blob: f8f3bfc435eec5bf0144798f6b890e6deacee0f6 [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
Mel Gorman0b423ca2016-05-19 17:14:27 -0700355/* Return a pointer to the bitmap storing bits affecting a block of pages */
356static inline unsigned long *get_pageblock_bitmap(struct page *page,
357 unsigned long pfn)
358{
359#ifdef CONFIG_SPARSEMEM
360 return __pfn_to_section(pfn)->pageblock_flags;
361#else
362 return page_zone(page)->pageblock_flags;
363#endif /* CONFIG_SPARSEMEM */
364}
365
366static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
367{
368#ifdef CONFIG_SPARSEMEM
369 pfn &= (PAGES_PER_SECTION-1);
370 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
371#else
372 pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
373 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
374#endif /* CONFIG_SPARSEMEM */
375}
376
377/**
378 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
379 * @page: The page within the block of interest
380 * @pfn: The target page frame number
381 * @end_bitidx: The last bit of interest to retrieve
382 * @mask: mask of bits that the caller is interested in
383 *
384 * Return: pageblock_bits flags
385 */
386static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page,
387 unsigned long pfn,
388 unsigned long end_bitidx,
389 unsigned long mask)
390{
391 unsigned long *bitmap;
392 unsigned long bitidx, word_bitidx;
393 unsigned long word;
394
395 bitmap = get_pageblock_bitmap(page, pfn);
396 bitidx = pfn_to_bitidx(page, pfn);
397 word_bitidx = bitidx / BITS_PER_LONG;
398 bitidx &= (BITS_PER_LONG-1);
399
400 word = bitmap[word_bitidx];
401 bitidx += end_bitidx;
402 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
403}
404
405unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
406 unsigned long end_bitidx,
407 unsigned long mask)
408{
409 return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask);
410}
411
412static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
413{
414 return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK);
415}
416
417/**
418 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
419 * @page: The page within the block of interest
420 * @flags: The flags to set
421 * @pfn: The target page frame number
422 * @end_bitidx: The last bit of interest
423 * @mask: mask of bits that the caller is interested in
424 */
425void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
426 unsigned long pfn,
427 unsigned long end_bitidx,
428 unsigned long mask)
429{
430 unsigned long *bitmap;
431 unsigned long bitidx, word_bitidx;
432 unsigned long old_word, word;
433
434 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
435
436 bitmap = get_pageblock_bitmap(page, pfn);
437 bitidx = pfn_to_bitidx(page, pfn);
438 word_bitidx = bitidx / BITS_PER_LONG;
439 bitidx &= (BITS_PER_LONG-1);
440
441 VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
442
443 bitidx += end_bitidx;
444 mask <<= (BITS_PER_LONG - bitidx - 1);
445 flags <<= (BITS_PER_LONG - bitidx - 1);
446
447 word = READ_ONCE(bitmap[word_bitidx]);
448 for (;;) {
449 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
450 if (word == old_word)
451 break;
452 word = old_word;
453 }
454}
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700455
Minchan Kimee6f5092012-07-31 16:43:50 -0700456void set_pageblock_migratetype(struct page *page, int migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700457{
KOSAKI Motohiro5d0f3f72013-11-12 15:08:18 -0800458 if (unlikely(page_group_by_mobility_disabled &&
459 migratetype < MIGRATE_PCPTYPES))
Mel Gorman49255c62009-06-16 15:31:58 -0700460 migratetype = MIGRATE_UNMOVABLE;
461
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700462 set_pageblock_flags_group(page, (unsigned long)migratetype,
463 PB_migrate, PB_migrate_end);
464}
465
Nick Piggin13e74442006-01-06 00:10:58 -0800466#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700467static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700469 int ret = 0;
470 unsigned seq;
471 unsigned long pfn = page_to_pfn(page);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800472 unsigned long sp, start_pfn;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700473
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700474 do {
475 seq = zone_span_seqbegin(zone);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800476 start_pfn = zone->zone_start_pfn;
477 sp = zone->spanned_pages;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800478 if (!zone_spans_pfn(zone, pfn))
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700479 ret = 1;
480 } while (zone_span_seqretry(zone, seq));
481
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800482 if (ret)
Dave Hansen613813e2014-06-04 16:07:27 -0700483 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
484 pfn, zone_to_nid(zone), zone->name,
485 start_pfn, start_pfn + sp);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800486
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700487 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700488}
489
490static int page_is_consistent(struct zone *zone, struct page *page)
491{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700492 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700493 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700495 return 0;
496
497 return 1;
498}
499/*
500 * Temporary debugging check for pages not lying within a given zone.
501 */
502static int bad_range(struct zone *zone, struct page *page)
503{
504 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700506 if (!page_is_consistent(zone, page))
507 return 1;
508
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 return 0;
510}
Nick Piggin13e74442006-01-06 00:10:58 -0800511#else
512static inline int bad_range(struct zone *zone, struct page *page)
513{
514 return 0;
515}
516#endif
517
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700518static void bad_page(struct page *page, const char *reason,
519 unsigned long bad_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800521 static unsigned long resume;
522 static unsigned long nr_shown;
523 static unsigned long nr_unshown;
524
525 /*
526 * Allow a burst of 60 reports, then keep quiet for that minute;
527 * or allow a steady drip of one report per second.
528 */
529 if (nr_shown == 60) {
530 if (time_before(jiffies, resume)) {
531 nr_unshown++;
532 goto out;
533 }
534 if (nr_unshown) {
Vlastimil Babkaff8e8112016-03-15 14:56:24 -0700535 pr_alert(
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800536 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800537 nr_unshown);
538 nr_unshown = 0;
539 }
540 nr_shown = 0;
541 }
542 if (nr_shown++ == 0)
543 resume = jiffies + 60 * HZ;
544
Vlastimil Babkaff8e8112016-03-15 14:56:24 -0700545 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800546 current->comm, page_to_pfn(page));
Vlastimil Babkaff8e8112016-03-15 14:56:24 -0700547 __dump_page(page, reason);
548 bad_flags &= page->flags;
549 if (bad_flags)
550 pr_alert("bad because of flags: %#lx(%pGp)\n",
551 bad_flags, &bad_flags);
Vlastimil Babka4e462112016-03-15 14:56:21 -0700552 dump_page_owner(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700553
Dave Jones4f318882011-10-31 17:07:24 -0700554 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800556out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800557 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800558 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030559 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560}
561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562/*
563 * Higher-order pages are called "compound pages". They are structured thusly:
564 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800565 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800567 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
568 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800570 * The first tail page's ->compound_dtor holds the offset in array of compound
571 * page destructors. See compound_page_dtors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800573 * The first tail page's ->compound_order holds the order of allocation.
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800574 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800576
Kirill A. Shutemov9a982252016-01-15 16:54:17 -0800577void free_compound_page(struct page *page)
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800578{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700579 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800580}
581
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -0800582void prep_compound_page(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583{
584 int i;
585 int nr_pages = 1 << order;
586
Kirill A. Shutemovf1e61552015-11-06 16:29:50 -0800587 set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700588 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700589 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800590 for (i = 1; i < nr_pages; i++) {
591 struct page *p = page + i;
Youquan Song58a84aa2011-12-08 14:34:18 -0800592 set_page_count(p, 0);
Kirill A. Shutemov1c290f62016-01-15 16:52:07 -0800593 p->mapping = TAIL_MAPPING;
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800594 set_compound_head(p, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 }
Kirill A. Shutemov53f92632016-01-15 16:53:42 -0800596 atomic_set(compound_mapcount_ptr(page), -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597}
598
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800599#ifdef CONFIG_DEBUG_PAGEALLOC
600unsigned int _debug_guardpage_minorder;
Christian Borntraegerea6eabb2016-03-15 14:55:30 -0700601bool _debug_pagealloc_enabled __read_mostly
602 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
Joonsoo Kim505f6d22016-03-17 14:17:56 -0700603EXPORT_SYMBOL(_debug_pagealloc_enabled);
Joonsoo Kime30825f2014-12-12 16:55:49 -0800604bool _debug_guardpage_enabled __read_mostly;
605
Joonsoo Kim031bc572014-12-12 16:55:52 -0800606static int __init early_debug_pagealloc(char *buf)
607{
608 if (!buf)
609 return -EINVAL;
Minfei Huang2a138dc2016-05-20 16:58:13 -0700610 return kstrtobool(buf, &_debug_pagealloc_enabled);
Joonsoo Kim031bc572014-12-12 16:55:52 -0800611}
612early_param("debug_pagealloc", early_debug_pagealloc);
613
Joonsoo Kime30825f2014-12-12 16:55:49 -0800614static bool need_debug_guardpage(void)
615{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800616 /* If we don't use debug_pagealloc, we don't need guard page */
617 if (!debug_pagealloc_enabled())
618 return false;
619
Joonsoo Kime30825f2014-12-12 16:55:49 -0800620 return true;
621}
622
623static void init_debug_guardpage(void)
624{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800625 if (!debug_pagealloc_enabled())
626 return;
627
Joonsoo Kime30825f2014-12-12 16:55:49 -0800628 _debug_guardpage_enabled = true;
629}
630
631struct page_ext_operations debug_guardpage_ops = {
632 .need = need_debug_guardpage,
633 .init = init_debug_guardpage,
634};
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800635
636static int __init debug_guardpage_minorder_setup(char *buf)
637{
638 unsigned long res;
639
640 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
Joe Perches11705322016-03-17 14:19:50 -0700641 pr_err("Bad debug_guardpage_minorder value\n");
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800642 return 0;
643 }
644 _debug_guardpage_minorder = res;
Joe Perches11705322016-03-17 14:19:50 -0700645 pr_info("Setting debug_guardpage_minorder to %lu\n", res);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800646 return 0;
647}
648__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
649
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800650static inline void set_page_guard(struct zone *zone, struct page *page,
651 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800652{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800653 struct page_ext *page_ext;
654
655 if (!debug_guardpage_enabled())
656 return;
657
658 page_ext = lookup_page_ext(page);
659 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
660
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800661 INIT_LIST_HEAD(&page->lru);
662 set_page_private(page, order);
663 /* Guard pages are not available for any usage */
664 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800665}
666
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800667static inline void clear_page_guard(struct zone *zone, struct page *page,
668 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800669{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800670 struct page_ext *page_ext;
671
672 if (!debug_guardpage_enabled())
673 return;
674
675 page_ext = lookup_page_ext(page);
676 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
677
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800678 set_page_private(page, 0);
679 if (!is_migrate_isolate(migratetype))
680 __mod_zone_freepage_state(zone, (1 << order), migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800681}
682#else
Joonsoo Kime30825f2014-12-12 16:55:49 -0800683struct page_ext_operations debug_guardpage_ops = { NULL, };
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800684static inline void set_page_guard(struct zone *zone, struct page *page,
685 unsigned int order, int migratetype) {}
686static inline void clear_page_guard(struct zone *zone, struct page *page,
687 unsigned int order, int migratetype) {}
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800688#endif
689
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700690static inline void set_page_order(struct page *page, unsigned int order)
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700691{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700692 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000693 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694}
695
696static inline void rmv_page_order(struct page *page)
697{
Nick Piggin676165a2006-04-10 11:21:48 +1000698 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700699 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700}
701
702/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 * This function checks whether a page is free && is the buddy
704 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800705 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000706 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700707 * (c) a page and its buddy have the same order &&
708 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 *
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700710 * For recording whether a page is in the buddy system, we set ->_mapcount
711 * PAGE_BUDDY_MAPCOUNT_VALUE.
712 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
713 * serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000714 *
715 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700717static inline int page_is_buddy(struct page *page, struct page *buddy,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700718 unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700720 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800721 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800722
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800723 if (page_is_guard(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700724 if (page_zone_id(page) != page_zone_id(buddy))
725 return 0;
726
Weijie Yang4c5018c2015-02-10 14:11:39 -0800727 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
728
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800729 return 1;
730 }
731
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700732 if (PageBuddy(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700733 /*
734 * zone check is done late to avoid uselessly
735 * calculating zone/node ids for pages that could
736 * never merge.
737 */
738 if (page_zone_id(page) != page_zone_id(buddy))
739 return 0;
740
Weijie Yang4c5018c2015-02-10 14:11:39 -0800741 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
742
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700743 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000744 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700745 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746}
747
748/*
749 * Freeing function for a buddy system allocator.
750 *
751 * The concept of a buddy system is to maintain direct-mapped table
752 * (containing bit values) for memory blocks of various "orders".
753 * The bottom level table contains the map for the smallest allocatable
754 * units of memory (here, pages), and each level above it describes
755 * pairs of units from the levels below, hence, "buddies".
756 * At a high level, all that happens here is marking the table entry
757 * at the bottom level available, and propagating the changes upward
758 * as necessary, plus some accounting needed to play nicely with other
759 * parts of the VM system.
760 * At each level, we keep a list of pages, which are heads of continuous
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700761 * free pages of length of (1 << order) and marked with _mapcount
762 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
763 * field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100765 * other. That is, if we allocate a small block, and both were
766 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100768 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100770 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 */
772
Nick Piggin48db57f2006-01-08 01:00:42 -0800773static inline void __free_one_page(struct page *page,
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700774 unsigned long pfn,
Mel Gormaned0ae212009-06-16 15:32:07 -0700775 struct zone *zone, unsigned int order,
776 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777{
778 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700779 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800780 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700781 struct page *buddy;
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700782 unsigned int max_order;
783
784 max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
Cody P Schaferd29bb972013-02-22 16:35:25 -0800786 VM_BUG_ON(!zone_is_initialized(zone));
Kirill A. Shutemov6e9f0d52015-02-11 15:25:50 -0800787 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
Mel Gormaned0ae212009-06-16 15:32:07 -0700789 VM_BUG_ON(migratetype == -1);
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700790 if (likely(!is_migrate_isolate(migratetype)))
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800791 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Mel Gormaned0ae212009-06-16 15:32:07 -0700792
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700793 page_idx = pfn & ((1 << MAX_ORDER) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794
Sasha Levin309381fea2014-01-23 15:52:54 -0800795 VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
796 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700798continue_merging:
Joonsoo Kim3c605092014-11-13 15:19:21 -0800799 while (order < max_order - 1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800800 buddy_idx = __find_buddy_index(page_idx, order);
801 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700802 if (!page_is_buddy(page, buddy, order))
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700803 goto done_merging;
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800804 /*
805 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
806 * merge with it and move up one order.
807 */
808 if (page_is_guard(buddy)) {
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800809 clear_page_guard(zone, buddy, order, migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800810 } else {
811 list_del(&buddy->lru);
812 zone->free_area[order].nr_free--;
813 rmv_page_order(buddy);
814 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800815 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 page = page + (combined_idx - page_idx);
817 page_idx = combined_idx;
818 order++;
819 }
Vlastimil Babkad9dddbf2016-03-25 14:21:50 -0700820 if (max_order < MAX_ORDER) {
821 /* If we are here, it means order is >= pageblock_order.
822 * We want to prevent merge between freepages on isolate
823 * pageblock and normal pageblock. Without this, pageblock
824 * isolation could cause incorrect freepage or CMA accounting.
825 *
826 * We don't want to hit this code for the more frequent
827 * low-order merging.
828 */
829 if (unlikely(has_isolate_pageblock(zone))) {
830 int buddy_mt;
831
832 buddy_idx = __find_buddy_index(page_idx, order);
833 buddy = page + (buddy_idx - page_idx);
834 buddy_mt = get_pageblock_migratetype(buddy);
835
836 if (migratetype != buddy_mt
837 && (is_migrate_isolate(migratetype) ||
838 is_migrate_isolate(buddy_mt)))
839 goto done_merging;
840 }
841 max_order++;
842 goto continue_merging;
843 }
844
845done_merging:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700847
848 /*
849 * If this is not the largest possible page, check if the buddy
850 * of the next-highest order is free. If it is, it's possible
851 * that pages are being freed that will coalesce soon. In case,
852 * that is happening, add the free page to the tail of the list
853 * so it's less likely to be used soon and more likely to be merged
854 * as a higher order page
855 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700856 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700857 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800858 combined_idx = buddy_idx & page_idx;
859 higher_page = page + (combined_idx - page_idx);
860 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700861 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700862 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
863 list_add_tail(&page->lru,
864 &zone->free_area[order].free_list[migratetype]);
865 goto out;
866 }
867 }
868
869 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
870out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 zone->free_area[order].nr_free++;
872}
873
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700874/*
875 * A bad page could be due to a number of fields. Instead of multiple branches,
876 * try and check multiple fields with one check. The caller must do a detailed
877 * check if necessary.
878 */
879static inline bool page_expected_state(struct page *page,
880 unsigned long check_flags)
881{
882 if (unlikely(atomic_read(&page->_mapcount) != -1))
883 return false;
884
885 if (unlikely((unsigned long)page->mapping |
886 page_ref_count(page) |
887#ifdef CONFIG_MEMCG
888 (unsigned long)page->mem_cgroup |
889#endif
890 (page->flags & check_flags)))
891 return false;
892
893 return true;
894}
895
Mel Gormanbb552ac2016-05-19 17:14:18 -0700896static void free_pages_check_bad(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897{
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700898 const char *bad_reason;
899 unsigned long bad_flags;
900
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700901 bad_reason = NULL;
902 bad_flags = 0;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800903
Kirill A. Shutemov53f92632016-01-15 16:53:42 -0800904 if (unlikely(atomic_read(&page->_mapcount) != -1))
Dave Hansenf0b791a2014-01-23 15:52:49 -0800905 bad_reason = "nonzero mapcount";
906 if (unlikely(page->mapping != NULL))
907 bad_reason = "non-NULL mapping";
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700908 if (unlikely(page_ref_count(page) != 0))
Joonsoo Kim0139aa72016-05-19 17:10:49 -0700909 bad_reason = "nonzero _refcount";
Dave Hansenf0b791a2014-01-23 15:52:49 -0800910 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
911 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
912 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
913 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -0800914#ifdef CONFIG_MEMCG
915 if (unlikely(page->mem_cgroup))
916 bad_reason = "page still charged to cgroup";
917#endif
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700918 bad_page(page, bad_reason, bad_flags);
Mel Gormanbb552ac2016-05-19 17:14:18 -0700919}
920
921static inline int free_pages_check(struct page *page)
922{
Mel Gormanda838d42016-05-19 17:14:21 -0700923 if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
Mel Gormanbb552ac2016-05-19 17:14:18 -0700924 return 0;
Mel Gormanbb552ac2016-05-19 17:14:18 -0700925
926 /* Something has gone sideways, find it */
927 free_pages_check_bad(page);
Mel Gorman7bfec6f2016-05-19 17:14:15 -0700928 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929}
930
Mel Gorman4db75482016-05-19 17:14:32 -0700931static int free_tail_pages_check(struct page *head_page, struct page *page)
932{
933 int ret = 1;
934
935 /*
936 * We rely page->lru.next never has bit 0 set, unless the page
937 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
938 */
939 BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
940
941 if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
942 ret = 0;
943 goto out;
944 }
945 switch (page - head_page) {
946 case 1:
947 /* the first tail page: ->mapping is compound_mapcount() */
948 if (unlikely(compound_mapcount(page))) {
949 bad_page(page, "nonzero compound_mapcount", 0);
950 goto out;
951 }
952 break;
953 case 2:
954 /*
955 * the second tail page: ->mapping is
956 * page_deferred_list().next -- ignore value.
957 */
958 break;
959 default:
960 if (page->mapping != TAIL_MAPPING) {
961 bad_page(page, "corrupted mapping in tail page", 0);
962 goto out;
963 }
964 break;
965 }
966 if (unlikely(!PageTail(page))) {
967 bad_page(page, "PageTail not set", 0);
968 goto out;
969 }
970 if (unlikely(compound_head(page) != head_page)) {
971 bad_page(page, "compound_head not consistent", 0);
972 goto out;
973 }
974 ret = 0;
975out:
976 page->mapping = NULL;
977 clear_compound_head(page);
978 return ret;
979}
980
Mel Gormane2769db2016-05-19 17:14:38 -0700981static __always_inline bool free_pages_prepare(struct page *page,
982 unsigned int order, bool check_free)
983{
984 int bad = 0;
985
986 VM_BUG_ON_PAGE(PageTail(page), page);
987
988 trace_mm_page_free(page, order);
989 kmemcheck_free_shadow(page, order);
Mel Gormane2769db2016-05-19 17:14:38 -0700990
991 /*
992 * Check tail pages before head page information is cleared to
993 * avoid checking PageCompound for order-0 pages.
994 */
995 if (unlikely(order)) {
996 bool compound = PageCompound(page);
997 int i;
998
999 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1000
1001 for (i = 1; i < (1 << order); i++) {
1002 if (compound)
1003 bad += free_tail_pages_check(page, page + i);
1004 if (unlikely(free_pages_check(page + i))) {
1005 bad++;
1006 continue;
1007 }
1008 (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1009 }
1010 }
1011 if (PageAnonHead(page))
1012 page->mapping = NULL;
1013 if (check_free)
1014 bad += free_pages_check(page);
1015 if (bad)
1016 return false;
1017
1018 page_cpupid_reset_last(page);
1019 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1020 reset_page_owner(page, order);
1021
1022 if (!PageHighMem(page)) {
1023 debug_check_no_locks_freed(page_address(page),
1024 PAGE_SIZE << order);
1025 debug_check_no_obj_freed(page_address(page),
1026 PAGE_SIZE << order);
1027 }
1028 arch_free_page(page, order);
1029 kernel_poison_pages(page, 1 << order, 0);
1030 kernel_map_pages(page, 1 << order, 0);
seokhoon.yoon29b52de2016-05-20 16:58:47 -07001031 kasan_free_pages(page, order);
Mel Gormane2769db2016-05-19 17:14:38 -07001032
1033 return true;
1034}
Mel Gorman4db75482016-05-19 17:14:32 -07001035
1036#ifdef CONFIG_DEBUG_VM
1037static inline bool free_pcp_prepare(struct page *page)
1038{
Mel Gormane2769db2016-05-19 17:14:38 -07001039 return free_pages_prepare(page, 0, true);
Mel Gorman4db75482016-05-19 17:14:32 -07001040}
1041
1042static inline bool bulkfree_pcp_prepare(struct page *page)
1043{
1044 return false;
1045}
1046#else
1047static bool free_pcp_prepare(struct page *page)
1048{
Mel Gormane2769db2016-05-19 17:14:38 -07001049 return free_pages_prepare(page, 0, false);
Mel Gorman4db75482016-05-19 17:14:32 -07001050}
1051
1052static bool bulkfree_pcp_prepare(struct page *page)
1053{
1054 return free_pages_check(page);
1055}
1056#endif /* CONFIG_DEBUG_VM */
1057
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001059 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -07001061 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 *
1063 * If the zone was previously in an "all pages pinned" state then look to
1064 * see if this freeing clears that state.
1065 *
1066 * And clear the zone's pages_scanned counter, to hold off the "all pages are
1067 * pinned" detection logic.
1068 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001069static void free_pcppages_bulk(struct zone *zone, int count,
1070 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071{
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001072 int migratetype = 0;
Mel Gormana6f9edd62009-09-21 17:03:20 -07001073 int batch_free = 0;
Mel Gorman0d5d8232014-08-06 16:07:16 -07001074 unsigned long nr_scanned;
Mel Gorman37779992016-05-19 17:13:58 -07001075 bool isolated_pageblocks;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001076
Nick Pigginc54ad302006-01-06 00:10:56 -08001077 spin_lock(&zone->lock);
Mel Gorman37779992016-05-19 17:13:58 -07001078 isolated_pageblocks = has_isolate_pageblock(zone);
Mel Gorman0d5d8232014-08-06 16:07:16 -07001079 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
1080 if (nr_scanned)
1081 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -07001082
Mel Gormane5b31ac2016-05-19 17:14:24 -07001083 while (count) {
Nick Piggin48db57f2006-01-08 01:00:42 -08001084 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001085 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -08001086
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001087 /*
Mel Gormana6f9edd62009-09-21 17:03:20 -07001088 * Remove pages from lists in a round-robin fashion. A
1089 * batch_free count is maintained that is incremented when an
1090 * empty list is encountered. This is so more pages are freed
1091 * off fuller lists instead of spinning excessively around empty
1092 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001093 */
1094 do {
Mel Gormana6f9edd62009-09-21 17:03:20 -07001095 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001096 if (++migratetype == MIGRATE_PCPTYPES)
1097 migratetype = 0;
1098 list = &pcp->lists[migratetype];
1099 } while (list_empty(list));
1100
Namhyung Kim1d168712011-03-22 16:32:45 -07001101 /* This is the only non-empty list. Free them all. */
1102 if (batch_free == MIGRATE_PCPTYPES)
Mel Gormane5b31ac2016-05-19 17:14:24 -07001103 batch_free = count;
Namhyung Kim1d168712011-03-22 16:32:45 -07001104
Mel Gormana6f9edd62009-09-21 17:03:20 -07001105 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -07001106 int mt; /* migratetype of the to-be-freed page */
1107
Geliang Tanga16601c2016-01-14 15:20:30 -08001108 page = list_last_entry(list, struct page, lru);
Mel Gormana6f9edd62009-09-21 17:03:20 -07001109 /* must delete as __free_one_page list manipulates */
1110 list_del(&page->lru);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -07001111
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001112 mt = get_pcppage_migratetype(page);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -07001113 /* MIGRATE_ISOLATE page should not go to pcplists */
1114 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1115 /* Pageblock could have been isolated meanwhile */
Mel Gorman37779992016-05-19 17:13:58 -07001116 if (unlikely(isolated_pageblocks))
Joonsoo Kim51bb1a42014-11-13 15:19:14 -08001117 mt = get_pageblock_migratetype(page);
Joonsoo Kim51bb1a42014-11-13 15:19:14 -08001118
Mel Gorman4db75482016-05-19 17:14:32 -07001119 if (bulkfree_pcp_prepare(page))
1120 continue;
1121
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001122 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -07001123 trace_mm_page_pcpu_drain(page, 0, mt);
Mel Gormane5b31ac2016-05-19 17:14:24 -07001124 } while (--count && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 }
Nick Pigginc54ad302006-01-06 00:10:56 -08001126 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127}
1128
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001129static void free_one_page(struct zone *zone,
1130 struct page *page, unsigned long pfn,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001131 unsigned int order,
Mel Gormaned0ae212009-06-16 15:32:07 -07001132 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -08001133{
Mel Gorman0d5d8232014-08-06 16:07:16 -07001134 unsigned long nr_scanned;
Christoph Lameter006d22d2006-09-25 23:31:48 -07001135 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -07001136 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
1137 if (nr_scanned)
1138 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -07001139
Joonsoo Kimad53f922014-11-13 15:19:11 -08001140 if (unlikely(has_isolate_pageblock(zone) ||
1141 is_migrate_isolate(migratetype))) {
1142 migratetype = get_pfnblock_migratetype(page, pfn);
Joonsoo Kimad53f922014-11-13 15:19:11 -08001143 }
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001144 __free_one_page(page, pfn, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -07001145 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -08001146}
1147
Robin Holt1e8ce832015-06-30 14:56:45 -07001148static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1149 unsigned long zone, int nid)
1150{
Robin Holt1e8ce832015-06-30 14:56:45 -07001151 set_page_links(page, zone, nid, pfn);
Robin Holt1e8ce832015-06-30 14:56:45 -07001152 init_page_count(page);
1153 page_mapcount_reset(page);
1154 page_cpupid_reset_last(page);
Robin Holt1e8ce832015-06-30 14:56:45 -07001155
Robin Holt1e8ce832015-06-30 14:56:45 -07001156 INIT_LIST_HEAD(&page->lru);
1157#ifdef WANT_PAGE_VIRTUAL
1158 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1159 if (!is_highmem_idx(zone))
1160 set_page_address(page, __va(pfn << PAGE_SHIFT));
1161#endif
1162}
1163
1164static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
1165 int nid)
1166{
1167 return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
1168}
1169
Mel Gorman7e18adb2015-06-30 14:57:05 -07001170#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1171static void init_reserved_page(unsigned long pfn)
1172{
1173 pg_data_t *pgdat;
1174 int nid, zid;
1175
1176 if (!early_page_uninitialised(pfn))
1177 return;
1178
1179 nid = early_pfn_to_nid(pfn);
1180 pgdat = NODE_DATA(nid);
1181
1182 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1183 struct zone *zone = &pgdat->node_zones[zid];
1184
1185 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1186 break;
1187 }
1188 __init_single_pfn(pfn, zid, nid);
1189}
1190#else
1191static inline void init_reserved_page(unsigned long pfn)
1192{
1193}
1194#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1195
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001196/*
1197 * Initialised pages do not have PageReserved set. This function is
1198 * called for each range allocated by the bootmem allocator and
1199 * marks the pages PageReserved. The remaining valid pages are later
1200 * sent to the buddy page allocator.
1201 */
Stefan Bader4b50bcc2016-05-20 16:58:38 -07001202void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001203{
1204 unsigned long start_pfn = PFN_DOWN(start);
1205 unsigned long end_pfn = PFN_UP(end);
1206
Mel Gorman7e18adb2015-06-30 14:57:05 -07001207 for (; start_pfn < end_pfn; start_pfn++) {
1208 if (pfn_valid(start_pfn)) {
1209 struct page *page = pfn_to_page(start_pfn);
1210
1211 init_reserved_page(start_pfn);
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -08001212
1213 /* Avoid false-positive PageTail() */
1214 INIT_LIST_HEAD(&page->lru);
1215
Mel Gorman7e18adb2015-06-30 14:57:05 -07001216 SetPageReserved(page);
1217 }
1218 }
Nathan Zimmer92923ca2015-06-30 14:56:48 -07001219}
1220
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001221static void __free_pages_ok(struct page *page, unsigned int order)
1222{
1223 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -07001224 int migratetype;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001225 unsigned long pfn = page_to_pfn(page);
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001226
Mel Gormane2769db2016-05-19 17:14:38 -07001227 if (!free_pages_prepare(page, order, true))
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001228 return;
1229
Mel Gormancfc47a22014-06-04 16:10:19 -07001230 migratetype = get_pfnblock_migratetype(page, pfn);
Nick Pigginc54ad302006-01-06 00:10:56 -08001231 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001232 __count_vm_events(PGFREE, 1 << order);
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001233 free_one_page(page_zone(page), page, pfn, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -08001234 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235}
1236
Li Zhang949698a2016-05-19 17:11:37 -07001237static void __init __free_pages_boot_core(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -08001238{
Johannes Weinerc3993072012-01-10 15:08:10 -08001239 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001240 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -08001241 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -08001242
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001243 prefetchw(p);
1244 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1245 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -08001246 __ClearPageReserved(p);
1247 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -08001248 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001249 __ClearPageReserved(p);
1250 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -08001251
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001252 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -08001253 set_page_refcounted(page);
1254 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -08001255}
1256
Mel Gorman75a592a2015-06-30 14:56:59 -07001257#if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1258 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
Mel Gorman7ace9912015-08-06 15:46:13 -07001259
Mel Gorman75a592a2015-06-30 14:56:59 -07001260static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1261
1262int __meminit early_pfn_to_nid(unsigned long pfn)
1263{
Mel Gorman7ace9912015-08-06 15:46:13 -07001264 static DEFINE_SPINLOCK(early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -07001265 int nid;
1266
Mel Gorman7ace9912015-08-06 15:46:13 -07001267 spin_lock(&early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -07001268 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
Mel Gorman7ace9912015-08-06 15:46:13 -07001269 if (nid < 0)
1270 nid = 0;
1271 spin_unlock(&early_pfn_lock);
1272
1273 return nid;
Mel Gorman75a592a2015-06-30 14:56:59 -07001274}
1275#endif
1276
1277#ifdef CONFIG_NODES_SPAN_OTHER_NODES
1278static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1279 struct mminit_pfnnid_cache *state)
1280{
1281 int nid;
1282
1283 nid = __early_pfn_to_nid(pfn, state);
1284 if (nid >= 0 && nid != node)
1285 return false;
1286 return true;
1287}
1288
1289/* Only safe to use early in boot when initialisation is single-threaded */
1290static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1291{
1292 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1293}
1294
1295#else
1296
1297static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1298{
1299 return true;
1300}
1301static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1302 struct mminit_pfnnid_cache *state)
1303{
1304 return true;
1305}
1306#endif
1307
1308
Mel Gorman0e1cc952015-06-30 14:57:27 -07001309void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001310 unsigned int order)
1311{
1312 if (early_page_uninitialised(pfn))
1313 return;
Li Zhang949698a2016-05-19 17:11:37 -07001314 return __free_pages_boot_core(page, order);
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001315}
1316
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001317/*
1318 * Check that the whole (or subset of) a pageblock given by the interval of
1319 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1320 * with the migration of free compaction scanner. The scanners then need to
1321 * use only pfn_valid_within() check for arches that allow holes within
1322 * pageblocks.
