blob: 17a3c66639a9cd6625f333b4f64ab012e2cb56c4 [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>
Yasunori Goto68113782006-06-23 02:03:11 -070046#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070047#include <linux/sort.h>
48#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070049#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080050#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070051#include <linux/page-isolation.h>
Joonsoo Kimeefa8642014-12-12 16:55:46 -080052#include <linux/page_ext.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070053#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010054#include <linux/kmemleak.h>
Mel Gorman56de7262010-05-24 14:32:30 -070055#include <linux/compaction.h>
Mel Gorman0d3d0622009-09-21 17:02:44 -070056#include <trace/events/kmem.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070057#include <linux/prefetch.h>
Lisa Du6e543d52013-09-11 14:22:36 -070058#include <linux/mm_inline.h>
Michal Nazarewicz041d3a82011-12-29 13:09:50 +010059#include <linux/migrate.h>
Joonsoo Kime30825f2014-12-12 16:55:49 -080060#include <linux/page_ext.h>
David Rientjes949f7ec2013-04-29 15:07:48 -070061#include <linux/hugetlb.h>
Clark Williams8bd75c72013-02-07 09:47:07 -060062#include <linux/sched/rt.h>
Joonsoo Kim48c96a32014-12-12 16:56:01 -080063#include <linux/page_owner.h>
Mel Gorman0e1cc952015-06-30 14:57:27 -070064#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Jiang Liu7ee3d4e2013-07-03 15:03:41 -070066#include <asm/sections.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070068#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070069#include "internal.h"
70
Cody P Schaferc8e251f2013-07-03 15:01:29 -070071/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
72static DEFINE_MUTEX(pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -070073#define MIN_PERCPU_PAGELIST_FRACTION (8)
Cody P Schaferc8e251f2013-07-03 15:01:29 -070074
Lee Schermerhorn72812012010-05-26 14:44:56 -070075#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
76DEFINE_PER_CPU(int, numa_node);
77EXPORT_PER_CPU_SYMBOL(numa_node);
78#endif
79
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070080#ifdef CONFIG_HAVE_MEMORYLESS_NODES
81/*
82 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
83 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
84 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
85 * defined in <linux/topology.h>.
86 */
87DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
88EXPORT_PER_CPU_SYMBOL(_numa_mem_);
Joonsoo Kimad2c8142014-10-09 15:26:13 -070089int _node_numa_mem_[MAX_NUMNODES];
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070090#endif
91
Linus Torvalds1da177e2005-04-16 15:20:36 -070092/*
Christoph Lameter13808912007-10-16 01:25:27 -070093 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 */
Christoph Lameter13808912007-10-16 01:25:27 -070095nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
96 [N_POSSIBLE] = NODE_MASK_ALL,
97 [N_ONLINE] = { { [0] = 1UL } },
98#ifndef CONFIG_NUMA
99 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
100#ifdef CONFIG_HIGHMEM
101 [N_HIGH_MEMORY] = { { [0] = 1UL } },
102#endif
Lai Jiangshan20b2f522012-12-12 13:52:00 -0800103#ifdef CONFIG_MOVABLE_NODE
104 [N_MEMORY] = { { [0] = 1UL } },
105#endif
Christoph Lameter13808912007-10-16 01:25:27 -0700106 [N_CPU] = { { [0] = 1UL } },
107#endif /* NUMA */
108};
109EXPORT_SYMBOL(node_states);
110
Jiang Liuc3d5f5f2013-07-03 15:03:14 -0700111/* Protect totalram_pages and zone->managed_pages */
112static DEFINE_SPINLOCK(managed_page_count_lock);
113
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700114unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -0700115unsigned long totalreserve_pages __read_mostly;
Pintu Kumare48322a2014-12-18 16:17:15 -0800116unsigned long totalcma_pages __read_mostly;
Johannes Weinerab8fabd2012-01-10 15:07:42 -0800117/*
118 * When calculating the number of globally allowed dirty pages, there
119 * is a certain number of per-zone reserves that should not be
120 * considered dirtyable memory. This is the sum of those reserves
121 * over all existing zones that contribute dirtyable memory.
122 */
123unsigned long dirty_balance_reserve __read_mostly;
124
Hugh Dickins1b76b022012-05-11 01:00:07 -0700125int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000126gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -0700128/*
129 * A cached value of the page's pageblock's migratetype, used when the page is
130 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
131 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
132 * Also the migratetype set in the page does not necessarily match the pcplist
133 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
134 * other index - this ensures that it will be put on the correct CMA freelist.
135 */
136static inline int get_pcppage_migratetype(struct page *page)
137{
138 return page->index;
139}
140
141static inline void set_pcppage_migratetype(struct page *page, int migratetype)
142{
143 page->index = migratetype;
144}
145
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800146#ifdef CONFIG_PM_SLEEP
147/*
148 * The following functions are used by the suspend/hibernate code to temporarily
149 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
150 * while devices are suspended. To avoid races with the suspend/hibernate code,
151 * they should always be called with pm_mutex held (gfp_allowed_mask also should
152 * only be modified with pm_mutex held, unless the suspend/hibernate code is
153 * guaranteed not to run in parallel with that modification).
154 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100155
156static gfp_t saved_gfp_mask;
157
158void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800159{
160 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100161 if (saved_gfp_mask) {
162 gfp_allowed_mask = saved_gfp_mask;
163 saved_gfp_mask = 0;
164 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800165}
166
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100167void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800168{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800169 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100170 WARN_ON(saved_gfp_mask);
171 saved_gfp_mask = gfp_allowed_mask;
Mel Gormand0164ad2015-11-06 16:28:21 -0800172 gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800173}
Mel Gormanf90ac392012-01-10 15:07:15 -0800174
175bool pm_suspended_storage(void)
176{
Mel Gormand0164ad2015-11-06 16:28:21 -0800177 if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
Mel Gormanf90ac392012-01-10 15:07:15 -0800178 return false;
179 return true;
180}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800181#endif /* CONFIG_PM_SLEEP */
182
Mel Gormand9c23402007-10-16 01:26:01 -0700183#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -0800184unsigned int pageblock_order __read_mostly;
Mel Gormand9c23402007-10-16 01:26:01 -0700185#endif
186
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800187static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189/*
190 * results with 256, 32 in the lowmem_reserve sysctl:
191 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
192 * 1G machine -> (16M dma, 784M normal, 224M high)
193 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
194 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
Yaowei Bai84109e12015-02-12 15:00:22 -0800195 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100196 *
197 * TBD: should special case ZONE_DMA32 machines here - in those we normally
198 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700200int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800201#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700202 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800203#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700204#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700205 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700206#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700207#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700208 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700209#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700210 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700211};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
213EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214
Helge Deller15ad7cd2006-12-06 20:40:36 -0800215static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800216#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700217 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800218#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700219#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700220 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700221#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700222 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700223#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700224 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700225#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700226 "Movable",
Dan Williams033fbae2015-08-09 15:29:06 -0400227#ifdef CONFIG_ZONE_DEVICE
228 "Device",
229#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700230};
231
Kirill A. Shutemovf1e61552015-11-06 16:29:50 -0800232static void free_compound_page(struct page *page);
233compound_page_dtor * const compound_page_dtors[] = {
234 NULL,
235 free_compound_page,
236#ifdef CONFIG_HUGETLB_PAGE
237 free_huge_page,
238#endif
239};
240
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241int min_free_kbytes = 1024;
Han Pingtian42aa83c2014-01-23 15:53:28 -0800242int user_min_free_kbytes = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Jan Beulich2c85f512009-09-21 17:03:07 -0700244static unsigned long __meminitdata nr_kernel_pages;
245static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700246static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
Tejun Heo0ee332c2011-12-08 10:22:09 -0800248#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
249static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
250static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
251static unsigned long __initdata required_kernelcore;
252static unsigned long __initdata required_movablecore;
253static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gormanc7132162006-09-27 01:49:43 -0700254
Tejun Heo0ee332c2011-12-08 10:22:09 -0800255/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
256int movable_zone;
257EXPORT_SYMBOL(movable_zone);
258#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700259
Miklos Szeredi418508c2007-05-23 13:57:55 -0700260#if MAX_NUMNODES > 1
261int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700262int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700263EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700264EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700265#endif
266
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700267int page_group_by_mobility_disabled __read_mostly;
268
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700269#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
270static inline void reset_deferred_meminit(pg_data_t *pgdat)
271{
272 pgdat->first_deferred_pfn = ULONG_MAX;
273}
274
275/* Returns true if the struct page for the pfn is uninitialised */
Mel Gorman0e1cc952015-06-30 14:57:27 -0700276static inline bool __meminit early_page_uninitialised(unsigned long pfn)
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700277{
Mel Gormanae026b22015-07-17 16:23:48 -0700278 if (pfn >= NODE_DATA(early_pfn_to_nid(pfn))->first_deferred_pfn)
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700279 return true;
280
281 return false;
282}
283
Mel Gorman7e18adb2015-06-30 14:57:05 -0700284static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
285{
286 if (pfn >= NODE_DATA(nid)->first_deferred_pfn)
287 return true;
288
289 return false;
290}
291
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700292/*
293 * Returns false when the remaining initialisation should be deferred until
294 * later in the boot cycle when it can be parallelised.
295 */
296static inline bool update_defer_init(pg_data_t *pgdat,
297 unsigned long pfn, unsigned long zone_end,
298 unsigned long *nr_initialised)
299{
300 /* Always populate low zones for address-contrained allocations */
301 if (zone_end < pgdat_end_pfn(pgdat))
302 return true;
303
304 /* Initialise at least 2G of the highest zone */
305 (*nr_initialised)++;
306 if (*nr_initialised > (2UL << (30 - PAGE_SHIFT)) &&
307 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
308 pgdat->first_deferred_pfn = pfn;
309 return false;
310 }
311
312 return true;
313}
314#else
315static inline void reset_deferred_meminit(pg_data_t *pgdat)
316{
317}
318
319static inline bool early_page_uninitialised(unsigned long pfn)
320{
321 return false;
322}
323
Mel Gorman7e18adb2015-06-30 14:57:05 -0700324static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
325{
326 return false;
327}
328
Mel Gorman3a80a7f2015-06-30 14:57:02 -0700329static inline bool update_defer_init(pg_data_t *pgdat,
330 unsigned long pfn, unsigned long zone_end,
331 unsigned long *nr_initialised)
332{
333 return true;
334}
335#endif
336
337
Minchan Kimee6f5092012-07-31 16:43:50 -0700338void set_pageblock_migratetype(struct page *page, int migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700339{
KOSAKI Motohiro5d0f3f72013-11-12 15:08:18 -0800340 if (unlikely(page_group_by_mobility_disabled &&
341 migratetype < MIGRATE_PCPTYPES))
Mel Gorman49255c62009-06-16 15:31:58 -0700342 migratetype = MIGRATE_UNMOVABLE;
343
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700344 set_pageblock_flags_group(page, (unsigned long)migratetype,
345 PB_migrate, PB_migrate_end);
346}
347
Nick Piggin13e74442006-01-06 00:10:58 -0800348#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700349static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700351 int ret = 0;
352 unsigned seq;
353 unsigned long pfn = page_to_pfn(page);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800354 unsigned long sp, start_pfn;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700355
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700356 do {
357 seq = zone_span_seqbegin(zone);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800358 start_pfn = zone->zone_start_pfn;
359 sp = zone->spanned_pages;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800360 if (!zone_spans_pfn(zone, pfn))
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700361 ret = 1;
362 } while (zone_span_seqretry(zone, seq));
363
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800364 if (ret)
Dave Hansen613813e2014-06-04 16:07:27 -0700365 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
366 pfn, zone_to_nid(zone), zone->name,
367 start_pfn, start_pfn + sp);
Cody P Schaferb5e6a5a2013-02-22 16:35:28 -0800368
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700369 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700370}
371
372static int page_is_consistent(struct zone *zone, struct page *page)
373{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700374 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700375 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700377 return 0;
378
379 return 1;
380}
381/*
382 * Temporary debugging check for pages not lying within a given zone.
383 */
384static int bad_range(struct zone *zone, struct page *page)
385{
386 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700388 if (!page_is_consistent(zone, page))
389 return 1;
390
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 return 0;
392}
Nick Piggin13e74442006-01-06 00:10:58 -0800393#else
394static inline int bad_range(struct zone *zone, struct page *page)
395{
396 return 0;
397}
398#endif
399
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700400static void bad_page(struct page *page, const char *reason,
401 unsigned long bad_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800403 static unsigned long resume;
404 static unsigned long nr_shown;
405 static unsigned long nr_unshown;
406
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200407 /* Don't complain about poisoned pages */
408 if (PageHWPoison(page)) {
Mel Gorman22b751c2013-02-22 16:34:59 -0800409 page_mapcount_reset(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200410 return;
411 }
412
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800413 /*
414 * Allow a burst of 60 reports, then keep quiet for that minute;
415 * or allow a steady drip of one report per second.
416 */
417 if (nr_shown == 60) {
418 if (time_before(jiffies, resume)) {
419 nr_unshown++;
420 goto out;
421 }
422 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800423 printk(KERN_ALERT
424 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800425 nr_unshown);
426 nr_unshown = 0;
427 }
428 nr_shown = 0;
429 }
430 if (nr_shown++ == 0)
431 resume = jiffies + 60 * HZ;
432
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800433 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800434 current->comm, page_to_pfn(page));
Dave Hansenf0b791a2014-01-23 15:52:49 -0800435 dump_page_badflags(page, reason, bad_flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700436
Dave Jones4f318882011-10-31 17:07:24 -0700437 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800439out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800440 /* Leave bad fields for debug, except PageBuddy could make trouble */
Mel Gorman22b751c2013-02-22 16:34:59 -0800441 page_mapcount_reset(page); /* remove PageBuddy */
Rusty Russell373d4d02013-01-21 17:17:39 +1030442 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443}
444
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445/*
446 * Higher-order pages are called "compound pages". They are structured thusly:
447 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800448 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800450 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
451 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800453 * The first tail page's ->compound_dtor holds the offset in array of compound
454 * page destructors. See compound_page_dtors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 *
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800456 * The first tail page's ->compound_order holds the order of allocation.
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800457 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800459
460static void free_compound_page(struct page *page)
461{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700462 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800463}
464
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -0800465void prep_compound_page(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466{
467 int i;
468 int nr_pages = 1 << order;
469
Kirill A. Shutemovf1e61552015-11-06 16:29:50 -0800470 set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700471 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700472 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800473 for (i = 1; i < nr_pages; i++) {
474 struct page *p = page + i;
Youquan Song58a84aa2011-12-08 14:34:18 -0800475 set_page_count(p, 0);
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800476 set_compound_head(p, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 }
478}
479
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800480#ifdef CONFIG_DEBUG_PAGEALLOC
481unsigned int _debug_guardpage_minorder;
Joonsoo Kim031bc572014-12-12 16:55:52 -0800482bool _debug_pagealloc_enabled __read_mostly;
Joonsoo Kime30825f2014-12-12 16:55:49 -0800483bool _debug_guardpage_enabled __read_mostly;
484
Joonsoo Kim031bc572014-12-12 16:55:52 -0800485static int __init early_debug_pagealloc(char *buf)
486{
487 if (!buf)
488 return -EINVAL;
489
490 if (strcmp(buf, "on") == 0)
491 _debug_pagealloc_enabled = true;
492
493 return 0;
494}
495early_param("debug_pagealloc", early_debug_pagealloc);
496
Joonsoo Kime30825f2014-12-12 16:55:49 -0800497static bool need_debug_guardpage(void)
498{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800499 /* If we don't use debug_pagealloc, we don't need guard page */
500 if (!debug_pagealloc_enabled())
501 return false;
502
Joonsoo Kime30825f2014-12-12 16:55:49 -0800503 return true;
504}
505
506static void init_debug_guardpage(void)
507{
Joonsoo Kim031bc572014-12-12 16:55:52 -0800508 if (!debug_pagealloc_enabled())
509 return;
510
Joonsoo Kime30825f2014-12-12 16:55:49 -0800511 _debug_guardpage_enabled = true;
512}
513
514struct page_ext_operations debug_guardpage_ops = {
515 .need = need_debug_guardpage,
516 .init = init_debug_guardpage,
517};
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800518
519static int __init debug_guardpage_minorder_setup(char *buf)
520{
521 unsigned long res;
522
523 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
524 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
525 return 0;
526 }
527 _debug_guardpage_minorder = res;
528 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
529 return 0;
530}
531__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
532
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800533static inline void set_page_guard(struct zone *zone, struct page *page,
534 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800535{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800536 struct page_ext *page_ext;
537
538 if (!debug_guardpage_enabled())
539 return;
540
541 page_ext = lookup_page_ext(page);
542 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
543
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800544 INIT_LIST_HEAD(&page->lru);
545 set_page_private(page, order);
546 /* Guard pages are not available for any usage */
547 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800548}
549
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800550static inline void clear_page_guard(struct zone *zone, struct page *page,
551 unsigned int order, int migratetype)
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800552{
Joonsoo Kime30825f2014-12-12 16:55:49 -0800553 struct page_ext *page_ext;
554
555 if (!debug_guardpage_enabled())
556 return;
557
558 page_ext = lookup_page_ext(page);
559 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
560
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800561 set_page_private(page, 0);
562 if (!is_migrate_isolate(migratetype))
563 __mod_zone_freepage_state(zone, (1 << order), migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800564}
565#else
Joonsoo Kime30825f2014-12-12 16:55:49 -0800566struct page_ext_operations debug_guardpage_ops = { NULL, };
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800567static inline void set_page_guard(struct zone *zone, struct page *page,
568 unsigned int order, int migratetype) {}
569static inline void clear_page_guard(struct zone *zone, struct page *page,
570 unsigned int order, int migratetype) {}
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800571#endif
572
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700573static inline void set_page_order(struct page *page, unsigned int order)
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700574{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700575 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000576 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577}
578
579static inline void rmv_page_order(struct page *page)
580{
Nick Piggin676165a2006-04-10 11:21:48 +1000581 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700582 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583}
584
585/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 * This function checks whether a page is free && is the buddy
587 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800588 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000589 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700590 * (c) a page and its buddy have the same order &&
591 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 *
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700593 * For recording whether a page is in the buddy system, we set ->_mapcount
594 * PAGE_BUDDY_MAPCOUNT_VALUE.
595 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
596 * serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000597 *
598 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700600static inline int page_is_buddy(struct page *page, struct page *buddy,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700601 unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700603 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800604 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800605
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800606 if (page_is_guard(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700607 if (page_zone_id(page) != page_zone_id(buddy))
608 return 0;
609
Weijie Yang4c5018c2015-02-10 14:11:39 -0800610 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
611
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800612 return 1;
613 }
614
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700615 if (PageBuddy(buddy) && page_order(buddy) == order) {
Mel Gormand34c5fa2014-06-04 16:10:10 -0700616 /*
617 * zone check is done late to avoid uselessly
618 * calculating zone/node ids for pages that could
619 * never merge.
620 */
621 if (page_zone_id(page) != page_zone_id(buddy))
622 return 0;
623
Weijie Yang4c5018c2015-02-10 14:11:39 -0800624 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
625
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700626 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000627 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700628 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629}
630
631/*
632 * Freeing function for a buddy system allocator.
633 *
634 * The concept of a buddy system is to maintain direct-mapped table
635 * (containing bit values) for memory blocks of various "orders".
636 * The bottom level table contains the map for the smallest allocatable
637 * units of memory (here, pages), and each level above it describes
638 * pairs of units from the levels below, hence, "buddies".
639 * At a high level, all that happens here is marking the table entry
640 * at the bottom level available, and propagating the changes upward
641 * as necessary, plus some accounting needed to play nicely with other
642 * parts of the VM system.
643 * At each level, we keep a list of pages, which are heads of continuous
Wang Sheng-Huicf6fe942013-09-11 14:22:48 -0700644 * free pages of length of (1 << order) and marked with _mapcount
645 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
646 * field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100648 * other. That is, if we allocate a small block, and both were
649 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 * If a block is freed, and its buddy is also free, then this
Michal Nazarewicz5f63b722012-01-11 15:16:11 +0100651 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +0100653 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 */
655
Nick Piggin48db57f2006-01-08 01:00:42 -0800656static inline void __free_one_page(struct page *page,
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700657 unsigned long pfn,
Mel Gormaned0ae212009-06-16 15:32:07 -0700658 struct zone *zone, unsigned int order,
659 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660{
661 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700662 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800663 unsigned long uninitialized_var(buddy_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700664 struct page *buddy;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -0800665 unsigned int max_order = MAX_ORDER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666
Cody P Schaferd29bb972013-02-22 16:35:25 -0800667 VM_BUG_ON(!zone_is_initialized(zone));
Kirill A. Shutemov6e9f0d52015-02-11 15:25:50 -0800668 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669
Mel Gormaned0ae212009-06-16 15:32:07 -0700670 VM_BUG_ON(migratetype == -1);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800671 if (is_migrate_isolate(migratetype)) {
672 /*
673 * We restrict max order of merging to prevent merge
674 * between freepages on isolate pageblock and normal
675 * pageblock. Without this, pageblock isolation
676 * could cause incorrect freepage accounting.
677 */
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -0800678 max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800679 } else {
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800680 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Joonsoo Kim3c605092014-11-13 15:19:21 -0800681 }
Mel Gormaned0ae212009-06-16 15:32:07 -0700682
Joonsoo Kim3c605092014-11-13 15:19:21 -0800683 page_idx = pfn & ((1 << max_order) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
Sasha Levin309381fea2014-01-23 15:52:54 -0800685 VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
686 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687
Joonsoo Kim3c605092014-11-13 15:19:21 -0800688 while (order < max_order - 1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800689 buddy_idx = __find_buddy_index(page_idx, order);
690 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700691 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700692 break;
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800693 /*
694 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
695 * merge with it and move up one order.
696 */
697 if (page_is_guard(buddy)) {
Joonsoo Kim2847cf92014-12-12 16:55:01 -0800698 clear_page_guard(zone, buddy, order, migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -0800699 } else {
700 list_del(&buddy->lru);
701 zone->free_area[order].nr_free--;
702 rmv_page_order(buddy);
703 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800704 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 page = page + (combined_idx - page_idx);
706 page_idx = combined_idx;
707 order++;
708 }
709 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700710
711 /*
712 * If this is not the largest possible page, check if the buddy
713 * of the next-highest order is free. If it is, it's possible
714 * that pages are being freed that will coalesce soon. In case,
715 * that is happening, add the free page to the tail of the list
716 * so it's less likely to be used soon and more likely to be merged
717 * as a higher order page
718 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700719 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700720 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800721 combined_idx = buddy_idx & page_idx;
722 higher_page = page + (combined_idx - page_idx);
723 buddy_idx = __find_buddy_index(combined_idx, order + 1);
Li Haifeng0ba8f2d2012-09-17 14:09:21 -0700724 higher_buddy = higher_page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700725 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
726 list_add_tail(&page->lru,
727 &zone->free_area[order].free_list[migratetype]);
728 goto out;
729 }
730 }
731
732 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
733out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 zone->free_area[order].nr_free++;
735}
736
Nick Piggin224abf92006-01-06 00:11:11 -0800737static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -0700739 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -0800740 unsigned long bad_flags = 0;
741
742 if (unlikely(page_mapcount(page)))
743 bad_reason = "nonzero mapcount";
744 if (unlikely(page->mapping != NULL))
745 bad_reason = "non-NULL mapping";
746 if (unlikely(atomic_read(&page->_count) != 0))
747 bad_reason = "nonzero _count";
748 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
749 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
750 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
751 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -0800752#ifdef CONFIG_MEMCG
753 if (unlikely(page->mem_cgroup))
754 bad_reason = "page still charged to cgroup";
755#endif
Dave Hansenf0b791a2014-01-23 15:52:49 -0800756 if (unlikely(bad_reason)) {
757 bad_page(page, bad_reason, bad_flags);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800758 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800759 }
Peter Zijlstra90572892013-10-07 11:29:20 +0100760 page_cpupid_reset_last(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800761 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
762 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
763 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764}
765
766/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700767 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700769 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 *
771 * If the zone was previously in an "all pages pinned" state then look to
772 * see if this freeing clears that state.
773 *
774 * And clear the zone's pages_scanned counter, to hold off the "all pages are
775 * pinned" detection logic.
776 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700777static void free_pcppages_bulk(struct zone *zone, int count,
778 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700780 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700781 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700782 int to_free = count;
Mel Gorman0d5d8232014-08-06 16:07:16 -0700783 unsigned long nr_scanned;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700784
Nick Pigginc54ad302006-01-06 00:10:56 -0800785 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700786 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
787 if (nr_scanned)
788 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700789
Mel Gorman72853e22010-09-09 16:38:16 -0700790 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800791 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700792 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800793
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700794 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700795 * Remove pages from lists in a round-robin fashion. A
796 * batch_free count is maintained that is incremented when an
797 * empty list is encountered. This is so more pages are freed
798 * off fuller lists instead of spinning excessively around empty
799 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700800 */
801 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700802 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700803 if (++migratetype == MIGRATE_PCPTYPES)
804 migratetype = 0;
805 list = &pcp->lists[migratetype];
806 } while (list_empty(list));
807
Namhyung Kim1d168712011-03-22 16:32:45 -0700808 /* This is the only non-empty list. Free them all. */
809 if (batch_free == MIGRATE_PCPTYPES)
810 batch_free = to_free;
811
Mel Gormana6f9edd2009-09-21 17:03:20 -0700812 do {
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700813 int mt; /* migratetype of the to-be-freed page */
814
Mel Gormana6f9edd2009-09-21 17:03:20 -0700815 page = list_entry(list->prev, struct page, lru);
816 /* must delete as __free_one_page list manipulates */
817 list_del(&page->lru);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -0700818
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -0700819 mt = get_pcppage_migratetype(page);
Vlastimil Babkaaa016d12015-09-08 15:01:22 -0700820 /* MIGRATE_ISOLATE page should not go to pcplists */
821 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
822 /* Pageblock could have been isolated meanwhile */
Joonsoo Kim8f82b552014-11-13 15:19:18 -0800823 if (unlikely(has_isolate_pageblock(zone)))
Joonsoo Kim51bb1a42014-11-13 15:19:14 -0800824 mt = get_pageblock_migratetype(page);
Joonsoo Kim51bb1a42014-11-13 15:19:14 -0800825
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700826 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
Bartlomiej Zolnierkiewicz770c8aa2012-10-08 16:31:57 -0700827 trace_mm_page_pcpu_drain(page, 0, mt);
Mel Gorman72853e22010-09-09 16:38:16 -0700828 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800830 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831}
832
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700833static void free_one_page(struct zone *zone,
834 struct page *page, unsigned long pfn,
Mel Gorman7aeb09f2014-06-04 16:10:21 -0700835 unsigned int order,
Mel Gormaned0ae212009-06-16 15:32:07 -0700836 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800837{
Mel Gorman0d5d8232014-08-06 16:07:16 -0700838 unsigned long nr_scanned;
Christoph Lameter006d22d2006-09-25 23:31:48 -0700839 spin_lock(&zone->lock);
Mel Gorman0d5d8232014-08-06 16:07:16 -0700840 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
841 if (nr_scanned)
842 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
Mel Gormanf2260e62009-06-16 15:32:13 -0700843
Joonsoo Kimad53f922014-11-13 15:19:11 -0800844 if (unlikely(has_isolate_pageblock(zone) ||
845 is_migrate_isolate(migratetype))) {
846 migratetype = get_pfnblock_migratetype(page, pfn);
Joonsoo Kimad53f922014-11-13 15:19:11 -0800847 }
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700848 __free_one_page(page, pfn, zone, order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700849 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800850}
851
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800852static int free_tail_pages_check(struct page *head_page, struct page *page)
853{
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800854 int ret = 1;
855
856 /*
857 * We rely page->lru.next never has bit 0 set, unless the page
858 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
859 */
860 BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
861
862 if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
863 ret = 0;
864 goto out;
865 }
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800866 if (unlikely(!PageTail(page))) {
867 bad_page(page, "PageTail not set", 0);
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800868 goto out;
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800869 }
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800870 if (unlikely(compound_head(page) != head_page)) {
871 bad_page(page, "compound_head not consistent", 0);
872 goto out;
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800873 }
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800874 ret = 0;
875out:
876 clear_compound_head(page);
877 return ret;
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800878}
879
Robin Holt1e8ce832015-06-30 14:56:45 -0700880static void __meminit __init_single_page(struct page *page, unsigned long pfn,
881 unsigned long zone, int nid)
882{
Robin Holt1e8ce832015-06-30 14:56:45 -0700883 set_page_links(page, zone, nid, pfn);
Robin Holt1e8ce832015-06-30 14:56:45 -0700884 init_page_count(page);
885 page_mapcount_reset(page);
886 page_cpupid_reset_last(page);
Robin Holt1e8ce832015-06-30 14:56:45 -0700887
Robin Holt1e8ce832015-06-30 14:56:45 -0700888 INIT_LIST_HEAD(&page->lru);
889#ifdef WANT_PAGE_VIRTUAL
890 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
891 if (!is_highmem_idx(zone))
892 set_page_address(page, __va(pfn << PAGE_SHIFT));
893#endif
894}
895
896static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
897 int nid)
898{
899 return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
900}
901
Mel Gorman7e18adb2015-06-30 14:57:05 -0700902#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
903static void init_reserved_page(unsigned long pfn)
904{
905 pg_data_t *pgdat;
906 int nid, zid;
907
908 if (!early_page_uninitialised(pfn))
909 return;
910
911 nid = early_pfn_to_nid(pfn);
912 pgdat = NODE_DATA(nid);
913
914 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
915 struct zone *zone = &pgdat->node_zones[zid];
916
917 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
918 break;
919 }
920 __init_single_pfn(pfn, zid, nid);
921}
922#else
923static inline void init_reserved_page(unsigned long pfn)
924{
925}
926#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
927
Nathan Zimmer92923ca2015-06-30 14:56:48 -0700928/*
929 * Initialised pages do not have PageReserved set. This function is
930 * called for each range allocated by the bootmem allocator and
931 * marks the pages PageReserved. The remaining valid pages are later
932 * sent to the buddy page allocator.
