaligned, original range are
7042 * put back to page allocator so that buddy can use them.
7043 */
7044
7045 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08007046 pfn_max_align_up(end), migratetype,
7047 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007048 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07007049 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007050
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007051 /*
7052 * In case of -EBUSY, we'd like to know which page causes problem.
7053 * So, just fall through. We will check it in test_pages_isolated().
7054 */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007055 ret = __alloc_contig_migrate_range(&cc, start, end);
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007056 if (ret && ret != -EBUSY)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007057 goto done;
7058
7059 /*
7060 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7061 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7062 * more, all pages in [start, end) are free in page allocator.
7063 * What we are going to do is to allocate all pages from
7064 * [start, end) (that is remove them from page allocator).
7065 *
7066 * The only problem is that pages at the beginning and at the
7067 * end of interesting range may be not aligned with pages that
7068 * page allocator holds, ie. they can be part of higher order
7069 * pages. Because of this, we reserve the bigger range and
7070 * once this is done free the pages we are not interested in.
7071 *
7072 * We don't have to hold zone->lock here because the pages are
7073 * isolated thus they won't get removed from buddy.
7074 */
7075
7076 lru_add_drain_all();
Vlastimil Babka510f5502014-12-10 15:43:07 -08007077 drain_all_pages(cc.zone);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007078
7079 order = 0;
7080 outer_start = start;
7081 while (!PageBuddy(pfn_to_page(outer_start))) {
7082 if (++order >= MAX_ORDER) {
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007083 outer_start = start;
7084 break;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007085 }
7086 outer_start &= ~0UL << order;
7087 }
7088
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007089 if (outer_start != start) {
7090 order = page_order(pfn_to_page(outer_start));
7091
7092 /*
7093 * outer_start page could be small order buddy page and
7094 * it doesn't include start page. Adjust outer_start
7095 * in this case to report failed page properly
7096 * on tracepoint in test_pages_isolated()
7097 */
7098 if (outer_start + (1UL << order) <= start)
7099 outer_start = start;
7100 }
7101
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007102 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08007103 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewiczdae803e2014-11-13 15:19:27 -08007104 pr_info("%s: [%lx, %lx) PFNs busy\n",
7105 __func__, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007106 ret = -EBUSY;
7107 goto done;
7108 }
7109
Marek Szyprowski49f223a2012-01-25 12:49:24 +01007110 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007111 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007112 if (!outer_end) {
7113 ret = -EBUSY;
7114 goto done;
7115 }
7116
7117 /* Free head and tail (if any) */
7118 if (start != outer_start)
7119 free_contig_range(outer_start, start - outer_start);
7120 if (end != outer_end)
7121 free_contig_range(end, outer_end - end);
7122
7123done:
7124 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02007125 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007126 return ret;
7127}
7128
7129void free_contig_range(unsigned long pfn, unsigned nr_pages)
7130{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08007131 unsigned int count = 0;
7132
7133 for (; nr_pages--; pfn++) {
7134 struct page *page = pfn_to_page(pfn);
7135
7136 count += page_count(page) != 1;
7137 __free_page(page);
7138 }
7139 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007140}
7141#endif
7142
Jiang Liu4ed7e022012-07-31 16:43:35 -07007143#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07007144/*
7145 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7146 * page high values need to be recalulated.
7147 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07007148void __meminit zone_pcp_update(struct zone *zone)
7149{
Cody P Schafer0a647f32013-07-03 15:01:33 -07007150 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007151 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07007152 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07007153 pageset_set_high_and_batch(zone,
7154 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007155 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07007156}
7157#endif
7158
Jiang Liu340175b2012-07-31 16:43:32 -07007159void zone_pcp_reset(struct zone *zone)
7160{
7161 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07007162 int cpu;
7163 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07007164
7165 /* avoid races with drain_pages() */
7166 local_irq_save(flags);
7167 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07007168 for_each_online_cpu(cpu) {
7169 pset = per_cpu_ptr(zone->pageset, cpu);
7170 drain_zonestat(zone, pset);
7171 }
Jiang Liu340175b2012-07-31 16:43:32 -07007172 free_percpu(zone->pageset);
7173 zone->pageset = &boot_pageset;
7174 }
7175 local_irq_restore(flags);
7176}
7177
Wen Congyang6dcd73d2012-12-11 16:01:01 -08007178#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007179/*
7180 * All pages in the range must be isolated before calling this.
7181 */
7182void
7183__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
7184{
7185 struct page *page;
7186 struct zone *zone;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07007187 unsigned int order, i;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007188 unsigned long pfn;
7189 unsigned long flags;
7190 /* find the first valid pfn */
7191 for (pfn = start_pfn; pfn < end_pfn; pfn++)
7192 if (pfn_valid(pfn))
7193 break;
7194 if (pfn == end_pfn)
7195 return;
7196 zone = page_zone(pfn_to_page(pfn));
7197 spin_lock_irqsave(&zone->lock, flags);
7198 pfn = start_pfn;
7199 while (pfn < end_pfn) {
7200 if (!pfn_valid(pfn)) {
7201 pfn++;
7202 continue;
7203 }
7204 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08007205 /*
7206 * The HWPoisoned page may be not in buddy system, and
7207 * page_count() is not 0.
7208 */
7209 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
7210 pfn++;
7211 SetPageReserved(page);
7212 continue;
7213 }
7214
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007215 BUG_ON(page_count(page));
7216 BUG_ON(!PageBuddy(page));
7217 order = page_order(page);
7218#ifdef CONFIG_DEBUG_VM
Joe Perches11705322016-03-17 14:19:50 -07007219 pr_info("remove from free list %lx %d %lx\n",
7220 pfn, 1 << order, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007221#endif
7222 list_del(&page->lru);
7223 rmv_page_order(page);
7224 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007225 for (i = 0; i < (1 << order); i++)
7226 SetPageReserved((page+i));
7227 pfn += (1 << order);
7228 }
7229 spin_unlock_irqrestore(&zone->lock, flags);
7230}
7231#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007232
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007233bool is_free_buddy_page(struct page *page)
7234{
7235 struct zone *zone = page_zone(page);
7236 unsigned long pfn = page_to_pfn(page);
7237 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07007238 unsigned int order;
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007239
7240 spin_lock_irqsave(&zone->lock, flags);
7241 for (order = 0; order < MAX_ORDER; order++) {
7242 struct page *page_head = page - (pfn & ((1 << order) - 1));
7243
7244 if (PageBuddy(page_head) && page_order(page_head) >= order)
7245 break;
7246 }
7247 spin_unlock_irqrestore(&zone->lock, flags);
7248
7249 return order < MAX_ORDER;
7250}