1323 *
1324 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1325 *
1326 * It's possible on some configurations to have a setup like node0 node1 node0
1327 * i.e. it's possible that all pages within a zones range of pages do not
1328 * belong to a single zone. We assume that a border between node0 and node1
1329 * can occur within a single pageblock, but not a node0 node1 node0
1330 * interleaving within a single pageblock. It is therefore sufficient to check
1331 * the first and last page of a pageblock and avoid checking each individual
1332 * page in a pageblock.
1333 */
1334struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1335 unsigned long end_pfn, struct zone *zone)
1336{
1337 struct page *start_page;
1338 struct page *end_page;
1339
1340 /* end_pfn is one past the range we are checking */
1341 end_pfn--;
1342
1343 if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1344 return NULL;
1345
1346 start_page = pfn_to_page(start_pfn);
1347
1348 if (page_zone(start_page) != zone)
1349 return NULL;
1350
1351 end_page = pfn_to_page(end_pfn);
1352
1353 /* This gives a shorter code than deriving page_zone(end_page) */
1354 if (page_zone_id(start_page) != page_zone_id(end_page))
1355 return NULL;
1356
1357 return start_page;
1358}
1359
1360void set_zone_contiguous(struct zone *zone)
1361{
1362 unsigned long block_start_pfn = zone->zone_start_pfn;
1363 unsigned long block_end_pfn;
1364
1365 block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1366 for (; block_start_pfn < zone_end_pfn(zone);
1367 block_start_pfn = block_end_pfn,
1368 block_end_pfn += pageblock_nr_pages) {
1369
1370 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1371
1372 if (!__pageblock_pfn_to_page(block_start_pfn,
1373 block_end_pfn, zone))
1374 return;
1375 }
1376
1377 /* We confirm that there is no hole */
1378 zone->contiguous = true;
1379}
1380
1381void clear_zone_contiguous(struct zone *zone)
1382{
1383 zone->contiguous = false;
1384}
1385
Mel Gorman7e18adb2015-06-30 14:57:05 -07001386#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
Mel Gorman0e1cc952015-06-30 14:57:27 -07001387static void __init deferred_free_range(struct page *page,
Mel Gormana4de83d2015-06-30 14:57:16 -07001388 unsigned long pfn, int nr_pages)
1389{
1390 int i;
1391
1392 if (!page)
1393 return;
1394
1395 /* Free a large naturally-aligned chunk if possible */
1396 if (nr_pages == MAX_ORDER_NR_PAGES &&
1397 (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
Mel Gormanac5d2532015-06-30 14:57:20 -07001398 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Li Zhang949698a2016-05-19 17:11:37 -07001399 __free_pages_boot_core(page, MAX_ORDER-1);
Mel Gormana4de83d2015-06-30 14:57:16 -07001400 return;
1401 }
1402
Li Zhang949698a2016-05-19 17:11:37 -07001403 for (i = 0; i < nr_pages; i++, page++)
1404 __free_pages_boot_core(page, 0);
Mel Gormana4de83d2015-06-30 14:57:16 -07001405}
1406
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001407/* Completion tracking for deferred_init_memmap() threads */
1408static atomic_t pgdat_init_n_undone __initdata;
1409static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1410
1411static inline void __init pgdat_init_report_one_done(void)
1412{
1413 if (atomic_dec_and_test(&pgdat_init_n_undone))
1414 complete(&pgdat_init_all_done_comp);
1415}
Mel Gorman0e1cc952015-06-30 14:57:27 -07001416
Mel Gorman7e18adb2015-06-30 14:57:05 -07001417/* Initialise remaining memory on a node */
Mel Gorman0e1cc952015-06-30 14:57:27 -07001418static int __init deferred_init_memmap(void *data)
Mel Gorman7e18adb2015-06-30 14:57:05 -07001419{
Mel Gorman0e1cc952015-06-30 14:57:27 -07001420 pg_data_t *pgdat = data;
1421 int nid = pgdat->node_id;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001422 struct mminit_pfnnid_cache nid_init_state = { };
1423 unsigned long start = jiffies;
1424 unsigned long nr_pages = 0;
1425 unsigned long walk_start, walk_end;
1426 int i, zid;
1427 struct zone *zone;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001428 unsigned long first_init_pfn = pgdat->first_deferred_pfn;
Mel Gorman0e1cc952015-06-30 14:57:27 -07001429 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001430
Mel Gorman0e1cc952015-06-30 14:57:27 -07001431 if (first_init_pfn == ULONG_MAX) {
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001432 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001433 return 0;
1434 }
1435
1436 /* Bind memory initialisation thread to a local node if possible */
1437 if (!cpumask_empty(cpumask))
1438 set_cpus_allowed_ptr(current, cpumask);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001439
1440 /* Sanity check boundaries */
1441 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1442 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1443 pgdat->first_deferred_pfn = ULONG_MAX;
1444
1445 /* Only the highest zone is deferred so find it */
1446 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1447 zone = pgdat->node_zones + zid;
1448 if (first_init_pfn < zone_end_pfn(zone))
1449 break;
1450 }
1451
1452 for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1453 unsigned long pfn, end_pfn;
Mel Gorman54608c32015-06-30 14:57:09 -07001454 struct page *page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001455 struct page *free_base_page = NULL;
1456 unsigned long free_base_pfn = 0;
1457 int nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001458
1459 end_pfn = min(walk_end, zone_end_pfn(zone));
1460 pfn = first_init_pfn;
1461 if (pfn < walk_start)
1462 pfn = walk_start;
1463 if (pfn < zone->zone_start_pfn)
1464 pfn = zone->zone_start_pfn;
1465
1466 for (; pfn < end_pfn; pfn++) {
Mel Gorman54608c32015-06-30 14:57:09 -07001467 if (!pfn_valid_within(pfn))
Mel Gormana4de83d2015-06-30 14:57:16 -07001468 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001469
Mel Gorman54608c32015-06-30 14:57:09 -07001470 /*
1471 * Ensure pfn_valid is checked every
1472 * MAX_ORDER_NR_PAGES for memory holes
1473 */
1474 if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1475 if (!pfn_valid(pfn)) {
1476 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001477 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001478 }
1479 }
1480
1481 if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1482 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001483 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001484 }
1485
1486 /* Minimise pfn page lookups and scheduler checks */
1487 if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1488 page++;
1489 } else {
Mel Gormana4de83d2015-06-30 14:57:16 -07001490 nr_pages += nr_to_free;
1491 deferred_free_range(free_base_page,
1492 free_base_pfn, nr_to_free);
1493 free_base_page = NULL;
1494 free_base_pfn = nr_to_free = 0;
1495
Mel Gorman54608c32015-06-30 14:57:09 -07001496 page = pfn_to_page(pfn);
1497 cond_resched();
1498 }
Mel Gorman7e18adb2015-06-30 14:57:05 -07001499
1500 if (page->flags) {
1501 VM_BUG_ON(page_zone(page) != zone);
Mel Gormana4de83d2015-06-30 14:57:16 -07001502 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001503 }
1504
1505 __init_single_page(page, pfn, zid, nid);
Mel Gormana4de83d2015-06-30 14:57:16 -07001506 if (!free_base_page) {
1507 free_base_page = page;
1508 free_base_pfn = pfn;
1509 nr_to_free = 0;
1510 }
1511 nr_to_free++;
1512
1513 /* Where possible, batch up pages for a single free */
1514 continue;
1515free_range:
1516 /* Free the current block of pages to allocator */
1517 nr_pages += nr_to_free;
1518 deferred_free_range(free_base_page, free_base_pfn,
1519 nr_to_free);
1520 free_base_page = NULL;
1521 free_base_pfn = nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001522 }
Mel Gormana4de83d2015-06-30 14:57:16 -07001523
Mel Gorman7e18adb2015-06-30 14:57:05 -07001524 first_init_pfn = max(end_pfn, first_init_pfn);
1525 }
1526
1527 /* Sanity check that the next zone really is unpopulated */
1528 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1529
Mel Gorman0e1cc952015-06-30 14:57:27 -07001530 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
Mel Gorman7e18adb2015-06-30 14:57:05 -07001531 jiffies_to_msecs(jiffies - start));
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001532
1533 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001534 return 0;
1535}
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001536#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
Mel Gorman0e1cc952015-06-30 14:57:27 -07001537
1538void __init page_alloc_init_late(void)
1539{
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001540 struct zone *zone;
1541
1542#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
Mel Gorman0e1cc952015-06-30 14:57:27 -07001543 int nid;
1544
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001545 /* There will be num_node_state(N_MEMORY) threads */
1546 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
Mel Gorman0e1cc952015-06-30 14:57:27 -07001547 for_each_node_state(nid, N_MEMORY) {
Mel Gorman0e1cc952015-06-30 14:57:27 -07001548 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1549 }
1550
1551 /* Block until all are initialised */
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001552 wait_for_completion(&pgdat_init_all_done_comp);
Mel Gorman4248b0d2015-08-06 15:46:20 -07001553
1554 /* Reinit limits that are based on free pages after the kernel is up */
1555 files_maxfiles_init();
Joonsoo Kim7cf91a92016-03-15 14:57:51 -07001556#endif
1557
1558 for_each_populated_zone(zone)
1559 set_zone_contiguous(zone);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001560}
Mel Gorman7e18adb2015-06-30 14:57:05 -07001561
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001562#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +08001563/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001564void __init init_cma_reserved_pageblock(struct page *page)
1565{
1566 unsigned i = pageblock_nr_pages;
1567 struct page *p = page;
1568
1569 do {
1570 __ClearPageReserved(p);
1571 set_page_count(p, 0);
1572 } while (++p, --i);
1573
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001574 set_pageblock_migratetype(page, MIGRATE_CMA);
Michal Nazarewiczdc783272014-07-02 15:22:35 -07001575
1576 if (pageblock_order >= MAX_ORDER) {
1577 i = pageblock_nr_pages;
1578 p = page;
1579 do {
1580 set_page_refcounted(p);
1581 __free_pages(p, MAX_ORDER - 1);
1582 p += MAX_ORDER_NR_PAGES;
1583 } while (i -= MAX_ORDER_NR_PAGES);
1584 } else {
1585 set_page_refcounted(page);
1586 __free_pages(page, pageblock_order);
1587 }
1588
Jiang Liu3dcc0572013-07-03 15:03:21 -07001589 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001590}
1591#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592
1593/*
1594 * The order of subdivision here is critical for the IO subsystem.
1595 * Please do not alter this order without good reasons and regression
1596 * testing. Specifically, as large blocks of memory are subdivided,
1597 * the order in which smaller blocks are delivered depends on the order
1598 * they're subdivided in this function. This is the primary factor
1599 * influencing the order in which pages are delivered to the IO
1600 * subsystem according to empirical testing, and this is also justified
1601 * by considering the behavior of a buddy system containing a single
1602 * large block of memory acted on by a series of small allocations.
1603 * This behavior is a critical factor in sglist merging's success.
1604 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +01001605 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 */
Nick Piggin085cc7d2006-01-06 00:11:01 -08001607static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001608 int low, int high, struct free_area *area,
1609 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610{
1611 unsigned long size = 1 << high;
1612
1613 while (high > low) {
1614 area--;
1615 high--;
1616 size >>= 1;
Sasha Levin309381fea2014-01-23 15:52:54 -08001617 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08001618
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001619 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
Joonsoo Kime30825f2014-12-12 16:55:49 -08001620 debug_guardpage_enabled() &&
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001621 high < debug_guardpage_minorder()) {
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08001622 /*
1623 * Mark as guard pages (or page), that will allow to
1624 * merge back to allocator when buddy will be freed.
1625 * Corresponding page table entries will not be touched,
1626 * pages will stay not present in virtual address space
1627 */
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001628 set_page_guard(zone, &page[size], high, migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08001629 continue;
1630 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001631 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 area->nr_free++;
1633 set_page_order(&page[size], high);
1634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635}
1636
Vlastimil Babka4e611802016-05-19 17:14:41 -07001637static void check_new_page_bad(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638{
Vlastimil Babka4e611802016-05-19 17:14:41 -07001639 const char *bad_reason = NULL;
1640 unsigned long bad_flags = 0;
Dave Hansenf0b791a2014-01-23 15:52:49 -08001641
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001642 if (unlikely(atomic_read(&page->_mapcount) != -1))
Dave Hansenf0b791a2014-01-23 15:52:49 -08001643 bad_reason = "nonzero mapcount";
1644 if (unlikely(page->mapping != NULL))
1645 bad_reason = "non-NULL mapping";
Joonsoo Kimfe896d12016-03-17 14:19:26 -07001646 if (unlikely(page_ref_count(page) != 0))
Dave Hansenf0b791a2014-01-23 15:52:49 -08001647 bad_reason = "nonzero _count";
Naoya Horiguchif4c18e62015-08-06 15:47:08 -07001648 if (unlikely(page->flags & __PG_HWPOISON)) {
1649 bad_reason = "HWPoisoned (hardware-corrupted)";
1650 bad_flags = __PG_HWPOISON;
Naoya Horiguchie570f562016-05-20 16:58:50 -07001651 /* Don't complain about hwpoisoned pages */
1652 page_mapcount_reset(page); /* remove PageBuddy */
1653 return;
Naoya Horiguchif4c18e62015-08-06 15:47:08 -07001654 }
Dave Hansenf0b791a2014-01-23 15:52:49 -08001655 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1656 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1657 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1658 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -08001659#ifdef CONFIG_MEMCG
1660 if (unlikely(page->mem_cgroup))
1661 bad_reason = "page still charged to cgroup";
1662#endif
Vlastimil Babka4e611802016-05-19 17:14:41 -07001663 bad_page(page, bad_reason, bad_flags);
1664}
1665
1666/*
1667 * This page is about to be returned from the page allocator
1668 */
1669static inline int check_new_page(struct page *page)
1670{
1671 if (likely(page_expected_state(page,
1672 PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
1673 return 0;
1674
1675 check_new_page_bad(page);
1676 return 1;
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001677}
1678
Laura Abbott1414c7f2016-03-15 14:56:30 -07001679static inline bool free_pages_prezeroed(bool poisoned)
1680{
1681 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1682 page_poisoning_enabled() && poisoned;
1683}
1684
Mel Gorman479f8542016-05-19 17:14:35 -07001685#ifdef CONFIG_DEBUG_VM
1686static bool check_pcp_refill(struct page *page)
1687{
1688 return false;
1689}
1690
1691static bool check_new_pcp(struct page *page)
1692{
1693 return check_new_page(page);
1694}
1695#else
1696static bool check_pcp_refill(struct page *page)
1697{
1698 return check_new_page(page);
1699}
1700static bool check_new_pcp(struct page *page)
1701{
1702 return false;
1703}
1704#endif /* CONFIG_DEBUG_VM */
1705
1706static bool check_new_pages(struct page *page, unsigned int order)
1707{
1708 int i;
1709 for (i = 0; i < (1 << order); i++) {
1710 struct page *p = page + i;
1711
1712 if (unlikely(check_new_page(p)))
1713 return true;
1714 }
1715
1716 return false;
1717}
1718
1719static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
Mel Gormanc6038442016-05-19 17:13:38 -07001720 unsigned int alloc_flags)
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001721{
1722 int i;
Laura Abbott1414c7f2016-03-15 14:56:30 -07001723 bool poisoned = true;
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001724
1725 for (i = 0; i < (1 << order); i++) {
1726 struct page *p = page + i;
Laura Abbott1414c7f2016-03-15 14:56:30 -07001727 if (poisoned)
1728 poisoned &= page_is_poisoned(p);
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001729 }
Hugh Dickins689bceb2005-11-21 21:32:20 -08001730
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001731 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -08001732 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -08001733
1734 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 kernel_map_pages(page, 1 << order, 1);
Laura Abbott8823b1d2016-03-15 14:56:27 -07001736 kernel_poison_pages(page, 1 << order, 1);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -08001737 kasan_alloc_pages(page, order);
Nick Piggin17cf4402006-03-22 00:08:41 -08001738
Laura Abbott1414c7f2016-03-15 14:56:30 -07001739 if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO))
Anisse Astierf4d28972015-06-24 16:56:36 -07001740 for (i = 0; i < (1 << order); i++)
1741 clear_highpage(page + i);
Nick Piggin17cf4402006-03-22 00:08:41 -08001742
1743 if (order && (gfp_flags & __GFP_COMP))
1744 prep_compound_page(page, order);
1745
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001746 set_page_owner(page, order, gfp_flags);
1747
Vlastimil Babka75379192015-02-11 15:25:38 -08001748 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001749 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
Vlastimil Babka75379192015-02-11 15:25:38 -08001750 * allocate the page. The expectation is that the caller is taking
1751 * steps that will free more memory. The caller should avoid the page
1752 * being used for !PFMEMALLOC purposes.
1753 */
Michal Hocko2f064f32015-08-21 14:11:51 -07001754 if (alloc_flags & ALLOC_NO_WATERMARKS)
1755 set_page_pfmemalloc(page);
1756 else
1757 clear_page_pfmemalloc(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758}
1759
Mel Gorman56fd56b2007-10-16 01:25:58 -07001760/*
1761 * Go through the free lists for the given migratetype and remove
1762 * the smallest available page from the freelists
1763 */
Mel Gorman728ec982009-06-16 15:32:04 -07001764static inline
1765struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -07001766 int migratetype)
1767{
1768 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -07001769 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001770 struct page *page;
1771
1772 /* Find a page of the appropriate size in the preferred list */
1773 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1774 area = &(zone->free_area[current_order]);
Geliang Tanga16601c2016-01-14 15:20:30 -08001775 page = list_first_entry_or_null(&area->free_list[migratetype],
Mel Gorman56fd56b2007-10-16 01:25:58 -07001776 struct page, lru);
Geliang Tanga16601c2016-01-14 15:20:30 -08001777 if (!page)
1778 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001779 list_del(&page->lru);
1780 rmv_page_order(page);
1781 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001782 expand(zone, page, order, current_order, area, migratetype);
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001783 set_pcppage_migratetype(page, migratetype);
Mel Gorman56fd56b2007-10-16 01:25:58 -07001784 return page;
1785 }
1786
1787 return NULL;
1788}
1789
1790
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001791/*
1792 * This array describes the order lists are fallen back to when
1793 * the free lists for the desirable migrate type are depleted
1794 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001795static int fallbacks[MIGRATE_TYPES][4] = {
Mel Gorman974a7862015-11-06 16:28:34 -08001796 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1797 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1798 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
Joonsoo Kimdc676472015-04-14 15:45:15 -07001799#ifdef CONFIG_CMA
Mel Gorman974a7862015-11-06 16:28:34 -08001800 [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001801#endif
Minchan Kim194159f2013-02-22 16:33:58 -08001802#ifdef CONFIG_MEMORY_ISOLATION
Mel Gorman974a7862015-11-06 16:28:34 -08001803 [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001804#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001805};
1806
Joonsoo Kimdc676472015-04-14 15:45:15 -07001807#ifdef CONFIG_CMA
1808static struct page *__rmqueue_cma_fallback(struct zone *zone,
1809 unsigned int order)
1810{
1811 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1812}
1813#else
1814static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1815 unsigned int order) { return NULL; }
1816#endif
1817
Mel Gormanc361be52007-10-16 01:25:51 -07001818/*
1819 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -07001820 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -07001821 * boundary. If alignment is required, use move_freepages_block()
1822 */
Minchan Kim435b4052012-10-08 16:32:16 -07001823int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -07001824 struct page *start_page, struct page *end_page,
1825 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001826{
1827 struct page *page;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08001828 unsigned int order;
Mel Gormand1003132007-10-16 01:26:00 -07001829 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -07001830
1831#ifndef CONFIG_HOLES_IN_ZONE
1832 /*
1833 * page_zone is not safe to call in this context when
1834 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1835 * anyway as we check zone boundaries in move_freepages_block().
1836 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -07001837 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -07001838 */
Mel Gorman97ee4ba2014-10-09 15:28:28 -07001839 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
Mel Gormanc361be52007-10-16 01:25:51 -07001840#endif
1841
1842 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -07001843 /* Make sure we are not inadvertently changing nodes */
Sasha Levin309381fea2014-01-23 15:52:54 -08001844 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
Adam Litke344c7902008-09-02 14:35:38 -07001845
Mel Gormanc361be52007-10-16 01:25:51 -07001846 if (!pfn_valid_within(page_to_pfn(page))) {
1847 page++;
1848 continue;
1849 }
1850
1851 if (!PageBuddy(page)) {
1852 page++;
1853 continue;
1854 }
1855
1856 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001857 list_move(&page->lru,
1858 &zone->free_area[order].free_list[migratetype]);
Mel Gormanc361be52007-10-16 01:25:51 -07001859 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001860 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001861 }
1862
Mel Gormand1003132007-10-16 01:26:00 -07001863 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001864}
1865
Minchan Kimee6f5092012-07-31 16:43:50 -07001866int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001867 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001868{
1869 unsigned long start_pfn, end_pfn;
1870 struct page *start_page, *end_page;
1871
1872 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001873 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001874 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001875 end_page = start_page + pageblock_nr_pages - 1;
1876 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001877
1878 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001879 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001880 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001881 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001882 return 0;
1883
1884 return move_freepages(zone, start_page, end_page, migratetype);
1885}
1886
Mel Gorman2f66a682009-09-21 17:02:31 -07001887static void change_pageblock_range(struct page *pageblock_page,
1888 int start_order, int migratetype)
1889{
1890 int nr_pageblocks = 1 << (start_order - pageblock_order);
1891
1892 while (nr_pageblocks--) {
1893 set_pageblock_migratetype(pageblock_page, migratetype);
1894 pageblock_page += pageblock_nr_pages;
1895 }
1896}
1897
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001898/*
Vlastimil Babka9c0415e2015-02-11 15:28:21 -08001899 * When we are falling back to another migratetype during allocation, try to
1900 * steal extra free pages from the same pageblocks to satisfy further
1901 * allocations, instead of polluting multiple pageblocks.
1902 *
1903 * If we are stealing a relatively large buddy page, it is likely there will
1904 * be more free pages in the pageblock, so try to steal them all. For
1905 * reclaimable and unmovable allocations, we steal regardless of page size,
1906 * as fragmentation caused by those allocations polluting movable pageblocks
1907 * is worse than movable allocations stealing from unmovable and reclaimable
1908 * pageblocks.
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001909 */
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001910static bool can_steal_fallback(unsigned int order, int start_mt)
1911{
1912 /*
1913 * Leaving this order check is intended, although there is
1914 * relaxed order check in next check. The reason is that
1915 * we can actually steal whole pageblock if this condition met,
1916 * but, below check doesn't guarantee it and that is just heuristic
1917 * so could be changed anytime.
1918 */
1919 if (order >= pageblock_order)
1920 return true;
1921
1922 if (order >= pageblock_order / 2 ||
1923 start_mt == MIGRATE_RECLAIMABLE ||
1924 start_mt == MIGRATE_UNMOVABLE ||
1925 page_group_by_mobility_disabled)
1926 return true;
1927
1928 return false;
1929}
1930
1931/*
1932 * This function implements actual steal behaviour. If order is large enough,
1933 * we can steal whole pageblock. If not, we first move freepages in this
1934 * pageblock and check whether half of pages are moved or not. If half of
1935 * pages are moved, we can change migratetype of pageblock and permanently
1936 * use it's pages as requested migratetype in the future.
1937 */
1938static void steal_suitable_fallback(struct zone *zone, struct page *page,
1939 int start_type)
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001940{
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08001941 unsigned int current_order = page_order(page);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001942 int pages;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001943
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001944 /* Take ownership for orders >= pageblock_order */
1945 if (current_order >= pageblock_order) {
1946 change_pageblock_range(page, current_order, start_type);
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001947 return;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001948 }
1949
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001950 pages = move_freepages_block(zone, page, start_type);
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001951
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001952 /* Claim the whole block if over half of it is free */
1953 if (pages >= (1 << (pageblock_order-1)) ||
1954 page_group_by_mobility_disabled)
1955 set_pageblock_migratetype(page, start_type);
1956}
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001957
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001958/*
1959 * Check whether there is a suitable fallback freepage with requested order.
1960 * If only_stealable is true, this function returns fallback_mt only if
1961 * we can steal other freepages all together. This would help to reduce
1962 * fragmentation due to mixed migratetype pages in one pageblock.
1963 */
1964int find_suitable_fallback(struct free_area *area, unsigned int order,
1965 int migratetype, bool only_stealable, bool *can_steal)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001966{
1967 int i;
1968 int fallback_mt;
1969
1970 if (area->nr_free == 0)
1971 return -1;
1972
1973 *can_steal = false;
1974 for (i = 0;; i++) {
1975 fallback_mt = fallbacks[migratetype][i];
Mel Gorman974a7862015-11-06 16:28:34 -08001976 if (fallback_mt == MIGRATE_TYPES)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001977 break;
1978
1979 if (list_empty(&area->free_list[fallback_mt]))
1980 continue;
1981
1982 if (can_steal_fallback(order, migratetype))
1983 *can_steal = true;
1984
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001985 if (!only_stealable)
1986 return fallback_mt;
1987
1988 if (*can_steal)
1989 return fallback_mt;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001990 }
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001991
1992 return -1;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001993}
1994
Mel Gorman0aaa29a2015-11-06 16:28:37 -08001995/*
1996 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1997 * there are no empty page blocks that contain a page with a suitable order
1998 */
1999static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2000 unsigned int alloc_order)
2001{
2002 int mt;
2003 unsigned long max_managed, flags;
2004
2005 /*
2006 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2007 * Check is race-prone but harmless.
2008 */
2009 max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
2010 if (zone->nr_reserved_highatomic >= max_managed)
2011 return;
2012
2013 spin_lock_irqsave(&zone->lock, flags);
2014
2015 /* Recheck the nr_reserved_highatomic limit under the lock */
2016 if (zone->nr_reserved_highatomic >= max_managed)
2017 goto out_unlock;
2018
2019 /* Yoink! */
2020 mt = get_pageblock_migratetype(page);
2021 if (mt != MIGRATE_HIGHATOMIC &&
2022 !is_migrate_isolate(mt) && !is_migrate_cma(mt)) {
2023 zone->nr_reserved_highatomic += pageblock_nr_pages;
2024 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2025 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC);
2026 }
2027
2028out_unlock:
2029 spin_unlock_irqrestore(&zone->lock, flags);
2030}
2031
2032/*
2033 * Used when an allocation is about to fail under memory pressure. This
2034 * potentially hurts the reliability of high-order allocations when under
2035 * intense memory pressure but failed atomic allocations should be easier
2036 * to recover from than an OOM.
2037 */
2038static void unreserve_highatomic_pageblock(const struct alloc_context *ac)
2039{
2040 struct zonelist *zonelist = ac->zonelist;
2041 unsigned long flags;
2042 struct zoneref *z;
2043 struct zone *zone;
2044 struct page *page;
2045 int order;
2046
2047 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2048 ac->nodemask) {
2049 /* Preserve at least one pageblock */
2050 if (zone->nr_reserved_highatomic <= pageblock_nr_pages)
2051 continue;
2052
2053 spin_lock_irqsave(&zone->lock, flags);
2054 for (order = 0; order < MAX_ORDER; order++) {
2055 struct free_area *area = &(zone->free_area[order]);
2056
Geliang Tanga16601c2016-01-14 15:20:30 -08002057 page = list_first_entry_or_null(
2058 &area->free_list[MIGRATE_HIGHATOMIC],
2059 struct page, lru);
2060 if (!page)
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002061 continue;
2062
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002063 /*
2064 * It should never happen but changes to locking could
2065 * inadvertently allow a per-cpu drain to add pages
2066 * to MIGRATE_HIGHATOMIC while unreserving so be safe
2067 * and watch for underflows.
2068 */
2069 zone->nr_reserved_highatomic -= min(pageblock_nr_pages,
2070 zone->nr_reserved_highatomic);
2071
2072 /*
2073 * Convert to ac->migratetype and avoid the normal
2074 * pageblock stealing heuristics. Minimally, the caller
2075 * is doing the work and needs the pages. More
2076 * importantly, if the block was always converted to
2077 * MIGRATE_UNMOVABLE or another type then the number
2078 * of pageblocks that cannot be completely freed
2079 * may increase.
2080 */
2081 set_pageblock_migratetype(page, ac->migratetype);
2082 move_freepages_block(zone, page, ac->migratetype);
2083 spin_unlock_irqrestore(&zone->lock, flags);
2084 return;
2085 }
2086 spin_unlock_irqrestore(&zone->lock, flags);
2087 }
2088}
2089
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002090/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07002091static inline struct page *
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002092__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002093{
Pintu Kumarb8af2942013-09-11 14:20:34 -07002094 struct free_area *area;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002095 unsigned int current_order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002096 struct page *page;
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002097 int fallback_mt;
2098 bool can_steal;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002099
2100 /* Find the largest possible block of pages in the other list */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002101 for (current_order = MAX_ORDER-1;
2102 current_order >= order && current_order <= MAX_ORDER-1;
2103 --current_order) {
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002104 area = &(zone->free_area[current_order]);
2105 fallback_mt = find_suitable_fallback(area, current_order,
Joonsoo Kim2149cda2015-04-14 15:45:21 -07002106 start_migratetype, false, &can_steal);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002107 if (fallback_mt == -1)
2108 continue;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002109
Geliang Tanga16601c2016-01-14 15:20:30 -08002110 page = list_first_entry(&area->free_list[fallback_mt],
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002111 struct page, lru);
2112 if (can_steal)
2113 steal_suitable_fallback(zone, page, start_migratetype);
Mel Gormane0104872007-10-16 01:25:53 -07002114
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002115 /* Remove the page from the freelists */
2116 area->nr_free--;
2117 list_del(&page->lru);
2118 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002119
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002120 expand(zone, page, order, current_order, area,
2121 start_migratetype);
2122 /*
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002123 * The pcppage_migratetype may differ from pageblock's
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002124 * migratetype depending on the decisions in
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002125 * find_suitable_fallback(). This is OK as long as it does not
2126 * differ for MIGRATE_CMA pageblocks. Those can be used as
2127 * fallback only via special __rmqueue_cma_fallback() function
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002128 */
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002129 set_pcppage_migratetype(page, start_migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002130
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002131 trace_mm_page_alloc_extfrag(page, order, current_order,
2132 start_migratetype, fallback_mt);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002133
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07002134 return page;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002135 }
2136
Mel Gorman728ec982009-06-16 15:32:04 -07002137 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002138}
2139
Mel Gorman56fd56b2007-10-16 01:25:58 -07002140/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 * Do the hard work of removing an element from the buddy allocator.
2142 * Call me with the zone->lock already held.
2143 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002144static struct page *__rmqueue(struct zone *zone, unsigned int order,
Mel Gorman6ac02062016-01-14 15:20:28 -08002145 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 struct page *page;
2148
Mel Gorman56fd56b2007-10-16 01:25:58 -07002149 page = __rmqueue_smallest(zone, order, migratetype);
Mel Gorman974a7862015-11-06 16:28:34 -08002150 if (unlikely(!page)) {
Joonsoo Kimdc676472015-04-14 15:45:15 -07002151 if (migratetype == MIGRATE_MOVABLE)
2152 page = __rmqueue_cma_fallback(zone, order);
2153
2154 if (!page)
2155 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gorman728ec982009-06-16 15:32:04 -07002156 }
2157
Mel Gorman0d3d0622009-09-21 17:02:44 -07002158 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002159 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160}
2161
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01002162/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 * Obtain a specified number of elements from the buddy allocator, all under
2164 * a single hold of the lock, for efficiency. Add them to the supplied list.
2165 * Returns the number of new pages which were placed at *list.