933 */
Mel Gorman7e18adb2015-06-30 14:57:05 -0700934void __meminit reserve_bootmem_region(unsigned long start, unsigned long end)
Nathan Zimmer92923ca2015-06-30 14:56:48 -0700935{
936 unsigned long start_pfn = PFN_DOWN(start);
937 unsigned long end_pfn = PFN_UP(end);
938
Mel Gorman7e18adb2015-06-30 14:57:05 -0700939 for (; start_pfn < end_pfn; start_pfn++) {
940 if (pfn_valid(start_pfn)) {
941 struct page *page = pfn_to_page(start_pfn);
942
943 init_reserved_page(start_pfn);
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800944
945 /* Avoid false-positive PageTail() */
946 INIT_LIST_HEAD(&page->lru);
947
Mel Gorman7e18adb2015-06-30 14:57:05 -0700948 SetPageReserved(page);
949 }
950 }
Nathan Zimmer92923ca2015-06-30 14:56:48 -0700951}
952
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700953static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954{
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800955 bool compound = PageCompound(page);
956 int i, bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Yu Zhaoab1f3062014-12-10 15:43:17 -0800958 VM_BUG_ON_PAGE(PageTail(page), page);
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800959 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
Yu Zhaoab1f3062014-12-10 15:43:17 -0800960
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -0800961 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100962 kmemcheck_free_shadow(page, order);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -0800963 kasan_free_pages(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100964
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800965 if (PageAnon(page))
966 page->mapping = NULL;
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800967 bad += free_pages_check(page);
968 for (i = 1; i < (1 << order); i++) {
969 if (compound)
970 bad += free_tail_pages_check(page, page + i);
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800971 bad += free_pages_check(page + i);
Kirill A. Shutemov81422f22015-02-11 15:25:52 -0800972 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800973 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700974 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800975
Joonsoo Kim48c96a32014-12-12 16:56:01 -0800976 reset_page_owner(page, order);
977
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700978 if (!PageHighMem(page)) {
Pintu Kumarb8af2942013-09-11 14:20:34 -0700979 debug_check_no_locks_freed(page_address(page),
980 PAGE_SIZE << order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700981 debug_check_no_obj_freed(page_address(page),
982 PAGE_SIZE << order);
983 }
Nick Piggindafb1362006-10-11 01:21:30 -0700984 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800985 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700986
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700987 return true;
988}
989
990static void __free_pages_ok(struct page *page, unsigned int order)
991{
992 unsigned long flags;
Minchan Kim95e34412012-10-08 16:32:11 -0700993 int migratetype;
Mel Gormandc4b0ca2014-06-04 16:10:17 -0700994 unsigned long pfn = page_to_pfn(page);
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700995
996 if (!free_pages_prepare(page, order))
997 return;
998
Mel Gormancfc47a22014-06-04 16:10:19 -0700999 migratetype = get_pfnblock_migratetype(page, pfn);
Nick Pigginc54ad302006-01-06 00:10:56 -08001000 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001001 __count_vm_events(PGFREE, 1 << order);
Mel Gormandc4b0ca2014-06-04 16:10:17 -07001002 free_one_page(page_zone(page), page, pfn, order, migratetype);
Nick Pigginc54ad302006-01-06 00:10:56 -08001003 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004}
1005
Mel Gorman0e1cc952015-06-30 14:57:27 -07001006static void __init __free_pages_boot_core(struct page *page,
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001007 unsigned long pfn, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -08001008{
Johannes Weinerc3993072012-01-10 15:08:10 -08001009 unsigned int nr_pages = 1 << order;
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001010 struct page *p = page;
Johannes Weinerc3993072012-01-10 15:08:10 -08001011 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -08001012
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001013 prefetchw(p);
1014 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1015 prefetchw(p + 1);
Johannes Weinerc3993072012-01-10 15:08:10 -08001016 __ClearPageReserved(p);
1017 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -08001018 }
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001019 __ClearPageReserved(p);
1020 set_page_count(p, 0);
Johannes Weinerc3993072012-01-10 15:08:10 -08001021
Yinghai Lue2d0bd22013-09-11 14:20:37 -07001022 page_zone(page)->managed_pages += nr_pages;
Johannes Weinerc3993072012-01-10 15:08:10 -08001023 set_page_refcounted(page);
1024 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -08001025}
1026
Mel Gorman75a592a2015-06-30 14:56:59 -07001027#if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1028 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
Mel Gorman7ace9912015-08-06 15:46:13 -07001029
Mel Gorman75a592a2015-06-30 14:56:59 -07001030static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1031
1032int __meminit early_pfn_to_nid(unsigned long pfn)
1033{
Mel Gorman7ace9912015-08-06 15:46:13 -07001034 static DEFINE_SPINLOCK(early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -07001035 int nid;
1036
Mel Gorman7ace9912015-08-06 15:46:13 -07001037 spin_lock(&early_pfn_lock);
Mel Gorman75a592a2015-06-30 14:56:59 -07001038 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
Mel Gorman7ace9912015-08-06 15:46:13 -07001039 if (nid < 0)
1040 nid = 0;
1041 spin_unlock(&early_pfn_lock);
1042
1043 return nid;
Mel Gorman75a592a2015-06-30 14:56:59 -07001044}
1045#endif
1046
1047#ifdef CONFIG_NODES_SPAN_OTHER_NODES
1048static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1049 struct mminit_pfnnid_cache *state)
1050{
1051 int nid;
1052
1053 nid = __early_pfn_to_nid(pfn, state);
1054 if (nid >= 0 && nid != node)
1055 return false;
1056 return true;
1057}
1058
1059/* Only safe to use early in boot when initialisation is single-threaded */
1060static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1061{
1062 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1063}
1064
1065#else
1066
1067static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1068{
1069 return true;
1070}
1071static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1072 struct mminit_pfnnid_cache *state)
1073{
1074 return true;
1075}
1076#endif
1077
1078
Mel Gorman0e1cc952015-06-30 14:57:27 -07001079void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
Mel Gorman3a80a7f2015-06-30 14:57:02 -07001080 unsigned int order)
1081{
1082 if (early_page_uninitialised(pfn))
1083 return;
1084 return __free_pages_boot_core(page, pfn, order);
1085}
1086
Mel Gorman7e18adb2015-06-30 14:57:05 -07001087#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
Mel Gorman0e1cc952015-06-30 14:57:27 -07001088static void __init deferred_free_range(struct page *page,
Mel Gormana4de83d2015-06-30 14:57:16 -07001089 unsigned long pfn, int nr_pages)
1090{
1091 int i;
1092
1093 if (!page)
1094 return;
1095
1096 /* Free a large naturally-aligned chunk if possible */
1097 if (nr_pages == MAX_ORDER_NR_PAGES &&
1098 (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
Mel Gormanac5d2532015-06-30 14:57:20 -07001099 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormana4de83d2015-06-30 14:57:16 -07001100 __free_pages_boot_core(page, pfn, MAX_ORDER-1);
1101 return;
1102 }
1103
1104 for (i = 0; i < nr_pages; i++, page++, pfn++)
1105 __free_pages_boot_core(page, pfn, 0);
1106}
1107
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001108/* Completion tracking for deferred_init_memmap() threads */
1109static atomic_t pgdat_init_n_undone __initdata;
1110static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1111
1112static inline void __init pgdat_init_report_one_done(void)
1113{
1114 if (atomic_dec_and_test(&pgdat_init_n_undone))
1115 complete(&pgdat_init_all_done_comp);
1116}
Mel Gorman0e1cc952015-06-30 14:57:27 -07001117
Mel Gorman7e18adb2015-06-30 14:57:05 -07001118/* Initialise remaining memory on a node */
Mel Gorman0e1cc952015-06-30 14:57:27 -07001119static int __init deferred_init_memmap(void *data)
Mel Gorman7e18adb2015-06-30 14:57:05 -07001120{
Mel Gorman0e1cc952015-06-30 14:57:27 -07001121 pg_data_t *pgdat = data;
1122 int nid = pgdat->node_id;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001123 struct mminit_pfnnid_cache nid_init_state = { };
1124 unsigned long start = jiffies;
1125 unsigned long nr_pages = 0;
1126 unsigned long walk_start, walk_end;
1127 int i, zid;
1128 struct zone *zone;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001129 unsigned long first_init_pfn = pgdat->first_deferred_pfn;
Mel Gorman0e1cc952015-06-30 14:57:27 -07001130 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001131
Mel Gorman0e1cc952015-06-30 14:57:27 -07001132 if (first_init_pfn == ULONG_MAX) {
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001133 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001134 return 0;
1135 }
1136
1137 /* Bind memory initialisation thread to a local node if possible */
1138 if (!cpumask_empty(cpumask))
1139 set_cpus_allowed_ptr(current, cpumask);
Mel Gorman7e18adb2015-06-30 14:57:05 -07001140
1141 /* Sanity check boundaries */
1142 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1143 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1144 pgdat->first_deferred_pfn = ULONG_MAX;
1145
1146 /* Only the highest zone is deferred so find it */
1147 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1148 zone = pgdat->node_zones + zid;
1149 if (first_init_pfn < zone_end_pfn(zone))
1150 break;
1151 }
1152
1153 for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1154 unsigned long pfn, end_pfn;
Mel Gorman54608c32015-06-30 14:57:09 -07001155 struct page *page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001156 struct page *free_base_page = NULL;
1157 unsigned long free_base_pfn = 0;
1158 int nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001159
1160 end_pfn = min(walk_end, zone_end_pfn(zone));
1161 pfn = first_init_pfn;
1162 if (pfn < walk_start)
1163 pfn = walk_start;
1164 if (pfn < zone->zone_start_pfn)
1165 pfn = zone->zone_start_pfn;
1166
1167 for (; pfn < end_pfn; pfn++) {
Mel Gorman54608c32015-06-30 14:57:09 -07001168 if (!pfn_valid_within(pfn))
Mel Gormana4de83d2015-06-30 14:57:16 -07001169 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001170
Mel Gorman54608c32015-06-30 14:57:09 -07001171 /*
1172 * Ensure pfn_valid is checked every
1173 * MAX_ORDER_NR_PAGES for memory holes
1174 */
1175 if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1176 if (!pfn_valid(pfn)) {
1177 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001178 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001179 }
1180 }
1181
1182 if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1183 page = NULL;
Mel Gormana4de83d2015-06-30 14:57:16 -07001184 goto free_range;
Mel Gorman54608c32015-06-30 14:57:09 -07001185 }
1186
1187 /* Minimise pfn page lookups and scheduler checks */
1188 if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1189 page++;
1190 } else {
Mel Gormana4de83d2015-06-30 14:57:16 -07001191 nr_pages += nr_to_free;
1192 deferred_free_range(free_base_page,
1193 free_base_pfn, nr_to_free);
1194 free_base_page = NULL;
1195 free_base_pfn = nr_to_free = 0;
1196
Mel Gorman54608c32015-06-30 14:57:09 -07001197 page = pfn_to_page(pfn);
1198 cond_resched();
1199 }
Mel Gorman7e18adb2015-06-30 14:57:05 -07001200
1201 if (page->flags) {
1202 VM_BUG_ON(page_zone(page) != zone);
Mel Gormana4de83d2015-06-30 14:57:16 -07001203 goto free_range;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001204 }
1205
1206 __init_single_page(page, pfn, zid, nid);
Mel Gormana4de83d2015-06-30 14:57:16 -07001207 if (!free_base_page) {
1208 free_base_page = page;
1209 free_base_pfn = pfn;
1210 nr_to_free = 0;
1211 }
1212 nr_to_free++;
1213
1214 /* Where possible, batch up pages for a single free */
1215 continue;
1216free_range:
1217 /* Free the current block of pages to allocator */
1218 nr_pages += nr_to_free;
1219 deferred_free_range(free_base_page, free_base_pfn,
1220 nr_to_free);
1221 free_base_page = NULL;
1222 free_base_pfn = nr_to_free = 0;
Mel Gorman7e18adb2015-06-30 14:57:05 -07001223 }
Mel Gormana4de83d2015-06-30 14:57:16 -07001224
Mel Gorman7e18adb2015-06-30 14:57:05 -07001225 first_init_pfn = max(end_pfn, first_init_pfn);
1226 }
1227
1228 /* Sanity check that the next zone really is unpopulated */
1229 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1230
Mel Gorman0e1cc952015-06-30 14:57:27 -07001231 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
Mel Gorman7e18adb2015-06-30 14:57:05 -07001232 jiffies_to_msecs(jiffies - start));
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001233
1234 pgdat_init_report_one_done();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001235 return 0;
1236}
1237
1238void __init page_alloc_init_late(void)
1239{
1240 int nid;
1241
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001242 /* There will be num_node_state(N_MEMORY) threads */
1243 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
Mel Gorman0e1cc952015-06-30 14:57:27 -07001244 for_each_node_state(nid, N_MEMORY) {
Mel Gorman0e1cc952015-06-30 14:57:27 -07001245 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1246 }
1247
1248 /* Block until all are initialised */
Nicolai Stanged3cd1312015-08-06 15:46:16 -07001249 wait_for_completion(&pgdat_init_all_done_comp);
Mel Gorman4248b0d2015-08-06 15:46:20 -07001250
1251 /* Reinit limits that are based on free pages after the kernel is up */
1252 files_maxfiles_init();
Mel Gorman7e18adb2015-06-30 14:57:05 -07001253}
1254#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1255
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001256#ifdef CONFIG_CMA
Li Zhong9cf510a2013-08-23 13:52:52 +08001257/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001258void __init init_cma_reserved_pageblock(struct page *page)
1259{
1260 unsigned i = pageblock_nr_pages;
1261 struct page *p = page;
1262
1263 do {
1264 __ClearPageReserved(p);
1265 set_page_count(p, 0);
1266 } while (++p, --i);
1267
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001268 set_pageblock_migratetype(page, MIGRATE_CMA);
Michal Nazarewiczdc783272014-07-02 15:22:35 -07001269
1270 if (pageblock_order >= MAX_ORDER) {
1271 i = pageblock_nr_pages;
1272 p = page;
1273 do {
1274 set_page_refcounted(p);
1275 __free_pages(p, MAX_ORDER - 1);
1276 p += MAX_ORDER_NR_PAGES;
1277 } while (i -= MAX_ORDER_NR_PAGES);
1278 } else {
1279 set_page_refcounted(page);
1280 __free_pages(page, pageblock_order);
1281 }
1282
Jiang Liu3dcc0572013-07-03 15:03:21 -07001283 adjust_managed_page_count(page, pageblock_nr_pages);
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001284}
1285#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
1287/*
1288 * The order of subdivision here is critical for the IO subsystem.
1289 * Please do not alter this order without good reasons and regression
1290 * testing. Specifically, as large blocks of memory are subdivided,
1291 * the order in which smaller blocks are delivered depends on the order
1292 * they're subdivided in this function. This is the primary factor
1293 * influencing the order in which pages are delivered to the IO
1294 * subsystem according to empirical testing, and this is also justified
1295 * by considering the behavior of a buddy system containing a single
1296 * large block of memory acted on by a series of small allocations.
1297 * This behavior is a critical factor in sglist merging's success.
1298 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +01001299 * -- nyc
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 */
Nick Piggin085cc7d2006-01-06 00:11:01 -08001301static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001302 int low, int high, struct free_area *area,
1303 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304{
1305 unsigned long size = 1 << high;
1306
1307 while (high > low) {
1308 area--;
1309 high--;
1310 size >>= 1;
Sasha Levin309381fea2014-01-23 15:52:54 -08001311 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08001312
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001313 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
Joonsoo Kime30825f2014-12-12 16:55:49 -08001314 debug_guardpage_enabled() &&
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001315 high < debug_guardpage_minorder()) {
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08001316 /*
1317 * Mark as guard pages (or page), that will allow to
1318 * merge back to allocator when buddy will be freed.
1319 * Corresponding page table entries will not be touched,
1320 * pages will stay not present in virtual address space
1321 */
Joonsoo Kim2847cf92014-12-12 16:55:01 -08001322 set_page_guard(zone, &page[size], high, migratetype);
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08001323 continue;
1324 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001325 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 area->nr_free++;
1327 set_page_order(&page[size], high);
1328 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329}
1330
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331/*
1332 * This page is about to be returned from the page allocator
1333 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001334static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335{
Kirill A. Shutemovd230dec2014-04-07 15:37:38 -07001336 const char *bad_reason = NULL;
Dave Hansenf0b791a2014-01-23 15:52:49 -08001337 unsigned long bad_flags = 0;
1338
1339 if (unlikely(page_mapcount(page)))
1340 bad_reason = "nonzero mapcount";
1341 if (unlikely(page->mapping != NULL))
1342 bad_reason = "non-NULL mapping";
1343 if (unlikely(atomic_read(&page->_count) != 0))
1344 bad_reason = "nonzero _count";
Naoya Horiguchif4c18e62015-08-06 15:47:08 -07001345 if (unlikely(page->flags & __PG_HWPOISON)) {
1346 bad_reason = "HWPoisoned (hardware-corrupted)";
1347 bad_flags = __PG_HWPOISON;
1348 }
Dave Hansenf0b791a2014-01-23 15:52:49 -08001349 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1350 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1351 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1352 }
Johannes Weiner9edad6e2014-12-10 15:44:58 -08001353#ifdef CONFIG_MEMCG
1354 if (unlikely(page->mem_cgroup))
1355 bad_reason = "page still charged to cgroup";
1356#endif
Dave Hansenf0b791a2014-01-23 15:52:49 -08001357 if (unlikely(bad_reason)) {
1358 bad_page(page, bad_reason, bad_flags);
Hugh Dickins689bceb2005-11-21 21:32:20 -08001359 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -08001360 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001361 return 0;
1362}
1363
Vlastimil Babka75379192015-02-11 15:25:38 -08001364static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1365 int alloc_flags)
Wu Fengguang2a7684a2009-09-16 11:50:12 +02001366{
1367 int i;
1368
1369 for (i = 0; i < (1 << order); i++) {
1370 struct page *p = page + i;
1371 if (unlikely(check_new_page(p)))
1372 return 1;
1373 }
Hugh Dickins689bceb2005-11-21 21:32:20 -08001374
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001375 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -08001376 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -08001377
1378 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 kernel_map_pages(page, 1 << order, 1);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -08001380 kasan_alloc_pages(page, order);
Nick Piggin17cf4402006-03-22 00:08:41 -08001381
1382 if (gfp_flags & __GFP_ZERO)
Anisse Astierf4d28972015-06-24 16:56:36 -07001383 for (i = 0; i < (1 << order); i++)
1384 clear_highpage(page + i);
Nick Piggin17cf4402006-03-22 00:08:41 -08001385
1386 if (order && (gfp_flags & __GFP_COMP))
1387 prep_compound_page(page, order);
1388
Joonsoo Kim48c96a32014-12-12 16:56:01 -08001389 set_page_owner(page, order, gfp_flags);
1390
Vlastimil Babka75379192015-02-11 15:25:38 -08001391 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001392 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
Vlastimil Babka75379192015-02-11 15:25:38 -08001393 * allocate the page. The expectation is that the caller is taking
1394 * steps that will free more memory. The caller should avoid the page
1395 * being used for !PFMEMALLOC purposes.
1396 */
Michal Hocko2f064f32015-08-21 14:11:51 -07001397 if (alloc_flags & ALLOC_NO_WATERMARKS)
1398 set_page_pfmemalloc(page);
1399 else
1400 clear_page_pfmemalloc(page);
Vlastimil Babka75379192015-02-11 15:25:38 -08001401
Hugh Dickins689bceb2005-11-21 21:32:20 -08001402 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403}
1404
Mel Gorman56fd56b2007-10-16 01:25:58 -07001405/*
1406 * Go through the free lists for the given migratetype and remove
1407 * the smallest available page from the freelists
1408 */
Mel Gorman728ec982009-06-16 15:32:04 -07001409static inline
1410struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -07001411 int migratetype)
1412{
1413 unsigned int current_order;
Pintu Kumarb8af2942013-09-11 14:20:34 -07001414 struct free_area *area;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001415 struct page *page;
1416
1417 /* Find a page of the appropriate size in the preferred list */
1418 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1419 area = &(zone->free_area[current_order]);
1420 if (list_empty(&area->free_list[migratetype]))
1421 continue;
1422
1423 page = list_entry(area->free_list[migratetype].next,
1424 struct page, lru);
1425 list_del(&page->lru);
1426 rmv_page_order(page);
1427 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -07001428 expand(zone, page, order, current_order, area, migratetype);
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001429 set_pcppage_migratetype(page, migratetype);
Mel Gorman56fd56b2007-10-16 01:25:58 -07001430 return page;
1431 }
1432
1433 return NULL;
1434}
1435
1436
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001437/*
1438 * This array describes the order lists are fallen back to when
1439 * the free lists for the desirable migrate type are depleted
1440 */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001441static int fallbacks[MIGRATE_TYPES][4] = {
Mel Gorman974a7862015-11-06 16:28:34 -08001442 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1443 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1444 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
Joonsoo Kimdc676472015-04-14 15:45:15 -07001445#ifdef CONFIG_CMA
Mel Gorman974a7862015-11-06 16:28:34 -08001446 [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */
Michal Nazarewicz47118af2011-12-29 13:09:50 +01001447#endif
Minchan Kim194159f2013-02-22 16:33:58 -08001448#ifdef CONFIG_MEMORY_ISOLATION
Mel Gorman974a7862015-11-06 16:28:34 -08001449 [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */
Minchan Kim194159f2013-02-22 16:33:58 -08001450#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001451};
1452
Joonsoo Kimdc676472015-04-14 15:45:15 -07001453#ifdef CONFIG_CMA
1454static struct page *__rmqueue_cma_fallback(struct zone *zone,
1455 unsigned int order)
1456{
1457 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1458}
1459#else
1460static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1461 unsigned int order) { return NULL; }
1462#endif
1463
Mel Gormanc361be52007-10-16 01:25:51 -07001464/*
1465 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -07001466 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -07001467 * boundary. If alignment is required, use move_freepages_block()
1468 */
Minchan Kim435b4052012-10-08 16:32:16 -07001469int move_freepages(struct zone *zone,
Adrian Bunkb69a7282008-07-23 21:28:12 -07001470 struct page *start_page, struct page *end_page,
1471 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001472{
1473 struct page *page;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08001474 unsigned int order;
Mel Gormand1003132007-10-16 01:26:00 -07001475 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -07001476
1477#ifndef CONFIG_HOLES_IN_ZONE
1478 /*
1479 * page_zone is not safe to call in this context when
1480 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1481 * anyway as we check zone boundaries in move_freepages_block().
1482 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -07001483 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -07001484 */
Mel Gorman97ee4ba2014-10-09 15:28:28 -07001485 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
Mel Gormanc361be52007-10-16 01:25:51 -07001486#endif
1487
1488 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -07001489 /* Make sure we are not inadvertently changing nodes */
Sasha Levin309381fea2014-01-23 15:52:54 -08001490 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
Adam Litke344c7902008-09-02 14:35:38 -07001491
Mel Gormanc361be52007-10-16 01:25:51 -07001492 if (!pfn_valid_within(page_to_pfn(page))) {
1493 page++;
1494 continue;
1495 }
1496
1497 if (!PageBuddy(page)) {
1498 page++;
1499 continue;
1500 }
1501
1502 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -07001503 list_move(&page->lru,
1504 &zone->free_area[order].free_list[migratetype]);
Mel Gormanc361be52007-10-16 01:25:51 -07001505 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -07001506 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001507 }
1508
Mel Gormand1003132007-10-16 01:26:00 -07001509 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001510}
1511
Minchan Kimee6f5092012-07-31 16:43:50 -07001512int move_freepages_block(struct zone *zone, struct page *page,
Linus Torvalds68e3e922012-06-03 20:05:57 -07001513 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001514{
1515 unsigned long start_pfn, end_pfn;
1516 struct page *start_page, *end_page;
1517
1518 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001519 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001520 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001521 end_page = start_page + pageblock_nr_pages - 1;
1522 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001523
1524 /* Do not cross zone boundaries */
Cody P Schafer108bcc92013-02-22 16:35:23 -08001525 if (!zone_spans_pfn(zone, start_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001526 start_page = page;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001527 if (!zone_spans_pfn(zone, end_pfn))
Mel Gormanc361be52007-10-16 01:25:51 -07001528 return 0;
1529
1530 return move_freepages(zone, start_page, end_page, migratetype);
1531}
1532
Mel Gorman2f66a682009-09-21 17:02:31 -07001533static void change_pageblock_range(struct page *pageblock_page,
1534 int start_order, int migratetype)
1535{
1536 int nr_pageblocks = 1 << (start_order - pageblock_order);
1537
1538 while (nr_pageblocks--) {
1539 set_pageblock_migratetype(pageblock_page, migratetype);
1540 pageblock_page += pageblock_nr_pages;
1541 }
1542}
1543
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001544/*
Vlastimil Babka9c0415e2015-02-11 15:28:21 -08001545 * When we are falling back to another migratetype during allocation, try to
1546 * steal extra free pages from the same pageblocks to satisfy further
1547 * allocations, instead of polluting multiple pageblocks.
1548 *
1549 * If we are stealing a relatively large buddy page, it is likely there will
1550 * be more free pages in the pageblock, so try to steal them all. For
1551 * reclaimable and unmovable allocations, we steal regardless of page size,
1552 * as fragmentation caused by those allocations polluting movable pageblocks
1553 * is worse than movable allocations stealing from unmovable and reclaimable
1554 * pageblocks.
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001555 */
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001556static bool can_steal_fallback(unsigned int order, int start_mt)
1557{
1558 /*
1559 * Leaving this order check is intended, although there is
1560 * relaxed order check in next check. The reason is that
1561 * we can actually steal whole pageblock if this condition met,
1562 * but, below check doesn't guarantee it and that is just heuristic
1563 * so could be changed anytime.
1564 */
1565 if (order >= pageblock_order)
1566 return true;
1567
1568 if (order >= pageblock_order / 2 ||
1569 start_mt == MIGRATE_RECLAIMABLE ||
1570 start_mt == MIGRATE_UNMOVABLE ||
1571 page_group_by_mobility_disabled)
1572 return true;
1573
1574 return false;
1575}
1576
1577/*
1578 * This function implements actual steal behaviour. If order is large enough,
1579 * we can steal whole pageblock. If not, we first move freepages in this
1580 * pageblock and check whether half of pages are moved or not. If half of
1581 * pages are moved, we can change migratetype of pageblock and permanently
1582 * use it's pages as requested migratetype in the future.
1583 */
1584static void steal_suitable_fallback(struct zone *zone, struct page *page,
1585 int start_type)
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001586{
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08001587 unsigned int current_order = page_order(page);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001588 int pages;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001589
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001590 /* Take ownership for orders >= pageblock_order */
1591 if (current_order >= pageblock_order) {
1592 change_pageblock_range(page, current_order, start_type);
Vlastimil Babka3a1086f2015-02-11 15:28:18 -08001593 return;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001594 }
1595
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001596 pages = move_freepages_block(zone, page, start_type);
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001597
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001598 /* Claim the whole block if over half of it is free */
1599 if (pages >= (1 << (pageblock_order-1)) ||
1600 page_group_by_mobility_disabled)
1601 set_pageblock_migratetype(page, start_type);
1602}
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001603
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001604/*
1605 * Check whether there is a suitable fallback freepage with requested order.
1606 * If only_stealable is true, this function returns fallback_mt only if
1607 * we can steal other freepages all together. This would help to reduce
1608 * fragmentation due to mixed migratetype pages in one pageblock.