2166 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01002167static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002168 unsigned long count, struct list_head *list,
Mel Gormanb745bc82014-06-04 16:10:22 -07002169 int migratetype, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170{
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07002171 int i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01002172
Nick Pigginc54ad302006-01-06 00:10:56 -08002173 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 for (i = 0; i < count; ++i) {
Mel Gorman6ac02062016-01-14 15:20:28 -08002175 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08002176 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08002178
Mel Gorman479f8542016-05-19 17:14:35 -07002179 if (unlikely(check_pcp_refill(page)))
2180 continue;
2181
Mel Gorman81eabcb2007-12-17 16:20:05 -08002182 /*
2183 * Split buddy pages returned by expand() are received here
2184 * in physical page order. The page is added to the callers and
2185 * list and the list head then moves forward. From the callers
2186 * perspective, the linked list is ordered by page number in
2187 * some conditions. This is useful for IO devices that can
2188 * merge IO requests if the physical pages are ordered
2189 * properly.
2190 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002191 if (likely(!cold))
Mel Gormane084b2d2009-07-29 15:02:04 -07002192 list_add(&page->lru, list);
2193 else
2194 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08002195 list = &page->lru;
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002196 if (is_migrate_cma(get_pcppage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07002197 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2198 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 }
Mel Gormanf2260e62009-06-16 15:32:13 -07002200 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08002201 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08002202 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203}
2204
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002205#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08002206/*
Christoph Lameter4037d452007-05-09 02:35:14 -07002207 * Called from the vmstat counter updater to drain pagesets of this
2208 * currently executing processor on remote nodes after they have
2209 * expired.
2210 *
Christoph Lameter879336c2006-03-22 00:09:08 -08002211 * Note that this function must be called with the thread pinned to
2212 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08002213 */
Christoph Lameter4037d452007-05-09 02:35:14 -07002214void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002215{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002216 unsigned long flags;
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07002217 int to_drain, batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002218
Christoph Lameter4037d452007-05-09 02:35:14 -07002219 local_irq_save(flags);
Jason Low4db0c3c2015-04-15 16:14:08 -07002220 batch = READ_ONCE(pcp->batch);
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07002221 to_drain = min(pcp->count, batch);
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07002222 if (to_drain > 0) {
2223 free_pcppages_bulk(zone, to_drain, pcp);
2224 pcp->count -= to_drain;
2225 }
Christoph Lameter4037d452007-05-09 02:35:14 -07002226 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07002227}
2228#endif
2229
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002230/*
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002231 * Drain pcplists of the indicated processor and zone.
2232 *
2233 * The processor must either be the current processor and the
2234 * thread pinned to the current processor or a processor that
2235 * is not online.
2236 */
2237static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2238{
2239 unsigned long flags;
2240 struct per_cpu_pageset *pset;
2241 struct per_cpu_pages *pcp;
2242
2243 local_irq_save(flags);
2244 pset = per_cpu_ptr(zone->pageset, cpu);
2245
2246 pcp = &pset->pcp;
2247 if (pcp->count) {
2248 free_pcppages_bulk(zone, pcp->count, pcp);
2249 pcp->count = 0;
2250 }
2251 local_irq_restore(flags);
2252}
2253
2254/*
2255 * Drain pcplists of all zones on the indicated processor.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002256 *
2257 * The processor must either be the current processor and the
2258 * thread pinned to the current processor or a processor that
2259 * is not online.
2260 */
2261static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262{
2263 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002265 for_each_populated_zone(zone) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002266 drain_pages_zone(cpu, zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 }
2268}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002270/*
2271 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002272 *
2273 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2274 * the single zone's pages.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002275 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002276void drain_local_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002277{
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002278 int cpu = smp_processor_id();
2279
2280 if (zone)
2281 drain_pages_zone(cpu, zone);
2282 else
2283 drain_pages(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002284}
2285
2286/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002287 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2288 *
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002289 * When zone parameter is non-NULL, spill just the single zone's pages.
2290 *
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002291 * Note that this code is protected against sending an IPI to an offline
2292 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2293 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2294 * nothing keeps CPUs from showing up after we populated the cpumask and
2295 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002296 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002297void drain_all_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002298{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002299 int cpu;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002300
2301 /*
2302 * Allocate in the BSS so we wont require allocation in
2303 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2304 */
2305 static cpumask_t cpus_with_pcps;
2306
2307 /*
2308 * We don't care about racing with CPU hotplug event
2309 * as offline notification will cause the notified
2310 * cpu to drain that CPU pcps and on_each_cpu_mask
2311 * disables preemption as part of its processing
2312 */
2313 for_each_online_cpu(cpu) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002314 struct per_cpu_pageset *pcp;
2315 struct zone *z;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002316 bool has_pcps = false;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002317
2318 if (zone) {
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002319 pcp = per_cpu_ptr(zone->pageset, cpu);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002320 if (pcp->pcp.count)
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002321 has_pcps = true;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002322 } else {
2323 for_each_populated_zone(z) {
2324 pcp = per_cpu_ptr(z->pageset, cpu);
2325 if (pcp->pcp.count) {
2326 has_pcps = true;
2327 break;
2328 }
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002329 }
2330 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002331
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07002332 if (has_pcps)
2333 cpumask_set_cpu(cpu, &cpus_with_pcps);
2334 else
2335 cpumask_clear_cpu(cpu, &cpus_with_pcps);
2336 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002337 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
2338 zone, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08002339}
2340
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02002341#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342
2343void mark_free_pages(struct zone *zone)
2344{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002345 unsigned long pfn, max_zone_pfn;
2346 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002347 unsigned int order, t;
Geliang Tang86760a22016-01-14 15:20:33 -08002348 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349
Xishi Qiu8080fc02013-09-11 14:21:45 -07002350 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 return;
2352
2353 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002354
Cody P Schafer108bcc92013-02-22 16:35:23 -08002355 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002356 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2357 if (pfn_valid(pfn)) {
Geliang Tang86760a22016-01-14 15:20:33 -08002358 page = pfn_to_page(pfn);
Joonsoo Kimba6b0972016-05-19 17:12:16 -07002359
2360 if (page_zone(page) != zone)
2361 continue;
2362
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07002363 if (!swsusp_page_is_forbidden(page))
2364 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002365 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002367 for_each_migratetype_order(order, t) {
Geliang Tang86760a22016-01-14 15:20:33 -08002368 list_for_each_entry(page,
2369 &zone->free_area[order].free_list[t], lru) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002370 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371
Geliang Tang86760a22016-01-14 15:20:33 -08002372 pfn = page_to_pfn(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002373 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07002374 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002375 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002376 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 spin_unlock_irqrestore(&zone->lock, flags);
2378}
Mel Gormane2c55dc2007-10-16 01:25:50 -07002379#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380
2381/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 * Free a 0-order page
Mel Gormanb745bc82014-06-04 16:10:22 -07002383 * cold == true ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002385void free_hot_cold_page(struct page *page, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386{
2387 struct zone *zone = page_zone(page);
2388 struct per_cpu_pages *pcp;
2389 unsigned long flags;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002390 unsigned long pfn = page_to_pfn(page);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002391 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392
Mel Gorman4db75482016-05-19 17:14:32 -07002393 if (!free_pcp_prepare(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -08002394 return;
2395
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002396 migratetype = get_pfnblock_migratetype(page, pfn);
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002397 set_pcppage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07002399 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07002400
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002401 /*
2402 * We only track unmovable, reclaimable and movable on pcp lists.
2403 * Free ISOLATE pages back to the allocator because they are being
2404 * offlined but treat RESERVE as movable pages so we can get those
2405 * areas back if necessary. Otherwise, we may have to free
2406 * excessively into the page allocator
2407 */
2408 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08002409 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002410 free_one_page(zone, page, pfn, 0, migratetype);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002411 goto out;
2412 }
2413 migratetype = MIGRATE_MOVABLE;
2414 }
2415
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002416 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Mel Gormanb745bc82014-06-04 16:10:22 -07002417 if (!cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002418 list_add(&page->lru, &pcp->lists[migratetype]);
Mel Gormanb745bc82014-06-04 16:10:22 -07002419 else
2420 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08002422 if (pcp->count >= pcp->high) {
Jason Low4db0c3c2015-04-15 16:14:08 -07002423 unsigned long batch = READ_ONCE(pcp->batch);
Cody P Schafer998d39c2013-07-03 15:01:32 -07002424 free_pcppages_bulk(zone, batch, pcp);
2425 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08002426 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002427
2428out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430}
2431
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002432/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002433 * Free a list of 0-order pages
2434 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002435void free_hot_cold_page_list(struct list_head *list, bool cold)
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002436{
2437 struct page *page, *next;
2438
2439 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08002440 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002441 free_hot_cold_page(page, cold);
2442 }
2443}
2444
2445/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002446 * split_page takes a non-compound higher-order page, and splits it into
2447 * n (1<<order) sub-pages: page[0..n]
2448 * Each sub-page must be freed individually.
2449 *
2450 * Note: this is probably too low level an operation for use in drivers.
2451 * Please consult with lkml before using this in your driver.
2452 */
2453void split_page(struct page *page, unsigned int order)
2454{
2455 int i;
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002456 gfp_t gfp_mask;
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002457
Sasha Levin309381fea2014-01-23 15:52:54 -08002458 VM_BUG_ON_PAGE(PageCompound(page), page);
2459 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002460
2461#ifdef CONFIG_KMEMCHECK
2462 /*
2463 * Split shadow pages too, because free(page[0]) would
2464 * otherwise free the whole shadow.
2465 */
2466 if (kmemcheck_page_is_tracked(page))
2467 split_page(virt_to_page(page[0].shadow), order);
2468#endif
2469
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002470 gfp_mask = get_page_owner_gfp(page);
2471 set_page_owner(page, 0, gfp_mask);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08002472 for (i = 1; i < (1 << order); i++) {
Nick Piggin7835e982006-03-22 00:08:40 -08002473 set_page_refcounted(page + i);
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002474 set_page_owner(page + i, 0, gfp_mask);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08002475 }
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002476}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07002477EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002478
Joonsoo Kim3c605092014-11-13 15:19:21 -08002479int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07002480{
Mel Gorman748446b2010-05-24 14:32:27 -07002481 unsigned long watermark;
2482 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002483 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07002484
2485 BUG_ON(!PageBuddy(page));
2486
2487 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002488 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07002489
Minchan Kim194159f2013-02-22 16:33:58 -08002490 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002491 /* Obey watermarks as if the page was being allocated */
2492 watermark = low_wmark_pages(zone) + (1 << order);
2493 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2494 return 0;
2495
Mel Gorman8fb74b92013-01-11 14:32:16 -08002496 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002497 }
Mel Gorman748446b2010-05-24 14:32:27 -07002498
2499 /* Remove page from free list */
2500 list_del(&page->lru);
2501 zone->free_area[order].nr_free--;
2502 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002503
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002504 set_page_owner(page, order, __GFP_MOVABLE);
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002505
Mel Gorman8fb74b92013-01-11 14:32:16 -08002506 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07002507 if (order >= pageblock_order - 1) {
2508 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002509 for (; page < endpage; page += pageblock_nr_pages) {
2510 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08002511 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002512 set_pageblock_migratetype(page,
2513 MIGRATE_MOVABLE);
2514 }
Mel Gorman748446b2010-05-24 14:32:27 -07002515 }
2516
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002517
Mel Gorman8fb74b92013-01-11 14:32:16 -08002518 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002519}
2520
2521/*
2522 * Similar to split_page except the page is already free. As this is only
2523 * being used for migration, the migratetype of the block also changes.
2524 * As this is called with interrupts disabled, the caller is responsible
2525 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2526 * are enabled.
2527 *
2528 * Note: this is probably too low level an operation for use in drivers.
2529 * Please consult with lkml before using this in your driver.
2530 */
2531int split_free_page(struct page *page)
2532{
2533 unsigned int order;
2534 int nr_pages;
2535
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002536 order = page_order(page);
2537
Mel Gorman8fb74b92013-01-11 14:32:16 -08002538 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002539 if (!nr_pages)
2540 return 0;
2541
2542 /* Split into individual pages */
2543 set_page_refcounted(page);
2544 split_page(page, order);
2545 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07002546}
2547
2548/*
Mel Gorman060e7412016-05-19 17:13:27 -07002549 * Update NUMA hit/miss statistics
2550 *
2551 * Must be called with interrupts disabled.
2552 *
2553 * When __GFP_OTHER_NODE is set assume the node of the preferred
2554 * zone is the local node. This is useful for daemons who allocate
2555 * memory on behalf of other processes.
2556 */
2557static inline void zone_statistics(struct zone *preferred_zone, struct zone *z,
2558 gfp_t flags)
2559{
2560#ifdef CONFIG_NUMA
2561 int local_nid = numa_node_id();
2562 enum zone_stat_item local_stat = NUMA_LOCAL;
2563
2564 if (unlikely(flags & __GFP_OTHER_NODE)) {
2565 local_stat = NUMA_OTHER;
2566 local_nid = preferred_zone->node;
2567 }
2568
2569 if (z->node == local_nid) {
2570 __inc_zone_state(z, NUMA_HIT);
2571 __inc_zone_state(z, local_stat);
2572 } else {
2573 __inc_zone_state(z, NUMA_MISS);
2574 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
2575 }
2576#endif
2577}
2578
2579/*
Vlastimil Babka75379192015-02-11 15:25:38 -08002580 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07002582static inline
2583struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002584 struct zone *zone, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07002585 gfp_t gfp_flags, unsigned int alloc_flags,
2586 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587{
2588 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08002589 struct page *page;
Mel Gormanb745bc82014-06-04 16:10:22 -07002590 bool cold = ((gfp_flags & __GFP_COLD) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591
Nick Piggin48db57f2006-01-08 01:00:42 -08002592 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002594 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 local_irq_save(flags);
Mel Gorman479f8542016-05-19 17:14:35 -07002597 do {
2598 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2599 list = &pcp->lists[migratetype];
2600 if (list_empty(list)) {
2601 pcp->count += rmqueue_bulk(zone, 0,
2602 pcp->batch, list,
2603 migratetype, cold);
2604 if (unlikely(list_empty(list)))
2605 goto failed;
2606 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07002607
Mel Gorman479f8542016-05-19 17:14:35 -07002608 if (cold)
2609 page = list_last_entry(list, struct page, lru);
2610 else
2611 page = list_first_entry(list, struct page, lru);
2612 } while (page && check_new_pcp(page));
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002613
Mel Gorman754078e2016-05-19 17:13:33 -07002614 __dec_zone_state(zone, NR_ALLOC_BATCH);
Mel Gormanb92a6ed2007-10-16 01:25:50 -07002615 list_del(&page->lru);
2616 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002617 } else {
Michal Hocko0f352e52016-03-17 14:19:32 -07002618 /*
2619 * We most definitely don't want callers attempting to
2620 * allocate greater than order-1 page units with __GFP_NOFAIL.
2621 */
2622 WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 spin_lock_irqsave(&zone->lock, flags);
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002624
Mel Gorman479f8542016-05-19 17:14:35 -07002625 do {
2626 page = NULL;
2627 if (alloc_flags & ALLOC_HARDER) {
2628 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
2629 if (page)
2630 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2631 }
2632 if (!page)
2633 page = __rmqueue(zone, order, migratetype);
2634 } while (page && check_new_pages(page, order));
Nick Piggina74609f2006-01-06 00:11:20 -08002635 spin_unlock(&zone->lock);
2636 if (!page)
2637 goto failed;
Mel Gorman754078e2016-05-19 17:13:33 -07002638 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07002639 __mod_zone_freepage_state(zone, -(1 << order),
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002640 get_pcppage_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 }
2642
Johannes Weinerabe5f972014-10-02 16:21:10 -07002643 if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
Johannes Weiner57054652014-10-09 15:28:17 -07002644 !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
2645 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Johannes Weiner27329362014-03-03 15:38:41 -08002646
Christoph Lameterf8891e52006-06-30 01:55:45 -07002647 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07002648 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002649 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650
Sasha Levin309381fea2014-01-23 15:52:54 -08002651 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08002653
2654failed:
2655 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002656 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657}
2658
Akinobu Mita933e3122006-12-08 02:39:45 -08002659#ifdef CONFIG_FAIL_PAGE_ALLOC
2660
Akinobu Mitab2588c42011-07-26 16:09:03 -07002661static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08002662 struct fault_attr attr;
2663
Viresh Kumar621a5f72015-09-26 15:04:07 -07002664 bool ignore_gfp_highmem;
Mel Gorman71baba42015-11-06 16:28:28 -08002665 bool ignore_gfp_reclaim;
Akinobu Mita54114992007-07-15 23:40:23 -07002666 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08002667} fail_page_alloc = {
2668 .attr = FAULT_ATTR_INITIALIZER,
Mel Gorman71baba42015-11-06 16:28:28 -08002669 .ignore_gfp_reclaim = true,
Viresh Kumar621a5f72015-09-26 15:04:07 -07002670 .ignore_gfp_highmem = true,
Akinobu Mita54114992007-07-15 23:40:23 -07002671 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08002672};
2673
2674static int __init setup_fail_page_alloc(char *str)
2675{
2676 return setup_fault_attr(&fail_page_alloc.attr, str);
2677}
2678__setup("fail_page_alloc=", setup_fail_page_alloc);
2679
Gavin Shandeaf3862012-07-31 16:41:51 -07002680static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002681{
Akinobu Mita54114992007-07-15 23:40:23 -07002682 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07002683 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002684 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07002685 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002686 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002687 return false;
Mel Gorman71baba42015-11-06 16:28:28 -08002688 if (fail_page_alloc.ignore_gfp_reclaim &&
2689 (gfp_mask & __GFP_DIRECT_RECLAIM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002690 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002691
2692 return should_fail(&fail_page_alloc.attr, 1 << order);
2693}
2694
2695#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2696
2697static int __init fail_page_alloc_debugfs(void)
2698{
Al Virof4ae40a2011-07-24 04:33:43 -04002699 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08002700 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08002701
Akinobu Mitadd48c082011-08-03 16:21:01 -07002702 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2703 &fail_page_alloc.attr);
2704 if (IS_ERR(dir))
2705 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002706
Akinobu Mitab2588c42011-07-26 16:09:03 -07002707 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
Mel Gorman71baba42015-11-06 16:28:28 -08002708 &fail_page_alloc.ignore_gfp_reclaim))
Akinobu Mitab2588c42011-07-26 16:09:03 -07002709 goto fail;
2710 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2711 &fail_page_alloc.ignore_gfp_highmem))
2712 goto fail;
2713 if (!debugfs_create_u32("min-order", mode, dir,
2714 &fail_page_alloc.min_order))
2715 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08002716
Akinobu Mitab2588c42011-07-26 16:09:03 -07002717 return 0;
2718fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07002719 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002720
Akinobu Mitab2588c42011-07-26 16:09:03 -07002721 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08002722}
2723
2724late_initcall(fail_page_alloc_debugfs);
2725
2726#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2727
2728#else /* CONFIG_FAIL_PAGE_ALLOC */
2729
Gavin Shandeaf3862012-07-31 16:41:51 -07002730static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002731{
Gavin Shandeaf3862012-07-31 16:41:51 -07002732 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002733}
2734
2735#endif /* CONFIG_FAIL_PAGE_ALLOC */
2736
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737/*
Mel Gorman97a16fc2015-11-06 16:28:40 -08002738 * Return true if free base pages are above 'mark'. For high-order checks it
2739 * will return true of the order-0 watermark is reached and there is at least
2740 * one free page of a suitable size. Checking now avoids taking the zone lock
2741 * to check in the allocation paths if no pages are free.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 */
Michal Hocko86a294a2016-05-20 16:57:12 -07002743bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2744 int classzone_idx, unsigned int alloc_flags,
2745 long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746{
Christoph Lameterd23ad422007-02-10 01:43:02 -08002747 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 int o;
Mel Gormanc6038442016-05-19 17:13:38 -07002749 const bool alloc_harder = (alloc_flags & ALLOC_HARDER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002751 /* free_pages may go negative - that's OK */
Michal Hockodf0a6da2012-01-10 15:08:02 -08002752 free_pages -= (1 << order) - 1;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002753
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002754 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 min -= min / 2;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002756
2757 /*
2758 * If the caller does not have rights to ALLOC_HARDER then subtract
2759 * the high-atomic reserves. This will over-estimate the size of the
2760 * atomic reserve but it avoids a search.
2761 */
Mel Gorman97a16fc2015-11-06 16:28:40 -08002762 if (likely(!alloc_harder))
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002763 free_pages -= z->nr_reserved_highatomic;
2764 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 min -= min / 4;
Mel Gormane2b19192015-11-06 16:28:09 -08002766
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002767#ifdef CONFIG_CMA
2768 /* If allocation can't use CMA areas don't use free CMA pages */
2769 if (!(alloc_flags & ALLOC_CMA))
Mel Gorman97a16fc2015-11-06 16:28:40 -08002770 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002771#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002772
Mel Gorman97a16fc2015-11-06 16:28:40 -08002773 /*
2774 * Check watermarks for an order-0 allocation request. If these
2775 * are not met, then a high-order request also cannot go ahead
2776 * even if a suitable page happened to be free.
2777 */
2778 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08002779 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780
Mel Gorman97a16fc2015-11-06 16:28:40 -08002781 /* If this is an order-0 request then the watermark is fine */
2782 if (!order)
2783 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784
Mel Gorman97a16fc2015-11-06 16:28:40 -08002785 /* For a high-order request, check at least one suitable page is free */
2786 for (o = order; o < MAX_ORDER; o++) {
2787 struct free_area *area = &z->free_area[o];
2788 int mt;
2789
2790 if (!area->nr_free)
2791 continue;
2792
2793 if (alloc_harder)
2794 return true;
2795
2796 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
2797 if (!list_empty(&area->free_list[mt]))
2798 return true;
2799 }
2800
2801#ifdef CONFIG_CMA
2802 if ((alloc_flags & ALLOC_CMA) &&
2803 !list_empty(&area->free_list[MIGRATE_CMA])) {
2804 return true;
2805 }
2806#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807 }
Mel Gorman97a16fc2015-11-06 16:28:40 -08002808 return false;
Mel Gorman88f5acf2011-01-13 15:45:41 -08002809}
2810
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002811bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Mel Gormanc6038442016-05-19 17:13:38 -07002812 int classzone_idx, unsigned int alloc_flags)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002813{
2814 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2815 zone_page_state(z, NR_FREE_PAGES));
2816}
2817
Mel Gorman48ee5f32016-05-19 17:14:07 -07002818static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
2819 unsigned long mark, int classzone_idx, unsigned int alloc_flags)
2820{
2821 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2822 long cma_pages = 0;
2823
2824#ifdef CONFIG_CMA
2825 /* If allocation can't use CMA areas don't use free CMA pages */
2826 if (!(alloc_flags & ALLOC_CMA))
2827 cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES);
2828#endif
2829
2830 /*
2831 * Fast check for order-0 only. If this fails then the reserves
2832 * need to be calculated. There is a corner case where the check
2833 * passes but only the high-order atomic reserve are free. If
2834 * the caller is !atomic then it'll uselessly search the free
2835 * list. That corner case is then slower but it is harmless.
2836 */
2837 if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx])
2838 return true;
2839
2840 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2841 free_pages);
2842}
2843
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002844bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
Mel Gormane2b19192015-11-06 16:28:09 -08002845 unsigned long mark, int classzone_idx)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002846{
2847 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2848
2849 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2850 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2851
Mel Gormane2b19192015-11-06 16:28:09 -08002852 return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002853 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854}
2855
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002856#ifdef CONFIG_NUMA
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002857static bool zone_local(struct zone *local_zone, struct zone *zone)
2858{
Johannes Weinerfff40682013-12-20 14:54:12 +00002859 return local_zone->node == zone->node;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002860}
2861
David Rientjes957f8222012-10-08 16:33:24 -07002862static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2863{
Mel Gorman5f7a75a2014-06-04 16:07:15 -07002864 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2865 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07002866}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002867#else /* CONFIG_NUMA */
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002868static bool zone_local(struct zone *local_zone, struct zone *zone)
2869{
2870 return true;
2871}
2872
David Rientjes957f8222012-10-08 16:33:24 -07002873static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2874{
2875 return true;
2876}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002877#endif /* CONFIG_NUMA */
2878
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002879static void reset_alloc_batches(struct zone *preferred_zone)
2880{
2881 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2882
2883 do {
2884 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2885 high_wmark_pages(zone) - low_wmark_pages(zone) -
2886 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner57054652014-10-09 15:28:17 -07002887 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002888 } while (zone++ != preferred_zone);
2889}
2890
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002891/*
Paul Jackson0798e512006-12-06 20:31:38 -08002892 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002893 * a page.
2894 */
2895static struct page *
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002896get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2897 const struct alloc_context *ac)
Martin Hicks753ee722005-06-21 17:14:41 -07002898{
Mel Gormanc33d6c02016-05-19 17:14:10 -07002899 struct zoneref *z = ac->preferred_zoneref;
Mel Gorman5117f452009-06-16 15:31:59 -07002900 struct zone *zone;
Mel Gorman30534752016-05-19 17:14:04 -07002901 bool fair_skipped = false;
2902 bool apply_fair = (alloc_flags & ALLOC_FAIR);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002903
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002904zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002905 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002906 * Scan zonelist, looking for a zone with enough free.
Vladimir Davydov344736f2014-10-20 15:50:30 +04002907 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002908 */
Mel Gormanc33d6c02016-05-19 17:14:10 -07002909 for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002910 ac->nodemask) {
Mel Gormanbe06af02016-05-19 17:13:47 -07002911 struct page *page;
Johannes Weinere085dbc2013-09-11 14:20:46 -07002912 unsigned long mark;
2913
Mel Gorman664eedd2014-06-04 16:10:08 -07002914 if (cpusets_enabled() &&
2915 (alloc_flags & ALLOC_CPUSET) &&
Vlastimil Babka002f2902016-05-19 17:14:30 -07002916 !__cpuset_zone_allowed(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07002917 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08002918 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002919 * Distribute pages in proportion to the individual
2920 * zone size to ensure fair page aging. The zone a
2921 * page was allocated in should have no effect on the
2922 * time the page has in memory before being reclaimed.
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002923 */
Mel Gorman30534752016-05-19 17:14:04 -07002924 if (apply_fair) {
Johannes Weiner57054652014-10-09 15:28:17 -07002925 if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
Mel Gormanfa379b92016-05-19 17:13:41 -07002926 fair_skipped = true;
Johannes Weiner3a025762014-04-07 15:37:48 -07002927 continue;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002928 }
Mel Gormanc33d6c02016-05-19 17:14:10 -07002929 if (!zone_local(ac->preferred_zoneref->zone, zone)) {
Mel Gorman30534752016-05-19 17:14:04 -07002930 if (fair_skipped)
2931 goto reset_fair;
2932 apply_fair = false;
2933 }
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002934 }
2935 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08002936 * When allocating a page cache page for writing, we
2937 * want to get it from a zone that is within its dirty
2938 * limit, such that no single zone holds more than its
2939 * proportional share of globally allowed dirty pages.
2940 * The dirty limits take into account the zone's
2941 * lowmem reserves and high watermark so that kswapd
2942 * should be able to balance it without having to
2943 * write pages from its LRU list.
2944 *
2945 * This may look like it could increase pressure on
2946 * lower zones by failing allocations in higher zones
2947 * before they are full. But the pages that do spill
2948 * over are limited as the lower zones are protected
2949 * by this very same mechanism. It should not become
2950 * a practical burden to them.
2951 *
2952 * XXX: For now, allow allocations to potentially
2953 * exceed the per-zone dirty limit in the slowpath
Mel Gormanc9ab0c42015-11-06 16:28:12 -08002954 * (spread_dirty_pages unset) before going into reclaim,
Johannes Weinera756cf52012-01-10 15:07:49 -08002955 * which is important when on a NUMA setup the allowed
2956 * zones are together not big enough to reach the
2957 * global limit. The proper fix for these situations
2958 * will require awareness of zones in the
2959 * dirty-throttling and the flusher threads.
2960 */
Mel Gormanc9ab0c42015-11-06 16:28:12 -08002961 if (ac->spread_dirty_pages && !zone_dirty_ok(zone))
Mel Gorman800a1e72014-06-04 16:10:06 -07002962 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002963
Johannes Weinere085dbc2013-09-11 14:20:46 -07002964 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gorman48ee5f32016-05-19 17:14:07 -07002965 if (!zone_watermark_fast(zone, order, mark,
Mel Gorman93ea9962016-05-19 17:14:13 -07002966 ac_classzone_idx(ac), alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07002967 int ret;
2968
Mel Gorman5dab2912014-06-04 16:10:14 -07002969 /* Checked here to keep the fast path fast */
2970 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2971 if (alloc_flags & ALLOC_NO_WATERMARKS)
2972 goto try_this_zone;
2973
David Rientjes957f8222012-10-08 16:33:24 -07002974 if (zone_reclaim_mode == 0 ||
Mel Gormanc33d6c02016-05-19 17:14:10 -07002975 !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
Mel Gormancd38b112011-07-25 17:12:29 -07002976 continue;
2977
Mel Gormanfa5e0842009-06-16 15:33:22 -07002978 ret = zone_reclaim(zone, gfp_mask, order);
2979 switch (ret) {
2980 case ZONE_RECLAIM_NOSCAN:
2981 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07002982 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002983 case ZONE_RECLAIM_FULL:
2984 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002985 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002986 default:
2987 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002988 if (zone_watermark_ok(zone, order, mark,
Mel Gorman93ea9962016-05-19 17:14:13 -07002989 ac_classzone_idx(ac), alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002990 goto try_this_zone;
2991
Mel Gormanfed27192013-04-29 15:07:57 -07002992 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002993 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002994 }
2995
Mel Gormanfa5e0842009-06-16 15:33:22 -07002996try_this_zone:
Mel Gormanc33d6c02016-05-19 17:14:10 -07002997 page = buffered_rmqueue(ac->preferred_zoneref->zone, zone, order,
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002998 gfp_mask, alloc_flags, ac->migratetype);
Vlastimil Babka75379192015-02-11 15:25:38 -08002999 if (page) {
Mel Gorman479f8542016-05-19 17:14:35 -07003000 prep_new_page(page, order, gfp_mask, alloc_flags);
Mel Gorman0aaa29a2015-11-06 16:28:37 -08003001
3002 /*
3003 * If this is a high-order atomic allocation then check
3004 * if the pageblock should be reserved for the future
3005 */
3006 if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
3007 reserve_highatomic_pageblock(page, zone, order);
3008
Vlastimil Babka75379192015-02-11 15:25:38 -08003009 return page;
3010 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003011 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003012
Mel Gorman4ffeaf32014-08-06 16:07:22 -07003013 /*
3014 * The first pass makes sure allocations are spread fairly within the
3015 * local node. However, the local node might have free pages left
3016 * after the fairness batches are exhausted, and remote zones haven't
3017 * even been considered yet. Try once more without fairness, and
3018 * include remote zones now, before entering the slowpath and waking
3019 * kswapd: prefer spilling to a remote zone over swapping locally.
3020 */
Mel Gorman30534752016-05-19 17:14:04 -07003021 if (fair_skipped) {
3022reset_fair:
3023 apply_fair = false;
3024 fair_skipped = false;
Mel Gormanc33d6c02016-05-19 17:14:10 -07003025 reset_alloc_batches(ac->preferred_zoneref->zone);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07003026 goto zonelist_scan;
Mel Gorman30534752016-05-19 17:14:04 -07003027 }
Mel Gorman4ffeaf32014-08-06 16:07:22 -07003028
3029 return NULL;
Martin Hicks753ee722005-06-21 17:14:41 -07003030}
3031
David Rientjes29423e772011-03-22 16:30:47 -07003032/*
3033 * Large machines with many possible nodes should not always dump per-node
3034 * meminfo in irq context.
3035 */
3036static inline bool should_suppress_show_mem(void)
3037{
3038 bool ret = false;
3039
3040#if NODES_SHIFT > 8
3041 ret = in_interrupt();
3042#endif
3043 return ret;
3044}
3045
Dave Hansena238ab52011-05-24 17:12:16 -07003046static DEFINE_RATELIMIT_STATE(nopage_rs,
3047 DEFAULT_RATELIMIT_INTERVAL,
3048 DEFAULT_RATELIMIT_BURST);
3049
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08003050void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...)
Dave Hansena238ab52011-05-24 17:12:16 -07003051{
Dave Hansena238ab52011-05-24 17:12:16 -07003052 unsigned int filter = SHOW_MEM_FILTER_NODES;
3053
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08003054 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
3055 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07003056 return;
3057
3058 /*
3059 * This documents exceptions given to allocations in certain
3060 * contexts that are allowed to allocate outside current's set
3061 * of allowed nodes.
3062 */
3063 if (!(gfp_mask & __GFP_NOMEMALLOC))
3064 if (test_thread_flag(TIF_MEMDIE) ||
3065 (current->flags & (PF_MEMALLOC | PF_EXITING)))
3066 filter &= ~SHOW_MEM_FILTER_NODES;
Mel Gormand0164ad2015-11-06 16:28:21 -08003067 if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
Dave Hansena238ab52011-05-24 17:12:16 -07003068 filter &= ~SHOW_MEM_FILTER_NODES;
3069
3070 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07003071 struct va_format vaf;
3072 va_list args;
3073
Dave Hansena238ab52011-05-24 17:12:16 -07003074 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07003075
3076 vaf.fmt = fmt;
3077 vaf.va = &args;
3078
3079 pr_warn("%pV", &vaf);
3080
Dave Hansena238ab52011-05-24 17:12:16 -07003081 va_end(args);
3082 }
3083
Vlastimil Babkac5c990e2016-03-15 14:56:02 -07003084 pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n",
3085 current->comm, order, gfp_mask, &gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07003086 dump_stack();
3087 if (!should_suppress_show_mem())
3088 show_mem(filter);
3089}
3090
Mel Gorman11e33f62009-06-16 15:31:57 -07003091static inline struct page *
3092__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003093 const struct alloc_context *ac, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07003094{
David Rientjes6e0fc462015-09-08 15:00:36 -07003095 struct oom_control oc = {
3096 .zonelist = ac->zonelist,
3097 .nodemask = ac->nodemask,
3098 .gfp_mask = gfp_mask,
3099 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07003100 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102
Johannes Weiner9879de72015-01-26 12:58:32 -08003103 *did_some_progress = 0;
3104
Johannes Weiner9879de72015-01-26 12:58:32 -08003105 /*
Johannes Weinerdc564012015-06-24 16:57:19 -07003106 * Acquire the oom lock. If that fails, somebody else is
3107 * making progress for us.
Johannes Weiner9879de72015-01-26 12:58:32 -08003108 */
Johannes Weinerdc564012015-06-24 16:57:19 -07003109 if (!mutex_trylock(&oom_lock)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08003110 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07003111 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112 return NULL;
3113 }
Jens Axboe6b1de912005-11-17 21:35:02 +01003114
Mel Gorman11e33f62009-06-16 15:31:57 -07003115 /*
3116 * Go through the zonelist yet one more time, keep very high watermark
3117 * here, this is only to catch a parallel oom killing, we must fail if
3118 * we're still under heavy pressure.
3119 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003120 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
3121 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003122 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07003123 goto out;
3124
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08003125 if (!(gfp_mask & __GFP_NOFAIL)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08003126 /* Coredumps can quickly deplete all memory reserves */
3127 if (current->flags & PF_DUMPCORE)
3128 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08003129 /* The OOM killer will not help higher order allocs */
3130 if (order > PAGE_ALLOC_COSTLY_ORDER)
3131 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07003132 /* The OOM killer does not needlessly kill tasks for lowmem */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003133 if (ac->high_zoneidx < ZONE_NORMAL)
David Rientjes03668b32010-08-09 17:18:54 -07003134 goto out;
Johannes Weiner90839052015-06-24 16:57:21 -07003135 if (pm_suspended_storage())
3136 goto out;
Michal Hocko3da88fb2016-05-19 17:13:09 -07003137 /*
3138 * XXX: GFP_NOFS allocations should rather fail than rely on
3139 * other request to make a forward progress.
3140 * We are in an unfortunate situation where out_of_memory cannot
3141 * do much for this context but let's try it to at least get
3142 * access to memory reserved if the current task is killed (see
3143 * out_of_memory). Once filesystems are ready to handle allocation
3144 * failures more gracefully we should just bail out here.
3145 */
3146
David Rientjes4167e9b2015-04-14 15:46:55 -07003147 /* The OOM killer may not free memory on a specific node */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08003148 if (gfp_mask & __GFP_THISNODE)
3149 goto out;
3150 }
Mel Gorman11e33f62009-06-16 15:31:57 -07003151 /* Exhausted what can be done so it's blamo time */
Michal Hocko5020e282016-01-14 15:20:36 -08003152 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
Michal Hockoc32b3cb2015-02-11 15:26:24 -08003153 *did_some_progress = 1;
Michal Hocko5020e282016-01-14 15:20:36 -08003154
3155 if (gfp_mask & __GFP_NOFAIL) {
3156 page = get_page_from_freelist(gfp_mask, order,
3157 ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac);
3158 /*
3159 * fallback to ignore cpuset restriction if our nodes
3160 * are depleted
3161 */
3162 if (!page)
3163 page = get_page_from_freelist(gfp_mask, order,
3164 ALLOC_NO_WATERMARKS, ac);
3165 }
3166 }
Mel Gorman11e33f62009-06-16 15:31:57 -07003167out:
Johannes Weinerdc564012015-06-24 16:57:19 -07003168 mutex_unlock(&oom_lock);
Mel Gorman11e33f62009-06-16 15:31:57 -07003169 return page;
3170}
3171
Michal Hocko33c2d212016-05-20 16:57:06 -07003172
3173/*
3174 * Maximum number of compaction retries wit a progress before OOM
3175 * killer is consider as the only way to move forward.
3176 */
3177#define MAX_COMPACT_RETRIES 16
3178
Mel Gorman56de7262010-05-24 14:32:30 -07003179#ifdef CONFIG_COMPACTION
3180/* Try memory compaction for high-order allocations before reclaim */
3181static struct page *
3182__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07003183 unsigned int alloc_flags, const struct alloc_context *ac,
Michal Hockoc5d01d02016-05-20 16:56:53 -07003184 enum migrate_mode mode, enum compact_result *compact_result)
Mel Gorman56de7262010-05-24 14:32:30 -07003185{
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003186 struct page *page;
Michal Hockoc5d01d02016-05-20 16:56:53 -07003187 int contended_compaction;
Vlastimil Babka53853e22014-10-09 15:27:02 -07003188
Mel Gorman66199712012-01-12 17:19:41 -08003189 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07003190 return NULL;
3191
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003192 current->flags |= PF_MEMALLOC;
Michal Hockoc5d01d02016-05-20 16:56:53 -07003193 *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
3194 mode, &contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003195 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07003196
Michal Hockoc5d01d02016-05-20 16:56:53 -07003197 if (*compact_result <= COMPACT_INACTIVE)
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003198 return NULL;
Mel Gorman56de7262010-05-24 14:32:30 -07003199
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003200 /*
3201 * At least in one zone compaction wasn't deferred or skipped, so let's
3202 * count a compaction stall
3203 */
3204 count_vm_event(COMPACTSTALL);
3205
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003206 page = get_page_from_freelist(gfp_mask, order,
3207 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003208
3209 if (page) {
3210 struct zone *zone = page_zone(page);
3211
3212 zone->compact_blockskip_flush = false;
3213 compaction_defer_reset(zone, order, true);
3214 count_vm_event(COMPACTSUCCESS);
3215 return page;
3216 }
3217
3218 /*
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003219 * It's bad if compaction run occurs and fails. The most likely reason
3220 * is that pages exist, but not enough to satisfy watermarks.
3221 */
3222 count_vm_event(COMPACTFAIL);
3223
Michal Hockoc5d01d02016-05-20 16:56:53 -07003224 /*
3225 * In all zones where compaction was attempted (and not
3226 * deferred or skipped), lock contention has been detected.
3227 * For THP allocation we do not want to disrupt the others
3228 * so we fallback to base pages instead.
3229 */
3230 if (contended_compaction == COMPACT_CONTENDED_LOCK)
3231 *compact_result = COMPACT_CONTENDED;
3232
3233 /*
3234 * If compaction was aborted due to need_resched(), we do not
3235 * want to further increase allocation latency, unless it is
3236 * khugepaged trying to collapse.
3237 */
3238 if (contended_compaction == COMPACT_CONTENDED_SCHED
3239 && !(current->flags & PF_KTHREAD))
3240 *compact_result = COMPACT_CONTENDED;
3241
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07003242 cond_resched();
3243
Mel Gorman56de7262010-05-24 14:32:30 -07003244 return NULL;
3245}
Michal Hocko33c2d212016-05-20 16:57:06 -07003246
3247static inline bool
Michal Hocko86a294a2016-05-20 16:57:12 -07003248should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
3249 enum compact_result compact_result, enum migrate_mode *migrate_mode,
Michal Hocko33c2d212016-05-20 16:57:06 -07003250 int compaction_retries)
3251{
Michal Hocko7854ea62016-05-20 16:57:09 -07003252 int max_retries = MAX_COMPACT_RETRIES;
3253
Michal Hocko33c2d212016-05-20 16:57:06 -07003254 if (!order)
3255 return false;
3256
3257 /*
3258 * compaction considers all the zone as desperately out of memory
3259 * so it doesn't really make much sense to retry except when the
3260 * failure could be caused by weak migration mode.
3261 */
3262 if (compaction_failed(compact_result)) {
3263 if (*migrate_mode == MIGRATE_ASYNC) {
3264 *migrate_mode = MIGRATE_SYNC_LIGHT;
3265 return true;
3266 }
3267 return false;
3268 }
3269
3270 /*
Michal Hocko7854ea62016-05-20 16:57:09 -07003271 * make sure the compaction wasn't deferred or didn't bail out early
3272 * due to locks contention before we declare that we should give up.
Michal Hocko86a294a2016-05-20 16:57:12 -07003273 * But do not retry if the given zonelist is not suitable for
3274 * compaction.
Michal Hocko33c2d212016-05-20 16:57:06 -07003275 */
Michal Hocko7854ea62016-05-20 16:57:09 -07003276 if (compaction_withdrawn(compact_result))
Michal Hocko86a294a2016-05-20 16:57:12 -07003277 return compaction_zonelist_suitable(ac, order, alloc_flags);
Michal Hocko7854ea62016-05-20 16:57:09 -07003278
3279 /*
3280 * !costly requests are much more important than __GFP_REPEAT
3281 * costly ones because they are de facto nofail and invoke OOM
3282 * killer to move on while costly can fail and users are ready
3283 * to cope with that. 1/4 retries is rather arbitrary but we
3284 * would need much more detailed feedback from compaction to
3285 * make a better decision.
3286 */
3287 if (order > PAGE_ALLOC_COSTLY_ORDER)
3288 max_retries /= 4;
3289 if (compaction_retries <= max_retries)
3290 return true;
Michal Hocko33c2d212016-05-20 16:57:06 -07003291
3292 return false;
3293}
Mel Gorman56de7262010-05-24 14:32:30 -07003294#else
3295static inline struct page *
3296__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07003297 unsigned int alloc_flags, const struct alloc_context *ac,
Michal Hockoc5d01d02016-05-20 16:56:53 -07003298 enum migrate_mode mode, enum compact_result *compact_result)
Mel Gorman56de7262010-05-24 14:32:30 -07003299{
Michal Hocko33c2d212016-05-20 16:57:06 -07003300 *compact_result = COMPACT_SKIPPED;
Mel Gorman56de7262010-05-24 14:32:30 -07003301 return NULL;
3302}
Michal Hocko33c2d212016-05-20 16:57:06 -07003303
3304static inline bool
Michal Hocko86a294a2016-05-20 16:57:12 -07003305should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
3306 enum compact_result compact_result,
Michal Hocko33c2d212016-05-20 16:57:06 -07003307 enum migrate_mode *migrate_mode,
3308 int compaction_retries)
3309{
Michal Hocko31e49bf2016-05-20 16:57:15 -07003310 struct zone *zone;
3311 struct zoneref *z;
3312
3313 if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
3314 return false;
3315
3316 /*
3317 * There are setups with compaction disabled which would prefer to loop
3318 * inside the allocator rather than hit the oom killer prematurely.
3319 * Let's give them a good hope and keep retrying while the order-0
3320 * watermarks are OK.
3321 */
3322 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3323 ac->nodemask) {
3324 if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
3325 ac_classzone_idx(ac), alloc_flags))
3326 return true;
3327 }
Michal Hocko33c2d212016-05-20 16:57:06 -07003328 return false;
3329}
Mel Gorman56de7262010-05-24 14:32:30 -07003330#endif /* CONFIG_COMPACTION */
3331
Marek Szyprowskibba90712012-01-25 12:09:52 +01003332/* Perform direct synchronous page reclaim */
3333static int
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003334__perform_reclaim(gfp_t gfp_mask, unsigned int order,
3335 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003336{
Mel Gorman11e33f62009-06-16 15:31:57 -07003337 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01003338 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07003339
3340 cond_resched();
3341
3342 /* We now go into synchronous reclaim */
3343 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003344 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07003345 lockdep_set_current_reclaim_state(gfp_mask);
3346 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003347 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07003348
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003349 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
3350 ac->nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07003351
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003352 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07003353 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003354 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07003355
3356 cond_resched();
3357
Marek Szyprowskibba90712012-01-25 12:09:52 +01003358 return progress;
3359}
3360
3361/* The really slow allocator path where we enter direct reclaim */
3362static inline struct page *
3363__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
Mel Gormanc6038442016-05-19 17:13:38 -07003364 unsigned int alloc_flags, const struct alloc_context *ac,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003365 unsigned long *did_some_progress)
Marek Szyprowskibba90712012-01-25 12:09:52 +01003366{
3367 struct page *page = NULL;
3368 bool drained = false;
3369
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003370 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003371 if (unlikely(!(*did_some_progress)))
3372 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07003373
Mel Gorman9ee493c2010-09-09 16:38:18 -07003374retry:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003375 page = get_page_from_freelist(gfp_mask, order,
3376 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003377
3378 /*
3379 * If an allocation failed after direct reclaim, it could be because
Mel Gorman0aaa29a2015-11-06 16:28:37 -08003380 * pages are pinned on the per-cpu lists or in high alloc reserves.
3381 * Shrink them them and try again
Mel Gorman9ee493c2010-09-09 16:38:18 -07003382 */
3383 if (!page && !drained) {
Mel Gorman0aaa29a2015-11-06 16:28:37 -08003384 unreserve_highatomic_pageblock(ac);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08003385 drain_all_pages(NULL);
Mel Gorman9ee493c2010-09-09 16:38:18 -07003386 drained = true;
3387 goto retry;
3388 }
3389
Mel Gorman11e33f62009-06-16 15:31:57 -07003390 return page;
3391}
3392
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003393static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003394{
3395 struct zoneref *z;
3396 struct zone *zone;
3397
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003398 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
3399 ac->high_zoneidx, ac->nodemask)
Mel Gorman93ea9962016-05-19 17:14:13 -07003400 wakeup_kswapd(zone, order, ac_classzone_idx(ac));
Mel Gorman11e33f62009-06-16 15:31:57 -07003401}
3402
Mel Gormanc6038442016-05-19 17:13:38 -07003403static inline unsigned int
Peter Zijlstra341ce062009-06-16 15:32:02 -07003404gfp_to_alloc_flags(gfp_t gfp_mask)
3405{
Mel Gormanc6038442016-05-19 17:13:38 -07003406 unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
Peter Zijlstra341ce062009-06-16 15:32:02 -07003407
Mel Gormana56f57f2009-06-16 15:32:02 -07003408 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07003409 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07003410
Peter Zijlstra341ce062009-06-16 15:32:02 -07003411 /*
3412 * The caller may dip into page reserves a bit more if the caller
3413 * cannot run direct reclaim, or if the caller has realtime scheduling
3414 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
Mel Gormand0164ad2015-11-06 16:28:21 -08003415 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
Peter Zijlstra341ce062009-06-16 15:32:02 -07003416 */
Namhyung Kime6223a32010-10-26 14:21:59 -07003417 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07003418
Mel Gormand0164ad2015-11-06 16:28:21 -08003419 if (gfp_mask & __GFP_ATOMIC) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003420 /*
David Rientjesb104a352014-07-30 16:08:24 -07003421 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3422 * if it can't schedule.
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003423 */
David Rientjesb104a352014-07-30 16:08:24 -07003424 if (!(gfp_mask & __GFP_NOMEMALLOC))
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08003425 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07003426 /*
David Rientjesb104a352014-07-30 16:08:24 -07003427 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
Vladimir Davydov344736f2014-10-20 15:50:30 +04003428 * comment for __cpuset_node_allowed().
Peter Zijlstra341ce062009-06-16 15:32:02 -07003429 */
3430 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003431 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07003432 alloc_flags |= ALLOC_HARDER;
3433
Mel Gormanb37f1dd2012-07-31 16:44:03 -07003434 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
3435 if (gfp_mask & __GFP_MEMALLOC)
3436 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07003437 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
3438 alloc_flags |= ALLOC_NO_WATERMARKS;
3439 else if (!in_interrupt() &&
3440 ((current->flags & PF_MEMALLOC) ||
3441 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07003442 alloc_flags |= ALLOC_NO_WATERMARKS;
3443 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07003444#ifdef CONFIG_CMA
David Rientjes43e7a342014-10-09 15:27:25 -07003445 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07003446 alloc_flags |= ALLOC_CMA;
3447#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07003448 return alloc_flags;
3449}
3450
Mel Gorman072bb0a2012-07-31 16:43:58 -07003451bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
3452{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07003453 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07003454}
3455
Mel Gormand0164ad2015-11-06 16:28:21 -08003456static inline bool is_thp_gfp_mask(gfp_t gfp_mask)
3457{
3458 return (gfp_mask & (GFP_TRANSHUGE | __GFP_KSWAPD_RECLAIM)) == GFP_TRANSHUGE;
3459}
3460
Michal Hocko0a0337e2016-05-20 16:57:00 -07003461/*
3462 * Maximum number of reclaim retries without any progress before OOM killer
3463 * is consider as the only way to move forward.
3464 */
3465#define MAX_RECLAIM_RETRIES 16
3466
3467/*
3468 * Checks whether it makes sense to retry the reclaim to make a forward progress
3469 * for the given allocation request.
3470 * The reclaim feedback represented by did_some_progress (any progress during
Michal Hocko7854ea62016-05-20 16:57:09 -07003471 * the last reclaim round) and no_progress_loops (number of reclaim rounds without
3472 * any progress in a row) is considered as well as the reclaimable pages on the
3473 * applicable zone list (with a backoff mechanism which is a function of
3474 * no_progress_loops).
Michal Hocko0a0337e2016-05-20 16:57:00 -07003475 *
3476 * Returns true if a retry is viable or false to enter the oom path.
3477 */
3478static inline bool
3479should_reclaim_retry(gfp_t gfp_mask, unsigned order,
3480 struct alloc_context *ac, int alloc_flags,
Michal Hocko7854ea62016-05-20 16:57:09 -07003481 bool did_some_progress, int no_progress_loops)
Michal Hocko0a0337e2016-05-20 16:57:00 -07003482{
3483 struct zone *zone;
3484 struct zoneref *z;
3485
3486 /*
3487 * Make sure we converge to OOM if we cannot make any progress
3488 * several times in the row.
3489 */
3490 if (no_progress_loops > MAX_RECLAIM_RETRIES)
3491 return false;
3492
Michal Hocko0a0337e2016-05-20 16:57:00 -07003493 /*
3494 * Keep reclaiming pages while there is a chance this will lead somewhere.
3495 * If none of the target zones can satisfy our allocation request even
3496 * if all reclaimable pages are considered then we are screwed and have
3497 * to go OOM.
3498 */
3499 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3500 ac->nodemask) {
3501 unsigned long available;
Michal Hockoede37712016-05-20 16:57:03 -07003502 unsigned long reclaimable;
Michal Hocko0a0337e2016-05-20 16:57:00 -07003503
Michal Hockoede37712016-05-20 16:57:03 -07003504 available = reclaimable = zone_reclaimable_pages(zone);
Michal Hocko0a0337e2016-05-20 16:57:00 -07003505 available -= DIV_ROUND_UP(no_progress_loops * available,
3506 MAX_RECLAIM_RETRIES);
3507 available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
3508
3509 /*
3510 * Would the allocation succeed if we reclaimed the whole
3511 * available?
3512 */
3513 if (__zone_watermark_ok(zone, order, min_wmark_pages(zone),
Michal Hockoede37712016-05-20 16:57:03 -07003514 ac_classzone_idx(ac), alloc_flags, available)) {
3515 /*
3516 * If we didn't make any progress and have a lot of
3517 * dirty + writeback pages then we should wait for
3518 * an IO to complete to slow down the reclaim and
3519 * prevent from pre mature OOM
3520 */
3521 if (!did_some_progress) {
3522 unsigned long writeback;
3523 unsigned long dirty;
3524
3525 writeback = zone_page_state_snapshot(zone,
3526 NR_WRITEBACK);
3527 dirty = zone_page_state_snapshot(zone, NR_FILE_DIRTY);
3528
3529 if (2*(writeback + dirty) > reclaimable) {
3530 congestion_wait(BLK_RW_ASYNC, HZ/10);
3531 return true;
3532 }
3533 }
3534
3535 /*
3536 * Memory allocation/reclaim might be called from a WQ
3537 * context and the current implementation of the WQ
3538 * concurrency control doesn't recognize that
3539 * a particular WQ is congested if the worker thread is
3540 * looping without ever sleeping. Therefore we have to
3541 * do a short sleep here rather than calling
3542 * cond_resched().
3543 */
3544 if (current->flags & PF_WQ_WORKER)
3545 schedule_timeout_uninterruptible(1);
3546 else
3547 cond_resched();
3548
Michal Hocko0a0337e2016-05-20 16:57:00 -07003549 return true;
3550 }
3551 }
3552
3553 return false;
3554}
3555
Mel Gorman11e33f62009-06-16 15:31:57 -07003556static inline struct page *
3557__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003558 struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07003559{
Mel Gormand0164ad2015-11-06 16:28:21 -08003560 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
Mel Gorman11e33f62009-06-16 15:31:57 -07003561 struct page *page = NULL;
Mel Gormanc6038442016-05-19 17:13:38 -07003562 unsigned int alloc_flags;
Mel Gorman11e33f62009-06-16 15:31:57 -07003563 unsigned long did_some_progress;
David Rientjese0b9dae2014-06-04 16:08:28 -07003564 enum migrate_mode migration_mode = MIGRATE_ASYNC;
Michal Hockoc5d01d02016-05-20 16:56:53 -07003565 enum compact_result compact_result;
Michal Hocko33c2d212016-05-20 16:57:06 -07003566 int compaction_retries = 0;
Michal Hocko0a0337e2016-05-20 16:57:00 -07003567 int no_progress_loops = 0;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003568
Christoph Lameter952f3b52006-12-06 20:33:26 -08003569 /*
Mel Gorman72807a72009-06-16 15:32:18 -07003570 * In the slowpath, we sanity check order to avoid ever trying to
3571 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3572 * be using allocators in order of preference for an area that is
3573 * too large.
3574 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07003575 if (order >= MAX_ORDER) {
3576 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07003577 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07003578 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003579
Christoph Lameter952f3b52006-12-06 20:33:26 -08003580 /*
Mel Gormand0164ad2015-11-06 16:28:21 -08003581 * We also sanity check to catch abuse of atomic reserves being used by
3582 * callers that are not in atomic context.
3583 */
3584 if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
3585 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
3586 gfp_mask &= ~__GFP_ATOMIC;
3587
Johannes Weiner9879de72015-01-26 12:58:32 -08003588retry:
Mel Gormand0164ad2015-11-06 16:28:21 -08003589 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003590 wake_all_kswapds(order, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003591
Paul Jackson9bf22292005-09-06 15:18:12 -07003592 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003593 * OK, we're below the kswapd watermark and have kicked background
3594 * reclaim. Now things get more complex, so set up alloc_flags according
3595 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07003596 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07003597 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598
Peter Zijlstra341ce062009-06-16 15:32:02 -07003599 /* This is the last chance, in general, before the goto nopage. */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003600 page = get_page_from_freelist(gfp_mask, order,
3601 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003602 if (page)
3603 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003604
Mel Gorman11e33f62009-06-16 15:31:57 -07003605 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07003606 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07003607 /*
3608 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3609 * the allocation is high priority and these type of
3610 * allocations are system rather than user orientated
3611 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003612 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
Michal Hocko33d53102016-01-14 15:19:05 -08003613 page = get_page_from_freelist(gfp_mask, order,
3614 ALLOC_NO_WATERMARKS, ac);
3615 if (page)
3616 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003617 }
3618
Mel Gormand0164ad2015-11-06 16:28:21 -08003619 /* Caller is not willing to reclaim, we can't balance anything */
3620 if (!can_direct_reclaim) {
David Rientjesaed0a0e2014-01-21 15:51:12 -08003621 /*
Michal Hocko33d53102016-01-14 15:19:05 -08003622 * All existing users of the __GFP_NOFAIL are blockable, so warn
3623 * of any new users that actually allow this type of allocation
3624 * to fail.
David Rientjesaed0a0e2014-01-21 15:51:12 -08003625 */
3626 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627 goto nopage;
David Rientjesaed0a0e2014-01-21 15:51:12 -08003628 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629
Peter Zijlstra341ce062009-06-16 15:32:02 -07003630 /* Avoid recursion of direct reclaim */
Michal Hocko33d53102016-01-14 15:19:05 -08003631 if (current->flags & PF_MEMALLOC) {
3632 /*
3633 * __GFP_NOFAIL request from this context is rather bizarre
3634 * because we cannot reclaim anything and only can loop waiting
3635 * for somebody to do a work for us.
3636 */
3637 if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3638 cond_resched();
3639 goto retry;
3640 }
Peter Zijlstra341ce062009-06-16 15:32:02 -07003641 goto nopage;
Michal Hocko33d53102016-01-14 15:19:05 -08003642 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643
David Rientjes6583bb62009-07-29 15:02:06 -07003644 /* Avoid allocations with no watermarks from looping endlessly */
3645 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3646 goto nopage;
3647
Mel Gorman77f1fe62011-01-13 15:45:57 -08003648 /*
3649 * Try direct compaction. The first pass is asynchronous. Subsequent
3650 * attempts after direct reclaim are synchronous
3651 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003652 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3653 migration_mode,
Michal Hockoc5d01d02016-05-20 16:56:53 -07003654 &compact_result);
Mel Gorman56de7262010-05-24 14:32:30 -07003655 if (page)
3656 goto got_pg;
David Rientjes75f30862014-06-04 16:08:30 -07003657
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003658 /* Checks for THP-specific high-order allocations */
Mel Gormand0164ad2015-11-06 16:28:21 -08003659 if (is_thp_gfp_mask(gfp_mask)) {
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003660 /*
3661 * If compaction is deferred for high-order allocations, it is
3662 * because sync compaction recently failed. If this is the case
3663 * and the caller requested a THP allocation, we do not want
3664 * to heavily disrupt the system, so we fail the allocation
3665 * instead of entering direct reclaim.
3666 */
Michal Hockoc5d01d02016-05-20 16:56:53 -07003667 if (compact_result == COMPACT_DEFERRED)
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003668 goto nopage;
3669
3670 /*
Michal Hockoc5d01d02016-05-20 16:56:53 -07003671 * Compaction is contended so rather back off than cause
3672 * excessive stalls.
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003673 */
Michal Hockoc5d01d02016-05-20 16:56:53 -07003674 if(compact_result == COMPACT_CONTENDED)
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003675 goto nopage;
3676 }
Mel Gorman66199712012-01-12 17:19:41 -08003677
Michal Hocko33c2d212016-05-20 16:57:06 -07003678 if (order && compaction_made_progress(compact_result))
3679 compaction_retries++;
David Rientjes8fe78042014-08-06 16:07:54 -07003680
Mel Gorman11e33f62009-06-16 15:31:57 -07003681 /* Try direct reclaim and then allocating */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003682 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3683 &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07003684 if (page)
3685 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003686
Johannes Weiner90839052015-06-24 16:57:21 -07003687 /* Do not loop if specifically requested */
3688 if (gfp_mask & __GFP_NORETRY)
3689 goto noretry;
3690
Michal Hocko0a0337e2016-05-20 16:57:00 -07003691 /*
3692 * Do not retry costly high order allocations unless they are
3693 * __GFP_REPEAT
3694 */
3695 if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_REPEAT))
3696 goto noretry;
3697
Michal Hocko7854ea62016-05-20 16:57:09 -07003698 /*
3699 * Costly allocations might have made a progress but this doesn't mean
3700 * their order will become available due to high fragmentation so
3701 * always increment the no progress counter for them
3702 */
3703 if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
Michal Hocko0a0337e2016-05-20 16:57:00 -07003704 no_progress_loops = 0;
Michal Hocko7854ea62016-05-20 16:57:09 -07003705 else
Michal Hocko0a0337e2016-05-20 16:57:00 -07003706 no_progress_loops++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707
Michal Hocko0a0337e2016-05-20 16:57:00 -07003708 if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
Michal Hocko7854ea62016-05-20 16:57:09 -07003709 did_some_progress > 0, no_progress_loops))
Michal Hocko0a0337e2016-05-20 16:57:00 -07003710 goto retry;
3711
Michal Hocko33c2d212016-05-20 16:57:06 -07003712 /*
3713 * It doesn't make any sense to retry for the compaction if the order-0
3714 * reclaim is not able to make any progress because the current
3715 * implementation of the compaction depends on the sufficient amount
3716 * of free memory (see __compaction_suitable)
3717 */
3718 if (did_some_progress > 0 &&
Michal Hocko86a294a2016-05-20 16:57:12 -07003719 should_compact_retry(ac, order, alloc_flags,
3720 compact_result, &migration_mode,
3721 compaction_retries))
Michal Hocko33c2d212016-05-20 16:57:06 -07003722 goto retry;
3723
Johannes Weiner90839052015-06-24 16:57:21 -07003724 /* Reclaim has failed us, start killing things */
3725 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3726 if (page)
3727 goto got_pg;
3728
3729 /* Retry as long as the OOM killer is making progress */
Michal Hocko0a0337e2016-05-20 16:57:00 -07003730 if (did_some_progress) {
3731 no_progress_loops = 0;
Johannes Weiner90839052015-06-24 16:57:21 -07003732 goto retry;
Michal Hocko0a0337e2016-05-20 16:57:00 -07003733 }
Johannes Weiner90839052015-06-24 16:57:21 -07003734
3735noretry:
3736 /*
Michal Hocko33c2d212016-05-20 16:57:06 -07003737 * High-order allocations do not necessarily loop after direct reclaim
3738 * and reclaim/compaction depends on compaction being called after
3739 * reclaim so call directly if necessary.
3740 * It can become very expensive to allocate transparent hugepages at
3741 * fault, so use asynchronous memory compaction for THP unless it is
3742 * khugepaged trying to collapse. All other requests should tolerate
3743 * at least light sync migration.
Johannes Weiner90839052015-06-24 16:57:21 -07003744 */
Michal Hocko33c2d212016-05-20 16:57:06 -07003745 if (is_thp_gfp_mask(gfp_mask) && !(current->flags & PF_KTHREAD))
3746 migration_mode = MIGRATE_ASYNC;
3747 else
3748 migration_mode = MIGRATE_SYNC_LIGHT;
Johannes Weiner90839052015-06-24 16:57:21 -07003749 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags,
3750 ac, migration_mode,
Michal Hockoc5d01d02016-05-20 16:56:53 -07003751 &compact_result);
Johannes Weiner90839052015-06-24 16:57:21 -07003752 if (page)
3753 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07003755 warn_alloc_failed(gfp_mask, order, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756got_pg:
Mel Gorman072bb0a2012-07-31 16:43:58 -07003757 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003758}
Mel Gorman11e33f62009-06-16 15:31:57 -07003759
3760/*
3761 * This is the 'heart' of the zoned buddy allocator.