1609 */
1610int find_suitable_fallback(struct free_area *area, unsigned int order,
1611 int migratetype, bool only_stealable, bool *can_steal)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001612{
1613 int i;
1614 int fallback_mt;
1615
1616 if (area->nr_free == 0)
1617 return -1;
1618
1619 *can_steal = false;
1620 for (i = 0;; i++) {
1621 fallback_mt = fallbacks[migratetype][i];
Mel Gorman974a7862015-11-06 16:28:34 -08001622 if (fallback_mt == MIGRATE_TYPES)
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001623 break;
1624
1625 if (list_empty(&area->free_list[fallback_mt]))
1626 continue;
1627
1628 if (can_steal_fallback(order, migratetype))
1629 *can_steal = true;
1630
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001631 if (!only_stealable)
1632 return fallback_mt;
1633
1634 if (*can_steal)
1635 return fallback_mt;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001636 }
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001637
1638 return -1;
Srivatsa S. Bhatfef903e2013-09-11 14:20:35 -07001639}
1640
Mel Gorman0aaa29a2015-11-06 16:28:37 -08001641/*
1642 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1643 * there are no empty page blocks that contain a page with a suitable order
1644 */
1645static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
1646 unsigned int alloc_order)
1647{
1648 int mt;
1649 unsigned long max_managed, flags;
1650
1651 /*
1652 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
1653 * Check is race-prone but harmless.
1654 */
1655 max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
1656 if (zone->nr_reserved_highatomic >= max_managed)
1657 return;
1658
1659 spin_lock_irqsave(&zone->lock, flags);
1660
1661 /* Recheck the nr_reserved_highatomic limit under the lock */
1662 if (zone->nr_reserved_highatomic >= max_managed)
1663 goto out_unlock;
1664
1665 /* Yoink! */
1666 mt = get_pageblock_migratetype(page);
1667 if (mt != MIGRATE_HIGHATOMIC &&
1668 !is_migrate_isolate(mt) && !is_migrate_cma(mt)) {
1669 zone->nr_reserved_highatomic += pageblock_nr_pages;
1670 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
1671 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC);
1672 }
1673
1674out_unlock:
1675 spin_unlock_irqrestore(&zone->lock, flags);
1676}
1677
1678/*
1679 * Used when an allocation is about to fail under memory pressure. This
1680 * potentially hurts the reliability of high-order allocations when under
1681 * intense memory pressure but failed atomic allocations should be easier
1682 * to recover from than an OOM.
1683 */
1684static void unreserve_highatomic_pageblock(const struct alloc_context *ac)
1685{
1686 struct zonelist *zonelist = ac->zonelist;
1687 unsigned long flags;
1688 struct zoneref *z;
1689 struct zone *zone;
1690 struct page *page;
1691 int order;
1692
1693 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
1694 ac->nodemask) {
1695 /* Preserve at least one pageblock */
1696 if (zone->nr_reserved_highatomic <= pageblock_nr_pages)
1697 continue;
1698
1699 spin_lock_irqsave(&zone->lock, flags);
1700 for (order = 0; order < MAX_ORDER; order++) {
1701 struct free_area *area = &(zone->free_area[order]);
1702
1703 if (list_empty(&area->free_list[MIGRATE_HIGHATOMIC]))
1704 continue;
1705
1706 page = list_entry(area->free_list[MIGRATE_HIGHATOMIC].next,
1707 struct page, lru);
1708
1709 /*
1710 * It should never happen but changes to locking could
1711 * inadvertently allow a per-cpu drain to add pages
1712 * to MIGRATE_HIGHATOMIC while unreserving so be safe
1713 * and watch for underflows.
1714 */
1715 zone->nr_reserved_highatomic -= min(pageblock_nr_pages,
1716 zone->nr_reserved_highatomic);
1717
1718 /*
1719 * Convert to ac->migratetype and avoid the normal
1720 * pageblock stealing heuristics. Minimally, the caller
1721 * is doing the work and needs the pages. More
1722 * importantly, if the block was always converted to
1723 * MIGRATE_UNMOVABLE or another type then the number
1724 * of pageblocks that cannot be completely freed
1725 * may increase.
1726 */
1727 set_pageblock_migratetype(page, ac->migratetype);
1728 move_freepages_block(zone, page, ac->migratetype);
1729 spin_unlock_irqrestore(&zone->lock, flags);
1730 return;
1731 }
1732 spin_unlock_irqrestore(&zone->lock, flags);
1733 }
1734}
1735
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001736/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001737static inline struct page *
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001738__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001739{
Pintu Kumarb8af2942013-09-11 14:20:34 -07001740 struct free_area *area;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001741 unsigned int current_order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001742 struct page *page;
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001743 int fallback_mt;
1744 bool can_steal;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001745
1746 /* Find the largest possible block of pages in the other list */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001747 for (current_order = MAX_ORDER-1;
1748 current_order >= order && current_order <= MAX_ORDER-1;
1749 --current_order) {
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001750 area = &(zone->free_area[current_order]);
1751 fallback_mt = find_suitable_fallback(area, current_order,
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001752 start_migratetype, false, &can_steal);
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001753 if (fallback_mt == -1)
1754 continue;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001755
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001756 page = list_entry(area->free_list[fallback_mt].next,
1757 struct page, lru);
1758 if (can_steal)
1759 steal_suitable_fallback(zone, page, start_migratetype);
Mel Gormane0104872007-10-16 01:25:53 -07001760
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001761 /* Remove the page from the freelists */
1762 area->nr_free--;
1763 list_del(&page->lru);
1764 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001765
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001766 expand(zone, page, order, current_order, area,
1767 start_migratetype);
1768 /*
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001769 * The pcppage_migratetype may differ from pageblock's
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001770 * migratetype depending on the decisions in
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001771 * find_suitable_fallback(). This is OK as long as it does not
1772 * differ for MIGRATE_CMA pageblocks. Those can be used as
1773 * fallback only via special __rmqueue_cma_fallback() function
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001774 */
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001775 set_pcppage_migratetype(page, start_migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001776
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001777 trace_mm_page_alloc_extfrag(page, order, current_order,
1778 start_migratetype, fallback_mt);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001779
Joonsoo Kim4eb7dce2015-04-14 15:45:18 -07001780 return page;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001781 }
1782
Mel Gorman728ec982009-06-16 15:32:04 -07001783 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001784}
1785
Mel Gorman56fd56b2007-10-16 01:25:58 -07001786/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 * Do the hard work of removing an element from the buddy allocator.
1788 * Call me with the zone->lock already held.
1789 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001790static struct page *__rmqueue(struct zone *zone, unsigned int order,
Mel Gorman0aaa29a2015-11-06 16:28:37 -08001791 int migratetype, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 struct page *page;
1794
Mel Gorman56fd56b2007-10-16 01:25:58 -07001795 page = __rmqueue_smallest(zone, order, migratetype);
Mel Gorman974a7862015-11-06 16:28:34 -08001796 if (unlikely(!page)) {
Joonsoo Kimdc676472015-04-14 15:45:15 -07001797 if (migratetype == MIGRATE_MOVABLE)
1798 page = __rmqueue_cma_fallback(zone, order);
1799
1800 if (!page)
1801 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gorman728ec982009-06-16 15:32:04 -07001802 }
1803
Mel Gorman0d3d0622009-09-21 17:02:44 -07001804 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001805 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806}
1807
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001808/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 * Obtain a specified number of elements from the buddy allocator, all under
1810 * a single hold of the lock, for efficiency. Add them to the supplied list.
1811 * Returns the number of new pages which were placed at *list.
1812 */
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001813static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001814 unsigned long count, struct list_head *list,
Mel Gormanb745bc82014-06-04 16:10:22 -07001815 int migratetype, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816{
Vlastimil Babka5bcc9f82014-06-04 16:07:22 -07001817 int i;
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01001818
Nick Pigginc54ad302006-01-06 00:10:56 -08001819 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 for (i = 0; i < count; ++i) {
Mel Gorman0aaa29a2015-11-06 16:28:37 -08001821 struct page *page = __rmqueue(zone, order, migratetype, 0);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001822 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001824
1825 /*
1826 * Split buddy pages returned by expand() are received here
1827 * in physical page order. The page is added to the callers and
1828 * list and the list head then moves forward. From the callers
1829 * perspective, the linked list is ordered by page number in
1830 * some conditions. This is useful for IO devices that can
1831 * merge IO requests if the physical pages are ordered
1832 * properly.
1833 */
Mel Gormanb745bc82014-06-04 16:10:22 -07001834 if (likely(!cold))
Mel Gormane084b2d2009-07-29 15:02:04 -07001835 list_add(&page->lru, list);
1836 else
1837 list_add_tail(&page->lru, list);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001838 list = &page->lru;
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07001839 if (is_migrate_cma(get_pcppage_migratetype(page)))
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07001840 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1841 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001843 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001844 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001845 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846}
1847
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001848#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001849/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001850 * Called from the vmstat counter updater to drain pagesets of this
1851 * currently executing processor on remote nodes after they have
1852 * expired.
1853 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001854 * Note that this function must be called with the thread pinned to
1855 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001856 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001857void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001858{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001859 unsigned long flags;
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001860 int to_drain, batch;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001861
Christoph Lameter4037d452007-05-09 02:35:14 -07001862 local_irq_save(flags);
Jason Low4db0c3c2015-04-15 16:14:08 -07001863 batch = READ_ONCE(pcp->batch);
Michal Nazarewicz7be12fc2014-08-06 16:05:15 -07001864 to_drain = min(pcp->count, batch);
KOSAKI Motohiro2a135152012-07-31 16:42:53 -07001865 if (to_drain > 0) {
1866 free_pcppages_bulk(zone, to_drain, pcp);
1867 pcp->count -= to_drain;
1868 }
Christoph Lameter4037d452007-05-09 02:35:14 -07001869 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001870}
1871#endif
1872
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001873/*
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001874 * Drain pcplists of the indicated processor and zone.
1875 *
1876 * The processor must either be the current processor and the
1877 * thread pinned to the current processor or a processor that
1878 * is not online.
1879 */
1880static void drain_pages_zone(unsigned int cpu, struct zone *zone)
1881{
1882 unsigned long flags;
1883 struct per_cpu_pageset *pset;
1884 struct per_cpu_pages *pcp;
1885
1886 local_irq_save(flags);
1887 pset = per_cpu_ptr(zone->pageset, cpu);
1888
1889 pcp = &pset->pcp;
1890 if (pcp->count) {
1891 free_pcppages_bulk(zone, pcp->count, pcp);
1892 pcp->count = 0;
1893 }
1894 local_irq_restore(flags);
1895}
1896
1897/*
1898 * Drain pcplists of all zones on the indicated processor.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001899 *
1900 * The processor must either be the current processor and the
1901 * thread pinned to the current processor or a processor that
1902 * is not online.
1903 */
1904static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905{
1906 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001908 for_each_populated_zone(zone) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001909 drain_pages_zone(cpu, zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 }
1911}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001913/*
1914 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001915 *
1916 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1917 * the single zone's pages.
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001918 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001919void drain_local_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001920{
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001921 int cpu = smp_processor_id();
1922
1923 if (zone)
1924 drain_pages_zone(cpu, zone);
1925 else
1926 drain_pages(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001927}
1928
1929/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001930 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1931 *
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001932 * When zone parameter is non-NULL, spill just the single zone's pages.
1933 *
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001934 * Note that this code is protected against sending an IPI to an offline
1935 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1936 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1937 * nothing keeps CPUs from showing up after we populated the cpumask and
1938 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001939 */
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001940void drain_all_pages(struct zone *zone)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001941{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001942 int cpu;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001943
1944 /*
1945 * Allocate in the BSS so we wont require allocation in
1946 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1947 */
1948 static cpumask_t cpus_with_pcps;
1949
1950 /*
1951 * We don't care about racing with CPU hotplug event
1952 * as offline notification will cause the notified
1953 * cpu to drain that CPU pcps and on_each_cpu_mask
1954 * disables preemption as part of its processing
1955 */
1956 for_each_online_cpu(cpu) {
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001957 struct per_cpu_pageset *pcp;
1958 struct zone *z;
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001959 bool has_pcps = false;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001960
1961 if (zone) {
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001962 pcp = per_cpu_ptr(zone->pageset, cpu);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001963 if (pcp->pcp.count)
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001964 has_pcps = true;
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001965 } else {
1966 for_each_populated_zone(z) {
1967 pcp = per_cpu_ptr(z->pageset, cpu);
1968 if (pcp->pcp.count) {
1969 has_pcps = true;
1970 break;
1971 }
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001972 }
1973 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001974
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001975 if (has_pcps)
1976 cpumask_set_cpu(cpu, &cpus_with_pcps);
1977 else
1978 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1979 }
Vlastimil Babka93481ff2014-12-10 15:43:01 -08001980 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
1981 zone, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001982}
1983
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001984#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985
1986void mark_free_pages(struct zone *zone)
1987{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001988 unsigned long pfn, max_zone_pfn;
1989 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07001990 unsigned int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 struct list_head *curr;
1992
Xishi Qiu8080fc02013-09-11 14:21:45 -07001993 if (zone_is_empty(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 return;
1995
1996 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001997
Cody P Schafer108bcc92013-02-22 16:35:23 -08001998 max_zone_pfn = zone_end_pfn(zone);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001999 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2000 if (pfn_valid(pfn)) {
2001 struct page *page = pfn_to_page(pfn);
2002
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07002003 if (!swsusp_page_is_forbidden(page))
2004 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002005 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002007 for_each_migratetype_order(order, t) {
2008 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002009 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002011 pfn = page_to_pfn(list_entry(curr, struct page, lru));
2012 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07002013 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07002014 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002015 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 spin_unlock_irqrestore(&zone->lock, flags);
2017}
Mel Gormane2c55dc2007-10-16 01:25:50 -07002018#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019
2020/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 * Free a 0-order page
Mel Gormanb745bc82014-06-04 16:10:22 -07002022 * cold == true ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002024void free_hot_cold_page(struct page *page, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025{
2026 struct zone *zone = page_zone(page);
2027 struct per_cpu_pages *pcp;
2028 unsigned long flags;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002029 unsigned long pfn = page_to_pfn(page);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002030 int migratetype;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07002032 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08002033 return;
2034
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002035 migratetype = get_pfnblock_migratetype(page, pfn);
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002036 set_pcppage_migratetype(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07002038 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07002039
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002040 /*
2041 * We only track unmovable, reclaimable and movable on pcp lists.
2042 * Free ISOLATE pages back to the allocator because they are being
2043 * offlined but treat RESERVE as movable pages so we can get those
2044 * areas back if necessary. Otherwise, we may have to free
2045 * excessively into the page allocator
2046 */
2047 if (migratetype >= MIGRATE_PCPTYPES) {
Minchan Kim194159f2013-02-22 16:33:58 -08002048 if (unlikely(is_migrate_isolate(migratetype))) {
Mel Gormandc4b0ca2014-06-04 16:10:17 -07002049 free_one_page(zone, page, pfn, 0, migratetype);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002050 goto out;
2051 }
2052 migratetype = MIGRATE_MOVABLE;
2053 }
2054
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002055 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Mel Gormanb745bc82014-06-04 16:10:22 -07002056 if (!cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002057 list_add(&page->lru, &pcp->lists[migratetype]);
Mel Gormanb745bc82014-06-04 16:10:22 -07002058 else
2059 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08002061 if (pcp->count >= pcp->high) {
Jason Low4db0c3c2015-04-15 16:14:08 -07002062 unsigned long batch = READ_ONCE(pcp->batch);
Cody P Schafer998d39c2013-07-03 15:01:32 -07002063 free_pcppages_bulk(zone, batch, pcp);
2064 pcp->count -= batch;
Nick Piggin48db57f2006-01-08 01:00:42 -08002065 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002066
2067out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069}
2070
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002071/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002072 * Free a list of 0-order pages
2073 */
Mel Gormanb745bc82014-06-04 16:10:22 -07002074void free_hot_cold_page_list(struct list_head *list, bool cold)
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002075{
2076 struct page *page, *next;
2077
2078 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08002079 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08002080 free_hot_cold_page(page, cold);
2081 }
2082}
2083
2084/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002085 * split_page takes a non-compound higher-order page, and splits it into
2086 * n (1<<order) sub-pages: page[0..n]
2087 * Each sub-page must be freed individually.
2088 *
2089 * Note: this is probably too low level an operation for use in drivers.
2090 * Please consult with lkml before using this in your driver.
2091 */
2092void split_page(struct page *page, unsigned int order)
2093{
2094 int i;
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002095 gfp_t gfp_mask;
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002096
Sasha Levin309381fea2014-01-23 15:52:54 -08002097 VM_BUG_ON_PAGE(PageCompound(page), page);
2098 VM_BUG_ON_PAGE(!page_count(page), page);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002099
2100#ifdef CONFIG_KMEMCHECK
2101 /*
2102 * Split shadow pages too, because free(page[0]) would
2103 * otherwise free the whole shadow.
2104 */
2105 if (kmemcheck_page_is_tracked(page))
2106 split_page(virt_to_page(page[0].shadow), order);
2107#endif
2108
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002109 gfp_mask = get_page_owner_gfp(page);
2110 set_page_owner(page, 0, gfp_mask);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08002111 for (i = 1; i < (1 << order); i++) {
Nick Piggin7835e982006-03-22 00:08:40 -08002112 set_page_refcounted(page + i);
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002113 set_page_owner(page + i, 0, gfp_mask);
Joonsoo Kim48c96a32014-12-12 16:56:01 -08002114 }
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002115}
K. Y. Srinivasan5853ff22013-03-25 15:47:38 -07002116EXPORT_SYMBOL_GPL(split_page);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002117
Joonsoo Kim3c605092014-11-13 15:19:21 -08002118int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07002119{
Mel Gorman748446b2010-05-24 14:32:27 -07002120 unsigned long watermark;
2121 struct zone *zone;
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002122 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07002123
2124 BUG_ON(!PageBuddy(page));
2125
2126 zone = page_zone(page);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002127 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07002128
Minchan Kim194159f2013-02-22 16:33:58 -08002129 if (!is_migrate_isolate(mt)) {
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002130 /* Obey watermarks as if the page was being allocated */
2131 watermark = low_wmark_pages(zone) + (1 << order);
2132 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2133 return 0;
2134
Mel Gorman8fb74b92013-01-11 14:32:16 -08002135 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski2e30abd2012-12-11 16:02:57 -08002136 }
Mel Gorman748446b2010-05-24 14:32:27 -07002137
2138 /* Remove page from free list */
2139 list_del(&page->lru);
2140 zone->free_area[order].nr_free--;
2141 rmv_page_order(page);
Bartlomiej Zolnierkiewicz2139cbe2012-10-08 16:32:00 -07002142
Joonsoo Kime2cfc912015-07-17 16:24:18 -07002143 set_page_owner(page, order, __GFP_MOVABLE);
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002144
Mel Gorman8fb74b92013-01-11 14:32:16 -08002145 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07002146 if (order >= pageblock_order - 1) {
2147 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002148 for (; page < endpage; page += pageblock_nr_pages) {
2149 int mt = get_pageblock_migratetype(page);
Minchan Kim194159f2013-02-22 16:33:58 -08002150 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
Michal Nazarewicz47118af2011-12-29 13:09:50 +01002151 set_pageblock_migratetype(page,
2152 MIGRATE_MOVABLE);
2153 }
Mel Gorman748446b2010-05-24 14:32:27 -07002154 }
2155
Joonsoo Kimf3a14ce2015-07-17 16:24:15 -07002156
Mel Gorman8fb74b92013-01-11 14:32:16 -08002157 return 1UL << order;
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002158}
2159
2160/*
2161 * Similar to split_page except the page is already free. As this is only
2162 * being used for migration, the migratetype of the block also changes.
2163 * As this is called with interrupts disabled, the caller is responsible
2164 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2165 * are enabled.
2166 *
2167 * Note: this is probably too low level an operation for use in drivers.
2168 * Please consult with lkml before using this in your driver.
2169 */
2170int split_free_page(struct page *page)
2171{
2172 unsigned int order;
2173 int nr_pages;
2174
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002175 order = page_order(page);
2176
Mel Gorman8fb74b92013-01-11 14:32:16 -08002177 nr_pages = __isolate_free_page(page, order);
Mel Gorman1fb3f8c2012-10-08 16:29:12 -07002178 if (!nr_pages)
2179 return 0;
2180
2181 /* Split into individual pages */
2182 set_page_refcounted(page);
2183 split_page(page, order);
2184 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07002185}
2186
2187/*
Vlastimil Babka75379192015-02-11 15:25:38 -08002188 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07002190static inline
2191struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002192 struct zone *zone, unsigned int order,
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002193 gfp_t gfp_flags, int alloc_flags, int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194{
2195 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08002196 struct page *page;
Mel Gormanb745bc82014-06-04 16:10:22 -07002197 bool cold = ((gfp_flags & __GFP_COLD) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198
Nick Piggin48db57f2006-01-08 01:00:42 -08002199 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002201 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002204 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2205 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002206 if (list_empty(list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07002207 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002208 pcp->batch, list,
Mel Gormane084b2d2009-07-29 15:02:04 -07002209 migratetype, cold);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002210 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07002211 goto failed;
Mel Gorman535131e62007-10-16 01:25:49 -07002212 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07002213
Mel Gorman5f8dcc22009-09-21 17:03:19 -07002214 if (cold)
2215 page = list_entry(list->prev, struct page, lru);
2216 else
2217 page = list_entry(list->next, struct page, lru);
2218
Mel Gormanb92a6ed2007-10-16 01:25:50 -07002219 list_del(&page->lru);
2220 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002221 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07002222 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
2223 /*
2224 * __GFP_NOFAIL is not to be used in new code.
2225 *
2226 * All __GFP_NOFAIL callers should be fixed so that they
2227 * properly detect and handle allocation failures.
2228 *
2229 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07002230 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07002231 * __GFP_NOFAIL.
2232 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07002233 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07002234 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 spin_lock_irqsave(&zone->lock, flags);
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002236
2237 page = NULL;
2238 if (alloc_flags & ALLOC_HARDER) {
2239 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
2240 if (page)
2241 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2242 }
2243 if (!page)
2244 page = __rmqueue(zone, order, migratetype, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002245 spin_unlock(&zone->lock);
2246 if (!page)
2247 goto failed;
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07002248 __mod_zone_freepage_state(zone, -(1 << order),
Vlastimil Babkabb14c2c2015-09-08 15:01:25 -07002249 get_pcppage_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 }
2251
Johannes Weiner3a025762014-04-07 15:37:48 -07002252 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
Johannes Weinerabe5f972014-10-02 16:21:10 -07002253 if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
Johannes Weiner57054652014-10-09 15:28:17 -07002254 !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
2255 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Johannes Weiner27329362014-03-03 15:38:41 -08002256
Christoph Lameterf8891e52006-06-30 01:55:45 -07002257 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07002258 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002259 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260
Sasha Levin309381fea2014-01-23 15:52:54 -08002261 VM_BUG_ON_PAGE(bad_range(zone, page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08002263
2264failed:
2265 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08002266 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267}
2268
Akinobu Mita933e3122006-12-08 02:39:45 -08002269#ifdef CONFIG_FAIL_PAGE_ALLOC
2270
Akinobu Mitab2588c42011-07-26 16:09:03 -07002271static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08002272 struct fault_attr attr;
2273
Viresh Kumar621a5f72015-09-26 15:04:07 -07002274 bool ignore_gfp_highmem;
Mel Gorman71baba42015-11-06 16:28:28 -08002275 bool ignore_gfp_reclaim;
Akinobu Mita54114992007-07-15 23:40:23 -07002276 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08002277} fail_page_alloc = {
2278 .attr = FAULT_ATTR_INITIALIZER,
Mel Gorman71baba42015-11-06 16:28:28 -08002279 .ignore_gfp_reclaim = true,
Viresh Kumar621a5f72015-09-26 15:04:07 -07002280 .ignore_gfp_highmem = true,
Akinobu Mita54114992007-07-15 23:40:23 -07002281 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08002282};
2283
2284static int __init setup_fail_page_alloc(char *str)
2285{
2286 return setup_fault_attr(&fail_page_alloc.attr, str);
2287}
2288__setup("fail_page_alloc=", setup_fail_page_alloc);
2289
Gavin Shandeaf3862012-07-31 16:41:51 -07002290static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002291{
Akinobu Mita54114992007-07-15 23:40:23 -07002292 if (order < fail_page_alloc.min_order)
Gavin Shandeaf3862012-07-31 16:41:51 -07002293 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002294 if (gfp_mask & __GFP_NOFAIL)
Gavin Shandeaf3862012-07-31 16:41:51 -07002295 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002296 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002297 return false;
Mel Gorman71baba42015-11-06 16:28:28 -08002298 if (fail_page_alloc.ignore_gfp_reclaim &&
2299 (gfp_mask & __GFP_DIRECT_RECLAIM))
Gavin Shandeaf3862012-07-31 16:41:51 -07002300 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002301
2302 return should_fail(&fail_page_alloc.attr, 1 << order);
2303}
2304
2305#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2306
2307static int __init fail_page_alloc_debugfs(void)
2308{
Al Virof4ae40a2011-07-24 04:33:43 -04002309 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08002310 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08002311
Akinobu Mitadd48c082011-08-03 16:21:01 -07002312 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2313 &fail_page_alloc.attr);
2314 if (IS_ERR(dir))
2315 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002316
Akinobu Mitab2588c42011-07-26 16:09:03 -07002317 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
Mel Gorman71baba42015-11-06 16:28:28 -08002318 &fail_page_alloc.ignore_gfp_reclaim))
Akinobu Mitab2588c42011-07-26 16:09:03 -07002319 goto fail;
2320 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2321 &fail_page_alloc.ignore_gfp_highmem))
2322 goto fail;
2323 if (!debugfs_create_u32("min-order", mode, dir,
2324 &fail_page_alloc.min_order))
2325 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08002326
Akinobu Mitab2588c42011-07-26 16:09:03 -07002327 return 0;
2328fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07002329 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08002330
Akinobu Mitab2588c42011-07-26 16:09:03 -07002331 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08002332}
2333
2334late_initcall(fail_page_alloc_debugfs);
2335
2336#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2337
2338#else /* CONFIG_FAIL_PAGE_ALLOC */
2339
Gavin Shandeaf3862012-07-31 16:41:51 -07002340static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
Akinobu Mita933e3122006-12-08 02:39:45 -08002341{
Gavin Shandeaf3862012-07-31 16:41:51 -07002342 return false;
Akinobu Mita933e3122006-12-08 02:39:45 -08002343}
2344
2345#endif /* CONFIG_FAIL_PAGE_ALLOC */
2346
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347/*
Mel Gorman97a16fc2015-11-06 16:28:40 -08002348 * Return true if free base pages are above 'mark'. For high-order checks it
2349 * will return true of the order-0 watermark is reached and there is at least
2350 * one free page of a suitable size. Checking now avoids taking the zone lock
2351 * to check in the allocation paths if no pages are free.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 */
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002353static bool __zone_watermark_ok(struct zone *z, unsigned int order,
2354 unsigned long mark, int classzone_idx, int alloc_flags,
2355 long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356{
Christoph Lameterd23ad422007-02-10 01:43:02 -08002357 long min = mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 int o;
Mel Gorman97a16fc2015-11-06 16:28:40 -08002359 const int alloc_harder = (alloc_flags & ALLOC_HARDER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002361 /* free_pages may go negative - that's OK */
Michal Hockodf0a6da2012-01-10 15:08:02 -08002362 free_pages -= (1 << order) - 1;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002363
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002364 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 min -= min / 2;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002366
2367 /*
2368 * If the caller does not have rights to ALLOC_HARDER then subtract
2369 * the high-atomic reserves. This will over-estimate the size of the
2370 * atomic reserve but it avoids a search.
2371 */
Mel Gorman97a16fc2015-11-06 16:28:40 -08002372 if (likely(!alloc_harder))
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002373 free_pages -= z->nr_reserved_highatomic;
2374 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 min -= min / 4;
Mel Gormane2b19192015-11-06 16:28:09 -08002376
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002377#ifdef CONFIG_CMA
2378 /* If allocation can't use CMA areas don't use free CMA pages */
2379 if (!(alloc_flags & ALLOC_CMA))
Mel Gorman97a16fc2015-11-06 16:28:40 -08002380 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002381#endif
Tomasz Stanislawski026b0812013-06-12 14:05:02 -07002382
Mel Gorman97a16fc2015-11-06 16:28:40 -08002383 /*
2384 * Check watermarks for an order-0 allocation request. If these
2385 * are not met, then a high-order request also cannot go ahead
2386 * even if a suitable page happened to be free.