3762 */
3763struct page *
3764__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3765 struct zonelist *zonelist, nodemask_t *nodemask)
3766{
Mel Gorman5bb1b162016-05-19 17:13:50 -07003767 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003768 unsigned int cpuset_mems_cookie;
Mel Gormanc6038442016-05-19 17:13:38 -07003769 unsigned int alloc_flags = ALLOC_WMARK_LOW|ALLOC_FAIR;
Mel Gorman83d4ca82016-05-19 17:13:56 -07003770 gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003771 struct alloc_context ac = {
3772 .high_zoneidx = gfp_zone(gfp_mask),
Mel Gorman682a3382016-05-19 17:13:30 -07003773 .zonelist = zonelist,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003774 .nodemask = nodemask,
3775 .migratetype = gfpflags_to_migratetype(gfp_mask),
3776 };
Mel Gorman11e33f62009-06-16 15:31:57 -07003777
Mel Gorman682a3382016-05-19 17:13:30 -07003778 if (cpusets_enabled()) {
Mel Gorman83d4ca82016-05-19 17:13:56 -07003779 alloc_mask |= __GFP_HARDWALL;
Mel Gorman682a3382016-05-19 17:13:30 -07003780 alloc_flags |= ALLOC_CPUSET;
3781 if (!ac.nodemask)
3782 ac.nodemask = &cpuset_current_mems_allowed;
3783 }
3784
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003785 gfp_mask &= gfp_allowed_mask;
3786
Mel Gorman11e33f62009-06-16 15:31:57 -07003787 lockdep_trace_alloc(gfp_mask);
3788
Mel Gormand0164ad2015-11-06 16:28:21 -08003789 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
Mel Gorman11e33f62009-06-16 15:31:57 -07003790
3791 if (should_fail_alloc_page(gfp_mask, order))
3792 return NULL;
3793
3794 /*
3795 * Check the zones suitable for the gfp_mask contain at least one
3796 * valid zone. It's possible to have an empty zonelist as a result
David Rientjes4167e9b2015-04-14 15:46:55 -07003797 * of __GFP_THISNODE and a memoryless node
Mel Gorman11e33f62009-06-16 15:31:57 -07003798 */
3799 if (unlikely(!zonelist->_zonerefs->zone))
3800 return NULL;
3801
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003802 if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
Vlastimil Babka21bb9bd2014-10-09 15:26:51 -07003803 alloc_flags |= ALLOC_CMA;
3804
Mel Gormancc9a6c82012-03-21 16:34:11 -07003805retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07003806 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003807
Mel Gormanc9ab0c42015-11-06 16:28:12 -08003808 /* Dirty zone balancing only done in the fast path */
3809 ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE);
3810
Mel Gorman5117f452009-06-16 15:31:59 -07003811 /* The preferred zone is used for statistics later */
Mel Gormanc33d6c02016-05-19 17:14:10 -07003812 ac.preferred_zoneref = first_zones_zonelist(ac.zonelist,
3813 ac.high_zoneidx, ac.nodemask);
3814 if (!ac.preferred_zoneref) {
Mel Gorman5bb1b162016-05-19 17:13:50 -07003815 page = NULL;
Mel Gorman4fcb0972016-05-19 17:14:01 -07003816 goto no_zone;
Mel Gorman5bb1b162016-05-19 17:13:50 -07003817 }
3818
Mel Gorman5117f452009-06-16 15:31:59 -07003819 /* First allocation attempt */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003820 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
Mel Gorman4fcb0972016-05-19 17:14:01 -07003821 if (likely(page))
3822 goto out;
Andrew Morton91fbdc02015-02-11 15:25:04 -08003823
Mel Gorman4fcb0972016-05-19 17:14:01 -07003824 /*
3825 * Runtime PM, block IO and its error handling path can deadlock
3826 * because I/O on the device might not complete.
3827 */
3828 alloc_mask = memalloc_noio_flags(gfp_mask);
3829 ac.spread_dirty_pages = false;
Mel Gorman11e33f62009-06-16 15:31:57 -07003830
Mel Gorman47415262016-05-19 17:14:44 -07003831 /*
3832 * Restore the original nodemask if it was potentially replaced with
3833 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
3834 */
3835 if (cpusets_enabled())
3836 ac.nodemask = nodemask;
Mel Gorman4fcb0972016-05-19 17:14:01 -07003837 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
Xishi Qiu23f086f2015-02-11 15:25:07 -08003838
Mel Gorman4fcb0972016-05-19 17:14:01 -07003839no_zone:
Mel Gormancc9a6c82012-03-21 16:34:11 -07003840 /*
3841 * When updating a task's mems_allowed, it is possible to race with
3842 * parallel threads in such a way that an allocation can fail while
3843 * the mask is being updated. If a page allocation is about to fail,
3844 * check if the cpuset changed during allocation and if so, retry.
3845 */
Mel Gorman83d4ca82016-05-19 17:13:56 -07003846 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) {
3847 alloc_mask = gfp_mask;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003848 goto retry_cpuset;
Mel Gorman83d4ca82016-05-19 17:13:56 -07003849 }
Mel Gormancc9a6c82012-03-21 16:34:11 -07003850
Mel Gorman4fcb0972016-05-19 17:14:01 -07003851out:
3852 if (kmemcheck_enabled && page)
3853 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3854
3855 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
3856
Mel Gorman11e33f62009-06-16 15:31:57 -07003857 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003858}
Mel Gormand2391712009-06-16 15:31:52 -07003859EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003860
3861/*
3862 * Common helper functions.
3863 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08003864unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003865{
Akinobu Mita945a1112009-09-21 17:01:47 -07003866 struct page *page;
3867
3868 /*
3869 * __get_free_pages() returns a 32-bit address, which cannot represent
3870 * a highmem page
3871 */
3872 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3873
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874 page = alloc_pages(gfp_mask, order);
3875 if (!page)
3876 return 0;
3877 return (unsigned long) page_address(page);
3878}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003879EXPORT_SYMBOL(__get_free_pages);
3880
Harvey Harrison920c7a52008-02-04 22:29:26 -08003881unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882{
Akinobu Mita945a1112009-09-21 17:01:47 -07003883 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003884}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003885EXPORT_SYMBOL(get_zeroed_page);
3886
Harvey Harrison920c7a52008-02-04 22:29:26 -08003887void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888{
Nick Pigginb5810032005-10-29 18:16:12 -07003889 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003890 if (order == 0)
Mel Gormanb745bc82014-06-04 16:10:22 -07003891 free_hot_cold_page(page, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003892 else
3893 __free_pages_ok(page, order);
3894 }
3895}
3896
3897EXPORT_SYMBOL(__free_pages);
3898
Harvey Harrison920c7a52008-02-04 22:29:26 -08003899void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003900{
3901 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07003902 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903 __free_pages(virt_to_page((void *)addr), order);
3904 }
3905}
3906
3907EXPORT_SYMBOL(free_pages);
3908
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003909/*
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003910 * Page Fragment:
3911 * An arbitrary-length arbitrary-offset area of memory which resides
3912 * within a 0 or higher order page. Multiple fragments within that page
3913 * are individually refcounted, in the page's reference counter.
3914 *
3915 * The page_frag functions below provide a simple allocation framework for
3916 * page fragments. This is used by the network stack and network device
3917 * drivers to provide a backing region of memory for use as either an
3918 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3919 */
3920static struct page *__page_frag_refill(struct page_frag_cache *nc,
3921 gfp_t gfp_mask)
3922{
3923 struct page *page = NULL;
3924 gfp_t gfp = gfp_mask;
3925
3926#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3927 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3928 __GFP_NOMEMALLOC;
3929 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3930 PAGE_FRAG_CACHE_MAX_ORDER);
3931 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3932#endif
3933 if (unlikely(!page))
3934 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3935
3936 nc->va = page ? page_address(page) : NULL;
3937
3938 return page;
3939}
3940
3941void *__alloc_page_frag(struct page_frag_cache *nc,
3942 unsigned int fragsz, gfp_t gfp_mask)
3943{
3944 unsigned int size = PAGE_SIZE;
3945 struct page *page;
3946 int offset;
3947
3948 if (unlikely(!nc->va)) {
3949refill:
3950 page = __page_frag_refill(nc, gfp_mask);
3951 if (!page)
3952 return NULL;
3953
3954#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3955 /* if size can vary use size else just use PAGE_SIZE */
3956 size = nc->size;
3957#endif
3958 /* Even if we own the page, we do not use atomic_set().
3959 * This would break get_page_unless_zero() users.
3960 */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07003961 page_ref_add(page, size - 1);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003962
3963 /* reset page count bias and offset to start of new frag */
Michal Hocko2f064f32015-08-21 14:11:51 -07003964 nc->pfmemalloc = page_is_pfmemalloc(page);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003965 nc->pagecnt_bias = size;
3966 nc->offset = size;
3967 }
3968
3969 offset = nc->offset - fragsz;
3970 if (unlikely(offset < 0)) {
3971 page = virt_to_page(nc->va);
3972
Joonsoo Kimfe896d12016-03-17 14:19:26 -07003973 if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003974 goto refill;
3975
3976#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3977 /* if size can vary use size else just use PAGE_SIZE */
3978 size = nc->size;
3979#endif
3980 /* OK, page count is 0, we can safely set it */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07003981 set_page_count(page, size);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003982
3983 /* reset page count bias and offset to start of new frag */
3984 nc->pagecnt_bias = size;
3985 offset = size - fragsz;
3986 }
3987
3988 nc->pagecnt_bias--;
3989 nc->offset = offset;
3990
3991 return nc->va + offset;
3992}
3993EXPORT_SYMBOL(__alloc_page_frag);
3994
3995/*
3996 * Frees a page fragment allocated out of either a compound or order 0 page.
3997 */
3998void __free_page_frag(void *addr)
3999{
4000 struct page *page = virt_to_head_page(addr);
4001
4002 if (unlikely(put_page_testzero(page)))
4003 __free_pages_ok(page, compound_order(page));
4004}
4005EXPORT_SYMBOL(__free_page_frag);
4006
4007/*
Vladimir Davydov52383432014-06-04 16:06:39 -07004008 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
Vladimir Davydova9bb7e62016-01-14 15:18:12 -08004009 * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is
4010 * equivalent to alloc_pages.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08004011 *
Vladimir Davydov52383432014-06-04 16:06:39 -07004012 * It should be used when the caller would like to use kmalloc, but since the
4013 * allocation is large, it has to fall back to the page allocator.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08004014 */
Vladimir Davydov52383432014-06-04 16:06:39 -07004015struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
4016{
4017 struct page *page;
Vladimir Davydov52383432014-06-04 16:06:39 -07004018
Vladimir Davydov52383432014-06-04 16:06:39 -07004019 page = alloc_pages(gfp_mask, order);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08004020 if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
4021 __free_pages(page, order);
4022 page = NULL;
4023 }
Vladimir Davydov52383432014-06-04 16:06:39 -07004024 return page;
4025}
4026
4027struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
4028{
4029 struct page *page;
Vladimir Davydov52383432014-06-04 16:06:39 -07004030
Vladimir Davydov52383432014-06-04 16:06:39 -07004031 page = alloc_pages_node(nid, gfp_mask, order);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08004032 if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
4033 __free_pages(page, order);
4034 page = NULL;
4035 }
Vladimir Davydov52383432014-06-04 16:06:39 -07004036 return page;
4037}
4038
4039/*
4040 * __free_kmem_pages and free_kmem_pages will free pages allocated with
4041 * alloc_kmem_pages.
4042 */
4043void __free_kmem_pages(struct page *page, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08004044{
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08004045 memcg_kmem_uncharge(page, order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08004046 __free_pages(page, order);
4047}
4048
Vladimir Davydov52383432014-06-04 16:06:39 -07004049void free_kmem_pages(unsigned long addr, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08004050{
4051 if (addr != 0) {
4052 VM_BUG_ON(!virt_addr_valid((void *)addr));
Vladimir Davydov52383432014-06-04 16:06:39 -07004053 __free_kmem_pages(virt_to_page((void *)addr), order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08004054 }
4055}
4056
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004057static void *make_alloc_exact(unsigned long addr, unsigned int order,
4058 size_t size)
Andi Kleenee85c2e2011-05-11 15:13:34 -07004059{
4060 if (addr) {
4061 unsigned long alloc_end = addr + (PAGE_SIZE << order);
4062 unsigned long used = addr + PAGE_ALIGN(size);
4063
4064 split_page(virt_to_page((void *)addr), order);
4065 while (used < alloc_end) {
4066 free_page(used);
4067 used += PAGE_SIZE;
4068 }
4069 }
4070 return (void *)addr;
4071}
4072
Timur Tabi2be0ffe2008-07-23 21:28:11 -07004073/**
4074 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4075 * @size: the number of bytes to allocate
4076 * @gfp_mask: GFP flags for the allocation
4077 *
4078 * This function is similar to alloc_pages(), except that it allocates the
4079 * minimum number of pages to satisfy the request. alloc_pages() can only
4080 * allocate memory in power-of-two pages.
4081 *
4082 * This function is also limited by MAX_ORDER.
4083 *
4084 * Memory allocated by this function must be released by free_pages_exact().
4085 */
4086void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
4087{
4088 unsigned int order = get_order(size);
4089 unsigned long addr;
4090
4091 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07004092 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07004093}
4094EXPORT_SYMBOL(alloc_pages_exact);
4095
4096/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07004097 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4098 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07004099 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07004100 * @size: the number of bytes to allocate
4101 * @gfp_mask: GFP flags for the allocation
4102 *
4103 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4104 * back.
Andi Kleenee85c2e2011-05-11 15:13:34 -07004105 */
Fabian Fredericke1931812014-08-06 16:04:59 -07004106void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
Andi Kleenee85c2e2011-05-11 15:13:34 -07004107{
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004108 unsigned int order = get_order(size);
Andi Kleenee85c2e2011-05-11 15:13:34 -07004109 struct page *p = alloc_pages_node(nid, gfp_mask, order);
4110 if (!p)
4111 return NULL;
4112 return make_alloc_exact((unsigned long)page_address(p), order, size);
4113}
Andi Kleenee85c2e2011-05-11 15:13:34 -07004114
4115/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07004116 * free_pages_exact - release memory allocated via alloc_pages_exact()
4117 * @virt: the value returned by alloc_pages_exact.
4118 * @size: size of allocation, same value as passed to alloc_pages_exact().
4119 *
4120 * Release the memory allocated by a previous call to alloc_pages_exact.
4121 */
4122void free_pages_exact(void *virt, size_t size)
4123{
4124 unsigned long addr = (unsigned long)virt;
4125 unsigned long end = addr + PAGE_ALIGN(size);
4126
4127 while (addr < end) {
4128 free_page(addr);
4129 addr += PAGE_SIZE;
4130 }
4131}
4132EXPORT_SYMBOL(free_pages_exact);
4133
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004134/**
4135 * nr_free_zone_pages - count number of pages beyond high watermark
4136 * @offset: The zone index of the highest zone
4137 *
4138 * nr_free_zone_pages() counts the number of counts pages which are beyond the
4139 * high watermark within all zones at or below a given zone index. For each
4140 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07004141 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004142 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004143static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004144{
Mel Gormandd1a2392008-04-28 02:12:17 -07004145 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07004146 struct zone *zone;
4147
Martin J. Blighe310fd42005-07-29 22:59:18 -07004148 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004149 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004150
Mel Gorman0e884602008-04-28 02:12:14 -07004151 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004152
Mel Gorman54a6eb52008-04-28 02:12:16 -07004153 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08004154 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07004155 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07004156 if (size > high)
4157 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158 }
4159
4160 return sum;
4161}
4162
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004163/**
4164 * nr_free_buffer_pages - count number of pages beyond high watermark
4165 *
4166 * nr_free_buffer_pages() counts the number of pages which are beyond the high
4167 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004168 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004169unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004170{
Al Viroaf4ca452005-10-21 02:55:38 -04004171 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004172}
Meelap Shahc2f1a552007-07-17 04:04:39 -07004173EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004174
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08004175/**
4176 * nr_free_pagecache_pages - count number of pages beyond high watermark
4177 *
4178 * nr_free_pagecache_pages() counts the number of pages which are beyond the
4179 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004180 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08004181unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004182{
Mel Gorman2a1e2742007-07-17 04:03:12 -07004183 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004184}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07004185
4186static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004187{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08004188 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08004189 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004190}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004191
Igor Redkod02bd272016-03-17 14:19:05 -07004192long si_mem_available(void)
4193{
4194 long available;
4195 unsigned long pagecache;
4196 unsigned long wmark_low = 0;
4197 unsigned long pages[NR_LRU_LISTS];
4198 struct zone *zone;
4199 int lru;
4200
4201 for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
4202 pages[lru] = global_page_state(NR_LRU_BASE + lru);
4203
4204 for_each_zone(zone)
4205 wmark_low += zone->watermark[WMARK_LOW];
4206
4207 /*
4208 * Estimate the amount of memory available for userspace allocations,
4209 * without causing swapping.
4210 */
4211 available = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
4212
4213 /*
4214 * Not all the page cache can be freed, otherwise the system will
4215 * start swapping. Assume at least half of the page cache, or the
4216 * low watermark worth of cache, needs to stay.
4217 */
4218 pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
4219 pagecache -= min(pagecache / 2, wmark_low);
4220 available += pagecache;
4221
4222 /*
4223 * Part of the reclaimable slab consists of items that are in use,
4224 * and cannot be freed. Cap this estimate at the low watermark.
4225 */
4226 available += global_page_state(NR_SLAB_RECLAIMABLE) -
4227 min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low);
4228
4229 if (available < 0)
4230 available = 0;
4231 return available;
4232}
4233EXPORT_SYMBOL_GPL(si_mem_available);
4234
Linus Torvalds1da177e2005-04-16 15:20:36 -07004235void si_meminfo(struct sysinfo *val)
4236{
4237 val->totalram = totalram_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07004238 val->sharedram = global_page_state(NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08004239 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004240 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004241 val->totalhigh = totalhigh_pages;
4242 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004243 val->mem_unit = PAGE_SIZE;
4244}
4245
4246EXPORT_SYMBOL(si_meminfo);
4247
4248#ifdef CONFIG_NUMA
4249void si_meminfo_node(struct sysinfo *val, int nid)
4250{
Jiang Liucdd91a72013-07-03 15:03:27 -07004251 int zone_type; /* needs to be signed */
4252 unsigned long managed_pages = 0;
Joonsoo Kimfc2bd792016-05-19 17:12:23 -07004253 unsigned long managed_highpages = 0;
4254 unsigned long free_highpages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004255 pg_data_t *pgdat = NODE_DATA(nid);
4256
Jiang Liucdd91a72013-07-03 15:03:27 -07004257 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
4258 managed_pages += pgdat->node_zones[zone_type].managed_pages;
4259 val->totalram = managed_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07004260 val->sharedram = node_page_state(nid, NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08004261 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004262#ifdef CONFIG_HIGHMEM
Joonsoo Kimfc2bd792016-05-19 17:12:23 -07004263 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
4264 struct zone *zone = &pgdat->node_zones[zone_type];
4265
4266 if (is_highmem(zone)) {
4267 managed_highpages += zone->managed_pages;
4268 free_highpages += zone_page_state(zone, NR_FREE_PAGES);
4269 }
4270 }
4271 val->totalhigh = managed_highpages;
4272 val->freehigh = free_highpages;
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004273#else
Joonsoo Kimfc2bd792016-05-19 17:12:23 -07004274 val->totalhigh = managed_highpages;
4275 val->freehigh = free_highpages;
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004276#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004277 val->mem_unit = PAGE_SIZE;
4278}
4279#endif
4280
David Rientjesddd588b2011-03-22 16:30:46 -07004281/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07004282 * Determine whether the node should be displayed or not, depending on whether
4283 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07004284 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07004285bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07004286{
4287 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07004288 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07004289
4290 if (!(flags & SHOW_MEM_FILTER_NODES))
4291 goto out;
4292
Mel Gormancc9a6c82012-03-21 16:34:11 -07004293 do {
Mel Gormand26914d2014-04-03 14:47:24 -07004294 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07004295 ret = !node_isset(nid, cpuset_current_mems_allowed);
Mel Gormand26914d2014-04-03 14:47:24 -07004296 } while (read_mems_allowed_retry(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07004297out:
4298 return ret;
4299}
4300
Linus Torvalds1da177e2005-04-16 15:20:36 -07004301#define K(x) ((x) << (PAGE_SHIFT-10))
4302
Rabin Vincent377e4f12012-12-11 16:00:24 -08004303static void show_migration_types(unsigned char type)
4304{
4305 static const char types[MIGRATE_TYPES] = {
4306 [MIGRATE_UNMOVABLE] = 'U',
Rabin Vincent377e4f12012-12-11 16:00:24 -08004307 [MIGRATE_MOVABLE] = 'M',
Vlastimil Babka475a2f92015-12-11 13:40:29 -08004308 [MIGRATE_RECLAIMABLE] = 'E',
4309 [MIGRATE_HIGHATOMIC] = 'H',
Rabin Vincent377e4f12012-12-11 16:00:24 -08004310#ifdef CONFIG_CMA
4311 [MIGRATE_CMA] = 'C',
4312#endif
Minchan Kim194159f2013-02-22 16:33:58 -08004313#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08004314 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08004315#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08004316 };
4317 char tmp[MIGRATE_TYPES + 1];
4318 char *p = tmp;
4319 int i;
4320
4321 for (i = 0; i < MIGRATE_TYPES; i++) {
4322 if (type & (1 << i))
4323 *p++ = types[i];
4324 }
4325
4326 *p = '\0';
4327 printk("(%s) ", tmp);
4328}
4329
Linus Torvalds1da177e2005-04-16 15:20:36 -07004330/*
4331 * Show free area list (used inside shift_scroll-lock stuff)
4332 * We also calculate the percentage fragmentation. We do this by counting the
4333 * memory on each free list with the exception of the first item on the list.
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004334 *
4335 * Bits in @filter:
4336 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4337 * cpuset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004338 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07004339void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004340{
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004341 unsigned long free_pcp = 0;
Jes Sorensenc7241912006-09-27 01:50:05 -07004342 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004343 struct zone *zone;
4344
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07004345 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07004346 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07004347 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004348
Konstantin Khlebnikov761b0672015-04-14 15:45:32 -07004349 for_each_online_cpu(cpu)
4350 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004351 }
4352
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07004353 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4354 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004355 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4356 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004357 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004358 " free:%lu free_pcp:%lu free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07004359 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07004360 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07004361 global_page_state(NR_ISOLATED_ANON),
4362 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07004363 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07004364 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07004365 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07004366 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07004367 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07004368 global_page_state(NR_UNSTABLE_NFS),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07004369 global_page_state(NR_SLAB_RECLAIMABLE),
4370 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07004371 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07004372 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08004373 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004374 global_page_state(NR_BOUNCE),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004375 global_page_state(NR_FREE_PAGES),
4376 free_pcp,
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004377 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004378
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07004379 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004380 int i;
4381
David Rientjes7bf02ea2011-05-24 17:11:16 -07004382 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07004383 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004384
4385 free_pcp = 0;
4386 for_each_online_cpu(cpu)
4387 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
4388
Linus Torvalds1da177e2005-04-16 15:20:36 -07004389 show_node(zone);
4390 printk("%s"
4391 " free:%lukB"
4392 " min:%lukB"
4393 " low:%lukB"
4394 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07004395 " active_anon:%lukB"
4396 " inactive_anon:%lukB"
4397 " active_file:%lukB"
4398 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07004399 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07004400 " isolated(anon):%lukB"
4401 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07004402 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08004403 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004404 " mlocked:%lukB"
4405 " dirty:%lukB"
4406 " writeback:%lukB"
4407 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07004408 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004409 " slab_reclaimable:%lukB"
4410 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07004411 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004412 " pagetables:%lukB"
4413 " unstable:%lukB"
4414 " bounce:%lukB"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004415 " free_pcp:%lukB"
4416 " local_pcp:%ukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004417 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004418 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07004419 " pages_scanned:%lu"
4420 " all_unreclaimable? %s"
4421 "\n",
4422 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08004423 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07004424 K(min_wmark_pages(zone)),
4425 K(low_wmark_pages(zone)),
4426 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07004427 K(zone_page_state(zone, NR_ACTIVE_ANON)),
4428 K(zone_page_state(zone, NR_INACTIVE_ANON)),
4429 K(zone_page_state(zone, NR_ACTIVE_FILE)),
4430 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07004431 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07004432 K(zone_page_state(zone, NR_ISOLATED_ANON)),
4433 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07004434 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08004435 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004436 K(zone_page_state(zone, NR_MLOCK)),
4437 K(zone_page_state(zone, NR_FILE_DIRTY)),
4438 K(zone_page_state(zone, NR_WRITEBACK)),
4439 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07004440 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004441 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
4442 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07004443 zone_page_state(zone, NR_KERNEL_STACK) *
4444 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004445 K(zone_page_state(zone, NR_PAGETABLE)),
4446 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
4447 K(zone_page_state(zone, NR_BOUNCE)),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07004448 K(free_pcp),
4449 K(this_cpu_read(zone->pageset->pcp.count)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07004450 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07004451 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Mel Gorman0d5d8232014-08-06 16:07:16 -07004452 K(zone_page_state(zone, NR_PAGES_SCANNED)),
Lisa Du6e543d52013-09-11 14:22:36 -07004453 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07004454 );
4455 printk("lowmem_reserve[]:");
4456 for (i = 0; i < MAX_NR_ZONES; i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07004457 printk(" %ld", zone->lowmem_reserve[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004458 printk("\n");
4459 }
4460
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07004461 for_each_populated_zone(zone) {
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004462 unsigned int order;
4463 unsigned long nr[MAX_ORDER], flags, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08004464 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004465
David Rientjes7bf02ea2011-05-24 17:11:16 -07004466 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07004467 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004468 show_node(zone);
4469 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004470
4471 spin_lock_irqsave(&zone->lock, flags);
4472 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08004473 struct free_area *area = &zone->free_area[order];
4474 int type;
4475
4476 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07004477 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08004478
4479 types[order] = 0;
4480 for (type = 0; type < MIGRATE_TYPES; type++) {
4481 if (!list_empty(&area->free_list[type]))
4482 types[order] |= 1 << type;
4483 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004484 }
4485 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08004486 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07004487 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08004488 if (nr[order])
4489 show_migration_types(types[order]);
4490 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004491 printk("= %lukB\n", K(total));
4492 }
4493
David Rientjes949f7ec2013-04-29 15:07:48 -07004494 hugetlb_show_meminfo();
4495
Larry Woodmane6f36022008-02-04 22:29:30 -08004496 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
4497
Linus Torvalds1da177e2005-04-16 15:20:36 -07004498 show_swap_cache_info();
4499}
4500
Mel Gorman19770b32008-04-28 02:12:18 -07004501static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
4502{
4503 zoneref->zone = zone;
4504 zoneref->zone_idx = zone_idx(zone);
4505}
4506
Linus Torvalds1da177e2005-04-16 15:20:36 -07004507/*
4508 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08004509 *
4510 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004511 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004512static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004513 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004514{
Christoph Lameter1a932052006-01-06 00:11:16 -08004515 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004516 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08004517
4518 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004519 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08004520 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08004521 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07004522 zoneref_set_zone(zone,
4523 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08004524 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004525 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004526 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004527
Christoph Lameter070f8032006-01-06 00:11:19 -08004528 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004529}
4530
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004531
4532/*
4533 * zonelist_order:
4534 * 0 = automatic detection of better ordering.
4535 * 1 = order by ([node] distance, -zonetype)
4536 * 2 = order by (-zonetype, [node] distance)
4537 *
4538 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4539 * the same zonelist. So only NUMA can configure this param.
4540 */
4541#define ZONELIST_ORDER_DEFAULT 0
4542#define ZONELIST_ORDER_NODE 1
4543#define ZONELIST_ORDER_ZONE 2
4544
4545/* zonelist order in the kernel.
4546 * set_zonelist_order() will set this to NODE or ZONE.
4547 */
4548static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
4549static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
4550
4551
Linus Torvalds1da177e2005-04-16 15:20:36 -07004552#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004553/* The value user specified ....changed by config */
4554static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4555/* string for sysctl */
4556#define NUMA_ZONELIST_ORDER_LEN 16
4557char numa_zonelist_order[16] = "default";
4558
4559/*
4560 * interface for configure zonelist ordering.
4561 * command line option "numa_zonelist_order"
4562 * = "[dD]efault - default, automatic configuration.
4563 * = "[nN]ode - order by node locality, then by zone within node
4564 * = "[zZ]one - order by zone, then by locality within zone
4565 */
4566
4567static int __parse_numa_zonelist_order(char *s)
4568{
4569 if (*s == 'd' || *s == 'D') {
4570 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4571 } else if (*s == 'n' || *s == 'N') {
4572 user_zonelist_order = ZONELIST_ORDER_NODE;
4573 } else if (*s == 'z' || *s == 'Z') {
4574 user_zonelist_order = ZONELIST_ORDER_ZONE;
4575 } else {
Joe Perches11705322016-03-17 14:19:50 -07004576 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004577 return -EINVAL;
4578 }
4579 return 0;
4580}
4581
4582static __init int setup_numa_zonelist_order(char *s)
4583{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08004584 int ret;
4585
4586 if (!s)
4587 return 0;
4588
4589 ret = __parse_numa_zonelist_order(s);
4590 if (ret == 0)
4591 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
4592
4593 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004594}
4595early_param("numa_zonelist_order", setup_numa_zonelist_order);
4596
4597/*
4598 * sysctl handler for numa_zonelist_order
4599 */
Joe Perchescccad5b2014-06-06 14:38:09 -07004600int numa_zonelist_order_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004601 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004602 loff_t *ppos)
4603{
4604 char saved_string[NUMA_ZONELIST_ORDER_LEN];
4605 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01004606 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004607
Andi Kleen443c6f12009-12-23 21:00:47 +01004608 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07004609 if (write) {
4610 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
4611 ret = -EINVAL;
4612 goto out;
4613 }
4614 strcpy(saved_string, (char *)table->data);
4615 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07004616 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004617 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01004618 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004619 if (write) {
4620 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07004621
4622 ret = __parse_numa_zonelist_order((char *)table->data);
4623 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004624 /*
4625 * bogus value. restore saved string
4626 */
Chen Gangdacbde02013-07-03 15:02:35 -07004627 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004628 NUMA_ZONELIST_ORDER_LEN);
4629 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004630 } else if (oldval != user_zonelist_order) {
4631 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07004632 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004633 mutex_unlock(&zonelists_mutex);
4634 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004635 }
Andi Kleen443c6f12009-12-23 21:00:47 +01004636out:
4637 mutex_unlock(&zl_order_mutex);
4638 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004639}
4640
4641
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004642#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004643static int node_load[MAX_NUMNODES];
4644
Linus Torvalds1da177e2005-04-16 15:20:36 -07004645/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07004646 * 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 -07004647 * @node: node whose fallback list we're appending
4648 * @used_node_mask: nodemask_t of already used nodes
4649 *
4650 * We use a number of factors to determine which is the next node that should
4651 * appear on a given node's fallback list. The node should not have appeared
4652 * already in @node's fallback list, and it should be the next closest node
4653 * according to the distance array (which contains arbitrary distance values
4654 * from each node to each node in the system), and should also prefer nodes
4655 * with no CPUs, since presumably they'll have very little allocation pressure
4656 * on them otherwise.
4657 * It returns -1 if no node is found.
4658 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004659static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004660{
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01004661 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004662 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08004663 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10304664 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004665
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01004666 /* Use the local node if we haven't already */
4667 if (!node_isset(node, *used_node_mask)) {
4668 node_set(node, *used_node_mask);
4669 return node;
4670 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004671
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004672 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004673
4674 /* Don't want a node to appear more than once */
4675 if (node_isset(n, *used_node_mask))
4676 continue;
4677
Linus Torvalds1da177e2005-04-16 15:20:36 -07004678 /* Use the distance array to find the distance */
4679 val = node_distance(node, n);
4680
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01004681 /* Penalize nodes under us ("prefer the next node") */
4682 val += (n < node);
4683
Linus Torvalds1da177e2005-04-16 15:20:36 -07004684 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10304685 tmp = cpumask_of_node(n);
4686 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004687 val += PENALTY_FOR_NODE_WITH_CPUS;
4688
4689 /* Slight preference for less loaded node */
4690 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4691 val += node_load[n];
4692
4693 if (val < min_val) {
4694 min_val = val;
4695 best_node = n;
4696 }
4697 }
4698
4699 if (best_node >= 0)
4700 node_set(best_node, *used_node_mask);
4701
4702 return best_node;
4703}
4704
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004705
4706/*
4707 * Build zonelists ordered by node and zones within node.
4708 * This results in maximum locality--normal zone overflows into local
4709 * DMA zone, if any--but risks exhausting DMA zone.
4710 */
4711static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004712{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004713 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004714 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004715
Mel Gorman54a6eb52008-04-28 02:12:16 -07004716 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07004717 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07004718 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004719 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07004720 zonelist->_zonerefs[j].zone = NULL;
4721 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004722}
4723
4724/*
Christoph Lameter523b9452007-10-16 01:25:37 -07004725 * Build gfp_thisnode zonelists
4726 */
4727static void build_thisnode_zonelists(pg_data_t *pgdat)
4728{
Christoph Lameter523b9452007-10-16 01:25:37 -07004729 int j;
4730 struct zonelist *zonelist;
4731
Mel Gorman54a6eb52008-04-28 02:12:16 -07004732 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004733 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07004734 zonelist->_zonerefs[j].zone = NULL;
4735 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07004736}
4737
4738/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004739 * Build zonelists ordered by zone and nodes within zones.
4740 * This results in conserving DMA zone[s] until all Normal memory is
4741 * exhausted, but results in overflowing to remote node while memory
4742 * may still exist in local DMA zone.
4743 */
4744static int node_order[MAX_NUMNODES];
4745
4746static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4747{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004748 int pos, j, node;
4749 int zone_type; /* needs to be signed */
4750 struct zone *z;
4751 struct zonelist *zonelist;
4752
Mel Gorman54a6eb52008-04-28 02:12:16 -07004753 zonelist = &pgdat->node_zonelists[0];
4754 pos = 0;
4755 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4756 for (j = 0; j < nr_nodes; j++) {
4757 node = node_order[j];
4758 z = &NODE_DATA(node)->node_zones[zone_type];
4759 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07004760 zoneref_set_zone(z,
4761 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004762 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004763 }
4764 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004765 }
Mel Gormandd1a2392008-04-28 02:12:17 -07004766 zonelist->_zonerefs[pos].zone = NULL;
4767 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004768}
4769
Mel Gorman31939132014-10-09 15:28:30 -07004770#if defined(CONFIG_64BIT)
4771/*
4772 * Devices that require DMA32/DMA are relatively rare and do not justify a
4773 * penalty to every machine in case the specialised case applies. Default
4774 * to Node-ordering on 64-bit NUMA machines
4775 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004776static int default_zonelist_order(void)
4777{
Mel Gorman31939132014-10-09 15:28:30 -07004778 return ZONELIST_ORDER_NODE;
4779}
4780#else
4781/*
4782 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4783 * by the kernel. If processes running on node 0 deplete the low memory zone
4784 * then reclaim will occur more frequency increasing stalls and potentially
4785 * be easier to OOM if a large percentage of the zone is under writeback or
4786 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4787 * Hence, default to zone ordering on 32-bit.
4788 */
4789static int default_zonelist_order(void)
4790{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004791 return ZONELIST_ORDER_ZONE;
4792}
Mel Gorman31939132014-10-09 15:28:30 -07004793#endif /* CONFIG_64BIT */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004794
4795static void set_zonelist_order(void)
4796{
4797 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4798 current_zonelist_order = default_zonelist_order();
4799 else
4800 current_zonelist_order = user_zonelist_order;
4801}
4802
4803static void build_zonelists(pg_data_t *pgdat)
4804{
Yaowei Baic00eb152016-01-14 15:19:00 -08004805 int i, node, load;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004806 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004807 int local_node, prev_node;
4808 struct zonelist *zonelist;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004809 unsigned int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004810
4811 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07004812 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004813 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07004814 zonelist->_zonerefs[0].zone = NULL;
4815 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004816 }
4817
4818 /* NUMA-aware ordering of nodes */
4819 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004820 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004821 prev_node = local_node;
4822 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004823
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004824 memset(node_order, 0, sizeof(node_order));
Yaowei Baic00eb152016-01-14 15:19:00 -08004825 i = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004826
Linus Torvalds1da177e2005-04-16 15:20:36 -07004827 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4828 /*
4829 * We don't want to pressure a particular node.
4830 * So adding penalty to the first node in same
4831 * distance group to make it round-robin.
4832 */
David Rientjes957f8222012-10-08 16:33:24 -07004833 if (node_distance(local_node, node) !=
4834 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004835 node_load[node] = load;
4836
Linus Torvalds1da177e2005-04-16 15:20:36 -07004837 prev_node = node;
4838 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004839 if (order == ZONELIST_ORDER_NODE)
4840 build_zonelists_in_node_order(pgdat, node);
4841 else
Yaowei Baic00eb152016-01-14 15:19:00 -08004842 node_order[i++] = node; /* remember order */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004843 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004844
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004845 if (order == ZONELIST_ORDER_ZONE) {
4846 /* calculate node order -- i.e., DMA last! */
Yaowei Baic00eb152016-01-14 15:19:00 -08004847 build_zonelists_in_zone_order(pgdat, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004848 }
Christoph Lameter523b9452007-10-16 01:25:37 -07004849
4850 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004851}
4852
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004853#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4854/*
4855 * Return node id of node used for "local" allocations.
4856 * I.e., first node id of first zone in arg node's generic zonelist.
4857 * Used for initializing percpu 'numa_mem', which is used primarily
4858 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4859 */
4860int local_memory_node(int node)
4861{
Mel Gormanc33d6c02016-05-19 17:14:10 -07004862 struct zoneref *z;
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004863
Mel Gormanc33d6c02016-05-19 17:14:10 -07004864 z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004865 gfp_zone(GFP_KERNEL),
Mel Gormanc33d6c02016-05-19 17:14:10 -07004866 NULL);
4867 return z->zone->node;
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004868}
4869#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004870
Linus Torvalds1da177e2005-04-16 15:20:36 -07004871#else /* CONFIG_NUMA */
4872
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004873static void set_zonelist_order(void)
4874{
4875 current_zonelist_order = ZONELIST_ORDER_ZONE;
4876}
4877
4878static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004879{
Christoph Lameter19655d32006-09-25 23:31:19 -07004880 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07004881 enum zone_type j;
4882 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004883
4884 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004885
Mel Gorman54a6eb52008-04-28 02:12:16 -07004886 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004887 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004888
Mel Gorman54a6eb52008-04-28 02:12:16 -07004889 /*
4890 * Now we build the zonelist so that it contains the zones
4891 * of all the other nodes.
4892 * We don't want to pressure a particular node, so when
4893 * building the zones for node N, we make sure that the
4894 * zones coming right after the local ones are those from
4895 * node N+1 (modulo N)
4896 */
4897 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4898 if (!node_online(node))
4899 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004900 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004901 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07004902 for (node = 0; node < local_node; node++) {
4903 if (!node_online(node))
4904 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004905 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004906 }
4907
Mel Gormandd1a2392008-04-28 02:12:17 -07004908 zonelist->_zonerefs[j].zone = NULL;
4909 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004910}
4911
4912#endif /* CONFIG_NUMA */
4913
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004914/*
4915 * Boot pageset table. One per cpu which is going to be used for all
4916 * zones and all nodes. The parameters will be set in such a way
4917 * that an item put on a list will immediately be handed over to
4918 * the buddy list. This is safe since pageset manipulation is done
4919 * with interrupts disabled.
4920 *
4921 * The boot_pagesets must be kept even after bootup is complete for
4922 * unused processors and/or zones. They do play a role for bootstrapping
4923 * hotplugged processors.
4924 *
4925 * zoneinfo_show() and maybe other functions do
4926 * not check if the processor is online before following the pageset pointer.
4927 * Other parts of the kernel may not check if the zone is available.
4928 */
4929static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4930static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07004931static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004932
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004933/*
4934 * Global mutex to protect against size modification of zonelists
4935 * as well as to serialize pageset setup for the new populated zone.
4936 */
4937DEFINE_MUTEX(zonelists_mutex);
4938
Rusty Russell9b1a4d32008-07-28 12:16:30 -05004939/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07004940static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004941{
Yasunori Goto68113782006-06-23 02:03:11 -07004942 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004943 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07004944 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004945
Bo Liu7f9cfb32009-08-18 14:11:19 -07004946#ifdef CONFIG_NUMA
4947 memset(node_load, 0, sizeof(node_load));
4948#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07004949
4950 if (self && !node_online(self->node_id)) {
4951 build_zonelists(self);
Jiang Liu9adb62a2012-07-31 16:43:28 -07004952 }
4953
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004954 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07004955 pg_data_t *pgdat = NODE_DATA(nid);
4956
4957 build_zonelists(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004958 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004959
4960 /*
4961 * Initialize the boot_pagesets that are going to be used
4962 * for bootstrapping processors. The real pagesets for
4963 * each zone will be allocated later when the per cpu
4964 * allocator is available.
4965 *
4966 * boot_pagesets are used also for bootstrapping offline
4967 * cpus if the system is already booted because the pagesets
4968 * are needed to initialize allocators on a specific cpu too.
4969 * F.e. the percpu allocator needs the page allocator which
4970 * needs the percpu allocator in order to allocate its pagesets
4971 * (a chicken-egg dilemma).
4972 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004973 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004974 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4975
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004976#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4977 /*
4978 * We now know the "local memory node" for each node--
4979 * i.e., the node of the first zone in the generic zonelist.
4980 * Set up numa_mem percpu variable for on-line cpus. During
4981 * boot, only the boot cpu should be on-line; we'll init the
4982 * secondary cpus' numa_mem as they come on-line. During
4983 * node/memory hotplug, we'll fixup all on-line cpus.
4984 */
4985 if (cpu_online(cpu))
4986 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4987#endif
4988 }
4989
Yasunori Goto68113782006-06-23 02:03:11 -07004990 return 0;
4991}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004992
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004993static noinline void __init
4994build_all_zonelists_init(void)
4995{
4996 __build_all_zonelists(NULL);
4997 mminit_verify_zonelist();
4998 cpuset_init_current_mems_allowed();
4999}
5000
Haicheng Li4eaf3f62010-05-24 14:32:52 -07005001/*
5002 * Called with zonelists_mutex held always
5003 * unless system_state == SYSTEM_BOOTING.
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08005004 *
5005 * __ref due to (1) call of __meminit annotated setup_zone_pageset
5006 * [we're only called with non-NULL zone through __meminit paths] and
5007 * (2) call of __init annotated helper build_all_zonelists_init
5008 * [protected by SYSTEM_BOOTING].
Haicheng Li4eaf3f62010-05-24 14:32:52 -07005009 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07005010void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07005011{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07005012 set_zonelist_order();
5013
Yasunori Goto68113782006-06-23 02:03:11 -07005014 if (system_state == SYSTEM_BOOTING) {
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08005015 build_all_zonelists_init();
Yasunori Goto68113782006-06-23 02:03:11 -07005016 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08005017#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07005018 if (zone)
5019 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08005020#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07005021 /* we have to stop all cpus to guarantee there is no user
5022 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07005023 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07005024 /* cpuset refresh routine should be here */
5025 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07005026 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07005027 /*
5028 * Disable grouping by mobility if the number of pages in the
5029 * system is too low to allow the mechanism to work. It would be
5030 * more accurate, but expensive to check per-zone. This check is
5031 * made on memory-hotadd so a system can start with mobility
5032 * disabled and enable it later
5033 */
Mel Gormand9c23402007-10-16 01:26:01 -07005034 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07005035 page_group_by_mobility_disabled = 1;
5036 else
5037 page_group_by_mobility_disabled = 0;
5038
Joe Perches756a0252016-03-17 14:19:47 -07005039 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
5040 nr_online_nodes,
5041 zonelist_order_name[current_zonelist_order],
5042 page_group_by_mobility_disabled ? "off" : "on",
5043 vm_total_pages);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07005044#ifdef CONFIG_NUMA
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005045 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07005046#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005047}
5048
5049/*
5050 * Helper functions to size the waitqueue hash table.
5051 * Essentially these want to choose hash table sizes sufficiently
5052 * large so that collisions trying to wait on pages are rare.
5053 * But in fact, the number of active page waitqueues on typical
5054 * systems is ridiculously low, less than 200. So this is even
5055 * conservative, even though it seems large.
5056 *
5057 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
5058 * waitqueues, i.e. the size of the waitq table given the number of pages.
5059 */
5060#define PAGES_PER_WAITQUEUE 256
5061
Yasunori Gotocca448f2006-06-23 02:03:10 -07005062#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07005063static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005064{
5065 unsigned long size = 1;
5066
5067 pages /= PAGES_PER_WAITQUEUE;
5068
5069 while (size < pages)
5070 size <<= 1;
5071
5072 /*
5073 * Once we have dozens or even hundreds of threads sleeping
5074 * on IO we've got bigger problems than wait queue collision.
5075 * Limit the size of the wait table to a reasonable size.
5076 */
5077 size = min(size, 4096UL);
5078
5079 return max(size, 4UL);
5080}
Yasunori Gotocca448f2006-06-23 02:03:10 -07005081#else
5082/*
5083 * A zone's size might be changed by hot-add, so it is not possible to determine
5084 * a suitable size for its wait_table. So we use the maximum size now.
5085 *
5086 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
5087 *
5088 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
5089 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
5090 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
5091 *
5092 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
5093 * or more by the traditional way. (See above). It equals:
5094 *
5095 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
5096 * ia64(16K page size) : = ( 8G + 4M)byte.
5097 * powerpc (64K page size) : = (32G +16M)byte.
5098 */
5099static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
5100{
5101 return 4096UL;
5102}
5103#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005104
5105/*
5106 * This is an integer logarithm so that shifts can be used later
5107 * to extract the more random high bits from the multiplicative
5108 * hash function before the remainder is taken.
5109 */
5110static inline unsigned long wait_table_bits(unsigned long size)
5111{
5112 return ffz(~size);
5113}
5114
Mel Gorman56fd56b2007-10-16 01:25:58 -07005115/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005116 * Initially all pages are reserved - free ones are freed
5117 * up by free_all_bootmem() once the early boot process is
5118 * done. Non-atomic initialization, single-pass.
5119 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01005120void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08005121 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005122{
Dan Williams4b94ffd2016-01-15 16:56:22 -08005123 struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn));
Andy Whitcroft29751f62005-06-23 00:08:00 -07005124 unsigned long end_pfn = start_pfn + size;
Dan Williams4b94ffd2016-01-15 16:56:22 -08005125 pg_data_t *pgdat = NODE_DATA(nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -07005126 unsigned long pfn;
Mel Gorman3a80a7f2015-06-30 14:57:02 -07005127 unsigned long nr_initialised = 0;
Taku Izumi342332e2016-03-15 14:55:22 -07005128#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5129 struct memblock_region *r = NULL, *tmp;
5130#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005131
Hugh Dickins22b31ee2009-01-06 14:40:09 -08005132 if (highest_memmap_pfn < end_pfn - 1)
5133 highest_memmap_pfn = end_pfn - 1;
5134
Dan Williams4b94ffd2016-01-15 16:56:22 -08005135 /*
5136 * Honor reservation requested by the driver for this ZONE_DEVICE
5137 * memory
5138 */
5139 if (altmap && start_pfn == altmap->base_pfn)
5140 start_pfn += altmap->reserve;
5141
Greg Ungerercbe8dd42006-01-12 01:05:24 -08005142 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08005143 /*
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005144 * There can be holes in boot-time mem_map[]s handed to this
5145 * function. They do not exist on hotplugged memory.
Dave Hansena2f3aa022007-01-10 23:15:30 -08005146 */
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005147 if (context != MEMMAP_EARLY)
5148 goto not_early;
5149
5150 if (!early_pfn_valid(pfn))
5151 continue;
5152 if (!early_pfn_in_nid(pfn, nid))
5153 continue;
5154 if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised))
5155 break;
Taku Izumi342332e2016-03-15 14:55:22 -07005156
5157#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005158 /*
5159 * If not mirrored_kernelcore and ZONE_MOVABLE exists, range
5160 * from zone_movable_pfn[nid] to end of each node should be
5161 * ZONE_MOVABLE not ZONE_NORMAL. skip it.
5162 */
5163 if (!mirrored_kernelcore && zone_movable_pfn[nid])
5164 if (zone == ZONE_NORMAL && pfn >= zone_movable_pfn[nid])
5165 continue;
Taku Izumi342332e2016-03-15 14:55:22 -07005166
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005167 /*
5168 * Check given memblock attribute by firmware which can affect
5169 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
5170 * mirrored, it's an overlapped memmap init. skip it.
5171 */
5172 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
5173 if (!r || pfn >= memblock_region_memory_end_pfn(r)) {
5174 for_each_memblock(memory, tmp)
5175 if (pfn < memblock_region_memory_end_pfn(tmp))
5176 break;
5177 r = tmp;
Taku Izumi342332e2016-03-15 14:55:22 -07005178 }
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005179 if (pfn >= memblock_region_memory_base_pfn(r) &&
5180 memblock_is_mirror(r)) {
5181 /* already initialized as NORMAL */
5182 pfn = memblock_region_memory_end_pfn(r);
5183 continue;
5184 }
Dave Hansena2f3aa022007-01-10 23:15:30 -08005185 }
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005186#endif
Mel Gormanac5d2532015-06-30 14:57:20 -07005187
Andrew Mortonb72d0ff2016-03-15 14:55:25 -07005188not_early:
Mel Gormanac5d2532015-06-30 14:57:20 -07005189 /*
5190 * Mark the block movable so that blocks are reserved for
5191 * movable at startup. This will force kernel allocations
5192 * to reserve their blocks rather than leaking throughout
5193 * the address space during boot when many long-lived
Mel Gorman974a7862015-11-06 16:28:34 -08005194 * kernel allocations are made.
Mel Gormanac5d2532015-06-30 14:57:20 -07005195 *
5196 * bitmap is created for zone's valid pfn range. but memmap
5197 * can be created for invalid pages (for alignment)
5198 * check here not to call set_pageblock_migratetype() against
5199 * pfn out of zone.
5200 */
5201 if (!(pfn & (pageblock_nr_pages - 1))) {
5202 struct page *page = pfn_to_page(pfn);
5203
5204 __init_single_page(page, pfn, zone, nid);
5205 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5206 } else {
5207 __init_single_pfn(pfn, zone, nid);
5208 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005209 }
5210}
5211
Andi Kleen1e548de2008-02-04 22:29:26 -08005212static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005213{
Mel Gorman7aeb09f2014-06-04 16:10:21 -07005214 unsigned int order, t;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07005215 for_each_migratetype_order(order, t) {
5216 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005217 zone->free_area[order].nr_free = 0;
5218 }
5219}
5220
5221#ifndef __HAVE_ARCH_MEMMAP_INIT
5222#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08005223 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005224#endif
5225
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005226static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005227{
David Howells3a6be872009-05-06 16:03:03 -07005228#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005229 int batch;
5230
5231 /*
5232 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07005233 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005234 *
5235 * OK, so we don't know how big the cache is. So guess.
5236 */
Jiang Liub40da042013-02-22 16:33:52 -08005237 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07005238 if (batch * PAGE_SIZE > 512 * 1024)
5239 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005240 batch /= 4; /* We effectively *= 4 below */
5241 if (batch < 1)
5242 batch = 1;
5243
5244 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11005245 * Clamp the batch to a 2^n - 1 value. Having a power
5246 * of 2 value was found to be more likely to have
5247 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005248 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11005249 * For example if 2 tasks are alternately allocating
5250 * batches of pages, one task can end up with a lot
5251 * of pages of one half of the possible page colors
5252 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005253 */
David Howells91552032009-05-06 16:03:02 -07005254 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07005255
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005256 return batch;
David Howells3a6be872009-05-06 16:03:03 -07005257
5258#else
5259 /* The deferral and batching of frees should be suppressed under NOMMU
5260 * conditions.
5261 *
5262 * The problem is that NOMMU needs to be able to allocate large chunks
5263 * of contiguous memory as there's no hardware page translation to
5264 * assemble apparent contiguous memory from discontiguous pages.
5265 *
5266 * Queueing large contiguous runs of pages for batching, however,
5267 * causes the pages to actually be freed in smaller chunks. As there
5268 * can be a significant delay between the individual batches being
5269 * recycled, this leads to the once large chunks of space being
5270 * fragmented and becoming unavailable for high-order allocations.
5271 */
5272 return 0;
5273#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005274}
5275
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005276/*
5277 * pcp->high and pcp->batch values are related and dependent on one another:
5278 * ->batch must never be higher then ->high.
5279 * The following function updates them in a safe manner without read side
5280 * locking.
5281 *
5282 * Any new users of pcp->batch and pcp->high should ensure they can cope with
5283 * those fields changing asynchronously (acording the the above rule).
5284 *
5285 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5286 * outside of boot time (or some other assurance that no concurrent updaters
5287 * exist).
5288 */
5289static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
5290 unsigned long batch)
5291{
5292 /* start with a fail safe value for batch */
5293 pcp->batch = 1;
5294 smp_wmb();
5295
5296 /* Update high, then batch, in order */
5297 pcp->high = high;
5298 smp_wmb();
5299
5300 pcp->batch = batch;
5301}
5302
Cody P Schafer36640332013-07-03 15:01:40 -07005303/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07005304static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
5305{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005306 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07005307}
5308
Cody P Schafer88c90db2013-07-03 15:01:35 -07005309static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07005310{
5311 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07005312 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07005313
Magnus Damm1c6fe942005-10-26 01:58:59 -07005314 memset(p, 0, sizeof(*p));
5315
Christoph Lameter3dfa5722008-02-04 22:29:19 -08005316 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07005317 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07005318 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
5319 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07005320}
5321
Cody P Schafer88c90db2013-07-03 15:01:35 -07005322static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
5323{
5324 pageset_init(p);
5325 pageset_set_batch(p, batch);
5326}
5327
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005328/*
Cody P Schafer36640332013-07-03 15:01:40 -07005329 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005330 * to the value high for the pageset p.
5331 */
Cody P Schafer36640332013-07-03 15:01:40 -07005332static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005333 unsigned long high)
5334{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005335 unsigned long batch = max(1UL, high / 4);
5336 if ((high / 4) > (PAGE_SHIFT * 8))
5337 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005338
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07005339 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005340}
5341
David Rientjes7cd2b0a2014-06-23 13:22:04 -07005342static void pageset_set_high_and_batch(struct zone *zone,
5343 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005344{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005345 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07005346 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005347 (zone->managed_pages /
5348 percpu_pagelist_fraction));
5349 else
5350 pageset_set_batch(pcp, zone_batchsize(zone));
5351}
5352
Cody P Schafer169f6c12013-07-03 15:01:41 -07005353static void __meminit zone_pageset_init(struct zone *zone, int cpu)
5354{
5355 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
5356
5357 pageset_init(pcp);
5358 pageset_set_high_and_batch(zone, pcp);
5359}
5360
Jiang Liu4ed7e022012-07-31 16:43:35 -07005361static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07005362{
5363 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07005364 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07005365 for_each_possible_cpu(cpu)
5366 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07005367}
5368
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005369/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005370 * Allocate per cpu pagesets and initialize them.
5371 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07005372 */
Al Viro78d99552005-12-15 09:18:25 +00005373void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005374{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005375 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005376
Wu Fengguang319774e2010-05-24 14:32:49 -07005377 for_each_populated_zone(zone)
5378 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07005379}
5380
Sam Ravnborg577a32f2007-05-17 23:29:25 +02005381static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07005382int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07005383{
5384 int i;
Yasunori Gotocca448f2006-06-23 02:03:10 -07005385 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07005386
5387 /*
5388 * The per-page waitqueue mechanism uses hashed waitqueues
5389 * per zone.
5390 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07005391 zone->wait_table_hash_nr_entries =
5392 wait_table_hash_nr_entries(zone_size_pages);
5393 zone->wait_table_bits =
5394 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07005395 alloc_size = zone->wait_table_hash_nr_entries
5396 * sizeof(wait_queue_head_t);
5397
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07005398 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07005399 zone->wait_table = (wait_queue_head_t *)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005400 memblock_virt_alloc_node_nopanic(
5401 alloc_size, zone->zone_pgdat->node_id);
Yasunori Gotocca448f2006-06-23 02:03:10 -07005402 } else {
5403 /*
5404 * This case means that a zone whose size was 0 gets new memory
5405 * via memory hot-add.
5406 * But it may be the case that a new node was hot-added. In
5407 * this case vmalloc() will not be able to use this new node's
5408 * memory - this wait_table must be initialized to use this new
5409 * node itself as well.
5410 * To use this new node's memory, further consideration will be
5411 * necessary.
5412 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07005413 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07005414 }
5415 if (!zone->wait_table)
5416 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07005417
Pintu Kumarb8af2942013-09-11 14:20:34 -07005418 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07005419 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07005420
5421 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07005422}
5423
Matt Tolentinoc09b4242006-01-17 07:03:44 +01005424static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07005425{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005426 /*
5427 * per cpu subsystem is not up at this point. The following code
5428 * relies on the ability of the linker to provide the
5429 * offset of a (static) per cpu variable into the per cpu area.
5430 */
5431 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07005432
Xishi Qiub38a8722013-11-12 15:07:20 -08005433 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005434 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
5435 zone->name, zone->present_pages,
5436 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07005437}
5438
Jiang Liu4ed7e022012-07-31 16:43:35 -07005439int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07005440 unsigned long zone_start_pfn,
Yaowei Baib171e402015-11-05 18:47:06 -08005441 unsigned long size)
Dave Hansened8ece22005-10-29 18:16:50 -07005442{
5443 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07005444 int ret;
5445 ret = zone_wait_table_init(zone, size);
5446 if (ret)
5447 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07005448 pgdat->nr_zones = zone_idx(zone) + 1;
5449
Dave Hansened8ece22005-10-29 18:16:50 -07005450 zone->zone_start_pfn = zone_start_pfn;
5451
Mel Gorman708614e2008-07-23 21:26:51 -07005452 mminit_dprintk(MMINIT_TRACE, "memmap_init",
5453 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5454 pgdat->node_id,
5455 (unsigned long)zone_idx(zone),
5456 zone_start_pfn, (zone_start_pfn + size));
5457
Andi Kleen1e548de2008-02-04 22:29:26 -08005458 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07005459
5460 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07005461}
5462
Tejun Heo0ee332c2011-12-08 10:22:09 -08005463#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07005464#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
Mel Gorman8a942fd2015-06-30 14:56:55 -07005465
Mel Gormanc7132162006-09-27 01:49:43 -07005466/*
5467 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
Mel Gormanc7132162006-09-27 01:49:43 -07005468 */
Mel Gorman8a942fd2015-06-30 14:56:55 -07005469int __meminit __early_pfn_to_nid(unsigned long pfn,
5470 struct mminit_pfnnid_cache *state)
Mel Gormanc7132162006-09-27 01:49:43 -07005471{
Tejun Heoc13291a2011-07-12 10:46:30 +02005472 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07005473 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07005474
Mel Gorman8a942fd2015-06-30 14:56:55 -07005475 if (state->last_start <= pfn && pfn < state->last_end)
5476 return state->last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005477
Yinghai Lue76b63f2013-09-11 14:22:17 -07005478 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
5479 if (nid != -1) {
Mel Gorman8a942fd2015-06-30 14:56:55 -07005480 state->last_start = start_pfn;
5481 state->last_end = end_pfn;
5482 state->last_nid = nid;
Yinghai Lue76b63f2013-09-11 14:22:17 -07005483 }
5484
5485 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005486}
5487#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5488
Mel Gormanc7132162006-09-27 01:49:43 -07005489/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08005490 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005491 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08005492 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07005493 *
Zhang Zhen7d018172014-06-04 16:10:53 -07005494 * If an architecture guarantees that all ranges registered contain no holes
5495 * and may be freed, this this function may be used instead of calling
5496 * memblock_free_early_nid() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07005497 */
Tejun Heoc13291a2011-07-12 10:46:30 +02005498void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07005499{
Tejun Heoc13291a2011-07-12 10:46:30 +02005500 unsigned long start_pfn, end_pfn;
5501 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005502
Tejun Heoc13291a2011-07-12 10:46:30 +02005503 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
5504 start_pfn = min(start_pfn, max_low_pfn);
5505 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005506
Tejun Heoc13291a2011-07-12 10:46:30 +02005507 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005508 memblock_free_early_nid(PFN_PHYS(start_pfn),
5509 (end_pfn - start_pfn) << PAGE_SHIFT,
5510 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005511 }
5512}
5513
5514/**
5515 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005516 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07005517 *
Zhang Zhen7d018172014-06-04 16:10:53 -07005518 * If an architecture guarantees that all ranges registered contain no holes and may
5519 * be freed, this function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07005520 */
5521void __init sparse_memory_present_with_active_regions(int nid)
5522{
Tejun Heoc13291a2011-07-12 10:46:30 +02005523 unsigned long start_pfn, end_pfn;
5524 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07005525
Tejun Heoc13291a2011-07-12 10:46:30 +02005526 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
5527 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005528}
5529
5530/**
5531 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005532 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5533 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5534 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07005535 *
5536 * It returns the start and end page frame of a node based on information
Zhang Zhen7d018172014-06-04 16:10:53 -07005537 * provided by memblock_set_node(). If called for a node
Mel Gormanc7132162006-09-27 01:49:43 -07005538 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005539 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07005540 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005541void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005542 unsigned long *start_pfn, unsigned long *end_pfn)
5543{
Tejun Heoc13291a2011-07-12 10:46:30 +02005544 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005545 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02005546
Mel Gormanc7132162006-09-27 01:49:43 -07005547 *start_pfn = -1UL;
5548 *end_pfn = 0;
5549
Tejun Heoc13291a2011-07-12 10:46:30 +02005550 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5551 *start_pfn = min(*start_pfn, this_start_pfn);
5552 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005553 }
5554
Christoph Lameter633c0662007-10-16 01:25:37 -07005555 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07005556 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005557}
5558
5559/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07005560 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5561 * assumption is made that zones within a node are ordered in monotonic
5562 * increasing memory addresses so that the "highest" populated zone is used
5563 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005564static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005565{
5566 int zone_index;
5567 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5568 if (zone_index == ZONE_MOVABLE)
5569 continue;
5570
5571 if (arch_zone_highest_possible_pfn[zone_index] >
5572 arch_zone_lowest_possible_pfn[zone_index])
5573 break;
5574 }
5575
5576 VM_BUG_ON(zone_index == -1);
5577 movable_zone = zone_index;
5578}
5579
5580/*
5581 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03005582 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005583 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5584 * in each node depending on the size of each node and how evenly kernelcore
5585 * is distributed. This helper function adjusts the zone ranges
5586 * provided by the architecture for a given node by using the end of the
5587 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5588 * zones within a node are in order of monotonic increases memory addresses
5589 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005590static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07005591 unsigned long zone_type,
5592 unsigned long node_start_pfn,
5593 unsigned long node_end_pfn,
5594 unsigned long *zone_start_pfn,
5595 unsigned long *zone_end_pfn)
5596{
5597 /* Only adjust if ZONE_MOVABLE is on this node */
5598 if (zone_movable_pfn[nid]) {
5599 /* Size ZONE_MOVABLE */
5600 if (zone_type == ZONE_MOVABLE) {
5601 *zone_start_pfn = zone_movable_pfn[nid];
5602 *zone_end_pfn = min(node_end_pfn,
5603 arch_zone_highest_possible_pfn[movable_zone]);
5604
Mel Gorman2a1e2742007-07-17 04:03:12 -07005605 /* Check if this whole range is within ZONE_MOVABLE */
5606 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5607 *zone_start_pfn = *zone_end_pfn;
5608 }
5609}
5610
5611/*
Mel Gormanc7132162006-09-27 01:49:43 -07005612 * Return the number of pages a zone spans in a node, including holes
5613 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5614 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005615static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005616 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005617 unsigned long node_start_pfn,
5618 unsigned long node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005619 unsigned long *zone_start_pfn,
5620 unsigned long *zone_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005621 unsigned long *ignored)
5622{
Xishi Qiub5685e92015-09-08 15:04:16 -07005623 /* When hotadd a new node from cpu_up(), the node should be empty */
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005624 if (!node_start_pfn && !node_end_pfn)
5625 return 0;
5626
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005627 /* Get the start and end of the zone */
Taku Izumid91749c2016-03-15 14:55:18 -07005628 *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5629 *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07005630 adjust_zone_range_for_zone_movable(nid, zone_type,
5631 node_start_pfn, node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005632 zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005633
5634 /* Check that this node has pages within the zone's required range */
Taku Izumid91749c2016-03-15 14:55:18 -07005635 if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07005636 return 0;
5637
5638 /* Move the zone boundaries inside the node if necessary */
Taku Izumid91749c2016-03-15 14:55:18 -07005639 *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
5640 *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005641
5642 /* Return the spanned pages */
Taku Izumid91749c2016-03-15 14:55:18 -07005643 return *zone_end_pfn - *zone_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005644}
5645
5646/*
5647 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005648 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07005649 */
Yinghai Lu32996252009-12-15 17:59:02 -08005650unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005651 unsigned long range_start_pfn,
5652 unsigned long range_end_pfn)
5653{
Tejun Heo96e907d2011-07-12 10:46:29 +02005654 unsigned long nr_absent = range_end_pfn - range_start_pfn;
5655 unsigned long start_pfn, end_pfn;
5656 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07005657
Tejun Heo96e907d2011-07-12 10:46:29 +02005658 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5659 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5660 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5661 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005662 }
Tejun Heo96e907d2011-07-12 10:46:29 +02005663 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07005664}
5665
5666/**
5667 * absent_pages_in_range - Return number of page frames in holes within a range
5668 * @start_pfn: The start PFN to start searching for holes
5669 * @end_pfn: The end PFN to stop searching for holes
5670 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005671 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07005672 */
5673unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5674 unsigned long end_pfn)
5675{
5676 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5677}
5678
5679/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005680static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005681 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005682 unsigned long node_start_pfn,
5683 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005684 unsigned long *ignored)
5685{
Tejun Heo96e907d2011-07-12 10:46:29 +02005686 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5687 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07005688 unsigned long zone_start_pfn, zone_end_pfn;
Taku Izumi342332e2016-03-15 14:55:22 -07005689 unsigned long nr_absent;
Mel Gorman9c7cd682006-09-27 01:49:58 -07005690
Xishi Qiub5685e92015-09-08 15:04:16 -07005691 /* When hotadd a new node from cpu_up(), the node should be empty */
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005692 if (!node_start_pfn && !node_end_pfn)
5693 return 0;
5694
Tejun Heo96e907d2011-07-12 10:46:29 +02005695 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5696 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07005697
Mel Gorman2a1e2742007-07-17 04:03:12 -07005698 adjust_zone_range_for_zone_movable(nid, zone_type,
5699 node_start_pfn, node_end_pfn,
5700 &zone_start_pfn, &zone_end_pfn);
Taku Izumi342332e2016-03-15 14:55:22 -07005701 nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5702
5703 /*
5704 * ZONE_MOVABLE handling.