2387 */
2388 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
Mel Gorman88f5acf2011-01-13 15:45:41 -08002389 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390
Mel Gorman97a16fc2015-11-06 16:28:40 -08002391 /* If this is an order-0 request then the watermark is fine */
2392 if (!order)
2393 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394
Mel Gorman97a16fc2015-11-06 16:28:40 -08002395 /* For a high-order request, check at least one suitable page is free */
2396 for (o = order; o < MAX_ORDER; o++) {
2397 struct free_area *area = &z->free_area[o];
2398 int mt;
2399
2400 if (!area->nr_free)
2401 continue;
2402
2403 if (alloc_harder)
2404 return true;
2405
2406 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
2407 if (!list_empty(&area->free_list[mt]))
2408 return true;
2409 }
2410
2411#ifdef CONFIG_CMA
2412 if ((alloc_flags & ALLOC_CMA) &&
2413 !list_empty(&area->free_list[MIGRATE_CMA])) {
2414 return true;
2415 }
2416#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 }
Mel Gorman97a16fc2015-11-06 16:28:40 -08002418 return false;
Mel Gorman88f5acf2011-01-13 15:45:41 -08002419}
2420
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002421bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002422 int classzone_idx, int alloc_flags)
2423{
2424 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2425 zone_page_state(z, NR_FREE_PAGES));
2426}
2427
Mel Gorman7aeb09f2014-06-04 16:10:21 -07002428bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
Mel Gormane2b19192015-11-06 16:28:09 -08002429 unsigned long mark, int classzone_idx)
Mel Gorman88f5acf2011-01-13 15:45:41 -08002430{
2431 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2432
2433 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2434 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2435
Mel Gormane2b19192015-11-06 16:28:09 -08002436 return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
Mel Gorman88f5acf2011-01-13 15:45:41 -08002437 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438}
2439
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002440#ifdef CONFIG_NUMA
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002441static bool zone_local(struct zone *local_zone, struct zone *zone)
2442{
Johannes Weinerfff40682013-12-20 14:54:12 +00002443 return local_zone->node == zone->node;
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002444}
2445
David Rientjes957f8222012-10-08 16:33:24 -07002446static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2447{
Mel Gorman5f7a75a2014-06-04 16:07:15 -07002448 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2449 RECLAIM_DISTANCE;
David Rientjes957f8222012-10-08 16:33:24 -07002450}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002451#else /* CONFIG_NUMA */
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002452static bool zone_local(struct zone *local_zone, struct zone *zone)
2453{
2454 return true;
2455}
2456
David Rientjes957f8222012-10-08 16:33:24 -07002457static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2458{
2459 return true;
2460}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002461#endif /* CONFIG_NUMA */
2462
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002463static void reset_alloc_batches(struct zone *preferred_zone)
2464{
2465 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2466
2467 do {
2468 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2469 high_wmark_pages(zone) - low_wmark_pages(zone) -
2470 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner57054652014-10-09 15:28:17 -07002471 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002472 } while (zone++ != preferred_zone);
2473}
2474
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002475/*
Paul Jackson0798e512006-12-06 20:31:38 -08002476 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002477 * a page.
2478 */
2479static struct page *
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002480get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2481 const struct alloc_context *ac)
Martin Hicks753ee722005-06-21 17:14:41 -07002482{
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002483 struct zonelist *zonelist = ac->zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07002484 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002485 struct page *page = NULL;
Mel Gorman5117f452009-06-16 15:31:59 -07002486 struct zone *zone;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002487 int nr_fair_skipped = 0;
2488 bool zonelist_rescan;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002489
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002490zonelist_scan:
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002491 zonelist_rescan = false;
2492
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002493 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002494 * Scan zonelist, looking for a zone with enough free.
Vladimir Davydov344736f2014-10-20 15:50:30 +04002495 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002496 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002497 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2498 ac->nodemask) {
Johannes Weinere085dbc2013-09-11 14:20:46 -07002499 unsigned long mark;
2500
Mel Gorman664eedd2014-06-04 16:10:08 -07002501 if (cpusets_enabled() &&
2502 (alloc_flags & ALLOC_CPUSET) &&
Vladimir Davydov344736f2014-10-20 15:50:30 +04002503 !cpuset_zone_allowed(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07002504 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08002505 /*
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002506 * Distribute pages in proportion to the individual
2507 * zone size to ensure fair page aging. The zone a
2508 * page was allocated in should have no effect on the
2509 * time the page has in memory before being reclaimed.
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002510 */
Johannes Weiner3a025762014-04-07 15:37:48 -07002511 if (alloc_flags & ALLOC_FAIR) {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002512 if (!zone_local(ac->preferred_zone, zone))
Mel Gormanf7b5d642014-08-06 16:07:20 -07002513 break;
Johannes Weiner57054652014-10-09 15:28:17 -07002514 if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002515 nr_fair_skipped++;
Johannes Weiner3a025762014-04-07 15:37:48 -07002516 continue;
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002517 }
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07002518 }
2519 /*
Johannes Weinera756cf52012-01-10 15:07:49 -08002520 * When allocating a page cache page for writing, we
2521 * want to get it from a zone that is within its dirty
2522 * limit, such that no single zone holds more than its
2523 * proportional share of globally allowed dirty pages.
2524 * The dirty limits take into account the zone's
2525 * lowmem reserves and high watermark so that kswapd
2526 * should be able to balance it without having to
2527 * write pages from its LRU list.
2528 *
2529 * This may look like it could increase pressure on
2530 * lower zones by failing allocations in higher zones
2531 * before they are full. But the pages that do spill
2532 * over are limited as the lower zones are protected
2533 * by this very same mechanism. It should not become
2534 * a practical burden to them.
2535 *
2536 * XXX: For now, allow allocations to potentially
2537 * exceed the per-zone dirty limit in the slowpath
Mel Gormanc9ab0c42015-11-06 16:28:12 -08002538 * (spread_dirty_pages unset) before going into reclaim,
Johannes Weinera756cf52012-01-10 15:07:49 -08002539 * which is important when on a NUMA setup the allowed
2540 * zones are together not big enough to reach the
2541 * global limit. The proper fix for these situations
2542 * will require awareness of zones in the
2543 * dirty-throttling and the flusher threads.
2544 */
Mel Gormanc9ab0c42015-11-06 16:28:12 -08002545 if (ac->spread_dirty_pages && !zone_dirty_ok(zone))
Mel Gorman800a1e72014-06-04 16:10:06 -07002546 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002547
Johannes Weinere085dbc2013-09-11 14:20:46 -07002548 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2549 if (!zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002550 ac->classzone_idx, alloc_flags)) {
Mel Gormanfa5e0842009-06-16 15:33:22 -07002551 int ret;
2552
Mel Gorman5dab2912014-06-04 16:10:14 -07002553 /* Checked here to keep the fast path fast */
2554 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2555 if (alloc_flags & ALLOC_NO_WATERMARKS)
2556 goto try_this_zone;
2557
David Rientjes957f8222012-10-08 16:33:24 -07002558 if (zone_reclaim_mode == 0 ||
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002559 !zone_allows_reclaim(ac->preferred_zone, zone))
Mel Gormancd38b112011-07-25 17:12:29 -07002560 continue;
2561
Mel Gormanfa5e0842009-06-16 15:33:22 -07002562 ret = zone_reclaim(zone, gfp_mask, order);
2563 switch (ret) {
2564 case ZONE_RECLAIM_NOSCAN:
2565 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07002566 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002567 case ZONE_RECLAIM_FULL:
2568 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07002569 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07002570 default:
2571 /* did we reclaim enough */
Mel Gormanfed27192013-04-29 15:07:57 -07002572 if (zone_watermark_ok(zone, order, mark,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002573 ac->classzone_idx, alloc_flags))
Mel Gormanfed27192013-04-29 15:07:57 -07002574 goto try_this_zone;
2575
Mel Gormanfed27192013-04-29 15:07:57 -07002576 continue;
Paul Jackson0798e512006-12-06 20:31:38 -08002577 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002578 }
2579
Mel Gormanfa5e0842009-06-16 15:33:22 -07002580try_this_zone:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002581 page = buffered_rmqueue(ac->preferred_zone, zone, order,
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002582 gfp_mask, alloc_flags, ac->migratetype);
Vlastimil Babka75379192015-02-11 15:25:38 -08002583 if (page) {
2584 if (prep_new_page(page, order, gfp_mask, alloc_flags))
2585 goto try_this_zone;
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002586
2587 /*
2588 * If this is a high-order atomic allocation then check
2589 * if the pageblock should be reserved for the future
2590 */
2591 if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
2592 reserve_highatomic_pageblock(page, zone, order);
2593
Vlastimil Babka75379192015-02-11 15:25:38 -08002594 return page;
2595 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002596 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002597
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002598 /*
2599 * The first pass makes sure allocations are spread fairly within the
2600 * local node. However, the local node might have free pages left
2601 * after the fairness batches are exhausted, and remote zones haven't
2602 * even been considered yet. Try once more without fairness, and
2603 * include remote zones now, before entering the slowpath and waking
2604 * kswapd: prefer spilling to a remote zone over swapping locally.
2605 */
2606 if (alloc_flags & ALLOC_FAIR) {
2607 alloc_flags &= ~ALLOC_FAIR;
2608 if (nr_fair_skipped) {
2609 zonelist_rescan = true;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002610 reset_alloc_batches(ac->preferred_zone);
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002611 }
2612 if (nr_online_nodes > 1)
2613 zonelist_rescan = true;
2614 }
2615
Mel Gorman4ffeaf32014-08-06 16:07:22 -07002616 if (zonelist_rescan)
2617 goto zonelist_scan;
2618
2619 return NULL;
Martin Hicks753ee722005-06-21 17:14:41 -07002620}
2621
David Rientjes29423e772011-03-22 16:30:47 -07002622/*
2623 * Large machines with many possible nodes should not always dump per-node
2624 * meminfo in irq context.
2625 */
2626static inline bool should_suppress_show_mem(void)
2627{
2628 bool ret = false;
2629
2630#if NODES_SHIFT > 8
2631 ret = in_interrupt();
2632#endif
2633 return ret;
2634}
2635
Dave Hansena238ab52011-05-24 17:12:16 -07002636static DEFINE_RATELIMIT_STATE(nopage_rs,
2637 DEFAULT_RATELIMIT_INTERVAL,
2638 DEFAULT_RATELIMIT_BURST);
2639
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08002640void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...)
Dave Hansena238ab52011-05-24 17:12:16 -07002641{
Dave Hansena238ab52011-05-24 17:12:16 -07002642 unsigned int filter = SHOW_MEM_FILTER_NODES;
2643
Stanislaw Gruszkac0a32fc52012-01-10 15:07:28 -08002644 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2645 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002646 return;
2647
2648 /*
2649 * This documents exceptions given to allocations in certain
2650 * contexts that are allowed to allocate outside current's set
2651 * of allowed nodes.
2652 */
2653 if (!(gfp_mask & __GFP_NOMEMALLOC))
2654 if (test_thread_flag(TIF_MEMDIE) ||
2655 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2656 filter &= ~SHOW_MEM_FILTER_NODES;
Mel Gormand0164ad2015-11-06 16:28:21 -08002657 if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
Dave Hansena238ab52011-05-24 17:12:16 -07002658 filter &= ~SHOW_MEM_FILTER_NODES;
2659
2660 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002661 struct va_format vaf;
2662 va_list args;
2663
Dave Hansena238ab52011-05-24 17:12:16 -07002664 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002665
2666 vaf.fmt = fmt;
2667 vaf.va = &args;
2668
2669 pr_warn("%pV", &vaf);
2670
Dave Hansena238ab52011-05-24 17:12:16 -07002671 va_end(args);
2672 }
2673
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08002674 pr_warn("%s: page allocation failure: order:%u, mode:0x%x\n",
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002675 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002676
2677 dump_stack();
2678 if (!should_suppress_show_mem())
2679 show_mem(filter);
2680}
2681
Mel Gorman11e33f62009-06-16 15:31:57 -07002682static inline struct page *
2683__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002684 const struct alloc_context *ac, unsigned long *did_some_progress)
Mel Gorman11e33f62009-06-16 15:31:57 -07002685{
David Rientjes6e0fc462015-09-08 15:00:36 -07002686 struct oom_control oc = {
2687 .zonelist = ac->zonelist,
2688 .nodemask = ac->nodemask,
2689 .gfp_mask = gfp_mask,
2690 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07002691 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693
Johannes Weiner9879de72015-01-26 12:58:32 -08002694 *did_some_progress = 0;
2695
Johannes Weiner9879de72015-01-26 12:58:32 -08002696 /*
Johannes Weinerdc564012015-06-24 16:57:19 -07002697 * Acquire the oom lock. If that fails, somebody else is
2698 * making progress for us.
Johannes Weiner9879de72015-01-26 12:58:32 -08002699 */
Johannes Weinerdc564012015-06-24 16:57:19 -07002700 if (!mutex_trylock(&oom_lock)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002701 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002702 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703 return NULL;
2704 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002705
Mel Gorman11e33f62009-06-16 15:31:57 -07002706 /*
2707 * Go through the zonelist yet one more time, keep very high watermark
2708 * here, this is only to catch a parallel oom killing, we must fail if
2709 * we're still under heavy pressure.
2710 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002711 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2712 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002713 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002714 goto out;
2715
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002716 if (!(gfp_mask & __GFP_NOFAIL)) {
Johannes Weiner9879de72015-01-26 12:58:32 -08002717 /* Coredumps can quickly deplete all memory reserves */
2718 if (current->flags & PF_DUMPCORE)
2719 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002720 /* The OOM killer will not help higher order allocs */
2721 if (order > PAGE_ALLOC_COSTLY_ORDER)
2722 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002723 /* The OOM killer does not needlessly kill tasks for lowmem */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002724 if (ac->high_zoneidx < ZONE_NORMAL)
David Rientjes03668b32010-08-09 17:18:54 -07002725 goto out;
Johannes Weiner90839052015-06-24 16:57:21 -07002726 /* The OOM killer does not compensate for IO-less reclaim */
Johannes Weinercc873172015-02-27 15:52:09 -08002727 if (!(gfp_mask & __GFP_FS)) {
2728 /*
2729 * XXX: Page reclaim didn't yield anything,
2730 * and the OOM killer can't be invoked, but
Johannes Weiner90839052015-06-24 16:57:21 -07002731 * keep looping as per tradition.
Johannes Weinercc873172015-02-27 15:52:09 -08002732 */
2733 *did_some_progress = 1;
Johannes Weiner9879de72015-01-26 12:58:32 -08002734 goto out;
Johannes Weinercc873172015-02-27 15:52:09 -08002735 }
Johannes Weiner90839052015-06-24 16:57:21 -07002736 if (pm_suspended_storage())
2737 goto out;
David Rientjes4167e9b2015-04-14 15:46:55 -07002738 /* The OOM killer may not free memory on a specific node */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002739 if (gfp_mask & __GFP_THISNODE)
2740 goto out;
2741 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002742 /* Exhausted what can be done so it's blamo time */
David Rientjes6e0fc462015-09-08 15:00:36 -07002743 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL))
Michal Hockoc32b3cb2015-02-11 15:26:24 -08002744 *did_some_progress = 1;
Mel Gorman11e33f62009-06-16 15:31:57 -07002745out:
Johannes Weinerdc564012015-06-24 16:57:19 -07002746 mutex_unlock(&oom_lock);
Mel Gorman11e33f62009-06-16 15:31:57 -07002747 return page;
2748}
2749
Mel Gorman56de7262010-05-24 14:32:30 -07002750#ifdef CONFIG_COMPACTION
2751/* Try memory compaction for high-order allocations before reclaim */
2752static struct page *
2753__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002754 int alloc_flags, const struct alloc_context *ac,
2755 enum migrate_mode mode, int *contended_compaction,
2756 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002757{
Vlastimil Babka53853e22014-10-09 15:27:02 -07002758 unsigned long compact_result;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002759 struct page *page;
Vlastimil Babka53853e22014-10-09 15:27:02 -07002760
Mel Gorman66199712012-01-12 17:19:41 -08002761 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002762 return NULL;
2763
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002764 current->flags |= PF_MEMALLOC;
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08002765 compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2766 mode, contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002767 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002768
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002769 switch (compact_result) {
2770 case COMPACT_DEFERRED:
Vlastimil Babka53853e22014-10-09 15:27:02 -07002771 *deferred_compaction = true;
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002772 /* fall-through */
2773 case COMPACT_SKIPPED:
2774 return NULL;
2775 default:
2776 break;
Mel Gorman56de7262010-05-24 14:32:30 -07002777 }
2778
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002779 /*
2780 * At least in one zone compaction wasn't deferred or skipped, so let's
2781 * count a compaction stall
2782 */
2783 count_vm_event(COMPACTSTALL);
2784
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002785 page = get_page_from_freelist(gfp_mask, order,
2786 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002787
2788 if (page) {
2789 struct zone *zone = page_zone(page);
2790
2791 zone->compact_blockskip_flush = false;
2792 compaction_defer_reset(zone, order, true);
2793 count_vm_event(COMPACTSUCCESS);
2794 return page;
2795 }
2796
2797 /*
Vlastimil Babka98dd3b42014-10-09 15:27:04 -07002798 * It's bad if compaction run occurs and fails. The most likely reason
2799 * is that pages exist, but not enough to satisfy watermarks.
2800 */
2801 count_vm_event(COMPACTFAIL);
2802
2803 cond_resched();
2804
Mel Gorman56de7262010-05-24 14:32:30 -07002805 return NULL;
2806}
2807#else
2808static inline struct page *
2809__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002810 int alloc_flags, const struct alloc_context *ac,
2811 enum migrate_mode mode, int *contended_compaction,
2812 bool *deferred_compaction)
Mel Gorman56de7262010-05-24 14:32:30 -07002813{
2814 return NULL;
2815}
2816#endif /* CONFIG_COMPACTION */
2817
Marek Szyprowskibba90712012-01-25 12:09:52 +01002818/* Perform direct synchronous page reclaim */
2819static int
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002820__perform_reclaim(gfp_t gfp_mask, unsigned int order,
2821 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002822{
Mel Gorman11e33f62009-06-16 15:31:57 -07002823 struct reclaim_state reclaim_state;
Marek Szyprowskibba90712012-01-25 12:09:52 +01002824 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002825
2826 cond_resched();
2827
2828 /* We now go into synchronous reclaim */
2829 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002830 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002831 lockdep_set_current_reclaim_state(gfp_mask);
2832 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002833 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002834
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002835 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
2836 ac->nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002837
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002838 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002839 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002840 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002841
2842 cond_resched();
2843
Marek Szyprowskibba90712012-01-25 12:09:52 +01002844 return progress;
2845}
2846
2847/* The really slow allocator path where we enter direct reclaim */
2848static inline struct page *
2849__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002850 int alloc_flags, const struct alloc_context *ac,
2851 unsigned long *did_some_progress)
Marek Szyprowskibba90712012-01-25 12:09:52 +01002852{
2853 struct page *page = NULL;
2854 bool drained = false;
2855
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002856 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002857 if (unlikely(!(*did_some_progress)))
2858 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002859
Mel Gorman9ee493c2010-09-09 16:38:18 -07002860retry:
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002861 page = get_page_from_freelist(gfp_mask, order,
2862 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002863
2864 /*
2865 * If an allocation failed after direct reclaim, it could be because
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002866 * pages are pinned on the per-cpu lists or in high alloc reserves.
2867 * Shrink them them and try again
Mel Gorman9ee493c2010-09-09 16:38:18 -07002868 */
2869 if (!page && !drained) {
Mel Gorman0aaa29a2015-11-06 16:28:37 -08002870 unreserve_highatomic_pageblock(ac);
Vlastimil Babka93481ff2014-12-10 15:43:01 -08002871 drain_all_pages(NULL);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002872 drained = true;
2873 goto retry;
2874 }
2875
Mel Gorman11e33f62009-06-16 15:31:57 -07002876 return page;
2877}
2878
Mel Gorman11e33f62009-06-16 15:31:57 -07002879/*
2880 * This is called in the allocator slow-path if the allocation request is of
2881 * sufficient urgency to ignore watermarks and take other desperate measures
2882 */
2883static inline struct page *
2884__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002885 const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002886{
2887 struct page *page;
2888
2889 do {
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002890 page = get_page_from_freelist(gfp_mask, order,
2891 ALLOC_NO_WATERMARKS, ac);
Mel Gorman11e33f62009-06-16 15:31:57 -07002892
2893 if (!page && gfp_mask & __GFP_NOFAIL)
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002894 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC,
2895 HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002896 } while (!page && (gfp_mask & __GFP_NOFAIL));
2897
2898 return page;
2899}
2900
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002901static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002902{
2903 struct zoneref *z;
2904 struct zone *zone;
2905
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002906 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
2907 ac->high_zoneidx, ac->nodemask)
2908 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
Mel Gorman11e33f62009-06-16 15:31:57 -07002909}
2910
Peter Zijlstra341ce062009-06-16 15:32:02 -07002911static inline int
2912gfp_to_alloc_flags(gfp_t gfp_mask)
2913{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002914 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002915
Mel Gormana56f57f2009-06-16 15:32:02 -07002916 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002917 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002918
Peter Zijlstra341ce062009-06-16 15:32:02 -07002919 /*
2920 * The caller may dip into page reserves a bit more if the caller
2921 * cannot run direct reclaim, or if the caller has realtime scheduling
2922 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
Mel Gormand0164ad2015-11-06 16:28:21 -08002923 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
Peter Zijlstra341ce062009-06-16 15:32:02 -07002924 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002925 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002926
Mel Gormand0164ad2015-11-06 16:28:21 -08002927 if (gfp_mask & __GFP_ATOMIC) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002928 /*
David Rientjesb104a352014-07-30 16:08:24 -07002929 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2930 * if it can't schedule.
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002931 */
David Rientjesb104a352014-07-30 16:08:24 -07002932 if (!(gfp_mask & __GFP_NOMEMALLOC))
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002933 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002934 /*
David Rientjesb104a352014-07-30 16:08:24 -07002935 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
Vladimir Davydov344736f2014-10-20 15:50:30 +04002936 * comment for __cpuset_node_allowed().
Peter Zijlstra341ce062009-06-16 15:32:02 -07002937 */
2938 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002939 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002940 alloc_flags |= ALLOC_HARDER;
2941
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002942 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2943 if (gfp_mask & __GFP_MEMALLOC)
2944 alloc_flags |= ALLOC_NO_WATERMARKS;
Mel Gorman907aed42012-07-31 16:44:07 -07002945 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2946 alloc_flags |= ALLOC_NO_WATERMARKS;
2947 else if (!in_interrupt() &&
2948 ((current->flags & PF_MEMALLOC) ||
2949 unlikely(test_thread_flag(TIF_MEMDIE))))
Peter Zijlstra341ce062009-06-16 15:32:02 -07002950 alloc_flags |= ALLOC_NO_WATERMARKS;
2951 }
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002952#ifdef CONFIG_CMA
David Rientjes43e7a342014-10-09 15:27:25 -07002953 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
Bartlomiej Zolnierkiewiczd95ea5d2012-10-08 16:32:05 -07002954 alloc_flags |= ALLOC_CMA;
2955#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002956 return alloc_flags;
2957}
2958
Mel Gorman072bb0a2012-07-31 16:43:58 -07002959bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2960{
Mel Gormanb37f1dd2012-07-31 16:44:03 -07002961 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002962}
2963
Mel Gormand0164ad2015-11-06 16:28:21 -08002964static inline bool is_thp_gfp_mask(gfp_t gfp_mask)
2965{
2966 return (gfp_mask & (GFP_TRANSHUGE | __GFP_KSWAPD_RECLAIM)) == GFP_TRANSHUGE;
2967}
2968
Mel Gorman11e33f62009-06-16 15:31:57 -07002969static inline struct page *
2970__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
Vlastimil Babkaa9263752015-02-11 15:25:41 -08002971 struct alloc_context *ac)
Mel Gorman11e33f62009-06-16 15:31:57 -07002972{
Mel Gormand0164ad2015-11-06 16:28:21 -08002973 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
Mel Gorman11e33f62009-06-16 15:31:57 -07002974 struct page *page = NULL;
2975 int alloc_flags;
2976 unsigned long pages_reclaimed = 0;
2977 unsigned long did_some_progress;
David Rientjese0b9dae2014-06-04 16:08:28 -07002978 enum migrate_mode migration_mode = MIGRATE_ASYNC;
Mel Gorman66199712012-01-12 17:19:41 -08002979 bool deferred_compaction = false;
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07002980 int contended_compaction = COMPACT_CONTENDED_NONE;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002981
Christoph Lameter952f3b52006-12-06 20:33:26 -08002982 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002983 * In the slowpath, we sanity check order to avoid ever trying to
2984 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2985 * be using allocators in order of preference for an area that is
2986 * too large.
2987 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002988 if (order >= MAX_ORDER) {
2989 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002990 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002991 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002992
Christoph Lameter952f3b52006-12-06 20:33:26 -08002993 /*
Mel Gormand0164ad2015-11-06 16:28:21 -08002994 * We also sanity check to catch abuse of atomic reserves being used by
2995 * callers that are not in atomic context.
2996 */
2997 if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
2998 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
2999 gfp_mask &= ~__GFP_ATOMIC;
3000
3001 /*
David Rientjes4167e9b2015-04-14 15:46:55 -07003002 * If this allocation cannot block and it is for a specific node, then
3003 * fail early. There's no need to wakeup kswapd or retry for a
3004 * speculative node-specific allocation.
Christoph Lameter952f3b52006-12-06 20:33:26 -08003005 */
Mel Gormand0164ad2015-11-06 16:28:21 -08003006 if (IS_ENABLED(CONFIG_NUMA) && (gfp_mask & __GFP_THISNODE) && !can_direct_reclaim)
Christoph Lameter952f3b52006-12-06 20:33:26 -08003007 goto nopage;
3008
Johannes Weiner9879de72015-01-26 12:58:32 -08003009retry:
Mel Gormand0164ad2015-11-06 16:28:21 -08003010 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003011 wake_all_kswapds(order, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003012
Paul Jackson9bf22292005-09-06 15:18:12 -07003013 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003014 * OK, we're below the kswapd watermark and have kicked background
3015 * reclaim. Now things get more complex, so set up alloc_flags according
3016 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07003017 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07003018 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019
David Rientjesf33261d2011-01-25 15:07:20 -08003020 /*
3021 * Find the true preferred zone if the allocation is unconstrained by
3022 * cpusets.
3023 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003024 if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
Mel Gormand8846372014-06-04 16:10:33 -07003025 struct zoneref *preferred_zoneref;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003026 preferred_zoneref = first_zones_zonelist(ac->zonelist,
3027 ac->high_zoneidx, NULL, &ac->preferred_zone);
3028 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gormand8846372014-06-04 16:10:33 -07003029 }
David Rientjesf33261d2011-01-25 15:07:20 -08003030
Peter Zijlstra341ce062009-06-16 15:32:02 -07003031 /* This is the last chance, in general, before the goto nopage. */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003032 page = get_page_from_freelist(gfp_mask, order,
3033 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08003034 if (page)
3035 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036
Mel Gorman11e33f62009-06-16 15:31:57 -07003037 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07003038 if (alloc_flags & ALLOC_NO_WATERMARKS) {
Mel Gorman183f6372012-07-31 16:44:12 -07003039 /*
3040 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3041 * the allocation is high priority and these type of
3042 * allocations are system rather than user orientated
3043 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003044 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
Mel Gorman183f6372012-07-31 16:44:12 -07003045
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003046 page = __alloc_pages_high_priority(gfp_mask, order, ac);
3047
Mel Gormancfd19c52012-07-31 16:44:10 -07003048 if (page) {
Peter Zijlstra341ce062009-06-16 15:32:02 -07003049 goto got_pg;
Mel Gormancfd19c52012-07-31 16:44:10 -07003050 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051 }
3052
Mel Gormand0164ad2015-11-06 16:28:21 -08003053 /* Caller is not willing to reclaim, we can't balance anything */
3054 if (!can_direct_reclaim) {
David Rientjesaed0a0e2014-01-21 15:51:12 -08003055 /*
3056 * All existing users of the deprecated __GFP_NOFAIL are
3057 * blockable, so warn of any new users that actually allow this
3058 * type of allocation to fail.
3059 */
3060 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 goto nopage;
David Rientjesaed0a0e2014-01-21 15:51:12 -08003062 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063
Peter Zijlstra341ce062009-06-16 15:32:02 -07003064 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08003065 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07003066 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067
David Rientjes6583bb62009-07-29 15:02:06 -07003068 /* Avoid allocations with no watermarks from looping endlessly */
3069 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3070 goto nopage;
3071
Mel Gorman77f1fe62011-01-13 15:45:57 -08003072 /*
3073 * Try direct compaction. The first pass is asynchronous. Subsequent
3074 * attempts after direct reclaim are synchronous
3075 */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003076 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3077 migration_mode,
3078 &contended_compaction,
Vlastimil Babka53853e22014-10-09 15:27:02 -07003079 &deferred_compaction);
Mel Gorman56de7262010-05-24 14:32:30 -07003080 if (page)
3081 goto got_pg;
David Rientjes75f30862014-06-04 16:08:30 -07003082
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003083 /* Checks for THP-specific high-order allocations */
Mel Gormand0164ad2015-11-06 16:28:21 -08003084 if (is_thp_gfp_mask(gfp_mask)) {
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07003085 /*
3086 * If compaction is deferred for high-order allocations, it is
3087 * because sync compaction recently failed. If this is the case
3088 * and the caller requested a THP allocation, we do not want
3089 * to heavily disrupt the system, so we fail the allocation
3090 * instead of entering direct reclaim.
3091 */
3092 if (deferred_compaction)
3093 goto nopage;
3094
3095 /*
3096 * In all zones where compaction was attempted (and not
3097 * deferred or skipped), lock contention has been detected.
3098 * For THP allocation we do not want to disrupt the others
3099 * so we fallback to base pages instead.
3100 */
3101 if (contended_compaction == COMPACT_CONTENDED_LOCK)
3102 goto nopage;
3103
3104 /*
3105 * If compaction was aborted due to need_resched(), we do not
3106 * want to further increase allocation latency, unless it is
3107 * khugepaged trying to collapse.
3108 */
3109 if (contended_compaction == COMPACT_CONTENDED_SCHED
3110 && !(current->flags & PF_KTHREAD))
3111 goto nopage;
3112 }
Mel Gorman66199712012-01-12 17:19:41 -08003113
David Rientjes8fe78042014-08-06 16:07:54 -07003114 /*
3115 * It can become very expensive to allocate transparent hugepages at
3116 * fault, so use asynchronous memory compaction for THP unless it is
3117 * khugepaged trying to collapse.
3118 */
Mel Gormand0164ad2015-11-06 16:28:21 -08003119 if (!is_thp_gfp_mask(gfp_mask) || (current->flags & PF_KTHREAD))
David Rientjes8fe78042014-08-06 16:07:54 -07003120 migration_mode = MIGRATE_SYNC_LIGHT;
3121
Mel Gorman11e33f62009-06-16 15:31:57 -07003122 /* Try direct reclaim and then allocating */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003123 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3124 &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07003125 if (page)
3126 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127
Johannes Weiner90839052015-06-24 16:57:21 -07003128 /* Do not loop if specifically requested */
3129 if (gfp_mask & __GFP_NORETRY)
3130 goto noretry;
3131
3132 /* Keep reclaiming pages as long as there is reasonable progress */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07003133 pages_reclaimed += did_some_progress;
Johannes Weiner90839052015-06-24 16:57:21 -07003134 if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) ||
3135 ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) {
Mel Gorman11e33f62009-06-16 15:31:57 -07003136 /* Wait for some write requests to complete then retry */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003137 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
Johannes Weiner9879de72015-01-26 12:58:32 -08003138 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 }
3140
Johannes Weiner90839052015-06-24 16:57:21 -07003141 /* Reclaim has failed us, start killing things */
3142 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3143 if (page)
3144 goto got_pg;
3145
3146 /* Retry as long as the OOM killer is making progress */
3147 if (did_some_progress)
3148 goto retry;
3149
3150noretry:
3151 /*
3152 * High-order allocations do not necessarily loop after
3153 * direct reclaim and reclaim/compaction depends on compaction
3154 * being called after reclaim so call directly if necessary
3155 */
3156 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags,
3157 ac, migration_mode,
3158 &contended_compaction,
3159 &deferred_compaction);
3160 if (page)
3161 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07003163 warn_alloc_failed(gfp_mask, order, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164got_pg:
Mel Gorman072bb0a2012-07-31 16:43:58 -07003165 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166}
Mel Gorman11e33f62009-06-16 15:31:57 -07003167
3168/*
3169 * This is the 'heart' of the zoned buddy allocator.
3170 */
3171struct page *
3172__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3173 struct zonelist *zonelist, nodemask_t *nodemask)
3174{
Mel Gormand8846372014-06-04 16:10:33 -07003175 struct zoneref *preferred_zoneref;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003176 struct page *page = NULL;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003177 unsigned int cpuset_mems_cookie;
Johannes Weiner3a025762014-04-07 15:37:48 -07003178 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
Andrew Morton91fbdc02015-02-11 15:25:04 -08003179 gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003180 struct alloc_context ac = {
3181 .high_zoneidx = gfp_zone(gfp_mask),
3182 .nodemask = nodemask,
3183 .migratetype = gfpflags_to_migratetype(gfp_mask),
3184 };
Mel Gorman11e33f62009-06-16 15:31:57 -07003185
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003186 gfp_mask &= gfp_allowed_mask;
3187
Mel Gorman11e33f62009-06-16 15:31:57 -07003188 lockdep_trace_alloc(gfp_mask);
3189
Mel Gormand0164ad2015-11-06 16:28:21 -08003190 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
Mel Gorman11e33f62009-06-16 15:31:57 -07003191
3192 if (should_fail_alloc_page(gfp_mask, order))
3193 return NULL;
3194
3195 /*
3196 * Check the zones suitable for the gfp_mask contain at least one
3197 * valid zone. It's possible to have an empty zonelist as a result
David Rientjes4167e9b2015-04-14 15:46:55 -07003198 * of __GFP_THISNODE and a memoryless node
Mel Gorman11e33f62009-06-16 15:31:57 -07003199 */
3200 if (unlikely(!zonelist->_zonerefs->zone))
3201 return NULL;
3202
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003203 if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
Vlastimil Babka21bb9bd2014-10-09 15:26:51 -07003204 alloc_flags |= ALLOC_CMA;
3205
Mel Gormancc9a6c82012-03-21 16:34:11 -07003206retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07003207 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003208
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003209 /* We set it here, as __alloc_pages_slowpath might have changed it */
3210 ac.zonelist = zonelist;
Mel Gormanc9ab0c42015-11-06 16:28:12 -08003211
3212 /* Dirty zone balancing only done in the fast path */
3213 ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE);
3214
Mel Gorman5117f452009-06-16 15:31:59 -07003215 /* The preferred zone is used for statistics later */
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003216 preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
3217 ac.nodemask ? : &cpuset_current_mems_allowed,
3218 &ac.preferred_zone);
3219 if (!ac.preferred_zone)
Mel Gormancc9a6c82012-03-21 16:34:11 -07003220 goto out;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003221 ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
Mel Gorman5117f452009-06-16 15:31:59 -07003222
3223 /* First allocation attempt */
Andrew Morton91fbdc02015-02-11 15:25:04 -08003224 alloc_mask = gfp_mask|__GFP_HARDWALL;
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003225 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08003226 if (unlikely(!page)) {
3227 /*
3228 * Runtime PM, block IO and its error handling path
3229 * can deadlock because I/O on the device might not
3230 * complete.
3231 */
Andrew Morton91fbdc02015-02-11 15:25:04 -08003232 alloc_mask = memalloc_noio_flags(gfp_mask);
Mel Gormanc9ab0c42015-11-06 16:28:12 -08003233 ac.spread_dirty_pages = false;
Andrew Morton91fbdc02015-02-11 15:25:04 -08003234
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003235 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
Ming Lei21caf2f2013-02-22 16:34:08 -08003236 }
Mel Gorman11e33f62009-06-16 15:31:57 -07003237
Xishi Qiu23f086f2015-02-11 15:25:07 -08003238 if (kmemcheck_enabled && page)
3239 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3240
Vlastimil Babkaa9263752015-02-11 15:25:41 -08003241 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07003242
3243out:
3244 /*
3245 * When updating a task's mems_allowed, it is possible to race with
3246 * parallel threads in such a way that an allocation can fail while
3247 * the mask is being updated. If a page allocation is about to fail,
3248 * check if the cpuset changed during allocation and if so, retry.
3249 */
Mel Gormand26914d2014-04-03 14:47:24 -07003250 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07003251 goto retry_cpuset;
3252
Mel Gorman11e33f62009-06-16 15:31:57 -07003253 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254}
Mel Gormand2391712009-06-16 15:31:52 -07003255EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003256
3257/*
3258 * Common helper functions.
3259 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08003260unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261{
Akinobu Mita945a1112009-09-21 17:01:47 -07003262 struct page *page;
3263
3264 /*
3265 * __get_free_pages() returns a 32-bit address, which cannot represent
3266 * a highmem page
3267 */
3268 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3269
Linus Torvalds1da177e2005-04-16 15:20:36 -07003270 page = alloc_pages(gfp_mask, order);
3271 if (!page)
3272 return 0;
3273 return (unsigned long) page_address(page);
3274}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275EXPORT_SYMBOL(__get_free_pages);
3276
Harvey Harrison920c7a52008-02-04 22:29:26 -08003277unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278{
Akinobu Mita945a1112009-09-21 17:01:47 -07003279 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281EXPORT_SYMBOL(get_zeroed_page);
3282
Harvey Harrison920c7a52008-02-04 22:29:26 -08003283void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003284{
Nick Pigginb5810032005-10-29 18:16:12 -07003285 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286 if (order == 0)
Mel Gormanb745bc82014-06-04 16:10:22 -07003287 free_hot_cold_page(page, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288 else
3289 __free_pages_ok(page, order);
3290 }
3291}
3292
3293EXPORT_SYMBOL(__free_pages);
3294
Harvey Harrison920c7a52008-02-04 22:29:26 -08003295void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296{
3297 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07003298 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299 __free_pages(virt_to_page((void *)addr), order);
3300 }
3301}
3302
3303EXPORT_SYMBOL(free_pages);
3304
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003305/*
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003306 * Page Fragment:
3307 * An arbitrary-length arbitrary-offset area of memory which resides
3308 * within a 0 or higher order page. Multiple fragments within that page
3309 * are individually refcounted, in the page's reference counter.
3310 *
3311 * The page_frag functions below provide a simple allocation framework for
3312 * page fragments. This is used by the network stack and network device
3313 * drivers to provide a backing region of memory for use as either an
3314 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3315 */
3316static struct page *__page_frag_refill(struct page_frag_cache *nc,
3317 gfp_t gfp_mask)
3318{
3319 struct page *page = NULL;
3320 gfp_t gfp = gfp_mask;
3321
3322#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3323 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3324 __GFP_NOMEMALLOC;
3325 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3326 PAGE_FRAG_CACHE_MAX_ORDER);
3327 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3328#endif
3329 if (unlikely(!page))
3330 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3331
3332 nc->va = page ? page_address(page) : NULL;
3333
3334 return page;
3335}
3336
3337void *__alloc_page_frag(struct page_frag_cache *nc,
3338 unsigned int fragsz, gfp_t gfp_mask)
3339{
3340 unsigned int size = PAGE_SIZE;
3341 struct page *page;
3342 int offset;
3343
3344 if (unlikely(!nc->va)) {
3345refill:
3346 page = __page_frag_refill(nc, gfp_mask);
3347 if (!page)
3348 return NULL;
3349
3350#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3351 /* if size can vary use size else just use PAGE_SIZE */
3352 size = nc->size;
3353#endif
3354 /* Even if we own the page, we do not use atomic_set().
3355 * This would break get_page_unless_zero() users.
3356 */
3357 atomic_add(size - 1, &page->_count);
3358
3359 /* reset page count bias and offset to start of new frag */
Michal Hocko2f064f32015-08-21 14:11:51 -07003360 nc->pfmemalloc = page_is_pfmemalloc(page);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -07003361 nc->pagecnt_bias = size;
3362 nc->offset = size;
3363 }
3364
3365 offset = nc->offset - fragsz;
3366 if (unlikely(offset < 0)) {
3367 page = virt_to_page(nc->va);
3368
3369 if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count))
3370 goto refill;
3371
3372#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3373 /* if size can vary use size else just use PAGE_SIZE */
3374 size = nc->size;
3375#endif
3376 /* OK, page count is 0, we can safely set it */
3377 atomic_set(&page->_count, size);
3378
3379 /* reset page count bias and offset to start of new frag */
3380 nc->pagecnt_bias = size;
3381 offset = size - fragsz;
3382 }
3383
3384 nc->pagecnt_bias--;
3385 nc->offset = offset;
3386
3387 return nc->va + offset;
3388}
3389EXPORT_SYMBOL(__alloc_page_frag);
3390
3391/*
3392 * Frees a page fragment allocated out of either a compound or order 0 page.
3393 */
3394void __free_page_frag(void *addr)
3395{
3396 struct page *page = virt_to_head_page(addr);
3397
3398 if (unlikely(put_page_testzero(page)))
3399 __free_pages_ok(page, compound_order(page));
3400}
3401EXPORT_SYMBOL(__free_page_frag);
3402
3403/*
Vladimir Davydov52383432014-06-04 16:06:39 -07003404 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3405 * of the current memory cgroup.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003406 *
Vladimir Davydov52383432014-06-04 16:06:39 -07003407 * It should be used when the caller would like to use kmalloc, but since the
3408 * allocation is large, it has to fall back to the page allocator.
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003409 */
Vladimir Davydov52383432014-06-04 16:06:39 -07003410struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
3411{
3412 struct page *page;
Vladimir Davydov52383432014-06-04 16:06:39 -07003413
Vladimir Davydov52383432014-06-04 16:06:39 -07003414 page = alloc_pages(gfp_mask, order);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08003415 if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
3416 __free_pages(page, order);
3417 page = NULL;
3418 }
Vladimir Davydov52383432014-06-04 16:06:39 -07003419 return page;
3420}
3421
3422struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
3423{
3424 struct page *page;
Vladimir Davydov52383432014-06-04 16:06:39 -07003425
Vladimir Davydov52383432014-06-04 16:06:39 -07003426 page = alloc_pages_node(nid, gfp_mask, order);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08003427 if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
3428 __free_pages(page, order);
3429 page = NULL;
3430 }
Vladimir Davydov52383432014-06-04 16:06:39 -07003431 return page;
3432}
3433
3434/*
3435 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3436 * alloc_kmem_pages.
3437 */
3438void __free_kmem_pages(struct page *page, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003439{
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08003440 memcg_kmem_uncharge(page, order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003441 __free_pages(page, order);
3442}
3443
Vladimir Davydov52383432014-06-04 16:06:39 -07003444void free_kmem_pages(unsigned long addr, unsigned int order)
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003445{
3446 if (addr != 0) {
3447 VM_BUG_ON(!virt_addr_valid((void *)addr));
Vladimir Davydov52383432014-06-04 16:06:39 -07003448 __free_kmem_pages(virt_to_page((void *)addr), order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -08003449 }
3450}
3451
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08003452static void *make_alloc_exact(unsigned long addr, unsigned int order,
3453 size_t size)
Andi Kleenee85c2e2011-05-11 15:13:34 -07003454{
3455 if (addr) {
3456 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3457 unsigned long used = addr + PAGE_ALIGN(size);
3458
3459 split_page(virt_to_page((void *)addr), order);
3460 while (used < alloc_end) {
3461 free_page(used);
3462 used += PAGE_SIZE;
3463 }
3464 }
3465 return (void *)addr;
3466}
3467
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003468/**
3469 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3470 * @size: the number of bytes to allocate
3471 * @gfp_mask: GFP flags for the allocation
3472 *
3473 * This function is similar to alloc_pages(), except that it allocates the
3474 * minimum number of pages to satisfy the request. alloc_pages() can only
3475 * allocate memory in power-of-two pages.
3476 *
3477 * This function is also limited by MAX_ORDER.
3478 *
3479 * Memory allocated by this function must be released by free_pages_exact().
3480 */
3481void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3482{
3483 unsigned int order = get_order(size);
3484 unsigned long addr;
3485
3486 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07003487 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003488}
3489EXPORT_SYMBOL(alloc_pages_exact);
3490
3491/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07003492 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3493 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07003494 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07003495 * @size: the number of bytes to allocate
3496 * @gfp_mask: GFP flags for the allocation
3497 *
3498 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3499 * back.
Andi Kleenee85c2e2011-05-11 15:13:34 -07003500 */
Fabian Fredericke1931812014-08-06 16:04:59 -07003501void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
Andi Kleenee85c2e2011-05-11 15:13:34 -07003502{
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08003503 unsigned int order = get_order(size);
Andi Kleenee85c2e2011-05-11 15:13:34 -07003504 struct page *p = alloc_pages_node(nid, gfp_mask, order);
3505 if (!p)
3506 return NULL;
3507 return make_alloc_exact((unsigned long)page_address(p), order, size);
3508}
Andi Kleenee85c2e2011-05-11 15:13:34 -07003509
3510/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07003511 * free_pages_exact - release memory allocated via alloc_pages_exact()
3512 * @virt: the value returned by alloc_pages_exact.
3513 * @size: size of allocation, same value as passed to alloc_pages_exact().
3514 *
3515 * Release the memory allocated by a previous call to alloc_pages_exact.
3516 */
3517void free_pages_exact(void *virt, size_t size)
3518{
3519 unsigned long addr = (unsigned long)virt;
3520 unsigned long end = addr + PAGE_ALIGN(size);
3521
3522 while (addr < end) {
3523 free_page(addr);
3524 addr += PAGE_SIZE;
3525 }
3526}
3527EXPORT_SYMBOL(free_pages_exact);
3528
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003529/**
3530 * nr_free_zone_pages - count number of pages beyond high watermark
3531 * @offset: The zone index of the highest zone
3532 *
3533 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3534 * high watermark within all zones at or below a given zone index. For each
3535 * zone, the number of pages is calculated as:
Jiang Liu834405c2013-07-03 15:03:04 -07003536 * managed_pages - high_pages
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003537 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003538static unsigned long nr_free_zone_pages(int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003539{
Mel Gormandd1a2392008-04-28 02:12:17 -07003540 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003541 struct zone *zone;
3542
Martin J. Blighe310fd42005-07-29 22:59:18 -07003543 /* Just pick one node, since fallback list is circular */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003544 unsigned long sum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545
Mel Gorman0e884602008-04-28 02:12:14 -07003546 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547
Mel Gorman54a6eb52008-04-28 02:12:16 -07003548 for_each_zone_zonelist(zone, z, zonelist, offset) {
Jiang Liub40da042013-02-22 16:33:52 -08003549 unsigned long size = zone->managed_pages;
Mel Gorman41858962009-06-16 15:32:12 -07003550 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07003551 if (size > high)
3552 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003553 }
3554
3555 return sum;
3556}
3557
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003558/**
3559 * nr_free_buffer_pages - count number of pages beyond high watermark
3560 *
3561 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3562 * watermark within ZONE_DMA and ZONE_NORMAL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003563 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003564unsigned long nr_free_buffer_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565{
Al Viroaf4ca452005-10-21 02:55:38 -04003566 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567}
Meelap Shahc2f1a552007-07-17 04:04:39 -07003568EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003569
Zhang Yanfeie0fb5812013-02-22 16:35:54 -08003570/**
3571 * nr_free_pagecache_pages - count number of pages beyond high watermark
3572 *
3573 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3574 * high watermark within all zones.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003575 */
Zhang Yanfeiebec3862013-02-22 16:35:43 -08003576unsigned long nr_free_pagecache_pages(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003577{
Mel Gorman2a1e2742007-07-17 04:03:12 -07003578 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07003580
3581static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582{
Kirill A. Shutemove5adfff2012-12-11 16:00:29 -08003583 if (IS_ENABLED(CONFIG_NUMA))
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08003584 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003586
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587void si_meminfo(struct sysinfo *val)
3588{
3589 val->totalram = totalram_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003590 val->sharedram = global_page_state(NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003591 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003592 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593 val->totalhigh = totalhigh_pages;
3594 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595 val->mem_unit = PAGE_SIZE;
3596}
3597
3598EXPORT_SYMBOL(si_meminfo);
3599
3600#ifdef CONFIG_NUMA
3601void si_meminfo_node(struct sysinfo *val, int nid)
3602{
Jiang Liucdd91a72013-07-03 15:03:27 -07003603 int zone_type; /* needs to be signed */
3604 unsigned long managed_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003605 pg_data_t *pgdat = NODE_DATA(nid);
3606
Jiang Liucdd91a72013-07-03 15:03:27 -07003607 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3608 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3609 val->totalram = managed_pages;
Rafael Aquinicc7452b2014-08-06 16:06:38 -07003610 val->sharedram = node_page_state(nid, NR_SHMEM);
Christoph Lameterd23ad422007-02-10 01:43:02 -08003611 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003612#ifdef CONFIG_HIGHMEM
Jiang Liub40da042013-02-22 16:33:52 -08003613 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08003614 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3615 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003616#else
3617 val->totalhigh = 0;
3618 val->freehigh = 0;
3619#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003620 val->mem_unit = PAGE_SIZE;
3621}
3622#endif
3623
David Rientjesddd588b2011-03-22 16:30:46 -07003624/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07003625 * Determine whether the node should be displayed or not, depending on whether
3626 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07003627 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003628bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07003629{
3630 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003631 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07003632
3633 if (!(flags & SHOW_MEM_FILTER_NODES))
3634 goto out;
3635
Mel Gormancc9a6c82012-03-21 16:34:11 -07003636 do {
Mel Gormand26914d2014-04-03 14:47:24 -07003637 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07003638 ret = !node_isset(nid, cpuset_current_mems_allowed);
Mel Gormand26914d2014-04-03 14:47:24 -07003639 } while (read_mems_allowed_retry(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07003640out:
3641 return ret;
3642}
3643
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644#define K(x) ((x) << (PAGE_SHIFT-10))
3645
Rabin Vincent377e4f12012-12-11 16:00:24 -08003646static void show_migration_types(unsigned char type)
3647{
3648 static const char types[MIGRATE_TYPES] = {
3649 [MIGRATE_UNMOVABLE] = 'U',
3650 [MIGRATE_RECLAIMABLE] = 'E',
3651 [MIGRATE_MOVABLE] = 'M',
Rabin Vincent377e4f12012-12-11 16:00:24 -08003652#ifdef CONFIG_CMA
3653 [MIGRATE_CMA] = 'C',
3654#endif
Minchan Kim194159f2013-02-22 16:33:58 -08003655#ifdef CONFIG_MEMORY_ISOLATION
Rabin Vincent377e4f12012-12-11 16:00:24 -08003656 [MIGRATE_ISOLATE] = 'I',
Minchan Kim194159f2013-02-22 16:33:58 -08003657#endif
Rabin Vincent377e4f12012-12-11 16:00:24 -08003658 };
3659 char tmp[MIGRATE_TYPES + 1];
3660 char *p = tmp;
3661 int i;
3662
3663 for (i = 0; i < MIGRATE_TYPES; i++) {
3664 if (type & (1 << i))
3665 *p++ = types[i];
3666 }
3667
3668 *p = '\0';
3669 printk("(%s) ", tmp);
3670}
3671
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672/*
3673 * Show free area list (used inside shift_scroll-lock stuff)
3674 * We also calculate the percentage fragmentation. We do this by counting the
3675 * memory on each free list with the exception of the first item on the list.
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003676 *
3677 * Bits in @filter:
3678 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3679 * cpuset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07003681void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682{
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003683 unsigned long free_pcp = 0;
Jes Sorensenc7241912006-09-27 01:50:05 -07003684 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003685 struct zone *zone;
3686
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003687 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07003688 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003689 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003690
Konstantin Khlebnikov761b0672015-04-14 15:45:32 -07003691 for_each_online_cpu(cpu)
3692 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003693 }
3694
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003695 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3696 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003697 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3698 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003699 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003700 " free:%lu free_pcp:%lu free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003701 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003702 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003703 global_page_state(NR_ISOLATED_ANON),
3704 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003705 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003706 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003707 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07003708 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07003709 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07003710 global_page_state(NR_UNSTABLE_NFS),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07003711 global_page_state(NR_SLAB_RECLAIMABLE),
3712 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07003713 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003714 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08003715 global_page_state(NR_PAGETABLE),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003716 global_page_state(NR_BOUNCE),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003717 global_page_state(NR_FREE_PAGES),
3718 free_pcp,
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003719 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003720
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003721 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722 int i;
3723
David Rientjes7bf02ea2011-05-24 17:11:16 -07003724 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003725 continue;
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003726
3727 free_pcp = 0;
3728 for_each_online_cpu(cpu)
3729 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3730
Linus Torvalds1da177e2005-04-16 15:20:36 -07003731 show_node(zone);
3732 printk("%s"
3733 " free:%lukB"
3734 " min:%lukB"
3735 " low:%lukB"
3736 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003737 " active_anon:%lukB"
3738 " inactive_anon:%lukB"
3739 " active_file:%lukB"
3740 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003741 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003742 " isolated(anon):%lukB"
3743 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003744 " present:%lukB"
Jiang Liu9feedc92012-12-12 13:52:12 -08003745 " managed:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003746 " mlocked:%lukB"
3747 " dirty:%lukB"
3748 " writeback:%lukB"
3749 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003750 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003751 " slab_reclaimable:%lukB"
3752 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003753 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003754 " pagetables:%lukB"
3755 " unstable:%lukB"
3756 " bounce:%lukB"
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003757 " free_pcp:%lukB"
3758 " local_pcp:%ukB"
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003759 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003760 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003761 " pages_scanned:%lu"
3762 " all_unreclaimable? %s"
3763 "\n",
3764 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003765 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003766 K(min_wmark_pages(zone)),
3767 K(low_wmark_pages(zone)),
3768 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003769 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3770 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3771 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3772 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003773 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003774 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3775 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003776 K(zone->present_pages),
Jiang Liu9feedc92012-12-12 13:52:12 -08003777 K(zone->managed_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003778 K(zone_page_state(zone, NR_MLOCK)),
3779 K(zone_page_state(zone, NR_FILE_DIRTY)),
3780 K(zone_page_state(zone, NR_WRITEBACK)),
3781 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003782 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003783 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3784 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003785 zone_page_state(zone, NR_KERNEL_STACK) *
3786 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003787 K(zone_page_state(zone, NR_PAGETABLE)),
3788 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3789 K(zone_page_state(zone, NR_BOUNCE)),
Konstantin Khlebnikovd1bfcdb2015-04-14 15:45:30 -07003790 K(free_pcp),
3791 K(this_cpu_read(zone->pageset->pcp.count)),
Bartlomiej Zolnierkiewiczd1ce7492012-10-08 16:32:02 -07003792 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003793 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Mel Gorman0d5d8232014-08-06 16:07:16 -07003794 K(zone_page_state(zone, NR_PAGES_SCANNED)),
Lisa Du6e543d52013-09-11 14:22:36 -07003795 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003796 );
3797 printk("lowmem_reserve[]:");
3798 for (i = 0; i < MAX_NR_ZONES; i++)
Mel Gorman3484b2d2014-08-06 16:07:14 -07003799 printk(" %ld", zone->lowmem_reserve[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800 printk("\n");
3801 }
3802
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003803 for_each_populated_zone(zone) {
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08003804 unsigned int order;
3805 unsigned long nr[MAX_ORDER], flags, total = 0;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003806 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807
David Rientjes7bf02ea2011-05-24 17:11:16 -07003808 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003809 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003810 show_node(zone);
3811 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812
3813 spin_lock_irqsave(&zone->lock, flags);
3814 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent377e4f12012-12-11 16:00:24 -08003815 struct free_area *area = &zone->free_area[order];
3816 int type;
3817
3818 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003819 total += nr[order] << order;
Rabin Vincent377e4f12012-12-11 16:00:24 -08003820
3821 types[order] = 0;
3822 for (type = 0; type < MIGRATE_TYPES; type++) {
3823 if (!list_empty(&area->free_list[type]))
3824 types[order] |= 1 << type;
3825 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003826 }
3827 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003828 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003829 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent377e4f12012-12-11 16:00:24 -08003830 if (nr[order])
3831 show_migration_types(types[order]);
3832 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003833 printk("= %lukB\n", K(total));
3834 }
3835
David Rientjes949f7ec2013-04-29 15:07:48 -07003836 hugetlb_show_meminfo();
3837
Larry Woodmane6f36022008-02-04 22:29:30 -08003838 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3839
Linus Torvalds1da177e2005-04-16 15:20:36 -07003840 show_swap_cache_info();
3841}
3842
Mel Gorman19770b32008-04-28 02:12:18 -07003843static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3844{
3845 zoneref->zone = zone;
3846 zoneref->zone_idx = zone_idx(zone);
3847}
3848
Linus Torvalds1da177e2005-04-16 15:20:36 -07003849/*
3850 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003851 *
3852 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003854static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003855 int nr_zones)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003856{
Christoph Lameter1a932052006-01-06 00:11:16 -08003857 struct zone *zone;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003858 enum zone_type zone_type = MAX_NR_ZONES;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003859
3860 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003861 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003862 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003863 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003864 zoneref_set_zone(zone,
3865 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003866 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003867 }
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003868 } while (zone_type);
Zhang Yanfeibc732f12013-07-08 16:00:06 -07003869
Christoph Lameter070f8032006-01-06 00:11:19 -08003870 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003871}
3872
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003873
3874/*
3875 * zonelist_order:
3876 * 0 = automatic detection of better ordering.
3877 * 1 = order by ([node] distance, -zonetype)
3878 * 2 = order by (-zonetype, [node] distance)
3879 *
3880 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3881 * the same zonelist. So only NUMA can configure this param.
3882 */
3883#define ZONELIST_ORDER_DEFAULT 0
3884#define ZONELIST_ORDER_NODE 1
3885#define ZONELIST_ORDER_ZONE 2
3886
3887/* zonelist order in the kernel.
3888 * set_zonelist_order() will set this to NODE or ZONE.
3889 */
3890static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3891static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3892
3893
Linus Torvalds1da177e2005-04-16 15:20:36 -07003894#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003895/* The value user specified ....changed by config */
3896static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3897/* string for sysctl */
3898#define NUMA_ZONELIST_ORDER_LEN 16
3899char numa_zonelist_order[16] = "default";
3900
3901/*
3902 * interface for configure zonelist ordering.
3903 * command line option "numa_zonelist_order"
3904 * = "[dD]efault - default, automatic configuration.