5705 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5706 * and vice versa.
5707 */
5708 if (zone_movable_pfn[nid]) {
5709 if (mirrored_kernelcore) {
5710 unsigned long start_pfn, end_pfn;
5711 struct memblock_region *r;
5712
5713 for_each_memblock(memory, r) {
5714 start_pfn = clamp(memblock_region_memory_base_pfn(r),
5715 zone_start_pfn, zone_end_pfn);
5716 end_pfn = clamp(memblock_region_memory_end_pfn(r),
5717 zone_start_pfn, zone_end_pfn);
5718
5719 if (zone_type == ZONE_MOVABLE &&
5720 memblock_is_mirror(r))
5721 nr_absent += end_pfn - start_pfn;
5722
5723 if (zone_type == ZONE_NORMAL &&
5724 !memblock_is_mirror(r))
5725 nr_absent += end_pfn - start_pfn;
5726 }
5727 } else {
5728 if (zone_type == ZONE_NORMAL)
5729 nr_absent += node_end_pfn - zone_movable_pfn[nid];
5730 }
5731 }
5732
5733 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07005734}
Mel Gorman0e0b8642006-09-27 01:49:56 -07005735
Tejun Heo0ee332c2011-12-08 10:22:09 -08005736#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005737static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005738 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005739 unsigned long node_start_pfn,
5740 unsigned long node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005741 unsigned long *zone_start_pfn,
5742 unsigned long *zone_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005743 unsigned long *zones_size)
5744{
Taku Izumid91749c2016-03-15 14:55:18 -07005745 unsigned int zone;
5746
5747 *zone_start_pfn = node_start_pfn;
5748 for (zone = 0; zone < zone_type; zone++)
5749 *zone_start_pfn += zones_size[zone];
5750
5751 *zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
5752
Mel Gormanc7132162006-09-27 01:49:43 -07005753 return zones_size[zone_type];
5754}
5755
Paul Mundt6ea6e682007-07-15 23:38:20 -07005756static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005757 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005758 unsigned long node_start_pfn,
5759 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005760 unsigned long *zholes_size)
5761{
5762 if (!zholes_size)
5763 return 0;
5764
5765 return zholes_size[zone_type];
5766}
Yinghai Lu20e69262013-03-01 14:51:27 -08005767
Tejun Heo0ee332c2011-12-08 10:22:09 -08005768#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005769
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005770static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005771 unsigned long node_start_pfn,
5772 unsigned long node_end_pfn,
5773 unsigned long *zones_size,
5774 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07005775{
Gu Zhengfebd5942015-06-24 16:57:02 -07005776 unsigned long realtotalpages = 0, totalpages = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005777 enum zone_type i;
5778
Gu Zhengfebd5942015-06-24 16:57:02 -07005779 for (i = 0; i < MAX_NR_ZONES; i++) {
5780 struct zone *zone = pgdat->node_zones + i;
Taku Izumid91749c2016-03-15 14:55:18 -07005781 unsigned long zone_start_pfn, zone_end_pfn;
Gu Zhengfebd5942015-06-24 16:57:02 -07005782 unsigned long size, real_size;
Mel Gormanc7132162006-09-27 01:49:43 -07005783
Gu Zhengfebd5942015-06-24 16:57:02 -07005784 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5785 node_start_pfn,
5786 node_end_pfn,
Taku Izumid91749c2016-03-15 14:55:18 -07005787 &zone_start_pfn,
5788 &zone_end_pfn,
Gu Zhengfebd5942015-06-24 16:57:02 -07005789 zones_size);
5790 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005791 node_start_pfn, node_end_pfn,
5792 zholes_size);
Taku Izumid91749c2016-03-15 14:55:18 -07005793 if (size)
5794 zone->zone_start_pfn = zone_start_pfn;
5795 else
5796 zone->zone_start_pfn = 0;
Gu Zhengfebd5942015-06-24 16:57:02 -07005797 zone->spanned_pages = size;
5798 zone->present_pages = real_size;
5799
5800 totalpages += size;
5801 realtotalpages += real_size;
5802 }
5803
5804 pgdat->node_spanned_pages = totalpages;
Mel Gormanc7132162006-09-27 01:49:43 -07005805 pgdat->node_present_pages = realtotalpages;
5806 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5807 realtotalpages);
5808}
5809
Mel Gorman835c1342007-10-16 01:25:47 -07005810#ifndef CONFIG_SPARSEMEM
5811/*
5812 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07005813 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5814 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07005815 * round what is now in bits to nearest long in bits, then return it in
5816 * bytes.
5817 */
Linus Torvalds7c455122013-02-18 09:58:02 -08005818static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005819{
5820 unsigned long usemapsize;
5821
Linus Torvalds7c455122013-02-18 09:58:02 -08005822 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005823 usemapsize = roundup(zonesize, pageblock_nr_pages);
5824 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07005825 usemapsize *= NR_PAGEBLOCK_BITS;
5826 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5827
5828 return usemapsize / 8;
5829}
5830
5831static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08005832 struct zone *zone,
5833 unsigned long zone_start_pfn,
5834 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005835{
Linus Torvalds7c455122013-02-18 09:58:02 -08005836 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07005837 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08005838 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005839 zone->pageblock_flags =
5840 memblock_virt_alloc_node_nopanic(usemapsize,
5841 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07005842}
5843#else
Linus Torvalds7c455122013-02-18 09:58:02 -08005844static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5845 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07005846#endif /* CONFIG_SPARSEMEM */
5847
Mel Gormand9c23402007-10-16 01:26:01 -07005848#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08005849
Mel Gormand9c23402007-10-16 01:26:01 -07005850/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07005851void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07005852{
Andrew Morton955c1cd2012-05-29 15:06:31 -07005853 unsigned int order;
5854
Mel Gormand9c23402007-10-16 01:26:01 -07005855 /* Check that pageblock_nr_pages has not already been setup */
5856 if (pageblock_order)
5857 return;
5858
Andrew Morton955c1cd2012-05-29 15:06:31 -07005859 if (HPAGE_SHIFT > PAGE_SHIFT)
5860 order = HUGETLB_PAGE_ORDER;
5861 else
5862 order = MAX_ORDER - 1;
5863
Mel Gormand9c23402007-10-16 01:26:01 -07005864 /*
5865 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07005866 * This value may be variable depending on boot parameters on IA64 and
5867 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07005868 */
5869 pageblock_order = order;
5870}
5871#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5872
Mel Gormanba72cb82007-11-28 16:21:13 -08005873/*
5874 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07005875 * is unused as pageblock_order is set at compile-time. See
5876 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5877 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08005878 */
Chen Gang15ca2202013-09-11 14:20:27 -07005879void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08005880{
Mel Gormanba72cb82007-11-28 16:21:13 -08005881}
Mel Gormand9c23402007-10-16 01:26:01 -07005882
5883#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5884
Jiang Liu01cefae2012-12-12 13:52:19 -08005885static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5886 unsigned long present_pages)
5887{
5888 unsigned long pages = spanned_pages;
5889
5890 /*
5891 * Provide a more accurate estimation if there are holes within
5892 * the zone and SPARSEMEM is in use. If there are holes within the
5893 * zone, each populated memory region may cost us one or two extra
5894 * memmap pages due to alignment because memmap pages for each
5895 * populated regions may not naturally algined on page boundary.
5896 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5897 */
5898 if (spanned_pages > present_pages + (present_pages >> 4) &&
5899 IS_ENABLED(CONFIG_SPARSEMEM))
5900 pages = present_pages;
5901
5902 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5903}
5904
Linus Torvalds1da177e2005-04-16 15:20:36 -07005905/*
5906 * Set up the zone data structures:
5907 * - mark all pages reserved
5908 * - mark all memory queues empty
5909 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07005910 *
5911 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005912 */
Wei Yang7f3eb552015-09-08 14:59:50 -07005913static void __paginginit free_area_init_core(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005914{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07005915 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07005916 int nid = pgdat->node_id;
Yasunori Goto718127c2006-06-23 02:03:10 -07005917 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005918
Dave Hansen208d54e2005-10-29 18:16:52 -07005919 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01005920#ifdef CONFIG_NUMA_BALANCING
5921 spin_lock_init(&pgdat->numabalancing_migrate_lock);
5922 pgdat->numabalancing_migrate_nr_pages = 0;
5923 pgdat->numabalancing_migrate_next_window = jiffies;
5924#endif
Kirill A. Shutemova3d0a9182016-02-02 16:57:08 -08005925#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5926 spin_lock_init(&pgdat->split_queue_lock);
5927 INIT_LIST_HEAD(&pgdat->split_queue);
5928 pgdat->split_queue_len = 0;
5929#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005930 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07005931 init_waitqueue_head(&pgdat->pfmemalloc_wait);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07005932#ifdef CONFIG_COMPACTION
5933 init_waitqueue_head(&pgdat->kcompactd_wait);
5934#endif
Joonsoo Kimeefa864b2014-12-12 16:55:46 -08005935 pgdat_page_ext_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01005936
Linus Torvalds1da177e2005-04-16 15:20:36 -07005937 for (j = 0; j < MAX_NR_ZONES; j++) {
5938 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08005939 unsigned long size, realsize, freesize, memmap_pages;
Taku Izumid91749c2016-03-15 14:55:18 -07005940 unsigned long zone_start_pfn = zone->zone_start_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005941
Gu Zhengfebd5942015-06-24 16:57:02 -07005942 size = zone->spanned_pages;
5943 realsize = freesize = zone->present_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005944
Mel Gorman0e0b8642006-09-27 01:49:56 -07005945 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08005946 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07005947 * is used by this zone for memmap. This affects the watermark
5948 * and per-cpu initialisations
5949 */
Jiang Liu01cefae2012-12-12 13:52:19 -08005950 memmap_pages = calc_memmap_size(size, realsize);
Zhong Hongboba914f42014-12-12 16:56:21 -08005951 if (!is_highmem_idx(j)) {
5952 if (freesize >= memmap_pages) {
5953 freesize -= memmap_pages;
5954 if (memmap_pages)
5955 printk(KERN_DEBUG
5956 " %s zone: %lu pages used for memmap\n",
5957 zone_names[j], memmap_pages);
5958 } else
Joe Perches11705322016-03-17 14:19:50 -07005959 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
Zhong Hongboba914f42014-12-12 16:56:21 -08005960 zone_names[j], memmap_pages, freesize);
5961 }
Mel Gorman0e0b8642006-09-27 01:49:56 -07005962
Christoph Lameter62672762007-02-10 01:43:07 -08005963 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08005964 if (j == 0 && freesize > dma_reserve) {
5965 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07005966 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08005967 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07005968 }
5969
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07005970 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08005971 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08005972 /* Charge for highmem memmap if there are enough kernel pages */
5973 else if (nr_kernel_pages > memmap_pages * 2)
5974 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08005975 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005976
Jiang Liu9feedc92012-12-12 13:52:12 -08005977 /*
5978 * Set an approximate value for lowmem here, it will be adjusted
5979 * when the bootmem allocator frees pages into the buddy system.
5980 * And all highmem pages will be managed by the buddy system.
5981 */
5982 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07005983#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07005984 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08005985 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07005986 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08005987 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07005988#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005989 zone->name = zone_names[j];
5990 spin_lock_init(&zone->lock);
5991 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07005992 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005993 zone->zone_pgdat = pgdat;
Dave Hansened8ece22005-10-29 18:16:50 -07005994 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005995
5996 /* For bootup, initialized properly in watermark setup */
5997 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
5998
Hugh Dickinsbea8c152012-11-16 14:14:54 -08005999 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006000 if (!size)
6001 continue;
6002
Andrew Morton955c1cd2012-05-29 15:06:31 -07006003 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08006004 setup_usemap(pgdat, zone, zone_start_pfn, size);
Yaowei Baib171e402015-11-05 18:47:06 -08006005 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
Yasunori Goto718127c2006-06-23 02:03:10 -07006006 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07006007 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006008 }
6009}
6010
Sam Ravnborg577a32f2007-05-17 23:29:25 +02006011static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006012{
Tony Luckb0aeba72015-11-10 10:09:47 -08006013 unsigned long __maybe_unused start = 0;
Laura Abbotta1c34a32015-11-05 18:48:46 -08006014 unsigned long __maybe_unused offset = 0;
6015
Linus Torvalds1da177e2005-04-16 15:20:36 -07006016 /* Skip empty nodes */
6017 if (!pgdat->node_spanned_pages)
6018 return;
6019
Andy Whitcroftd41dee32005-06-23 00:07:54 -07006020#ifdef CONFIG_FLAT_NODE_MEM_MAP
Tony Luckb0aeba72015-11-10 10:09:47 -08006021 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
6022 offset = pgdat->node_start_pfn - start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006023 /* ia64 gets its own node_mem_map, before this, without bootmem */
6024 if (!pgdat->node_mem_map) {
Tony Luckb0aeba72015-11-10 10:09:47 -08006025 unsigned long size, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07006026 struct page *map;
6027
Bob Piccoe984bb42006-05-20 15:00:31 -07006028 /*
6029 * The zone's endpoints aren't required to be MAX_ORDER
6030 * aligned but the node_mem_map endpoints must be in order
6031 * for the buddy allocator to function correctly.
6032 */
Cody P Schafer108bcc92013-02-22 16:35:23 -08006033 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07006034 end = ALIGN(end, MAX_ORDER_NR_PAGES);
6035 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07006036 map = alloc_remap(pgdat->node_id, size);
6037 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08006038 map = memblock_virt_alloc_node_nopanic(size,
6039 pgdat->node_id);
Laura Abbotta1c34a32015-11-05 18:48:46 -08006040 pgdat->node_mem_map = map + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006041 }
Roman Zippel12d810c2007-05-31 00:40:54 -07006042#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07006043 /*
6044 * With no DISCONTIG, the global mem_map is just set as node 0's
6045 */
Mel Gormanc7132162006-09-27 01:49:43 -07006046 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006047 mem_map = NODE_DATA(0)->node_mem_map;
Laura Abbotta1c34a32015-11-05 18:48:46 -08006048#if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
Mel Gormanc7132162006-09-27 01:49:43 -07006049 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Laura Abbotta1c34a32015-11-05 18:48:46 -08006050 mem_map -= offset;
Tejun Heo0ee332c2011-12-08 10:22:09 -08006051#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07006052 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006053#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07006054#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006055}
6056
Johannes Weiner9109fb72008-07-23 21:27:20 -07006057void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
6058 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006059{
Johannes Weiner9109fb72008-07-23 21:27:20 -07006060 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07006061 unsigned long start_pfn = 0;
6062 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07006063
Minchan Kim88fdf752012-07-31 16:46:14 -07006064 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07006065 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07006066
Mel Gorman3a80a7f2015-06-30 14:57:02 -07006067 reset_deferred_meminit(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006068 pgdat->node_id = nid;
6069 pgdat->node_start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07006070#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6071 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
Juergen Gross8d29e182015-02-11 15:26:01 -08006072 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
Zhen Lei4ada0c52015-09-08 15:04:19 -07006073 (u64)start_pfn << PAGE_SHIFT,
6074 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
Taku Izumid91749c2016-03-15 14:55:18 -07006075#else
6076 start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07006077#endif
6078 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
6079 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006080
6081 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07006082#ifdef CONFIG_FLAT_NODE_MEM_MAP
6083 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
6084 nid, (unsigned long)pgdat,
6085 (unsigned long)pgdat->node_mem_map);
6086#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07006087
Wei Yang7f3eb552015-09-08 14:59:50 -07006088 free_area_init_core(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006089}
6090
Tejun Heo0ee332c2011-12-08 10:22:09 -08006091#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07006092
6093#if MAX_NUMNODES > 1
6094/*
6095 * Figure out the number of possible node ids.
6096 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07006097void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07006098{
Wei Yang904a9552015-09-08 14:59:48 -07006099 unsigned int highest;
Miklos Szeredi418508c2007-05-23 13:57:55 -07006100
Wei Yang904a9552015-09-08 14:59:48 -07006101 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
Miklos Szeredi418508c2007-05-23 13:57:55 -07006102 nr_node_ids = highest + 1;
6103}
Miklos Szeredi418508c2007-05-23 13:57:55 -07006104#endif
6105
Mel Gormanc7132162006-09-27 01:49:43 -07006106/**
Tejun Heo1e019792011-07-12 09:45:34 +02006107 * node_map_pfn_alignment - determine the maximum internode alignment
6108 *
6109 * This function should be called after node map is populated and sorted.
6110 * It calculates the maximum power of two alignment which can distinguish
6111 * all the nodes.
6112 *
6113 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6114 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
6115 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
6116 * shifted, 1GiB is enough and this function will indicate so.
6117 *
6118 * This is used to test whether pfn -> nid mapping of the chosen memory
6119 * model has fine enough granularity to avoid incorrect mapping for the
6120 * populated node map.
6121 *
6122 * Returns the determined alignment in pfn's. 0 if there is no alignment
6123 * requirement (single node).
6124 */
6125unsigned long __init node_map_pfn_alignment(void)
6126{
6127 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02006128 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02006129 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02006130 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02006131
Tejun Heoc13291a2011-07-12 10:46:30 +02006132 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02006133 if (!start || last_nid < 0 || last_nid == nid) {
6134 last_nid = nid;
6135 last_end = end;
6136 continue;
6137 }
6138
6139 /*
6140 * Start with a mask granular enough to pin-point to the
6141 * start pfn and tick off bits one-by-one until it becomes
6142 * too coarse to separate the current node from the last.
6143 */
6144 mask = ~((1 << __ffs(start)) - 1);
6145 while (mask && last_end <= (start & (mask << 1)))
6146 mask <<= 1;
6147
6148 /* accumulate all internode masks */
6149 accl_mask |= mask;
6150 }
6151
6152 /* convert mask to number of pages */
6153 return ~accl_mask + 1;
6154}
6155
Mel Gormana6af2bc2007-02-10 01:42:57 -08006156/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07006157static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07006158{
Mel Gormana6af2bc2007-02-10 01:42:57 -08006159 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02006160 unsigned long start_pfn;
6161 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00006162
Tejun Heoc13291a2011-07-12 10:46:30 +02006163 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
6164 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07006165
Mel Gormana6af2bc2007-02-10 01:42:57 -08006166 if (min_pfn == ULONG_MAX) {
Joe Perches11705322016-03-17 14:19:50 -07006167 pr_warn("Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08006168 return 0;
6169 }
6170
6171 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07006172}
6173
6174/**
6175 * find_min_pfn_with_active_regions - Find the minimum PFN registered
6176 *
6177 * It returns the minimum PFN based on information provided via
Zhang Zhen7d018172014-06-04 16:10:53 -07006178 * memblock_set_node().
Mel Gormanc7132162006-09-27 01:49:43 -07006179 */
6180unsigned long __init find_min_pfn_with_active_regions(void)
6181{
6182 return find_min_pfn_for_node(MAX_NUMNODES);
6183}
6184
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006185/*
6186 * early_calculate_totalpages()
6187 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006188 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006189 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07006190static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07006191{
Mel Gorman7e63efe2007-07-17 04:03:15 -07006192 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02006193 unsigned long start_pfn, end_pfn;
6194 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07006195
Tejun Heoc13291a2011-07-12 10:46:30 +02006196 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
6197 unsigned long pages = end_pfn - start_pfn;
6198
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006199 totalpages += pages;
6200 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006201 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006202 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07006203 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07006204}
6205
Mel Gorman2a1e2742007-07-17 04:03:12 -07006206/*
6207 * Find the PFN the Movable zone begins in each node. Kernel memory
6208 * is spread evenly between nodes as long as the nodes have enough
6209 * memory. When they don't, some nodes will have more kernelcore than
6210 * others
6211 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07006212static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07006213{
6214 int i, nid;
6215 unsigned long usable_startpfn;
6216 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07006217 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006218 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006219 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006220 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07006221 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006222
6223 /* Need to find movable_zone earlier when movable_node is specified. */
6224 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07006225
Mel Gorman7e63efe2007-07-17 04:03:15 -07006226 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006227 * If movable_node is specified, ignore kernelcore and movablecore
6228 * options.
6229 */
6230 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07006231 for_each_memblock(memory, r) {
6232 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006233 continue;
6234
Emil Medve136199f2014-04-07 15:37:52 -07006235 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006236
Emil Medve136199f2014-04-07 15:37:52 -07006237 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006238 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6239 min(usable_startpfn, zone_movable_pfn[nid]) :
6240 usable_startpfn;
6241 }
6242
6243 goto out2;
6244 }
6245
6246 /*
Taku Izumi342332e2016-03-15 14:55:22 -07006247 * If kernelcore=mirror is specified, ignore movablecore option
6248 */
6249 if (mirrored_kernelcore) {
6250 bool mem_below_4gb_not_mirrored = false;
6251
6252 for_each_memblock(memory, r) {
6253 if (memblock_is_mirror(r))
6254 continue;
6255
6256 nid = r->nid;
6257
6258 usable_startpfn = memblock_region_memory_base_pfn(r);
6259
6260 if (usable_startpfn < 0x100000) {
6261 mem_below_4gb_not_mirrored = true;
6262 continue;
6263 }
6264
6265 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6266 min(usable_startpfn, zone_movable_pfn[nid]) :
6267 usable_startpfn;
6268 }
6269
6270 if (mem_below_4gb_not_mirrored)
6271 pr_warn("This configuration results in unmirrored kernel memory.");
6272
6273 goto out2;
6274 }
6275
6276 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006277 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07006278 * kernelcore that corresponds so that memory usable for
6279 * any allocation type is evenly spread. If both kernelcore
6280 * and movablecore are specified, then the value of kernelcore
6281 * will be used for required_kernelcore if it's greater than
6282 * what movablecore would have allowed.
6283 */
6284 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07006285 unsigned long corepages;
6286
6287 /*
6288 * Round-up so that ZONE_MOVABLE is at least as large as what
6289 * was requested by the user
6290 */
6291 required_movablecore =
6292 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
Xishi Qiu9fd745d2015-11-05 18:48:11 -08006293 required_movablecore = min(totalpages, required_movablecore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07006294 corepages = totalpages - required_movablecore;
6295
6296 required_kernelcore = max(required_kernelcore, corepages);
6297 }
6298
Xishi Qiubde304b2015-11-05 18:48:56 -08006299 /*
6300 * If kernelcore was not specified or kernelcore size is larger
6301 * than totalpages, there is no ZONE_MOVABLE.
6302 */
6303 if (!required_kernelcore || required_kernelcore >= totalpages)
Yinghai Lu66918dc2009-06-30 11:41:37 -07006304 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07006305
6306 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07006307 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
6308
6309restart:
6310 /* Spread kernelcore memory as evenly as possible throughout nodes */
6311 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006312 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02006313 unsigned long start_pfn, end_pfn;
6314
Mel Gorman2a1e2742007-07-17 04:03:12 -07006315 /*
6316 * Recalculate kernelcore_node if the division per node
6317 * now exceeds what is necessary to satisfy the requested
6318 * amount of memory for the kernel
6319 */
6320 if (required_kernelcore < kernelcore_node)
6321 kernelcore_node = required_kernelcore / usable_nodes;
6322
6323 /*
6324 * As the map is walked, we track how much memory is usable
6325 * by the kernel using kernelcore_remaining. When it is
6326 * 0, the rest of the node is usable by ZONE_MOVABLE
6327 */
6328 kernelcore_remaining = kernelcore_node;
6329
6330 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02006331 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07006332 unsigned long size_pages;
6333
Tejun Heoc13291a2011-07-12 10:46:30 +02006334 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07006335 if (start_pfn >= end_pfn)
6336 continue;
6337
6338 /* Account for what is only usable for kernelcore */
6339 if (start_pfn < usable_startpfn) {
6340 unsigned long kernel_pages;
6341 kernel_pages = min(end_pfn, usable_startpfn)
6342 - start_pfn;
6343
6344 kernelcore_remaining -= min(kernel_pages,
6345 kernelcore_remaining);
6346 required_kernelcore -= min(kernel_pages,
6347 required_kernelcore);
6348
6349 /* Continue if range is now fully accounted */
6350 if (end_pfn <= usable_startpfn) {
6351
6352 /*
6353 * Push zone_movable_pfn to the end so
6354 * that if we have to rebalance
6355 * kernelcore across nodes, we will
6356 * not double account here
6357 */
6358 zone_movable_pfn[nid] = end_pfn;
6359 continue;
6360 }
6361 start_pfn = usable_startpfn;
6362 }
6363
6364 /*
6365 * The usable PFN range for ZONE_MOVABLE is from
6366 * start_pfn->end_pfn. Calculate size_pages as the
6367 * number of pages used as kernelcore
6368 */
6369 size_pages = end_pfn - start_pfn;
6370 if (size_pages > kernelcore_remaining)
6371 size_pages = kernelcore_remaining;
6372 zone_movable_pfn[nid] = start_pfn + size_pages;
6373
6374 /*
6375 * Some kernelcore has been met, update counts and
6376 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07006377 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07006378 */
6379 required_kernelcore -= min(required_kernelcore,
6380 size_pages);
6381 kernelcore_remaining -= size_pages;
6382 if (!kernelcore_remaining)
6383 break;
6384 }
6385 }
6386
6387 /*
6388 * If there is still required_kernelcore, we do another pass with one
6389 * less node in the count. This will push zone_movable_pfn[nid] further
6390 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07006391 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07006392 */
6393 usable_nodes--;
6394 if (usable_nodes && required_kernelcore > usable_nodes)
6395 goto restart;
6396
Tang Chenb2f3eeb2014-01-21 15:49:38 -08006397out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07006398 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6399 for (nid = 0; nid < MAX_NUMNODES; nid++)
6400 zone_movable_pfn[nid] =
6401 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07006402
Yinghai Lu20e69262013-03-01 14:51:27 -08006403out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07006404 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006405 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07006406}
6407
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006408/* Any regular or high memory on that node ? */
6409static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006410{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006411 enum zone_type zone_type;
6412
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006413 if (N_MEMORY == N_NORMAL_MEMORY)
6414 return;
6415
6416 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006417 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08006418 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006419 node_set_state(nid, N_HIGH_MEMORY);
6420 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
6421 zone_type <= ZONE_NORMAL)
6422 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08006423 break;
6424 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006425 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006426}
6427
Mel Gormanc7132162006-09-27 01:49:43 -07006428/**
6429 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006430 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07006431 *
6432 * This will call free_area_init_node() for each active node in the system.
Zhang Zhen7d018172014-06-04 16:10:53 -07006433 * Using the page ranges provided by memblock_set_node(), the size of each
Mel Gormanc7132162006-09-27 01:49:43 -07006434 * zone in each node and their holes is calculated. If the maximum PFN
6435 * between two adjacent zones match, it is assumed that the zone is empty.
6436 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6437 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6438 * starts where the previous one ended. For example, ZONE_DMA32 starts
6439 * at arch_max_dma_pfn.
6440 */
6441void __init free_area_init_nodes(unsigned long *max_zone_pfn)
6442{
Tejun Heoc13291a2011-07-12 10:46:30 +02006443 unsigned long start_pfn, end_pfn;
6444 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08006445
Mel Gormanc7132162006-09-27 01:49:43 -07006446 /* Record where the zone boundaries are */
6447 memset(arch_zone_lowest_possible_pfn, 0,
6448 sizeof(arch_zone_lowest_possible_pfn));
6449 memset(arch_zone_highest_possible_pfn, 0,
6450 sizeof(arch_zone_highest_possible_pfn));
6451 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
6452 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
6453 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07006454 if (i == ZONE_MOVABLE)
6455 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07006456 arch_zone_lowest_possible_pfn[i] =
6457 arch_zone_highest_possible_pfn[i-1];
6458 arch_zone_highest_possible_pfn[i] =
6459 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
6460 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07006461 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
6462 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
6463
6464 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6465 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07006466 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07006467
Mel Gormanc7132162006-09-27 01:49:43 -07006468 /* Print out the zone ranges */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006469 pr_info("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07006470 for (i = 0; i < MAX_NR_ZONES; i++) {
6471 if (i == ZONE_MOVABLE)
6472 continue;
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006473 pr_info(" %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08006474 if (arch_zone_lowest_possible_pfn[i] ==
6475 arch_zone_highest_possible_pfn[i])
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006476 pr_cont("empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08006477 else
Juergen Gross8d29e182015-02-11 15:26:01 -08006478 pr_cont("[mem %#018Lx-%#018Lx]\n",
6479 (u64)arch_zone_lowest_possible_pfn[i]
6480 << PAGE_SHIFT,
6481 ((u64)arch_zone_highest_possible_pfn[i]
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07006482 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07006483 }
6484
6485 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006486 pr_info("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07006487 for (i = 0; i < MAX_NUMNODES; i++) {
6488 if (zone_movable_pfn[i])
Juergen Gross8d29e182015-02-11 15:26:01 -08006489 pr_info(" Node %d: %#018Lx\n", i,
6490 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07006491 }
Mel Gormanc7132162006-09-27 01:49:43 -07006492
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07006493 /* Print out the early node map */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08006494 pr_info("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02006495 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Juergen Gross8d29e182015-02-11 15:26:01 -08006496 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
6497 (u64)start_pfn << PAGE_SHIFT,
6498 ((u64)end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07006499
6500 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07006501 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08006502 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07006503 for_each_online_node(nid) {
6504 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07006505 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07006506 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07006507
6508 /* Any memory on that node */
6509 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08006510 node_set_state(nid, N_MEMORY);
6511 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07006512 }
6513}
Mel Gorman2a1e2742007-07-17 04:03:12 -07006514
Mel Gorman7e63efe2007-07-17 04:03:15 -07006515static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07006516{
6517 unsigned long long coremem;
6518 if (!p)
6519 return -EINVAL;
6520
6521 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07006522 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07006523
Mel Gorman7e63efe2007-07-17 04:03:15 -07006524 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07006525 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
6526
6527 return 0;
6528}
Mel Gormaned7ed362007-07-17 04:03:14 -07006529
Mel Gorman7e63efe2007-07-17 04:03:15 -07006530/*
6531 * kernelcore=size sets the amount of memory for use for allocations that
6532 * cannot be reclaimed or migrated.
6533 */
6534static int __init cmdline_parse_kernelcore(char *p)
6535{
Taku Izumi342332e2016-03-15 14:55:22 -07006536 /* parse kernelcore=mirror */
6537 if (parse_option_str(p, "mirror")) {
6538 mirrored_kernelcore = true;
6539 return 0;
6540 }
6541
Mel Gorman7e63efe2007-07-17 04:03:15 -07006542 return cmdline_parse_core(p, &required_kernelcore);
6543}
6544
6545/*
6546 * movablecore=size sets the amount of memory for use for allocations that
6547 * can be reclaimed or migrated.
6548 */
6549static int __init cmdline_parse_movablecore(char *p)
6550{
6551 return cmdline_parse_core(p, &required_movablecore);
6552}
6553
Mel Gormaned7ed362007-07-17 04:03:14 -07006554early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07006555early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07006556
Tejun Heo0ee332c2011-12-08 10:22:09 -08006557#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07006558
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006559void adjust_managed_page_count(struct page *page, long count)
6560{
6561 spin_lock(&managed_page_count_lock);
6562 page_zone(page)->managed_pages += count;
6563 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07006564#ifdef CONFIG_HIGHMEM
6565 if (PageHighMem(page))
6566 totalhigh_pages += count;
6567#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006568 spin_unlock(&managed_page_count_lock);
6569}
Jiang Liu3dcc0572013-07-03 15:03:21 -07006570EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07006571
Jiang Liu11199692013-07-03 15:02:48 -07006572unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07006573{
Jiang Liu11199692013-07-03 15:02:48 -07006574 void *pos;
6575 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07006576
Jiang Liu11199692013-07-03 15:02:48 -07006577 start = (void *)PAGE_ALIGN((unsigned long)start);
6578 end = (void *)((unsigned long)end & PAGE_MASK);
6579 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07006580 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07006581 memset(pos, poison, PAGE_SIZE);
6582 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07006583 }
6584
6585 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07006586 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07006587 s, pages << (PAGE_SHIFT - 10), start, end);
6588
6589 return pages;
6590}
Jiang Liu11199692013-07-03 15:02:48 -07006591EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07006592
Jiang Liucfa11e02013-04-29 15:07:00 -07006593#ifdef CONFIG_HIGHMEM
6594void free_highmem_page(struct page *page)
6595{
6596 __free_reserved_page(page);
6597 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07006598 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07006599 totalhigh_pages++;
6600}
6601#endif
6602
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006603
6604void __init mem_init_print_info(const char *str)
6605{
6606 unsigned long physpages, codesize, datasize, rosize, bss_size;
6607 unsigned long init_code_size, init_data_size;
6608
6609 physpages = get_num_physpages();
6610 codesize = _etext - _stext;
6611 datasize = _edata - _sdata;
6612 rosize = __end_rodata - __start_rodata;
6613 bss_size = __bss_stop - __bss_start;
6614 init_data_size = __init_end - __init_begin;
6615 init_code_size = _einittext - _sinittext;
6616
6617 /*
6618 * Detect special cases and adjust section sizes accordingly:
6619 * 1) .init.* may be embedded into .data sections
6620 * 2) .init.text.* may be out of [__init_begin, __init_end],
6621 * please refer to arch/tile/kernel/vmlinux.lds.S.
6622 * 3) .rodata.* may be embedded into .text or .data sections.
6623 */
6624#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07006625 do { \
6626 if (start <= pos && pos < end && size > adj) \
6627 size -= adj; \
6628 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006629
6630 adj_init_size(__init_begin, __init_end, init_data_size,
6631 _sinittext, init_code_size);
6632 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
6633 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
6634 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
6635 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
6636
6637#undef adj_init_size
6638
Joe Perches756a0252016-03-17 14:19:47 -07006639 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 -07006640#ifdef CONFIG_HIGHMEM
Joe Perches756a0252016-03-17 14:19:47 -07006641 ", %luK highmem"
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006642#endif
Joe Perches756a0252016-03-17 14:19:47 -07006643 "%s%s)\n",
6644 nr_free_pages() << (PAGE_SHIFT - 10),
6645 physpages << (PAGE_SHIFT - 10),
6646 codesize >> 10, datasize >> 10, rosize >> 10,
6647 (init_data_size + init_code_size) >> 10, bss_size >> 10,
6648 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10),
6649 totalcma_pages << (PAGE_SHIFT - 10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006650#ifdef CONFIG_HIGHMEM
Joe Perches756a0252016-03-17 14:19:47 -07006651 totalhigh_pages << (PAGE_SHIFT - 10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006652#endif
Joe Perches756a0252016-03-17 14:19:47 -07006653 str ? ", " : "", str ? str : "");
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07006654}
6655
Mel Gorman0e0b8642006-09-27 01:49:56 -07006656/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006657 * set_dma_reserve - set the specified number of pages reserved in the first zone
6658 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07006659 *
Yaowei Bai013110a2015-09-08 15:04:10 -07006660 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
Mel Gorman0e0b8642006-09-27 01:49:56 -07006661 * In the DMA zone, a significant percentage may be consumed by kernel image
6662 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07006663 * function may optionally be used to account for unfreeable pages in the
6664 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6665 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07006666 */
6667void __init set_dma_reserve(unsigned long new_dma_reserve)
6668{
6669 dma_reserve = new_dma_reserve;
6670}
6671
Linus Torvalds1da177e2005-04-16 15:20:36 -07006672void __init free_area_init(unsigned long *zones_size)
6673{
Johannes Weiner9109fb72008-07-23 21:27:20 -07006674 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006675 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
6676}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006677
Linus Torvalds1da177e2005-04-16 15:20:36 -07006678static int page_alloc_cpu_notify(struct notifier_block *self,
6679 unsigned long action, void *hcpu)
6680{
6681 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006682
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07006683 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07006684 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006685 drain_pages(cpu);
6686
6687 /*
6688 * Spill the event counters of the dead processor
6689 * into the current processors event counters.
6690 * This artificially elevates the count of the current
6691 * processor.
6692 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07006693 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08006694
6695 /*
6696 * Zero the differential counters of the dead processor
6697 * so that the vm statistics are consistent.
6698 *
6699 * This is only okay since the processor is dead and cannot
6700 * race with what we are doing.
6701 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07006702 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006703 }
6704 return NOTIFY_OK;
6705}
Linus Torvalds1da177e2005-04-16 15:20:36 -07006706
6707void __init page_alloc_init(void)
6708{
6709 hotcpu_notifier(page_alloc_cpu_notify, 0);
6710}
6711
6712/*
Yaowei Bai34b10062015-09-08 15:04:13 -07006713 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006714 * or min_free_kbytes changes.
6715 */
6716static void calculate_totalreserve_pages(void)
6717{
6718 struct pglist_data *pgdat;
6719 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006720 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006721
6722 for_each_online_pgdat(pgdat) {
6723 for (i = 0; i < MAX_NR_ZONES; i++) {
6724 struct zone *zone = pgdat->node_zones + i;
Mel Gorman3484b2d2014-08-06 16:07:14 -07006725 long max = 0;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006726
6727 /* Find valid and maximum lowmem_reserve in the zone */
6728 for (j = i; j < MAX_NR_ZONES; j++) {
6729 if (zone->lowmem_reserve[j] > max)
6730 max = zone->lowmem_reserve[j];
6731 }
6732
Mel Gorman41858962009-06-16 15:32:12 -07006733 /* we treat the high watermark as reserved pages. */
6734 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006735
Jiang Liub40da042013-02-22 16:33:52 -08006736 if (max > zone->managed_pages)
6737 max = zone->managed_pages;
Johannes Weinera8d01432016-01-14 15:20:15 -08006738
6739 zone->totalreserve_pages = max;
6740
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006741 reserve_pages += max;
6742 }
6743 }
6744 totalreserve_pages = reserve_pages;
6745}
6746
6747/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07006748 * setup_per_zone_lowmem_reserve - called whenever
Yaowei Bai34b10062015-09-08 15:04:13 -07006749 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
Linus Torvalds1da177e2005-04-16 15:20:36 -07006750 * has a correct pages reserved value, so an adequate number of
6751 * pages are left in the zone after a successful __alloc_pages().
6752 */
6753static void setup_per_zone_lowmem_reserve(void)
6754{
6755 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006756 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006757
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08006758 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006759 for (j = 0; j < MAX_NR_ZONES; j++) {
6760 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08006761 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006762
6763 zone->lowmem_reserve[j] = 0;
6764
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006765 idx = j;
6766 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006767 struct zone *lower_zone;
6768
Christoph Lameter2f6726e2006-09-25 23:31:18 -07006769 idx--;
6770
Linus Torvalds1da177e2005-04-16 15:20:36 -07006771 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6772 sysctl_lowmem_reserve_ratio[idx] = 1;
6773
6774 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08006775 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07006776 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08006777 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006778 }
6779 }
6780 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006781
6782 /* update totalreserve_pages */
6783 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006784}
6785
Mel Gormancfd3da12011-04-25 21:36:42 +00006786static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006787{
6788 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6789 unsigned long lowmem_pages = 0;
6790 struct zone *zone;
6791 unsigned long flags;
6792
6793 /* Calculate total number of !ZONE_HIGHMEM pages */
6794 for_each_zone(zone) {
6795 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08006796 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006797 }
6798
6799 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07006800 u64 tmp;
6801
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006802 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08006803 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07006804 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006805 if (is_highmem(zone)) {
6806 /*
Nick Piggin669ed172005-11-13 16:06:45 -08006807 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6808 * need highmem pages, so cap pages_min to a small
6809 * value here.
6810 *
Mel Gorman41858962009-06-16 15:32:12 -07006811 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Yaowei Bai42ff2702015-04-14 15:47:14 -07006812 * deltas control asynch page reclaim, and so should
Nick Piggin669ed172005-11-13 16:06:45 -08006813 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006814 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08006815 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006816
Jiang Liub40da042013-02-22 16:33:52 -08006817 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08006818 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07006819 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006820 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08006821 /*
6822 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07006823 * proportionate to the zone's size.
6824 */
Mel Gorman41858962009-06-16 15:32:12 -07006825 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006826 }
6827
Johannes Weiner795ae7a2016-03-17 14:19:14 -07006828 /*
6829 * Set the kswapd watermarks distance according to the
6830 * scale factor in proportion to available memory, but
6831 * ensure a minimum size on small systems.
6832 */
6833 tmp = max_t(u64, tmp >> 2,
6834 mult_frac(zone->managed_pages,
6835 watermark_scale_factor, 10000));
6836
6837 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp;
6838 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006839
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006840 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
Johannes Weinerabe5f972014-10-02 16:21:10 -07006841 high_wmark_pages(zone) - low_wmark_pages(zone) -
6842 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006843
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006844 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006845 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006846
6847 /* update totalreserve_pages */
6848 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006849}
6850
Mel Gormancfd3da12011-04-25 21:36:42 +00006851/**
6852 * setup_per_zone_wmarks - called when min_free_kbytes changes
6853 * or when memory is hot-{added|removed}
6854 *
6855 * Ensures that the watermark[min,low,high] values for each zone are set
6856 * correctly with respect to min_free_kbytes.
6857 */
6858void setup_per_zone_wmarks(void)
6859{
6860 mutex_lock(&zonelists_mutex);
6861 __setup_per_zone_wmarks();
6862 mutex_unlock(&zonelists_mutex);
6863}
6864
Randy Dunlap55a44622009-09-21 17:01:20 -07006865/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07006866 * Initialise min_free_kbytes.
6867 *
6868 * For small machines we want it small (128k min). For large machines
6869 * we want it large (64MB max). But it is not linear, because network
6870 * bandwidth does not increase linearly with machine size. We use
6871 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006872 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07006873 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6874 *
6875 * which yields
6876 *
6877 * 16MB: 512k
6878 * 32MB: 724k
6879 * 64MB: 1024k
6880 * 128MB: 1448k
6881 * 256MB: 2048k
6882 * 512MB: 2896k
6883 * 1024MB: 4096k
6884 * 2048MB: 5792k
6885 * 4096MB: 8192k
6886 * 8192MB: 11584k
6887 * 16384MB: 16384k
6888 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07006889int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006890{
6891 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07006892 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006893
6894 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07006895 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006896
Michal Hocko5f127332013-07-08 16:00:40 -07006897 if (new_min_free_kbytes > user_min_free_kbytes) {
6898 min_free_kbytes = new_min_free_kbytes;
6899 if (min_free_kbytes < 128)
6900 min_free_kbytes = 128;
6901 if (min_free_kbytes > 65536)
6902 min_free_kbytes = 65536;
6903 } else {
6904 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6905 new_min_free_kbytes, user_min_free_kbytes);
6906 }
Minchan Kimbc75d332009-06-16 15:32:48 -07006907 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07006908 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006909 setup_per_zone_lowmem_reserve();
6910 return 0;
6911}
Jason Baronbc22af72016-05-05 16:22:12 -07006912core_initcall(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006913
6914/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07006915 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07006916 * that we can call two helper functions whenever min_free_kbytes
6917 * changes.
6918 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006919int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006920 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006921{
Han Pingtianda8c7572014-01-23 15:53:17 -08006922 int rc;
6923
6924 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6925 if (rc)
6926 return rc;
6927
Michal Hocko5f127332013-07-08 16:00:40 -07006928 if (write) {
6929 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07006930 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07006931 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006932 return 0;
6933}
6934
Johannes Weiner795ae7a2016-03-17 14:19:14 -07006935int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
6936 void __user *buffer, size_t *length, loff_t *ppos)
6937{
6938 int rc;
6939
6940 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6941 if (rc)
6942 return rc;
6943
6944 if (write)
6945 setup_per_zone_wmarks();
6946
6947 return 0;
6948}
6949
Christoph Lameter96146342006-07-03 00:24:13 -07006950#ifdef CONFIG_NUMA
Joe Perchescccad5b2014-06-06 14:38:09 -07006951int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006952 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07006953{
6954 struct zone *zone;
6955 int rc;
6956
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006957 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07006958 if (rc)
6959 return rc;
6960
6961 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08006962 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07006963 sysctl_min_unmapped_ratio) / 100;
6964 return 0;
6965}
Christoph Lameter0ff38492006-09-25 23:31:52 -07006966
Joe Perchescccad5b2014-06-06 14:38:09 -07006967int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006968 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07006969{
6970 struct zone *zone;
6971 int rc;
6972
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006973 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07006974 if (rc)
6975 return rc;
6976
6977 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08006978 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07006979 sysctl_min_slab_ratio) / 100;
6980 return 0;
6981}
Christoph Lameter96146342006-07-03 00:24:13 -07006982#endif
6983
Linus Torvalds1da177e2005-04-16 15:20:36 -07006984/*
6985 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6986 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6987 * whenever sysctl_lowmem_reserve_ratio changes.
6988 *
6989 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07006990 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07006991 * if in function of the boot time zone sizes.
6992 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006993int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006994 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006995{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006996 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006997 setup_per_zone_lowmem_reserve();
6998 return 0;
6999}
7000
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08007001/*
7002 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07007003 * cpu. It is the fraction of total pages in each zone that a hot per cpu
7004 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08007005 */
Joe Perchescccad5b2014-06-06 14:38:09 -07007006int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07007007 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08007008{
7009 struct zone *zone;
David Rientjes7cd2b0a2014-06-23 13:22:04 -07007010 int old_percpu_pagelist_fraction;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08007011 int ret;
7012
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007013 mutex_lock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07007014 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
7015
7016 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
7017 if (!write || ret < 0)
7018 goto out;
7019
7020 /* Sanity checking to avoid pcp imbalance */
7021 if (percpu_pagelist_fraction &&
7022 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
7023 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
7024 ret = -EINVAL;
7025 goto out;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08007026 }
David Rientjes7cd2b0a2014-06-23 13:22:04 -07007027
7028 /* No change? */
7029 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
7030 goto out;
7031
7032 for_each_populated_zone(zone) {
7033 unsigned int cpu;
7034
7035 for_each_possible_cpu(cpu)
7036 pageset_set_high_and_batch(zone,
7037 per_cpu_ptr(zone->pageset, cpu));
7038 }
7039out:
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007040 mutex_unlock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07007041 return ret;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08007042}
7043
Rasmus Villemoesa9919c72015-06-24 16:56:28 -07007044#ifdef CONFIG_NUMA
David S. Millerf034b5d2006-08-24 03:08:07 -07007045int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007046
Linus Torvalds1da177e2005-04-16 15:20:36 -07007047static int __init set_hashdist(char *str)
7048{
7049 if (!str)
7050 return 0;
7051 hashdist = simple_strtoul(str, &str, 0);
7052 return 1;
7053}
7054__setup("hashdist=", set_hashdist);
7055#endif
7056
7057/*
7058 * allocate a large system hash table from bootmem
7059 * - it is assumed that the hash table must contain an exact power-of-2
7060 * quantity of entries
7061 * - limit is the number of hash buckets, not the total allocation size
7062 */
7063void *__init alloc_large_system_hash(const char *tablename,
7064 unsigned long bucketsize,
7065 unsigned long numentries,
7066 int scale,
7067 int flags,
7068 unsigned int *_hash_shift,
7069 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00007070 unsigned long low_limit,
7071 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007072{
Tim Bird31fe62b2012-05-23 13:33:35 +00007073 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007074 unsigned long log2qty, size;
7075 void *table = NULL;
7076
7077 /* allow the kernel cmdline to have a say */
7078 if (!numentries) {
7079 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08007080 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07007081
7082 /* It isn't necessary when PAGE_SIZE >= 1MB */
7083 if (PAGE_SHIFT < 20)
7084 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007085
7086 /* limit to 1 bucket per 2^scale bytes of low memory */
7087 if (scale > PAGE_SHIFT)
7088 numentries >>= (scale - PAGE_SHIFT);
7089 else
7090 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08007091
7092 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07007093 if (unlikely(flags & HASH_SMALL)) {
7094 /* Makes no sense without HASH_EARLY */
7095 WARN_ON(!(flags & HASH_EARLY));
7096 if (!(numentries >> *_hash_shift)) {
7097 numentries = 1UL << *_hash_shift;
7098 BUG_ON(!numentries);
7099 }
7100 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08007101 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007102 }
John Hawkes6e692ed2006-03-25 03:08:02 -08007103 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007104
7105 /* limit allocation size to 1/16 total memory by default */
7106 if (max == 0) {
7107 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
7108 do_div(max, bucketsize);
7109 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08007110 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007111
Tim Bird31fe62b2012-05-23 13:33:35 +00007112 if (numentries < low_limit)
7113 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007114 if (numentries > max)
7115 numentries = max;
7116
David Howellsf0d1b0b2006-12-08 02:37:49 -08007117 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007118
7119 do {
7120 size = bucketsize << log2qty;
7121 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08007122 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007123 else if (hashdist)
7124 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
7125 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07007126 /*
7127 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07007128 * some pages at the end of hash table which
7129 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07007130 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01007131 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07007132 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01007133 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
7134 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007135 }
7136 } while (!table && size > PAGE_SIZE && --log2qty);
7137
7138 if (!table)
7139 panic("Failed to allocate %s hash table\n", tablename);
7140
Joe Perches11705322016-03-17 14:19:50 -07007141 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7142 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007143
7144 if (_hash_shift)
7145 *_hash_shift = log2qty;
7146 if (_hash_mask)
7147 *_hash_mask = (1 << log2qty) - 1;
7148
7149 return table;
7150}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08007151
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07007152/*
Minchan Kim80934512012-07-31 16:43:01 -07007153 * This function checks whether pageblock includes unmovable pages or not.
7154 * If @count is not zero, it is okay to include less @count unmovable pages
7155 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07007156 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07007157 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
7158 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07007159 */
Wen Congyangb023f462012-12-11 16:00:45 -08007160bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
7161 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007162{
7163 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01007164 int mt;
7165
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007166 /*
7167 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07007168 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007169 */
7170 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07007171 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01007172 mt = get_pageblock_migratetype(page);
7173 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07007174 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007175
7176 pfn = page_to_pfn(page);
7177 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
7178 unsigned long check = pfn + iter;
7179
Namhyung Kim29723fc2011-02-25 14:44:25 -08007180 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007181 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08007182
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007183 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07007184
7185 /*
7186 * Hugepages are not in LRU lists, but they're movable.
7187 * We need not scan over tail pages bacause we don't
7188 * handle each tail page individually in migration.
7189 */
7190 if (PageHuge(page)) {
7191 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
7192 continue;
7193 }
7194
Minchan Kim97d255c2012-07-31 16:42:59 -07007195 /*
7196 * We can't use page_count without pin a page
7197 * because another CPU can free compound page.
7198 * This check already skips compound tails of THP
Joonsoo Kim0139aa72016-05-19 17:10:49 -07007199 * because their page->_refcount is zero at all time.
Minchan Kim97d255c2012-07-31 16:42:59 -07007200 */
Joonsoo Kimfe896d12016-03-17 14:19:26 -07007201 if (!page_ref_count(page)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007202 if (PageBuddy(page))
7203 iter += (1 << page_order(page)) - 1;
7204 continue;
7205 }
Minchan Kim97d255c2012-07-31 16:42:59 -07007206
Wen Congyangb023f462012-12-11 16:00:45 -08007207 /*
7208 * The HWPoisoned page may be not in buddy system, and
7209 * page_count() is not 0.
7210 */
7211 if (skip_hwpoisoned_pages && PageHWPoison(page))
7212 continue;
7213
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007214 if (!PageLRU(page))
7215 found++;
7216 /*
Johannes Weiner6b4f7792014-12-12 16:56:13 -08007217 * If there are RECLAIMABLE pages, we need to check
7218 * it. But now, memory offline itself doesn't call
7219 * shrink_node_slabs() and it still to be fixed.
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007220 */
7221 /*
7222 * If the page is not RAM, page_count()should be 0.
7223 * we don't need more check. This is an _used_ not-movable page.
7224 *
7225 * The problematic thing here is PG_reserved pages. PG_reserved
7226 * is set to both of a memory hole page and a _used_ kernel
7227 * page at boot.
7228 */
7229 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07007230 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007231 }
Minchan Kim80934512012-07-31 16:43:01 -07007232 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07007233}
7234
7235bool is_pageblock_removable_nolock(struct page *page)
7236{
Michal Hocko656a0702012-01-20 14:33:58 -08007237 struct zone *zone;
7238 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08007239
7240 /*
7241 * We have to be careful here because we are iterating over memory
7242 * sections which are not zone aware so we might end up outside of
7243 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08007244 * We have to take care about the node as well. If the node is offline
7245 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08007246 */
Michal Hocko656a0702012-01-20 14:33:58 -08007247 if (!node_online(page_to_nid(page)))
7248 return false;
7249
7250 zone = page_zone(page);
7251 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08007252 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08007253 return false;
7254
Wen Congyangb023f462012-12-11 16:00:45 -08007255 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b542007-10-16 01:26:11 -07007256}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007257
Vlastimil Babka080fe202016-02-05 15:36:41 -08007258#if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007259
7260static unsigned long pfn_max_align_down(unsigned long pfn)
7261{
7262 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
7263 pageblock_nr_pages) - 1);
7264}
7265
7266static unsigned long pfn_max_align_up(unsigned long pfn)
7267{
7268 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
7269 pageblock_nr_pages));
7270}
7271
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007272/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007273static int __alloc_contig_migrate_range(struct compact_control *cc,
7274 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007275{
7276 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07007277 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007278 unsigned long pfn = start;
7279 unsigned int tries = 0;
7280 int ret = 0;
7281
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08007282 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007283
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007284 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007285 if (fatal_signal_pending(current)) {
7286 ret = -EINTR;
7287 break;
7288 }
7289
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007290 if (list_empty(&cc->migratepages)) {
7291 cc->nr_migratepages = 0;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07007292 pfn = isolate_migratepages_range(cc, pfn, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007293 if (!pfn) {
7294 ret = -EINTR;
7295 break;
7296 }
7297 tries = 0;
7298 } else if (++tries == 5) {
7299 ret = ret < 0 ? ret : -EBUSY;
7300 break;
7301 }
7302
Minchan Kimbeb51ea2012-10-08 16:33:51 -07007303 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
7304 &cc->migratepages);
7305 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07007306
Hugh Dickins9c620e22013-02-22 16:35:14 -08007307 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07007308 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007309 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08007310 if (ret < 0) {
7311 putback_movable_pages(&cc->migratepages);
7312 return ret;
7313 }
7314 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007315}
7316
7317/**
7318 * alloc_contig_range() -- tries to allocate given range of pages
7319 * @start: start PFN to allocate
7320 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02007321 * @migratetype: migratetype of the underlaying pageblocks (either
7322 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
7323 * in range must have the same migratetype and it must
7324 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007325 *
7326 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7327 * aligned, however it's the caller's responsibility to guarantee that
7328 * we are the only thread that changes migrate type of pageblocks the
7329 * pages fall in.
7330 *
7331 * The PFN range must belong to a single zone.
7332 *
7333 * Returns zero on success or negative error code. On success all
7334 * pages which PFN is in [start, end) are allocated for the caller and
7335 * need to be freed with free_contig_range().
7336 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02007337int alloc_contig_range(unsigned long start, unsigned long end,
7338 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007339{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007340 unsigned long outer_start, outer_end;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08007341 unsigned int order;
7342 int ret = 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007343
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007344 struct compact_control cc = {
7345 .nr_migratepages = 0,
7346 .order = -1,
7347 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07007348 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007349 .ignore_skip_hint = true,
7350 };
7351 INIT_LIST_HEAD(&cc.migratepages);
7352
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007353 /*
7354 * What we do here is we mark all pageblocks in range as
7355 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7356 * have different sizes, and due to the way page allocator
7357 * work, we align the range to biggest of the two pages so
7358 * that page allocator won't try to merge buddies from
7359 * different pageblocks and change MIGRATE_ISOLATE to some
7360 * other migration type.
7361 *
7362 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7363 * migrate the pages from an unaligned range (ie. pages that
7364 * we are interested in). This will put all the pages in
7365 * range back to page allocator as MIGRATE_ISOLATE.
7366 *
7367 * When this is done, we take the pages in range from page
7368 * allocator removing them from the buddy system. This way
7369 * page allocator will never consider using them.
7370 *
7371 * This lets us mark the pageblocks back as
7372 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7373 * aligned range but not in the unaligned, original range are
7374 * put back to page allocator so that buddy can use them.
7375 */
7376
7377 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08007378 pfn_max_align_up(end), migratetype,
7379 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007380 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07007381 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007382
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007383 /*
7384 * In case of -EBUSY, we'd like to know which page causes problem.
7385 * So, just fall through. We will check it in test_pages_isolated().
7386 */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007387 ret = __alloc_contig_migrate_range(&cc, start, end);
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007388 if (ret && ret != -EBUSY)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007389 goto done;
7390
7391 /*
7392 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7393 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7394 * more, all pages in [start, end) are free in page allocator.
7395 * What we are going to do is to allocate all pages from
7396 * [start, end) (that is remove them from page allocator).
7397 *
7398 * The only problem is that pages at the beginning and at the
7399 * end of interesting range may be not aligned with pages that
7400 * page allocator holds, ie. they can be part of higher order
7401 * pages. Because of this, we reserve the bigger range and
7402 * once this is done free the pages we are not interested in.
7403 *
7404 * We don't have to hold zone->lock here because the pages are
7405 * isolated thus they won't get removed from buddy.
7406 */
7407
7408 lru_add_drain_all();
Vlastimil Babka510f5502014-12-10 15:43:07 -08007409 drain_all_pages(cc.zone);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007410
7411 order = 0;
7412 outer_start = start;
7413 while (!PageBuddy(pfn_to_page(outer_start))) {
7414 if (++order >= MAX_ORDER) {
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007415 outer_start = start;
7416 break;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007417 }
7418 outer_start &= ~0UL << order;
7419 }
7420
Joonsoo Kim8ef58492016-01-14 15:18:45 -08007421 if (outer_start != start) {
7422 order = page_order(pfn_to_page(outer_start));
7423
7424 /*
7425 * outer_start page could be small order buddy page and
7426 * it doesn't include start page. Adjust outer_start
7427 * in this case to report failed page properly
7428 * on tracepoint in test_pages_isolated()
7429 */
7430 if (outer_start + (1UL << order) <= start)
7431 outer_start = start;
7432 }
7433
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007434 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08007435 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewiczdae803e2014-11-13 15:19:27 -08007436 pr_info("%s: [%lx, %lx) PFNs busy\n",
7437 __func__, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007438 ret = -EBUSY;
7439 goto done;
7440 }
7441
Marek Szyprowski49f223a2012-01-25 12:49:24 +01007442 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07007443 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007444 if (!outer_end) {
7445 ret = -EBUSY;
7446 goto done;
7447 }
7448
7449 /* Free head and tail (if any) */
7450 if (start != outer_start)
7451 free_contig_range(outer_start, start - outer_start);
7452 if (end != outer_end)
7453 free_contig_range(end, outer_end - end);
7454
7455done:
7456 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02007457 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007458 return ret;
7459}
7460
7461void free_contig_range(unsigned long pfn, unsigned nr_pages)
7462{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08007463 unsigned int count = 0;
7464
7465 for (; nr_pages--; pfn++) {
7466 struct page *page = pfn_to_page(pfn);
7467
7468 count += page_count(page) != 1;
7469 __free_page(page);
7470 }
7471 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01007472}
7473#endif
7474
Jiang Liu4ed7e022012-07-31 16:43:35 -07007475#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07007476/*
7477 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7478 * page high values need to be recalulated.
7479 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07007480void __meminit zone_pcp_update(struct zone *zone)
7481{
Cody P Schafer0a647f32013-07-03 15:01:33 -07007482 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007483 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07007484 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07007485 pageset_set_high_and_batch(zone,
7486 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07007487 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07007488}
7489#endif
7490
Jiang Liu340175b2012-07-31 16:43:32 -07007491void zone_pcp_reset(struct zone *zone)
7492{
7493 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07007494 int cpu;
7495 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07007496
7497 /* avoid races with drain_pages() */
7498 local_irq_save(flags);
7499 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07007500 for_each_online_cpu(cpu) {
7501 pset = per_cpu_ptr(zone->pageset, cpu);
7502 drain_zonestat(zone, pset);
7503 }
Jiang Liu340175b2012-07-31 16:43:32 -07007504 free_percpu(zone->pageset);
7505 zone->pageset = &boot_pageset;
7506 }
7507 local_irq_restore(flags);
7508}
7509
Wen Congyang6dcd73d2012-12-11 16:01:01 -08007510#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007511/*
Joonsoo Kimb9eb6312016-05-19 17:12:06 -07007512 * All pages in the range must be in a single zone and isolated
7513 * before calling this.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007514 */
7515void
7516__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
7517{
7518 struct page *page;
7519 struct zone *zone;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07007520 unsigned int order, i;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007521 unsigned long pfn;
7522 unsigned long flags;
7523 /* find the first valid pfn */
7524 for (pfn = start_pfn; pfn < end_pfn; pfn++)
7525 if (pfn_valid(pfn))
7526 break;
7527 if (pfn == end_pfn)
7528 return;
7529 zone = page_zone(pfn_to_page(pfn));
7530 spin_lock_irqsave(&zone->lock, flags);
7531 pfn = start_pfn;
7532 while (pfn < end_pfn) {
7533 if (!pfn_valid(pfn)) {
7534 pfn++;
7535 continue;
7536 }
7537 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08007538 /*
7539 * The HWPoisoned page may be not in buddy system, and
7540 * page_count() is not 0.
7541 */
7542 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
7543 pfn++;
7544 SetPageReserved(page);
7545 continue;
7546 }
7547
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007548 BUG_ON(page_count(page));
7549 BUG_ON(!PageBuddy(page));
7550 order = page_order(page);
7551#ifdef CONFIG_DEBUG_VM
Joe Perches11705322016-03-17 14:19:50 -07007552 pr_info("remove from free list %lx %d %lx\n",
7553 pfn, 1 << order, end_pfn);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007554#endif
7555 list_del(&page->lru);
7556 rmv_page_order(page);
7557 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07007558 for (i = 0; i < (1 << order); i++)
7559 SetPageReserved((page+i));
7560 pfn += (1 << order);
7561 }
7562 spin_unlock_irqrestore(&zone->lock, flags);
7563}
7564#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007565
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007566bool is_free_buddy_page(struct page *page)
7567{
7568 struct zone *zone = page_zone(page);
7569 unsigned long pfn = page_to_pfn(page);
7570 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07007571 unsigned int order;
Wu Fengguang8d22ba12009-12-16 12:19:58 +01007572
7573 spin_lock_irqsave(&zone->lock, flags);
7574 for (order = 0; order < MAX_ORDER; order++) {
7575 struct page *page_head = page - (pfn & ((1 << order) - 1));
7576
7577 if (PageBuddy(page_head) && page_order(page_head) >= order)
7578 break;
7579 }
7580 spin_unlock_irqrestore(&zone->lock, flags);
7581
7582 return order < MAX_ORDER;
7583}