3905 * = "[nN]ode - order by node locality, then by zone within node
3906 * = "[zZ]one - order by zone, then by locality within zone
3907 */
3908
3909static int __parse_numa_zonelist_order(char *s)
3910{
3911 if (*s == 'd' || *s == 'D') {
3912 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3913 } else if (*s == 'n' || *s == 'N') {
3914 user_zonelist_order = ZONELIST_ORDER_NODE;
3915 } else if (*s == 'z' || *s == 'Z') {
3916 user_zonelist_order = ZONELIST_ORDER_ZONE;
3917 } else {
3918 printk(KERN_WARNING
3919 "Ignoring invalid numa_zonelist_order value: "
3920 "%s\n", s);
3921 return -EINVAL;
3922 }
3923 return 0;
3924}
3925
3926static __init int setup_numa_zonelist_order(char *s)
3927{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003928 int ret;
3929
3930 if (!s)
3931 return 0;
3932
3933 ret = __parse_numa_zonelist_order(s);
3934 if (ret == 0)
3935 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3936
3937 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003938}
3939early_param("numa_zonelist_order", setup_numa_zonelist_order);
3940
3941/*
3942 * sysctl handler for numa_zonelist_order
3943 */
Joe Perchescccad5b2014-06-06 14:38:09 -07003944int numa_zonelist_order_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003945 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003946 loff_t *ppos)
3947{
3948 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3949 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003950 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003951
Andi Kleen443c6f12009-12-23 21:00:47 +01003952 mutex_lock(&zl_order_mutex);
Chen Gangdacbde02013-07-03 15:02:35 -07003953 if (write) {
3954 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3955 ret = -EINVAL;
3956 goto out;
3957 }
3958 strcpy(saved_string, (char *)table->data);
3959 }
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003960 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003961 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003962 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003963 if (write) {
3964 int oldval = user_zonelist_order;
Chen Gangdacbde02013-07-03 15:02:35 -07003965
3966 ret = __parse_numa_zonelist_order((char *)table->data);
3967 if (ret) {
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003968 /*
3969 * bogus value. restore saved string
3970 */
Chen Gangdacbde02013-07-03 15:02:35 -07003971 strncpy((char *)table->data, saved_string,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003972 NUMA_ZONELIST_ORDER_LEN);
3973 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003974 } else if (oldval != user_zonelist_order) {
3975 mutex_lock(&zonelists_mutex);
Jiang Liu9adb62a2012-07-31 16:43:28 -07003976 build_all_zonelists(NULL, NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003977 mutex_unlock(&zonelists_mutex);
3978 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003979 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003980out:
3981 mutex_unlock(&zl_order_mutex);
3982 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003983}
3984
3985
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003986#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003987static int node_load[MAX_NUMNODES];
3988
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07003990 * 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 -07003991 * @node: node whose fallback list we're appending
3992 * @used_node_mask: nodemask_t of already used nodes
3993 *
3994 * We use a number of factors to determine which is the next node that should
3995 * appear on a given node's fallback list. The node should not have appeared
3996 * already in @node's fallback list, and it should be the next closest node
3997 * according to the distance array (which contains arbitrary distance values
3998 * from each node to each node in the system), and should also prefer nodes
3999 * with no CPUs, since presumably they'll have very little allocation pressure
4000 * on them otherwise.
4001 * It returns -1 if no node is found.
4002 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004003static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004004{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01004005 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006 int min_val = INT_MAX;
David Rientjes00ef2d22013-02-22 16:35:36 -08004007 int best_node = NUMA_NO_NODE;
Rusty Russella70f7302009-03-13 14:49:46 +10304008 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009
Linus Torvalds4cf808e2006-02-17 20:38:21 +01004010 /* Use the local node if we haven't already */
4011 if (!node_isset(node, *used_node_mask)) {
4012 node_set(node, *used_node_mask);
4013 return node;
4014 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08004016 for_each_node_state(n, N_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004017
4018 /* Don't want a node to appear more than once */
4019 if (node_isset(n, *used_node_mask))
4020 continue;
4021
Linus Torvalds1da177e2005-04-16 15:20:36 -07004022 /* Use the distance array to find the distance */
4023 val = node_distance(node, n);
4024
Linus Torvalds4cf808e2006-02-17 20:38:21 +01004025 /* Penalize nodes under us ("prefer the next node") */
4026 val += (n < node);
4027
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10304029 tmp = cpumask_of_node(n);
4030 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004031 val += PENALTY_FOR_NODE_WITH_CPUS;
4032
4033 /* Slight preference for less loaded node */
4034 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4035 val += node_load[n];
4036
4037 if (val < min_val) {
4038 min_val = val;
4039 best_node = n;
4040 }
4041 }
4042
4043 if (best_node >= 0)
4044 node_set(best_node, *used_node_mask);
4045
4046 return best_node;
4047}
4048
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004049
4050/*
4051 * Build zonelists ordered by node and zones within node.
4052 * This results in maximum locality--normal zone overflows into local
4053 * DMA zone, if any--but risks exhausting DMA zone.
4054 */
4055static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004056{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004057 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004059
Mel Gorman54a6eb52008-04-28 02:12:16 -07004060 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07004061 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07004062 ;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004063 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gormandd1a2392008-04-28 02:12:17 -07004064 zonelist->_zonerefs[j].zone = NULL;
4065 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004066}
4067
4068/*
Christoph Lameter523b9452007-10-16 01:25:37 -07004069 * Build gfp_thisnode zonelists
4070 */
4071static void build_thisnode_zonelists(pg_data_t *pgdat)
4072{
Christoph Lameter523b9452007-10-16 01:25:37 -07004073 int j;
4074 struct zonelist *zonelist;
4075
Mel Gorman54a6eb52008-04-28 02:12:16 -07004076 zonelist = &pgdat->node_zonelists[1];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004077 j = build_zonelists_node(pgdat, zonelist, 0);
Mel Gormandd1a2392008-04-28 02:12:17 -07004078 zonelist->_zonerefs[j].zone = NULL;
4079 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07004080}
4081
4082/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004083 * Build zonelists ordered by zone and nodes within zones.
4084 * This results in conserving DMA zone[s] until all Normal memory is
4085 * exhausted, but results in overflowing to remote node while memory
4086 * may still exist in local DMA zone.
4087 */
4088static int node_order[MAX_NUMNODES];
4089
4090static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4091{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004092 int pos, j, node;
4093 int zone_type; /* needs to be signed */
4094 struct zone *z;
4095 struct zonelist *zonelist;
4096
Mel Gorman54a6eb52008-04-28 02:12:16 -07004097 zonelist = &pgdat->node_zonelists[0];
4098 pos = 0;
4099 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4100 for (j = 0; j < nr_nodes; j++) {
4101 node = node_order[j];
4102 z = &NODE_DATA(node)->node_zones[zone_type];
4103 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07004104 zoneref_set_zone(z,
4105 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004106 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004107 }
4108 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004109 }
Mel Gormandd1a2392008-04-28 02:12:17 -07004110 zonelist->_zonerefs[pos].zone = NULL;
4111 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004112}
4113
Mel Gorman31939132014-10-09 15:28:30 -07004114#if defined(CONFIG_64BIT)
4115/*
4116 * Devices that require DMA32/DMA are relatively rare and do not justify a
4117 * penalty to every machine in case the specialised case applies. Default
4118 * to Node-ordering on 64-bit NUMA machines
4119 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004120static int default_zonelist_order(void)
4121{
Mel Gorman31939132014-10-09 15:28:30 -07004122 return ZONELIST_ORDER_NODE;
4123}
4124#else
4125/*
4126 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4127 * by the kernel. If processes running on node 0 deplete the low memory zone
4128 * then reclaim will occur more frequency increasing stalls and potentially
4129 * be easier to OOM if a large percentage of the zone is under writeback or
4130 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4131 * Hence, default to zone ordering on 32-bit.
4132 */
4133static int default_zonelist_order(void)
4134{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004135 return ZONELIST_ORDER_ZONE;
4136}
Mel Gorman31939132014-10-09 15:28:30 -07004137#endif /* CONFIG_64BIT */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004138
4139static void set_zonelist_order(void)
4140{
4141 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4142 current_zonelist_order = default_zonelist_order();
4143 else
4144 current_zonelist_order = user_zonelist_order;
4145}
4146
4147static void build_zonelists(pg_data_t *pgdat)
4148{
4149 int j, node, load;
4150 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004151 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004152 int local_node, prev_node;
4153 struct zonelist *zonelist;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08004154 unsigned int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004155
4156 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07004157 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07004159 zonelist->_zonerefs[0].zone = NULL;
4160 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004161 }
4162
4163 /* NUMA-aware ordering of nodes */
4164 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004165 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166 prev_node = local_node;
4167 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004168
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004169 memset(node_order, 0, sizeof(node_order));
4170 j = 0;
4171
Linus Torvalds1da177e2005-04-16 15:20:36 -07004172 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4173 /*
4174 * We don't want to pressure a particular node.
4175 * So adding penalty to the first node in same
4176 * distance group to make it round-robin.
4177 */
David Rientjes957f8222012-10-08 16:33:24 -07004178 if (node_distance(local_node, node) !=
4179 node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004180 node_load[node] = load;
4181
Linus Torvalds1da177e2005-04-16 15:20:36 -07004182 prev_node = node;
4183 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004184 if (order == ZONELIST_ORDER_NODE)
4185 build_zonelists_in_node_order(pgdat, node);
4186 else
4187 node_order[j++] = node; /* remember order */
4188 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004189
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004190 if (order == ZONELIST_ORDER_ZONE) {
4191 /* calculate node order -- i.e., DMA last! */
4192 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004193 }
Christoph Lameter523b9452007-10-16 01:25:37 -07004194
4195 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004196}
4197
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004198#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4199/*
4200 * Return node id of node used for "local" allocations.
4201 * I.e., first node id of first zone in arg node's generic zonelist.
4202 * Used for initializing percpu 'numa_mem', which is used primarily
4203 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4204 */
4205int local_memory_node(int node)
4206{
4207 struct zone *zone;
4208
4209 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
4210 gfp_zone(GFP_KERNEL),
4211 NULL,
4212 &zone);
4213 return zone->node;
4214}
4215#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004216
Linus Torvalds1da177e2005-04-16 15:20:36 -07004217#else /* CONFIG_NUMA */
4218
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004219static void set_zonelist_order(void)
4220{
4221 current_zonelist_order = ZONELIST_ORDER_ZONE;
4222}
4223
4224static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004225{
Christoph Lameter19655d32006-09-25 23:31:19 -07004226 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07004227 enum zone_type j;
4228 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004229
4230 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004231
Mel Gorman54a6eb52008-04-28 02:12:16 -07004232 zonelist = &pgdat->node_zonelists[0];
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004233 j = build_zonelists_node(pgdat, zonelist, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004234
Mel Gorman54a6eb52008-04-28 02:12:16 -07004235 /*
4236 * Now we build the zonelist so that it contains the zones
4237 * of all the other nodes.
4238 * We don't want to pressure a particular node, so when
4239 * building the zones for node N, we make sure that the
4240 * zones coming right after the local ones are those from
4241 * node N+1 (modulo N)
4242 */
4243 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4244 if (!node_online(node))
4245 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004246 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004247 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07004248 for (node = 0; node < local_node; node++) {
4249 if (!node_online(node))
4250 continue;
Zhang Yanfeibc732f12013-07-08 16:00:06 -07004251 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
Mel Gorman54a6eb52008-04-28 02:12:16 -07004252 }
4253
Mel Gormandd1a2392008-04-28 02:12:17 -07004254 zonelist->_zonerefs[j].zone = NULL;
4255 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004256}
4257
4258#endif /* CONFIG_NUMA */
4259
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004260/*
4261 * Boot pageset table. One per cpu which is going to be used for all
4262 * zones and all nodes. The parameters will be set in such a way
4263 * that an item put on a list will immediately be handed over to
4264 * the buddy list. This is safe since pageset manipulation is done
4265 * with interrupts disabled.
4266 *
4267 * The boot_pagesets must be kept even after bootup is complete for
4268 * unused processors and/or zones. They do play a role for bootstrapping
4269 * hotplugged processors.
4270 *
4271 * zoneinfo_show() and maybe other functions do
4272 * not check if the processor is online before following the pageset pointer.
4273 * Other parts of the kernel may not check if the zone is available.
4274 */
4275static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4276static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07004277static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004278
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004279/*
4280 * Global mutex to protect against size modification of zonelists
4281 * as well as to serialize pageset setup for the new populated zone.
4282 */
4283DEFINE_MUTEX(zonelists_mutex);
4284
Rusty Russell9b1a4d32008-07-28 12:16:30 -05004285/* return values int ....just for stop_machine() */
Jiang Liu4ed7e022012-07-31 16:43:35 -07004286static int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004287{
Yasunori Goto68113782006-06-23 02:03:11 -07004288 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004289 int cpu;
Jiang Liu9adb62a2012-07-31 16:43:28 -07004290 pg_data_t *self = data;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004291
Bo Liu7f9cfb32009-08-18 14:11:19 -07004292#ifdef CONFIG_NUMA
4293 memset(node_load, 0, sizeof(node_load));
4294#endif
Jiang Liu9adb62a2012-07-31 16:43:28 -07004295
4296 if (self && !node_online(self->node_id)) {
4297 build_zonelists(self);
Jiang Liu9adb62a2012-07-31 16:43:28 -07004298 }
4299
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004300 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07004301 pg_data_t *pgdat = NODE_DATA(nid);
4302
4303 build_zonelists(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08004304 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004305
4306 /*
4307 * Initialize the boot_pagesets that are going to be used
4308 * for bootstrapping processors. The real pagesets for
4309 * each zone will be allocated later when the per cpu
4310 * allocator is available.
4311 *
4312 * boot_pagesets are used also for bootstrapping offline
4313 * cpus if the system is already booted because the pagesets
4314 * are needed to initialize allocators on a specific cpu too.
4315 * F.e. the percpu allocator needs the page allocator which
4316 * needs the percpu allocator in order to allocate its pagesets
4317 * (a chicken-egg dilemma).
4318 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004319 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004320 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4321
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07004322#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4323 /*
4324 * We now know the "local memory node" for each node--
4325 * i.e., the node of the first zone in the generic zonelist.
4326 * Set up numa_mem percpu variable for on-line cpus. During
4327 * boot, only the boot cpu should be on-line; we'll init the
4328 * secondary cpus' numa_mem as they come on-line. During
4329 * node/memory hotplug, we'll fixup all on-line cpus.
4330 */
4331 if (cpu_online(cpu))
4332 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4333#endif
4334 }
4335
Yasunori Goto68113782006-06-23 02:03:11 -07004336 return 0;
4337}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004338
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004339static noinline void __init
4340build_all_zonelists_init(void)
4341{
4342 __build_all_zonelists(NULL);
4343 mminit_verify_zonelist();
4344 cpuset_init_current_mems_allowed();
4345}
4346
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004347/*
4348 * Called with zonelists_mutex held always
4349 * unless system_state == SYSTEM_BOOTING.
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004350 *
4351 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4352 * [we're only called with non-NULL zone through __meminit paths] and
4353 * (2) call of __init annotated helper build_all_zonelists_init
4354 * [protected by SYSTEM_BOOTING].
Haicheng Li4eaf3f62010-05-24 14:32:52 -07004355 */
Jiang Liu9adb62a2012-07-31 16:43:28 -07004356void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
Yasunori Goto68113782006-06-23 02:03:11 -07004357{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004358 set_zonelist_order();
4359
Yasunori Goto68113782006-06-23 02:03:11 -07004360 if (system_state == SYSTEM_BOOTING) {
Rasmus Villemoes061f67b2015-02-12 15:00:06 -08004361 build_all_zonelists_init();
Yasunori Goto68113782006-06-23 02:03:11 -07004362 } else {
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08004363#ifdef CONFIG_MEMORY_HOTPLUG
Jiang Liu9adb62a2012-07-31 16:43:28 -07004364 if (zone)
4365 setup_zone_pageset(zone);
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08004366#endif
Cody P Schaferdd1895e2013-07-03 15:01:36 -07004367 /* we have to stop all cpus to guarantee there is no user
4368 of zonelist */
Jiang Liu9adb62a2012-07-31 16:43:28 -07004369 stop_machine(__build_all_zonelists, pgdat, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07004370 /* cpuset refresh routine should be here */
4371 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07004372 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004373 /*
4374 * Disable grouping by mobility if the number of pages in the
4375 * system is too low to allow the mechanism to work. It would be
4376 * more accurate, but expensive to check per-zone. This check is
4377 * made on memory-hotadd so a system can start with mobility
4378 * disabled and enable it later
4379 */
Mel Gormand9c23402007-10-16 01:26:01 -07004380 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004381 page_group_by_mobility_disabled = 1;
4382 else
4383 page_group_by_mobility_disabled = 0;
4384
Anton Blanchardf88dfff2014-12-10 15:42:53 -08004385 pr_info("Built %i zonelists in %s order, mobility grouping %s. "
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004386 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004387 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004388 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07004389 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004390 vm_total_pages);
4391#ifdef CONFIG_NUMA
Anton Blanchardf88dfff2014-12-10 15:42:53 -08004392 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07004393#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004394}
4395
4396/*
4397 * Helper functions to size the waitqueue hash table.
4398 * Essentially these want to choose hash table sizes sufficiently
4399 * large so that collisions trying to wait on pages are rare.
4400 * But in fact, the number of active page waitqueues on typical
4401 * systems is ridiculously low, less than 200. So this is even
4402 * conservative, even though it seems large.
4403 *
4404 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4405 * waitqueues, i.e. the size of the waitq table given the number of pages.
4406 */
4407#define PAGES_PER_WAITQUEUE 256
4408
Yasunori Gotocca448f2006-06-23 02:03:10 -07004409#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07004410static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004411{
4412 unsigned long size = 1;
4413
4414 pages /= PAGES_PER_WAITQUEUE;
4415
4416 while (size < pages)
4417 size <<= 1;
4418
4419 /*
4420 * Once we have dozens or even hundreds of threads sleeping
4421 * on IO we've got bigger problems than wait queue collision.
4422 * Limit the size of the wait table to a reasonable size.
4423 */
4424 size = min(size, 4096UL);
4425
4426 return max(size, 4UL);
4427}
Yasunori Gotocca448f2006-06-23 02:03:10 -07004428#else
4429/*
4430 * A zone's size might be changed by hot-add, so it is not possible to determine
4431 * a suitable size for its wait_table. So we use the maximum size now.
4432 *
4433 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4434 *
4435 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4436 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4437 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4438 *
4439 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4440 * or more by the traditional way. (See above). It equals:
4441 *
4442 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4443 * ia64(16K page size) : = ( 8G + 4M)byte.
4444 * powerpc (64K page size) : = (32G +16M)byte.
4445 */
4446static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4447{
4448 return 4096UL;
4449}
4450#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004451
4452/*
4453 * This is an integer logarithm so that shifts can be used later
4454 * to extract the more random high bits from the multiplicative
4455 * hash function before the remainder is taken.
4456 */
4457static inline unsigned long wait_table_bits(unsigned long size)
4458{
4459 return ffz(~size);
4460}
4461
Mel Gorman56fd56b2007-10-16 01:25:58 -07004462/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004463 * Initially all pages are reserved - free ones are freed
4464 * up by free_all_bootmem() once the early boot process is
4465 * done. Non-atomic initialization, single-pass.
4466 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004467void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08004468 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004469{
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004470 pg_data_t *pgdat = NODE_DATA(nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -07004471 unsigned long end_pfn = start_pfn + size;
4472 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004473 struct zone *z;
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004474 unsigned long nr_initialised = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004475
Hugh Dickins22b31ee2009-01-06 14:40:09 -08004476 if (highest_memmap_pfn < end_pfn - 1)
4477 highest_memmap_pfn = end_pfn - 1;
4478
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004479 z = &pgdat->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08004480 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08004481 /*
4482 * There can be holes in boot-time mem_map[]s
4483 * handed to this function. They do not
4484 * exist on hotplugged memory.
4485 */
4486 if (context == MEMMAP_EARLY) {
4487 if (!early_pfn_valid(pfn))
4488 continue;
4489 if (!early_pfn_in_nid(pfn, nid))
4490 continue;
Mel Gorman3a80a7f2015-06-30 14:57:02 -07004491 if (!update_defer_init(pgdat, pfn, end_pfn,
4492 &nr_initialised))
4493 break;
Dave Hansena2f3aa022007-01-10 23:15:30 -08004494 }
Mel Gormanac5d2532015-06-30 14:57:20 -07004495
4496 /*
4497 * Mark the block movable so that blocks are reserved for
4498 * movable at startup. This will force kernel allocations
4499 * to reserve their blocks rather than leaking throughout
4500 * the address space during boot when many long-lived
Mel Gorman974a7862015-11-06 16:28:34 -08004501 * kernel allocations are made.
Mel Gormanac5d2532015-06-30 14:57:20 -07004502 *
4503 * bitmap is created for zone's valid pfn range. but memmap
4504 * can be created for invalid pages (for alignment)
4505 * check here not to call set_pageblock_migratetype() against
4506 * pfn out of zone.
4507 */
4508 if (!(pfn & (pageblock_nr_pages - 1))) {
4509 struct page *page = pfn_to_page(pfn);
4510
4511 __init_single_page(page, pfn, zone, nid);
4512 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4513 } else {
4514 __init_single_pfn(pfn, zone, nid);
4515 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004516 }
4517}
4518
Andi Kleen1e548de2008-02-04 22:29:26 -08004519static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004520{
Mel Gorman7aeb09f2014-06-04 16:10:21 -07004521 unsigned int order, t;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07004522 for_each_migratetype_order(order, t) {
4523 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004524 zone->free_area[order].nr_free = 0;
4525 }
4526}
4527
4528#ifndef __HAVE_ARCH_MEMMAP_INIT
4529#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08004530 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004531#endif
4532
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004533static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004534{
David Howells3a6be872009-05-06 16:03:03 -07004535#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004536 int batch;
4537
4538 /*
4539 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07004540 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004541 *
4542 * OK, so we don't know how big the cache is. So guess.
4543 */
Jiang Liub40da042013-02-22 16:33:52 -08004544 batch = zone->managed_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07004545 if (batch * PAGE_SIZE > 512 * 1024)
4546 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004547 batch /= 4; /* We effectively *= 4 below */
4548 if (batch < 1)
4549 batch = 1;
4550
4551 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004552 * Clamp the batch to a 2^n - 1 value. Having a power
4553 * of 2 value was found to be more likely to have
4554 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004555 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11004556 * For example if 2 tasks are alternately allocating
4557 * batches of pages, one task can end up with a lot
4558 * of pages of one half of the possible page colors
4559 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004560 */
David Howells91552032009-05-06 16:03:02 -07004561 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07004562
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004563 return batch;
David Howells3a6be872009-05-06 16:03:03 -07004564
4565#else
4566 /* The deferral and batching of frees should be suppressed under NOMMU
4567 * conditions.
4568 *
4569 * The problem is that NOMMU needs to be able to allocate large chunks
4570 * of contiguous memory as there's no hardware page translation to
4571 * assemble apparent contiguous memory from discontiguous pages.
4572 *
4573 * Queueing large contiguous runs of pages for batching, however,
4574 * causes the pages to actually be freed in smaller chunks. As there
4575 * can be a significant delay between the individual batches being
4576 * recycled, this leads to the once large chunks of space being
4577 * fragmented and becoming unavailable for high-order allocations.
4578 */
4579 return 0;
4580#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004581}
4582
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004583/*
4584 * pcp->high and pcp->batch values are related and dependent on one another:
4585 * ->batch must never be higher then ->high.
4586 * The following function updates them in a safe manner without read side
4587 * locking.
4588 *
4589 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4590 * those fields changing asynchronously (acording the the above rule).
4591 *
4592 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4593 * outside of boot time (or some other assurance that no concurrent updaters
4594 * exist).
4595 */
4596static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4597 unsigned long batch)
4598{
4599 /* start with a fail safe value for batch */
4600 pcp->batch = 1;
4601 smp_wmb();
4602
4603 /* Update high, then batch, in order */
4604 pcp->high = high;
4605 smp_wmb();
4606
4607 pcp->batch = batch;
4608}
4609
Cody P Schafer36640332013-07-03 15:01:40 -07004610/* a companion to pageset_set_high() */
Cody P Schafer4008bab2013-07-03 15:01:28 -07004611static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4612{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004613 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
Cody P Schafer4008bab2013-07-03 15:01:28 -07004614}
4615
Cody P Schafer88c90db2013-07-03 15:01:35 -07004616static void pageset_init(struct per_cpu_pageset *p)
Christoph Lameter2caaad42005-06-21 17:15:00 -07004617{
4618 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004619 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004620
Magnus Damm1c6fe942005-10-26 01:58:59 -07004621 memset(p, 0, sizeof(*p));
4622
Christoph Lameter3dfa5722008-02-04 22:29:19 -08004623 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07004624 pcp->count = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004625 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4626 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07004627}
4628
Cody P Schafer88c90db2013-07-03 15:01:35 -07004629static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4630{
4631 pageset_init(p);
4632 pageset_set_batch(p, batch);
4633}
4634
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004635/*
Cody P Schafer36640332013-07-03 15:01:40 -07004636 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004637 * to the value high for the pageset p.
4638 */
Cody P Schafer36640332013-07-03 15:01:40 -07004639static void pageset_set_high(struct per_cpu_pageset *p,
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004640 unsigned long high)
4641{
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004642 unsigned long batch = max(1UL, high / 4);
4643 if ((high / 4) > (PAGE_SHIFT * 8))
4644 batch = PAGE_SHIFT * 8;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004645
Cody P Schafer8d7a8fa2013-07-03 15:01:31 -07004646 pageset_update(&p->pcp, high, batch);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004647}
4648
David Rientjes7cd2b0a2014-06-23 13:22:04 -07004649static void pageset_set_high_and_batch(struct zone *zone,
4650 struct per_cpu_pageset *pcp)
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004651{
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004652 if (percpu_pagelist_fraction)
Cody P Schafer36640332013-07-03 15:01:40 -07004653 pageset_set_high(pcp,
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004654 (zone->managed_pages /
4655 percpu_pagelist_fraction));
4656 else
4657 pageset_set_batch(pcp, zone_batchsize(zone));
4658}
4659
Cody P Schafer169f6c12013-07-03 15:01:41 -07004660static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4661{
4662 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4663
4664 pageset_init(pcp);
4665 pageset_set_high_and_batch(zone, pcp);
4666}
4667
Jiang Liu4ed7e022012-07-31 16:43:35 -07004668static void __meminit setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07004669{
4670 int cpu;
Wu Fengguang319774e2010-05-24 14:32:49 -07004671 zone->pageset = alloc_percpu(struct per_cpu_pageset);
Cody P Schafer56cef2b2013-07-03 15:01:38 -07004672 for_each_possible_cpu(cpu)
4673 zone_pageset_init(zone, cpu);
Wu Fengguang319774e2010-05-24 14:32:49 -07004674}
4675
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004676/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004677 * Allocate per cpu pagesets and initialize them.
4678 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004679 */
Al Viro78d99552005-12-15 09:18:25 +00004680void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004681{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004682 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004683
Wu Fengguang319774e2010-05-24 14:32:49 -07004684 for_each_populated_zone(zone)
4685 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004686}
4687
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004688static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004689int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004690{
4691 int i;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004692 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004693
4694 /*
4695 * The per-page waitqueue mechanism uses hashed waitqueues
4696 * per zone.
4697 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004698 zone->wait_table_hash_nr_entries =
4699 wait_table_hash_nr_entries(zone_size_pages);
4700 zone->wait_table_bits =
4701 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004702 alloc_size = zone->wait_table_hash_nr_entries
4703 * sizeof(wait_queue_head_t);
4704
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004705 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004706 zone->wait_table = (wait_queue_head_t *)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004707 memblock_virt_alloc_node_nopanic(
4708 alloc_size, zone->zone_pgdat->node_id);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004709 } else {
4710 /*
4711 * This case means that a zone whose size was 0 gets new memory
4712 * via memory hot-add.
4713 * But it may be the case that a new node was hot-added. In
4714 * this case vmalloc() will not be able to use this new node's
4715 * memory - this wait_table must be initialized to use this new
4716 * node itself as well.
4717 * To use this new node's memory, further consideration will be
4718 * necessary.
4719 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004720 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004721 }
4722 if (!zone->wait_table)
4723 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004724
Pintu Kumarb8af2942013-09-11 14:20:34 -07004725 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004726 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004727
4728 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004729}
4730
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004731static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004732{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004733 /*
4734 * per cpu subsystem is not up at this point. The following code
4735 * relies on the ability of the linker to provide the
4736 * offset of a (static) per cpu variable into the per cpu area.
4737 */
4738 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004739
Xishi Qiub38a8722013-11-12 15:07:20 -08004740 if (populated_zone(zone))
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004741 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4742 zone->name, zone->present_pages,
4743 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004744}
4745
Jiang Liu4ed7e022012-07-31 16:43:35 -07004746int __meminit init_currently_empty_zone(struct zone *zone,
Yasunori Goto718127c2006-06-23 02:03:10 -07004747 unsigned long zone_start_pfn,
Yaowei Baib171e402015-11-05 18:47:06 -08004748 unsigned long size)
Dave Hansened8ece22005-10-29 18:16:50 -07004749{
4750 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004751 int ret;
4752 ret = zone_wait_table_init(zone, size);
4753 if (ret)
4754 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004755 pgdat->nr_zones = zone_idx(zone) + 1;
4756
Dave Hansened8ece22005-10-29 18:16:50 -07004757 zone->zone_start_pfn = zone_start_pfn;
4758
Mel Gorman708614e2008-07-23 21:26:51 -07004759 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4760 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4761 pgdat->node_id,
4762 (unsigned long)zone_idx(zone),
4763 zone_start_pfn, (zone_start_pfn + size));
4764
Andi Kleen1e548de2008-02-04 22:29:26 -08004765 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004766
4767 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004768}
4769
Tejun Heo0ee332c2011-12-08 10:22:09 -08004770#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004771#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
Mel Gorman8a942fd2015-06-30 14:56:55 -07004772
Mel Gormanc7132162006-09-27 01:49:43 -07004773/*
4774 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
Mel Gormanc7132162006-09-27 01:49:43 -07004775 */
Mel Gorman8a942fd2015-06-30 14:56:55 -07004776int __meminit __early_pfn_to_nid(unsigned long pfn,
4777 struct mminit_pfnnid_cache *state)
Mel Gormanc7132162006-09-27 01:49:43 -07004778{
Tejun Heoc13291a2011-07-12 10:46:30 +02004779 unsigned long start_pfn, end_pfn;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004780 int nid;
Russ Anderson7c243c72013-04-29 15:07:59 -07004781
Mel Gorman8a942fd2015-06-30 14:56:55 -07004782 if (state->last_start <= pfn && pfn < state->last_end)
4783 return state->last_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004784
Yinghai Lue76b63f2013-09-11 14:22:17 -07004785 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4786 if (nid != -1) {
Mel Gorman8a942fd2015-06-30 14:56:55 -07004787 state->last_start = start_pfn;
4788 state->last_end = end_pfn;
4789 state->last_nid = nid;
Yinghai Lue76b63f2013-09-11 14:22:17 -07004790 }
4791
4792 return nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004793}
4794#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4795
Mel Gormanc7132162006-09-27 01:49:43 -07004796/**
Santosh Shilimkar67828322014-01-21 15:50:25 -08004797 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004798 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
Santosh Shilimkar67828322014-01-21 15:50:25 -08004799 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
Mel Gormanc7132162006-09-27 01:49:43 -07004800 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004801 * If an architecture guarantees that all ranges registered contain no holes
4802 * and may be freed, this this function may be used instead of calling
4803 * memblock_free_early_nid() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004804 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004805void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004806{
Tejun Heoc13291a2011-07-12 10:46:30 +02004807 unsigned long start_pfn, end_pfn;
4808 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004809
Tejun Heoc13291a2011-07-12 10:46:30 +02004810 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4811 start_pfn = min(start_pfn, max_low_pfn);
4812 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004813
Tejun Heoc13291a2011-07-12 10:46:30 +02004814 if (start_pfn < end_pfn)
Santosh Shilimkar67828322014-01-21 15:50:25 -08004815 memblock_free_early_nid(PFN_PHYS(start_pfn),
4816 (end_pfn - start_pfn) << PAGE_SHIFT,
4817 this_nid);
Mel Gormanc7132162006-09-27 01:49:43 -07004818 }
4819}
4820
4821/**
4822 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004823 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004824 *
Zhang Zhen7d018172014-06-04 16:10:53 -07004825 * If an architecture guarantees that all ranges registered contain no holes and may
4826 * be freed, this function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004827 */
4828void __init sparse_memory_present_with_active_regions(int nid)
4829{
Tejun Heoc13291a2011-07-12 10:46:30 +02004830 unsigned long start_pfn, end_pfn;
4831 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004832
Tejun Heoc13291a2011-07-12 10:46:30 +02004833 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4834 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004835}
4836
4837/**
4838 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004839 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4840 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4841 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004842 *
4843 * It returns the start and end page frame of a node based on information
Zhang Zhen7d018172014-06-04 16:10:53 -07004844 * provided by memblock_set_node(). If called for a node
Mel Gormanc7132162006-09-27 01:49:43 -07004845 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004846 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004847 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004848void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004849 unsigned long *start_pfn, unsigned long *end_pfn)
4850{
Tejun Heoc13291a2011-07-12 10:46:30 +02004851 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004852 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02004853
Mel Gormanc7132162006-09-27 01:49:43 -07004854 *start_pfn = -1UL;
4855 *end_pfn = 0;
4856
Tejun Heoc13291a2011-07-12 10:46:30 +02004857 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4858 *start_pfn = min(*start_pfn, this_start_pfn);
4859 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004860 }
4861
Christoph Lameter633c0662007-10-16 01:25:37 -07004862 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004863 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004864}
4865
4866/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07004867 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4868 * assumption is made that zones within a node are ordered in monotonic
4869 * increasing memory addresses so that the "highest" populated zone is used
4870 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004871static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004872{
4873 int zone_index;
4874 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4875 if (zone_index == ZONE_MOVABLE)
4876 continue;
4877
4878 if (arch_zone_highest_possible_pfn[zone_index] >
4879 arch_zone_lowest_possible_pfn[zone_index])
4880 break;
4881 }
4882
4883 VM_BUG_ON(zone_index == -1);
4884 movable_zone = zone_index;
4885}
4886
4887/*
4888 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004889 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004890 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4891 * in each node depending on the size of each node and how evenly kernelcore
4892 * is distributed. This helper function adjusts the zone ranges
4893 * provided by the architecture for a given node by using the end of the
4894 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4895 * zones within a node are in order of monotonic increases memory addresses
4896 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004897static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004898 unsigned long zone_type,
4899 unsigned long node_start_pfn,
4900 unsigned long node_end_pfn,
4901 unsigned long *zone_start_pfn,
4902 unsigned long *zone_end_pfn)
4903{
4904 /* Only adjust if ZONE_MOVABLE is on this node */
4905 if (zone_movable_pfn[nid]) {
4906 /* Size ZONE_MOVABLE */
4907 if (zone_type == ZONE_MOVABLE) {
4908 *zone_start_pfn = zone_movable_pfn[nid];
4909 *zone_end_pfn = min(node_end_pfn,
4910 arch_zone_highest_possible_pfn[movable_zone]);
4911
4912 /* Adjust for ZONE_MOVABLE starting within this range */
4913 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4914 *zone_end_pfn > zone_movable_pfn[nid]) {
4915 *zone_end_pfn = zone_movable_pfn[nid];
4916
4917 /* Check if this whole range is within ZONE_MOVABLE */
4918 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4919 *zone_start_pfn = *zone_end_pfn;
4920 }
4921}
4922
4923/*
Mel Gormanc7132162006-09-27 01:49:43 -07004924 * Return the number of pages a zone spans in a node, including holes
4925 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4926 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004927static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004928 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004929 unsigned long node_start_pfn,
4930 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004931 unsigned long *ignored)
4932{
Mel Gormanc7132162006-09-27 01:49:43 -07004933 unsigned long zone_start_pfn, zone_end_pfn;
4934
Xishi Qiub5685e92015-09-08 15:04:16 -07004935 /* When hotadd a new node from cpu_up(), the node should be empty */
Xishi Qiuf9126ab2015-08-14 15:35:16 -07004936 if (!node_start_pfn && !node_end_pfn)
4937 return 0;
4938
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004939 /* Get the start and end of the zone */
Mel Gormanc7132162006-09-27 01:49:43 -07004940 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4941 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004942 adjust_zone_range_for_zone_movable(nid, zone_type,
4943 node_start_pfn, node_end_pfn,
4944 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004945
4946 /* Check that this node has pages within the zone's required range */
4947 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4948 return 0;
4949
4950 /* Move the zone boundaries inside the node if necessary */
4951 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4952 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4953
4954 /* Return the spanned pages */
4955 return zone_end_pfn - zone_start_pfn;
4956}
4957
4958/*
4959 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004960 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004961 */
Yinghai Lu32996252009-12-15 17:59:02 -08004962unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004963 unsigned long range_start_pfn,
4964 unsigned long range_end_pfn)
4965{
Tejun Heo96e907d2011-07-12 10:46:29 +02004966 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4967 unsigned long start_pfn, end_pfn;
4968 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004969
Tejun Heo96e907d2011-07-12 10:46:29 +02004970 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4971 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4972 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4973 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004974 }
Tejun Heo96e907d2011-07-12 10:46:29 +02004975 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07004976}
4977
4978/**
4979 * absent_pages_in_range - Return number of page frames in holes within a range
4980 * @start_pfn: The start PFN to start searching for holes
4981 * @end_pfn: The end PFN to stop searching for holes
4982 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004983 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004984 */
4985unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4986 unsigned long end_pfn)
4987{
4988 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4989}
4990
4991/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004992static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004993 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07004994 unsigned long node_start_pfn,
4995 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07004996 unsigned long *ignored)
4997{
Tejun Heo96e907d2011-07-12 10:46:29 +02004998 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4999 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07005000 unsigned long zone_start_pfn, zone_end_pfn;
5001
Xishi Qiub5685e92015-09-08 15:04:16 -07005002 /* When hotadd a new node from cpu_up(), the node should be empty */
Xishi Qiuf9126ab2015-08-14 15:35:16 -07005003 if (!node_start_pfn && !node_end_pfn)
5004 return 0;
5005
Tejun Heo96e907d2011-07-12 10:46:29 +02005006 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5007 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07005008
Mel Gorman2a1e2742007-07-17 04:03:12 -07005009 adjust_zone_range_for_zone_movable(nid, zone_type,
5010 node_start_pfn, node_end_pfn,
5011 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07005012 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005013}
Mel Gorman0e0b8642006-09-27 01:49:56 -07005014
Tejun Heo0ee332c2011-12-08 10:22:09 -08005015#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07005016static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005017 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005018 unsigned long node_start_pfn,
5019 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005020 unsigned long *zones_size)
5021{
5022 return zones_size[zone_type];
5023}
5024
Paul Mundt6ea6e682007-07-15 23:38:20 -07005025static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07005026 unsigned long zone_type,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005027 unsigned long node_start_pfn,
5028 unsigned long node_end_pfn,
Mel Gormanc7132162006-09-27 01:49:43 -07005029 unsigned long *zholes_size)
5030{
5031 if (!zholes_size)
5032 return 0;
5033
5034 return zholes_size[zone_type];
5035}
Yinghai Lu20e69262013-03-01 14:51:27 -08005036
Tejun Heo0ee332c2011-12-08 10:22:09 -08005037#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005038
Yasunori Gotoa3142c82007-05-08 00:23:07 -07005039static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005040 unsigned long node_start_pfn,
5041 unsigned long node_end_pfn,
5042 unsigned long *zones_size,
5043 unsigned long *zholes_size)
Mel Gormanc7132162006-09-27 01:49:43 -07005044{
Gu Zhengfebd5942015-06-24 16:57:02 -07005045 unsigned long realtotalpages = 0, totalpages = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07005046 enum zone_type i;
5047
Gu Zhengfebd5942015-06-24 16:57:02 -07005048 for (i = 0; i < MAX_NR_ZONES; i++) {
5049 struct zone *zone = pgdat->node_zones + i;
5050 unsigned long size, real_size;
Mel Gormanc7132162006-09-27 01:49:43 -07005051
Gu Zhengfebd5942015-06-24 16:57:02 -07005052 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5053 node_start_pfn,
5054 node_end_pfn,
5055 zones_size);
5056 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005057 node_start_pfn, node_end_pfn,
5058 zholes_size);
Gu Zhengfebd5942015-06-24 16:57:02 -07005059 zone->spanned_pages = size;
5060 zone->present_pages = real_size;
5061
5062 totalpages += size;
5063 realtotalpages += real_size;
5064 }
5065
5066 pgdat->node_spanned_pages = totalpages;
Mel Gormanc7132162006-09-27 01:49:43 -07005067 pgdat->node_present_pages = realtotalpages;
5068 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5069 realtotalpages);
5070}
5071
Mel Gorman835c1342007-10-16 01:25:47 -07005072#ifndef CONFIG_SPARSEMEM
5073/*
5074 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07005075 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5076 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07005077 * round what is now in bits to nearest long in bits, then return it in
5078 * bytes.
5079 */
Linus Torvalds7c455122013-02-18 09:58:02 -08005080static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005081{
5082 unsigned long usemapsize;
5083
Linus Torvalds7c455122013-02-18 09:58:02 -08005084 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005085 usemapsize = roundup(zonesize, pageblock_nr_pages);
5086 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07005087 usemapsize *= NR_PAGEBLOCK_BITS;
5088 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5089
5090 return usemapsize / 8;
5091}
5092
5093static void __init setup_usemap(struct pglist_data *pgdat,
Linus Torvalds7c455122013-02-18 09:58:02 -08005094 struct zone *zone,
5095 unsigned long zone_start_pfn,
5096 unsigned long zonesize)
Mel Gorman835c1342007-10-16 01:25:47 -07005097{
Linus Torvalds7c455122013-02-18 09:58:02 -08005098 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
Mel Gorman835c1342007-10-16 01:25:47 -07005099 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08005100 if (usemapsize)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005101 zone->pageblock_flags =
5102 memblock_virt_alloc_node_nopanic(usemapsize,
5103 pgdat->node_id);
Mel Gorman835c1342007-10-16 01:25:47 -07005104}
5105#else
Linus Torvalds7c455122013-02-18 09:58:02 -08005106static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5107 unsigned long zone_start_pfn, unsigned long zonesize) {}
Mel Gorman835c1342007-10-16 01:25:47 -07005108#endif /* CONFIG_SPARSEMEM */
5109
Mel Gormand9c23402007-10-16 01:26:01 -07005110#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08005111
Mel Gormand9c23402007-10-16 01:26:01 -07005112/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
Chen Gang15ca2202013-09-11 14:20:27 -07005113void __paginginit set_pageblock_order(void)
Mel Gormand9c23402007-10-16 01:26:01 -07005114{
Andrew Morton955c1cd2012-05-29 15:06:31 -07005115 unsigned int order;
5116
Mel Gormand9c23402007-10-16 01:26:01 -07005117 /* Check that pageblock_nr_pages has not already been setup */
5118 if (pageblock_order)
5119 return;
5120
Andrew Morton955c1cd2012-05-29 15:06:31 -07005121 if (HPAGE_SHIFT > PAGE_SHIFT)
5122 order = HUGETLB_PAGE_ORDER;
5123 else
5124 order = MAX_ORDER - 1;
5125
Mel Gormand9c23402007-10-16 01:26:01 -07005126 /*
5127 * Assume the largest contiguous order of interest is a huge page.
Andrew Morton955c1cd2012-05-29 15:06:31 -07005128 * This value may be variable depending on boot parameters on IA64 and
5129 * powerpc.
Mel Gormand9c23402007-10-16 01:26:01 -07005130 */
5131 pageblock_order = order;
5132}
5133#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5134
Mel Gormanba72cb82007-11-28 16:21:13 -08005135/*
5136 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
Andrew Morton955c1cd2012-05-29 15:06:31 -07005137 * is unused as pageblock_order is set at compile-time. See
5138 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5139 * the kernel config
Mel Gormanba72cb82007-11-28 16:21:13 -08005140 */
Chen Gang15ca2202013-09-11 14:20:27 -07005141void __paginginit set_pageblock_order(void)
Mel Gormanba72cb82007-11-28 16:21:13 -08005142{
Mel Gormanba72cb82007-11-28 16:21:13 -08005143}
Mel Gormand9c23402007-10-16 01:26:01 -07005144
5145#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5146
Jiang Liu01cefae2012-12-12 13:52:19 -08005147static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5148 unsigned long present_pages)
5149{
5150 unsigned long pages = spanned_pages;
5151
5152 /*
5153 * Provide a more accurate estimation if there are holes within
5154 * the zone and SPARSEMEM is in use. If there are holes within the
5155 * zone, each populated memory region may cost us one or two extra
5156 * memmap pages due to alignment because memmap pages for each
5157 * populated regions may not naturally algined on page boundary.
5158 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5159 */
5160 if (spanned_pages > present_pages + (present_pages >> 4) &&
5161 IS_ENABLED(CONFIG_SPARSEMEM))
5162 pages = present_pages;
5163
5164 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5165}
5166
Linus Torvalds1da177e2005-04-16 15:20:36 -07005167/*
5168 * Set up the zone data structures:
5169 * - mark all pages reserved
5170 * - mark all memory queues empty
5171 * - clear the memory bitmaps
Minchan Kim6527af52012-07-31 16:46:16 -07005172 *
5173 * NOTE: pgdat should get zeroed by caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005174 */
Wei Yang7f3eb552015-09-08 14:59:50 -07005175static void __paginginit free_area_init_core(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005176{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07005177 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07005178 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005179 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07005180 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005181
Dave Hansen208d54e2005-10-29 18:16:52 -07005182 pgdat_resize_init(pgdat);
Andrea Arcangeli8177a422012-03-23 20:56:34 +01005183#ifdef CONFIG_NUMA_BALANCING
5184 spin_lock_init(&pgdat->numabalancing_migrate_lock);
5185 pgdat->numabalancing_migrate_nr_pages = 0;
5186 pgdat->numabalancing_migrate_next_window = jiffies;
5187#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005188 init_waitqueue_head(&pgdat->kswapd_wait);
Mel Gorman55150612012-07-31 16:44:35 -07005189 init_waitqueue_head(&pgdat->pfmemalloc_wait);
Joonsoo Kimeefa8642014-12-12 16:55:46 -08005190 pgdat_page_ext_init(pgdat);
Michal Nazarewicz5f63b722012-01-11 15:16:11 +01005191
Linus Torvalds1da177e2005-04-16 15:20:36 -07005192 for (j = 0; j < MAX_NR_ZONES; j++) {
5193 struct zone *zone = pgdat->node_zones + j;
Jiang Liu9feedc92012-12-12 13:52:12 -08005194 unsigned long size, realsize, freesize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005195
Gu Zhengfebd5942015-06-24 16:57:02 -07005196 size = zone->spanned_pages;
5197 realsize = freesize = zone->present_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005198
Mel Gorman0e0b8642006-09-27 01:49:56 -07005199 /*
Jiang Liu9feedc92012-12-12 13:52:12 -08005200 * Adjust freesize so that it accounts for how much memory
Mel Gorman0e0b8642006-09-27 01:49:56 -07005201 * is used by this zone for memmap. This affects the watermark
5202 * and per-cpu initialisations
5203 */
Jiang Liu01cefae2012-12-12 13:52:19 -08005204 memmap_pages = calc_memmap_size(size, realsize);
Zhong Hongboba914f42014-12-12 16:56:21 -08005205 if (!is_highmem_idx(j)) {
5206 if (freesize >= memmap_pages) {
5207 freesize -= memmap_pages;
5208 if (memmap_pages)
5209 printk(KERN_DEBUG
5210 " %s zone: %lu pages used for memmap\n",
5211 zone_names[j], memmap_pages);
5212 } else
5213 printk(KERN_WARNING
5214 " %s zone: %lu pages exceeds freesize %lu\n",
5215 zone_names[j], memmap_pages, freesize);
5216 }
Mel Gorman0e0b8642006-09-27 01:49:56 -07005217
Christoph Lameter62672762007-02-10 01:43:07 -08005218 /* Account for reserved pages */
Jiang Liu9feedc92012-12-12 13:52:12 -08005219 if (j == 0 && freesize > dma_reserve) {
5220 freesize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07005221 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08005222 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07005223 }
5224
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07005225 if (!is_highmem_idx(j))
Jiang Liu9feedc92012-12-12 13:52:12 -08005226 nr_kernel_pages += freesize;
Jiang Liu01cefae2012-12-12 13:52:19 -08005227 /* Charge for highmem memmap if there are enough kernel pages */
5228 else if (nr_kernel_pages > memmap_pages * 2)
5229 nr_kernel_pages -= memmap_pages;
Jiang Liu9feedc92012-12-12 13:52:12 -08005230 nr_all_pages += freesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005231
Jiang Liu9feedc92012-12-12 13:52:12 -08005232 /*
5233 * Set an approximate value for lowmem here, it will be adjusted
5234 * when the bootmem allocator frees pages into the buddy system.
5235 * And all highmem pages will be managed by the buddy system.
5236 */
5237 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
Christoph Lameter96146342006-07-03 00:24:13 -07005238#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07005239 zone->node = nid;
Jiang Liu9feedc92012-12-12 13:52:12 -08005240 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07005241 / 100;
Jiang Liu9feedc92012-12-12 13:52:12 -08005242 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07005243#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005244 zone->name = zone_names[j];
5245 spin_lock_init(&zone->lock);
5246 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07005247 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005248 zone->zone_pgdat = pgdat;
Dave Hansened8ece22005-10-29 18:16:50 -07005249 zone_pcp_init(zone);
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07005250
5251 /* For bootup, initialized properly in watermark setup */
5252 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
5253
Hugh Dickinsbea8c152012-11-16 14:14:54 -08005254 lruvec_init(&zone->lruvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005255 if (!size)
5256 continue;
5257
Andrew Morton955c1cd2012-05-29 15:06:31 -07005258 set_pageblock_order();
Linus Torvalds7c455122013-02-18 09:58:02 -08005259 setup_usemap(pgdat, zone, zone_start_pfn, size);
Yaowei Baib171e402015-11-05 18:47:06 -08005260 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
Yasunori Goto718127c2006-06-23 02:03:10 -07005261 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07005262 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005263 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005264 }
5265}
5266
Sam Ravnborg577a32f2007-05-17 23:29:25 +02005267static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005268{
Tony Luckb0aeba72015-11-10 10:09:47 -08005269 unsigned long __maybe_unused start = 0;
Laura Abbotta1c34a32015-11-05 18:48:46 -08005270 unsigned long __maybe_unused offset = 0;
5271
Linus Torvalds1da177e2005-04-16 15:20:36 -07005272 /* Skip empty nodes */
5273 if (!pgdat->node_spanned_pages)
5274 return;
5275
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005276#ifdef CONFIG_FLAT_NODE_MEM_MAP
Tony Luckb0aeba72015-11-10 10:09:47 -08005277 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
5278 offset = pgdat->node_start_pfn - start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005279 /* ia64 gets its own node_mem_map, before this, without bootmem */
5280 if (!pgdat->node_mem_map) {
Tony Luckb0aeba72015-11-10 10:09:47 -08005281 unsigned long size, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005282 struct page *map;
5283
Bob Piccoe984bb42006-05-20 15:00:31 -07005284 /*
5285 * The zone's endpoints aren't required to be MAX_ORDER
5286 * aligned but the node_mem_map endpoints must be in order
5287 * for the buddy allocator to function correctly.
5288 */
Cody P Schafer108bcc92013-02-22 16:35:23 -08005289 end = pgdat_end_pfn(pgdat);
Bob Piccoe984bb42006-05-20 15:00:31 -07005290 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5291 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07005292 map = alloc_remap(pgdat->node_id, size);
5293 if (!map)
Santosh Shilimkar67828322014-01-21 15:50:25 -08005294 map = memblock_virt_alloc_node_nopanic(size,
5295 pgdat->node_id);
Laura Abbotta1c34a32015-11-05 18:48:46 -08005296 pgdat->node_mem_map = map + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005297 }
Roman Zippel12d810c2007-05-31 00:40:54 -07005298#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07005299 /*
5300 * With no DISCONTIG, the global mem_map is just set as node 0's
5301 */
Mel Gormanc7132162006-09-27 01:49:43 -07005302 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005303 mem_map = NODE_DATA(0)->node_mem_map;
Laura Abbotta1c34a32015-11-05 18:48:46 -08005304#if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
Mel Gormanc7132162006-09-27 01:49:43 -07005305 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Laura Abbotta1c34a32015-11-05 18:48:46 -08005306 mem_map -= offset;
Tejun Heo0ee332c2011-12-08 10:22:09 -08005307#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005308 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005309#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005310#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005311}
5312
Johannes Weiner9109fb72008-07-23 21:27:20 -07005313void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5314 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005315{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005316 pg_data_t *pgdat = NODE_DATA(nid);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005317 unsigned long start_pfn = 0;
5318 unsigned long end_pfn = 0;
Johannes Weiner9109fb72008-07-23 21:27:20 -07005319
Minchan Kim88fdf752012-07-31 16:46:14 -07005320 /* pg_data_t should be reset to zero when it's allocated */
Linus Torvalds8783b6e2012-08-02 10:37:03 -07005321 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
Minchan Kim88fdf752012-07-31 16:46:14 -07005322
Mel Gorman3a80a7f2015-06-30 14:57:02 -07005323 reset_deferred_meminit(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005324 pgdat->node_id = nid;
5325 pgdat->node_start_pfn = node_start_pfn;
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005326#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5327 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
Juergen Gross8d29e182015-02-11 15:26:01 -08005328 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
Zhen Lei4ada0c52015-09-08 15:04:19 -07005329 (u64)start_pfn << PAGE_SHIFT,
5330 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
Zhang Yanfei7960aed2013-07-08 15:59:52 -07005331#endif
5332 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5333 zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005334
5335 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07005336#ifdef CONFIG_FLAT_NODE_MEM_MAP
5337 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5338 nid, (unsigned long)pgdat,
5339 (unsigned long)pgdat->node_mem_map);
5340#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07005341
Wei Yang7f3eb552015-09-08 14:59:50 -07005342 free_area_init_core(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005343}
5344
Tejun Heo0ee332c2011-12-08 10:22:09 -08005345#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07005346
5347#if MAX_NUMNODES > 1
5348/*
5349 * Figure out the number of possible node ids.
5350 */
Cody P Schaferf9872ca2013-04-29 15:08:01 -07005351void __init setup_nr_node_ids(void)
Miklos Szeredi418508c2007-05-23 13:57:55 -07005352{
Wei Yang904a9552015-09-08 14:59:48 -07005353 unsigned int highest;
Miklos Szeredi418508c2007-05-23 13:57:55 -07005354
Wei Yang904a9552015-09-08 14:59:48 -07005355 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
Miklos Szeredi418508c2007-05-23 13:57:55 -07005356 nr_node_ids = highest + 1;
5357}
Miklos Szeredi418508c2007-05-23 13:57:55 -07005358#endif
5359
Mel Gormanc7132162006-09-27 01:49:43 -07005360/**
Tejun Heo1e019792011-07-12 09:45:34 +02005361 * node_map_pfn_alignment - determine the maximum internode alignment
5362 *
5363 * This function should be called after node map is populated and sorted.
5364 * It calculates the maximum power of two alignment which can distinguish
5365 * all the nodes.
5366 *
5367 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5368 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5369 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5370 * shifted, 1GiB is enough and this function will indicate so.
5371 *
5372 * This is used to test whether pfn -> nid mapping of the chosen memory
5373 * model has fine enough granularity to avoid incorrect mapping for the
5374 * populated node map.
5375 *
5376 * Returns the determined alignment in pfn's. 0 if there is no alignment
5377 * requirement (single node).
5378 */
5379unsigned long __init node_map_pfn_alignment(void)
5380{
5381 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005382 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02005383 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02005384 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02005385
Tejun Heoc13291a2011-07-12 10:46:30 +02005386 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02005387 if (!start || last_nid < 0 || last_nid == nid) {
5388 last_nid = nid;
5389 last_end = end;
5390 continue;
5391 }
5392
5393 /*
5394 * Start with a mask granular enough to pin-point to the
5395 * start pfn and tick off bits one-by-one until it becomes
5396 * too coarse to separate the current node from the last.
5397 */
5398 mask = ~((1 << __ffs(start)) - 1);
5399 while (mask && last_end <= (start & (mask << 1)))
5400 mask <<= 1;
5401
5402 /* accumulate all internode masks */
5403 accl_mask |= mask;
5404 }
5405
5406 /* convert mask to number of pages */
5407 return ~accl_mask + 1;
5408}
5409
Mel Gormana6af2bc2007-02-10 01:42:57 -08005410/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07005411static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07005412{
Mel Gormana6af2bc2007-02-10 01:42:57 -08005413 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02005414 unsigned long start_pfn;
5415 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00005416
Tejun Heoc13291a2011-07-12 10:46:30 +02005417 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5418 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005419
Mel Gormana6af2bc2007-02-10 01:42:57 -08005420 if (min_pfn == ULONG_MAX) {
5421 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07005422 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08005423 return 0;
5424 }
5425
5426 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07005427}
5428
5429/**
5430 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5431 *
5432 * It returns the minimum PFN based on information provided via
Zhang Zhen7d018172014-06-04 16:10:53 -07005433 * memblock_set_node().
Mel Gormanc7132162006-09-27 01:49:43 -07005434 */
5435unsigned long __init find_min_pfn_with_active_regions(void)
5436{
5437 return find_min_pfn_for_node(MAX_NUMNODES);
5438}
5439
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005440/*
5441 * early_calculate_totalpages()
5442 * Sum pages in active regions for movable zone.
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005443 * Populate N_MEMORY for calculating usable_nodes.
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005444 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07005445static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07005446{
Mel Gorman7e63efe2007-07-17 04:03:15 -07005447 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02005448 unsigned long start_pfn, end_pfn;
5449 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005450
Tejun Heoc13291a2011-07-12 10:46:30 +02005451 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5452 unsigned long pages = end_pfn - start_pfn;
5453
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005454 totalpages += pages;
5455 if (pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005456 node_set_state(nid, N_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005457 }
Pintu Kumarb8af2942013-09-11 14:20:34 -07005458 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07005459}
5460
Mel Gorman2a1e2742007-07-17 04:03:12 -07005461/*
5462 * Find the PFN the Movable zone begins in each node. Kernel memory
5463 * is spread evenly between nodes as long as the nodes have enough
5464 * memory. When they don't, some nodes will have more kernelcore than
5465 * others
5466 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07005467static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005468{
5469 int i, nid;
5470 unsigned long usable_startpfn;
5471 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07005472 /* save the state before borrow the nodemask */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005473 nodemask_t saved_node_state = node_states[N_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005474 unsigned long totalpages = early_calculate_totalpages();
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005475 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
Emil Medve136199f2014-04-07 15:37:52 -07005476 struct memblock_region *r;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005477
5478 /* Need to find movable_zone earlier when movable_node is specified. */
5479 find_usable_zone_for_movable();
Mel Gorman2a1e2742007-07-17 04:03:12 -07005480
Mel Gorman7e63efe2007-07-17 04:03:15 -07005481 /*
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005482 * If movable_node is specified, ignore kernelcore and movablecore
5483 * options.
5484 */
5485 if (movable_node_is_enabled()) {
Emil Medve136199f2014-04-07 15:37:52 -07005486 for_each_memblock(memory, r) {
5487 if (!memblock_is_hotpluggable(r))
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005488 continue;
5489
Emil Medve136199f2014-04-07 15:37:52 -07005490 nid = r->nid;
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005491
Emil Medve136199f2014-04-07 15:37:52 -07005492 usable_startpfn = PFN_DOWN(r->base);
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005493 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5494 min(usable_startpfn, zone_movable_pfn[nid]) :
5495 usable_startpfn;
5496 }
5497
5498 goto out2;
5499 }
5500
5501 /*
5502 * If movablecore=nn[KMG] was specified, calculate what size of
Mel Gorman7e63efe2007-07-17 04:03:15 -07005503 * kernelcore that corresponds so that memory usable for
5504 * any allocation type is evenly spread. If both kernelcore
5505 * and movablecore are specified, then the value of kernelcore
5506 * will be used for required_kernelcore if it's greater than
5507 * what movablecore would have allowed.
5508 */
5509 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07005510 unsigned long corepages;
5511
5512 /*
5513 * Round-up so that ZONE_MOVABLE is at least as large as what
5514 * was requested by the user
5515 */
5516 required_movablecore =
5517 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
Xishi Qiu9fd745d2015-11-05 18:48:11 -08005518 required_movablecore = min(totalpages, required_movablecore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005519 corepages = totalpages - required_movablecore;
5520
5521 required_kernelcore = max(required_kernelcore, corepages);
5522 }
5523
Xishi Qiubde304b2015-11-05 18:48:56 -08005524 /*
5525 * If kernelcore was not specified or kernelcore size is larger
5526 * than totalpages, there is no ZONE_MOVABLE.
5527 */
5528 if (!required_kernelcore || required_kernelcore >= totalpages)
Yinghai Lu66918dc2009-06-30 11:41:37 -07005529 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005530
5531 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005532 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5533
5534restart:
5535 /* Spread kernelcore memory as evenly as possible throughout nodes */
5536 kernelcore_node = required_kernelcore / usable_nodes;
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005537 for_each_node_state(nid, N_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02005538 unsigned long start_pfn, end_pfn;
5539
Mel Gorman2a1e2742007-07-17 04:03:12 -07005540 /*
5541 * Recalculate kernelcore_node if the division per node
5542 * now exceeds what is necessary to satisfy the requested
5543 * amount of memory for the kernel
5544 */
5545 if (required_kernelcore < kernelcore_node)
5546 kernelcore_node = required_kernelcore / usable_nodes;
5547
5548 /*
5549 * As the map is walked, we track how much memory is usable
5550 * by the kernel using kernelcore_remaining. When it is
5551 * 0, the rest of the node is usable by ZONE_MOVABLE
5552 */
5553 kernelcore_remaining = kernelcore_node;
5554
5555 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02005556 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005557 unsigned long size_pages;
5558
Tejun Heoc13291a2011-07-12 10:46:30 +02005559 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005560 if (start_pfn >= end_pfn)
5561 continue;
5562
5563 /* Account for what is only usable for kernelcore */
5564 if (start_pfn < usable_startpfn) {
5565 unsigned long kernel_pages;
5566 kernel_pages = min(end_pfn, usable_startpfn)
5567 - start_pfn;
5568
5569 kernelcore_remaining -= min(kernel_pages,
5570 kernelcore_remaining);
5571 required_kernelcore -= min(kernel_pages,
5572 required_kernelcore);
5573
5574 /* Continue if range is now fully accounted */
5575 if (end_pfn <= usable_startpfn) {
5576
5577 /*
5578 * Push zone_movable_pfn to the end so
5579 * that if we have to rebalance
5580 * kernelcore across nodes, we will
5581 * not double account here
5582 */
5583 zone_movable_pfn[nid] = end_pfn;
5584 continue;
5585 }
5586 start_pfn = usable_startpfn;
5587 }
5588
5589 /*
5590 * The usable PFN range for ZONE_MOVABLE is from
5591 * start_pfn->end_pfn. Calculate size_pages as the
5592 * number of pages used as kernelcore
5593 */
5594 size_pages = end_pfn - start_pfn;
5595 if (size_pages > kernelcore_remaining)
5596 size_pages = kernelcore_remaining;
5597 zone_movable_pfn[nid] = start_pfn + size_pages;
5598
5599 /*
5600 * Some kernelcore has been met, update counts and
5601 * break if the kernelcore for this node has been
Pintu Kumarb8af2942013-09-11 14:20:34 -07005602 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005603 */
5604 required_kernelcore -= min(required_kernelcore,
5605 size_pages);
5606 kernelcore_remaining -= size_pages;
5607 if (!kernelcore_remaining)
5608 break;
5609 }
5610 }
5611
5612 /*
5613 * If there is still required_kernelcore, we do another pass with one
5614 * less node in the count. This will push zone_movable_pfn[nid] further
5615 * along on the nodes that still have memory until kernelcore is
Pintu Kumarb8af2942013-09-11 14:20:34 -07005616 * satisfied
Mel Gorman2a1e2742007-07-17 04:03:12 -07005617 */
5618 usable_nodes--;
5619 if (usable_nodes && required_kernelcore > usable_nodes)
5620 goto restart;
5621
Tang Chenb2f3eeb2014-01-21 15:49:38 -08005622out2:
Mel Gorman2a1e2742007-07-17 04:03:12 -07005623 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5624 for (nid = 0; nid < MAX_NUMNODES; nid++)
5625 zone_movable_pfn[nid] =
5626 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07005627
Yinghai Lu20e69262013-03-01 14:51:27 -08005628out:
Yinghai Lu66918dc2009-06-30 11:41:37 -07005629 /* restore the node_state */
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005630 node_states[N_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005631}
5632
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005633/* Any regular or high memory on that node ? */
5634static void check_for_memory(pg_data_t *pgdat, int nid)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005635{
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005636 enum zone_type zone_type;
5637
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005638 if (N_MEMORY == N_NORMAL_MEMORY)
5639 return;
5640
5641 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005642 struct zone *zone = &pgdat->node_zones[zone_type];
Xishi Qiub38a8722013-11-12 15:07:20 -08005643 if (populated_zone(zone)) {
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005644 node_set_state(nid, N_HIGH_MEMORY);
5645 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5646 zone_type <= ZONE_NORMAL)
5647 node_set_state(nid, N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08005648 break;
5649 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005650 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005651}
5652
Mel Gormanc7132162006-09-27 01:49:43 -07005653/**
5654 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005655 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07005656 *
5657 * This will call free_area_init_node() for each active node in the system.
Zhang Zhen7d018172014-06-04 16:10:53 -07005658 * Using the page ranges provided by memblock_set_node(), the size of each
Mel Gormanc7132162006-09-27 01:49:43 -07005659 * zone in each node and their holes is calculated. If the maximum PFN
5660 * between two adjacent zones match, it is assumed that the zone is empty.
5661 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5662 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5663 * starts where the previous one ended. For example, ZONE_DMA32 starts
5664 * at arch_max_dma_pfn.
5665 */
5666void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5667{
Tejun Heoc13291a2011-07-12 10:46:30 +02005668 unsigned long start_pfn, end_pfn;
5669 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08005670
Mel Gormanc7132162006-09-27 01:49:43 -07005671 /* Record where the zone boundaries are */
5672 memset(arch_zone_lowest_possible_pfn, 0,
5673 sizeof(arch_zone_lowest_possible_pfn));
5674 memset(arch_zone_highest_possible_pfn, 0,
5675 sizeof(arch_zone_highest_possible_pfn));
5676 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5677 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5678 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07005679 if (i == ZONE_MOVABLE)
5680 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07005681 arch_zone_lowest_possible_pfn[i] =
5682 arch_zone_highest_possible_pfn[i-1];
5683 arch_zone_highest_possible_pfn[i] =
5684 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5685 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07005686 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5687 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5688
5689 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5690 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07005691 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07005692
Mel Gormanc7132162006-09-27 01:49:43 -07005693 /* Print out the zone ranges */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005694 pr_info("Zone ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005695 for (i = 0; i < MAX_NR_ZONES; i++) {
5696 if (i == ZONE_MOVABLE)
5697 continue;
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005698 pr_info(" %-8s ", zone_names[i]);
David Rientjes72f0ba02010-03-05 13:42:14 -08005699 if (arch_zone_lowest_possible_pfn[i] ==
5700 arch_zone_highest_possible_pfn[i])
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005701 pr_cont("empty\n");
David Rientjes72f0ba02010-03-05 13:42:14 -08005702 else
Juergen Gross8d29e182015-02-11 15:26:01 -08005703 pr_cont("[mem %#018Lx-%#018Lx]\n",
5704 (u64)arch_zone_lowest_possible_pfn[i]
5705 << PAGE_SHIFT,
5706 ((u64)arch_zone_highest_possible_pfn[i]
Bjorn Helgaasa62e2f42012-05-29 15:06:30 -07005707 << PAGE_SHIFT) - 1);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005708 }
5709
5710 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005711 pr_info("Movable zone start for each node\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07005712 for (i = 0; i < MAX_NUMNODES; i++) {
5713 if (zone_movable_pfn[i])
Juergen Gross8d29e182015-02-11 15:26:01 -08005714 pr_info(" Node %d: %#018Lx\n", i,
5715 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005716 }
Mel Gormanc7132162006-09-27 01:49:43 -07005717
Wanpeng Lif2d52fe2012-10-08 16:32:24 -07005718 /* Print out the early node map */
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005719 pr_info("Early memory node ranges\n");
Tejun Heoc13291a2011-07-12 10:46:30 +02005720 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Juergen Gross8d29e182015-02-11 15:26:01 -08005721 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
5722 (u64)start_pfn << PAGE_SHIFT,
5723 ((u64)end_pfn << PAGE_SHIFT) - 1);
Mel Gormanc7132162006-09-27 01:49:43 -07005724
5725 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005726 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005727 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005728 for_each_online_node(nid) {
5729 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005730 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005731 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005732
5733 /* Any memory on that node */
5734 if (pgdat->node_present_pages)
Lai Jiangshan4b0ef1fe2012-12-12 13:51:46 -08005735 node_set_state(nid, N_MEMORY);
5736 check_for_memory(pgdat, nid);
Mel Gormanc7132162006-09-27 01:49:43 -07005737 }
5738}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005739
Mel Gorman7e63efe2007-07-17 04:03:15 -07005740static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005741{
5742 unsigned long long coremem;
5743 if (!p)
5744 return -EINVAL;
5745
5746 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005747 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005748
Mel Gorman7e63efe2007-07-17 04:03:15 -07005749 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005750 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5751
5752 return 0;
5753}
Mel Gormaned7ed362007-07-17 04:03:14 -07005754
Mel Gorman7e63efe2007-07-17 04:03:15 -07005755/*
5756 * kernelcore=size sets the amount of memory for use for allocations that
5757 * cannot be reclaimed or migrated.
5758 */
5759static int __init cmdline_parse_kernelcore(char *p)
5760{
5761 return cmdline_parse_core(p, &required_kernelcore);
5762}
5763
5764/*
5765 * movablecore=size sets the amount of memory for use for allocations that
5766 * can be reclaimed or migrated.
5767 */
5768static int __init cmdline_parse_movablecore(char *p)
5769{
5770 return cmdline_parse_core(p, &required_movablecore);
5771}
5772
Mel Gormaned7ed362007-07-17 04:03:14 -07005773early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005774early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005775
Tejun Heo0ee332c2011-12-08 10:22:09 -08005776#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005777
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005778void adjust_managed_page_count(struct page *page, long count)
5779{
5780 spin_lock(&managed_page_count_lock);
5781 page_zone(page)->managed_pages += count;
5782 totalram_pages += count;
Jiang Liu3dcc0572013-07-03 15:03:21 -07005783#ifdef CONFIG_HIGHMEM
5784 if (PageHighMem(page))
5785 totalhigh_pages += count;
5786#endif
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005787 spin_unlock(&managed_page_count_lock);
5788}
Jiang Liu3dcc0572013-07-03 15:03:21 -07005789EXPORT_SYMBOL(adjust_managed_page_count);
Jiang Liuc3d5f5f2013-07-03 15:03:14 -07005790
Jiang Liu11199692013-07-03 15:02:48 -07005791unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
Jiang Liu69afade2013-04-29 15:06:21 -07005792{
Jiang Liu11199692013-07-03 15:02:48 -07005793 void *pos;
5794 unsigned long pages = 0;
Jiang Liu69afade2013-04-29 15:06:21 -07005795
Jiang Liu11199692013-07-03 15:02:48 -07005796 start = (void *)PAGE_ALIGN((unsigned long)start);
5797 end = (void *)((unsigned long)end & PAGE_MASK);
5798 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
Jiang Liudbe67df2013-07-03 15:02:51 -07005799 if ((unsigned int)poison <= 0xFF)
Jiang Liu11199692013-07-03 15:02:48 -07005800 memset(pos, poison, PAGE_SIZE);
5801 free_reserved_page(virt_to_page(pos));
Jiang Liu69afade2013-04-29 15:06:21 -07005802 }
5803
5804 if (pages && s)
Jiang Liu11199692013-07-03 15:02:48 -07005805 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
Jiang Liu69afade2013-04-29 15:06:21 -07005806 s, pages << (PAGE_SHIFT - 10), start, end);
5807
5808 return pages;
5809}
Jiang Liu11199692013-07-03 15:02:48 -07005810EXPORT_SYMBOL(free_reserved_area);
Jiang Liu69afade2013-04-29 15:06:21 -07005811
Jiang Liucfa11e02013-04-29 15:07:00 -07005812#ifdef CONFIG_HIGHMEM
5813void free_highmem_page(struct page *page)
5814{
5815 __free_reserved_page(page);
5816 totalram_pages++;
Jiang Liu7b4b2a02013-07-03 15:03:11 -07005817 page_zone(page)->managed_pages++;
Jiang Liucfa11e02013-04-29 15:07:00 -07005818 totalhigh_pages++;
5819}
5820#endif
5821
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005822
5823void __init mem_init_print_info(const char *str)
5824{
5825 unsigned long physpages, codesize, datasize, rosize, bss_size;
5826 unsigned long init_code_size, init_data_size;
5827
5828 physpages = get_num_physpages();
5829 codesize = _etext - _stext;
5830 datasize = _edata - _sdata;
5831 rosize = __end_rodata - __start_rodata;
5832 bss_size = __bss_stop - __bss_start;
5833 init_data_size = __init_end - __init_begin;
5834 init_code_size = _einittext - _sinittext;
5835
5836 /*
5837 * Detect special cases and adjust section sizes accordingly:
5838 * 1) .init.* may be embedded into .data sections
5839 * 2) .init.text.* may be out of [__init_begin, __init_end],
5840 * please refer to arch/tile/kernel/vmlinux.lds.S.
5841 * 3) .rodata.* may be embedded into .text or .data sections.
5842 */
5843#define adj_init_size(start, end, size, pos, adj) \
Pintu Kumarb8af2942013-09-11 14:20:34 -07005844 do { \
5845 if (start <= pos && pos < end && size > adj) \
5846 size -= adj; \
5847 } while (0)
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005848
5849 adj_init_size(__init_begin, __init_end, init_data_size,
5850 _sinittext, init_code_size);
5851 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5852 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5853 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5854 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5855
5856#undef adj_init_size
5857
Anton Blanchardf88dfff2014-12-10 15:42:53 -08005858 pr_info("Memory: %luK/%luK available "
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005859 "(%luK kernel code, %luK rwdata, %luK rodata, "
Pintu Kumare48322a2014-12-18 16:17:15 -08005860 "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005861#ifdef CONFIG_HIGHMEM
5862 ", %luK highmem"
5863#endif
5864 "%s%s)\n",
5865 nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5866 codesize >> 10, datasize >> 10, rosize >> 10,
5867 (init_data_size + init_code_size) >> 10, bss_size >> 10,
Pintu Kumare48322a2014-12-18 16:17:15 -08005868 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10),
5869 totalcma_pages << (PAGE_SHIFT-10),
Jiang Liu7ee3d4e2013-07-03 15:03:41 -07005870#ifdef CONFIG_HIGHMEM
5871 totalhigh_pages << (PAGE_SHIFT-10),
5872#endif
5873 str ? ", " : "", str ? str : "");
5874}
5875
Mel Gorman0e0b8642006-09-27 01:49:56 -07005876/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005877 * set_dma_reserve - set the specified number of pages reserved in the first zone
5878 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07005879 *
Yaowei Bai013110a2015-09-08 15:04:10 -07005880 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005881 * In the DMA zone, a significant percentage may be consumed by kernel image
5882 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005883 * function may optionally be used to account for unfreeable pages in the
5884 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5885 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005886 */
5887void __init set_dma_reserve(unsigned long new_dma_reserve)
5888{
5889 dma_reserve = new_dma_reserve;
5890}
5891
Linus Torvalds1da177e2005-04-16 15:20:36 -07005892void __init free_area_init(unsigned long *zones_size)
5893{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005894 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005895 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5896}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005897
Linus Torvalds1da177e2005-04-16 15:20:36 -07005898static int page_alloc_cpu_notify(struct notifier_block *self,
5899 unsigned long action, void *hcpu)
5900{
5901 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005902
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005903 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07005904 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005905 drain_pages(cpu);
5906
5907 /*
5908 * Spill the event counters of the dead processor
5909 * into the current processors event counters.
5910 * This artificially elevates the count of the current
5911 * processor.
5912 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07005913 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005914
5915 /*
5916 * Zero the differential counters of the dead processor
5917 * so that the vm statistics are consistent.
5918 *
5919 * This is only okay since the processor is dead and cannot
5920 * race with what we are doing.
5921 */
Christoph Lameter2bb921e2013-09-11 14:21:30 -07005922 cpu_vm_stats_fold(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005923 }
5924 return NOTIFY_OK;
5925}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005926
5927void __init page_alloc_init(void)
5928{
5929 hotcpu_notifier(page_alloc_cpu_notify, 0);
5930}
5931
5932/*
Yaowei Bai34b10062015-09-08 15:04:13 -07005933 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005934 * or min_free_kbytes changes.
5935 */
5936static void calculate_totalreserve_pages(void)
5937{
5938 struct pglist_data *pgdat;
5939 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005940 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005941
5942 for_each_online_pgdat(pgdat) {
5943 for (i = 0; i < MAX_NR_ZONES; i++) {
5944 struct zone *zone = pgdat->node_zones + i;
Mel Gorman3484b2d2014-08-06 16:07:14 -07005945 long max = 0;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005946
5947 /* Find valid and maximum lowmem_reserve in the zone */
5948 for (j = i; j < MAX_NR_ZONES; j++) {
5949 if (zone->lowmem_reserve[j] > max)
5950 max = zone->lowmem_reserve[j];
5951 }
5952
Mel Gorman41858962009-06-16 15:32:12 -07005953 /* we treat the high watermark as reserved pages. */
5954 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005955
Jiang Liub40da042013-02-22 16:33:52 -08005956 if (max > zone->managed_pages)
5957 max = zone->managed_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005958 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005959 /*
5960 * Lowmem reserves are not available to
5961 * GFP_HIGHUSER page cache allocations and
5962 * kswapd tries to balance zones to their high
5963 * watermark. As a result, neither should be
5964 * regarded as dirtyable memory, to prevent a
5965 * situation where reclaim has to clean pages
5966 * in order to balance the zones.
5967 */
5968 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005969 }
5970 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005971 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005972 totalreserve_pages = reserve_pages;
5973}
5974
5975/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005976 * setup_per_zone_lowmem_reserve - called whenever
Yaowei Bai34b10062015-09-08 15:04:13 -07005977 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
Linus Torvalds1da177e2005-04-16 15:20:36 -07005978 * has a correct pages reserved value, so an adequate number of
5979 * pages are left in the zone after a successful __alloc_pages().
5980 */
5981static void setup_per_zone_lowmem_reserve(void)
5982{
5983 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005984 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005985
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005986 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005987 for (j = 0; j < MAX_NR_ZONES; j++) {
5988 struct zone *zone = pgdat->node_zones + j;
Jiang Liub40da042013-02-22 16:33:52 -08005989 unsigned long managed_pages = zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005990
5991 zone->lowmem_reserve[j] = 0;
5992
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005993 idx = j;
5994 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005995 struct zone *lower_zone;
5996
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005997 idx--;
5998
Linus Torvalds1da177e2005-04-16 15:20:36 -07005999 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6000 sysctl_lowmem_reserve_ratio[idx] = 1;
6001
6002 lower_zone = pgdat->node_zones + idx;
Jiang Liub40da042013-02-22 16:33:52 -08006003 lower_zone->lowmem_reserve[j] = managed_pages /
Linus Torvalds1da177e2005-04-16 15:20:36 -07006004 sysctl_lowmem_reserve_ratio[idx];
Jiang Liub40da042013-02-22 16:33:52 -08006005 managed_pages += lower_zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006006 }
6007 }
6008 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006009
6010 /* update totalreserve_pages */
6011 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006012}
6013
Mel Gormancfd3da12011-04-25 21:36:42 +00006014static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006015{
6016 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6017 unsigned long lowmem_pages = 0;
6018 struct zone *zone;
6019 unsigned long flags;
6020
6021 /* Calculate total number of !ZONE_HIGHMEM pages */
6022 for_each_zone(zone) {
6023 if (!is_highmem(zone))
Jiang Liub40da042013-02-22 16:33:52 -08006024 lowmem_pages += zone->managed_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006025 }
6026
6027 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07006028 u64 tmp;
6029
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006030 spin_lock_irqsave(&zone->lock, flags);
Jiang Liub40da042013-02-22 16:33:52 -08006031 tmp = (u64)pages_min * zone->managed_pages;
Andrew Mortonac924c62006-05-15 09:43:59 -07006032 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006033 if (is_highmem(zone)) {
6034 /*
Nick Piggin669ed172005-11-13 16:06:45 -08006035 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6036 * need highmem pages, so cap pages_min to a small
6037 * value here.
6038 *
Mel Gorman41858962009-06-16 15:32:12 -07006039 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Yaowei Bai42ff2702015-04-14 15:47:14 -07006040 * deltas control asynch page reclaim, and so should
Nick Piggin669ed172005-11-13 16:06:45 -08006041 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006042 */
Andrew Morton90ae8d62013-02-22 16:32:22 -08006043 unsigned long min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006044
Jiang Liub40da042013-02-22 16:33:52 -08006045 min_pages = zone->managed_pages / 1024;
Andrew Morton90ae8d62013-02-22 16:32:22 -08006046 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
Mel Gorman41858962009-06-16 15:32:12 -07006047 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006048 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08006049 /*
6050 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07006051 * proportionate to the zone's size.
6052 */
Mel Gorman41858962009-06-16 15:32:12 -07006053 zone->watermark[WMARK_MIN] = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006054 }
6055
Mel Gorman41858962009-06-16 15:32:12 -07006056 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
6057 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006058
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006059 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
Johannes Weinerabe5f972014-10-02 16:21:10 -07006060 high_wmark_pages(zone) - low_wmark_pages(zone) -
6061 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
Johannes Weiner81c0a2b2013-09-11 14:20:47 -07006062
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07006063 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006064 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07006065
6066 /* update totalreserve_pages */
6067 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006068}
6069
Mel Gormancfd3da12011-04-25 21:36:42 +00006070/**
6071 * setup_per_zone_wmarks - called when min_free_kbytes changes
6072 * or when memory is hot-{added|removed}
6073 *
6074 * Ensures that the watermark[min,low,high] values for each zone are set
6075 * correctly with respect to min_free_kbytes.
6076 */
6077void setup_per_zone_wmarks(void)
6078{
6079 mutex_lock(&zonelists_mutex);
6080 __setup_per_zone_wmarks();
6081 mutex_unlock(&zonelists_mutex);
6082}
6083
Randy Dunlap55a44622009-09-21 17:01:20 -07006084/*
Rik van Riel556adec2008-10-18 20:26:34 -07006085 * The inactive anon list should be small enough that the VM never has to
6086 * do too much work, but large enough that each inactive page has a chance
6087 * to be referenced again before it is swapped out.
6088 *
6089 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6090 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6091 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6092 * the anonymous pages are kept on the inactive list.
6093 *
6094 * total target max
6095 * memory ratio inactive anon
6096 * -------------------------------------
6097 * 10MB 1 5MB
6098 * 100MB 1 50MB
6099 * 1GB 3 250MB
6100 * 10GB 10 0.9GB
6101 * 100GB 31 3GB
6102 * 1TB 101 10GB
6103 * 10TB 320 32GB
6104 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07006105static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07006106{
6107 unsigned int gb, ratio;
6108
6109 /* Zone size in gigabytes */
Jiang Liub40da042013-02-22 16:33:52 -08006110 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
Minchan Kim96cb4df2009-06-16 15:32:49 -07006111 if (gb)
6112 ratio = int_sqrt(10 * gb);
6113 else
6114 ratio = 1;
6115
6116 zone->inactive_ratio = ratio;
6117}
6118
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07006119static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07006120{
6121 struct zone *zone;
6122
Minchan Kim96cb4df2009-06-16 15:32:49 -07006123 for_each_zone(zone)
6124 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07006125}
6126
Linus Torvalds1da177e2005-04-16 15:20:36 -07006127/*
6128 * Initialise min_free_kbytes.
6129 *
6130 * For small machines we want it small (128k min). For large machines
6131 * we want it large (64MB max). But it is not linear, because network
6132 * bandwidth does not increase linearly with machine size. We use
6133 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006134 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
Linus Torvalds1da177e2005-04-16 15:20:36 -07006135 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6136 *
6137 * which yields
6138 *
6139 * 16MB: 512k
6140 * 32MB: 724k
6141 * 64MB: 1024k
6142 * 128MB: 1448k
6143 * 256MB: 2048k
6144 * 512MB: 2896k
6145 * 1024MB: 4096k
6146 * 2048MB: 5792k
6147 * 4096MB: 8192k
6148 * 8192MB: 11584k
6149 * 16384MB: 16384k
6150 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07006151int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006152{
6153 unsigned long lowmem_kbytes;
Michal Hocko5f127332013-07-08 16:00:40 -07006154 int new_min_free_kbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006155
6156 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
Michal Hocko5f127332013-07-08 16:00:40 -07006157 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006158
Michal Hocko5f127332013-07-08 16:00:40 -07006159 if (new_min_free_kbytes > user_min_free_kbytes) {
6160 min_free_kbytes = new_min_free_kbytes;
6161 if (min_free_kbytes < 128)
6162 min_free_kbytes = 128;
6163 if (min_free_kbytes > 65536)
6164 min_free_kbytes = 65536;
6165 } else {
6166 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6167 new_min_free_kbytes, user_min_free_kbytes);
6168 }
Minchan Kimbc75d332009-06-16 15:32:48 -07006169 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07006170 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006171 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07006172 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006173 return 0;
6174}
Minchan Kimbc75d332009-06-16 15:32:48 -07006175module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006176
6177/*
Pintu Kumarb8af2942013-09-11 14:20:34 -07006178 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
Linus Torvalds1da177e2005-04-16 15:20:36 -07006179 * that we can call two helper functions whenever min_free_kbytes
6180 * changes.
6181 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006182int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006183 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006184{
Han Pingtianda8c7572014-01-23 15:53:17 -08006185 int rc;
6186
6187 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6188 if (rc)
6189 return rc;
6190
Michal Hocko5f127332013-07-08 16:00:40 -07006191 if (write) {
6192 user_min_free_kbytes = min_free_kbytes;
Minchan Kimbc75d332009-06-16 15:32:48 -07006193 setup_per_zone_wmarks();
Michal Hocko5f127332013-07-08 16:00:40 -07006194 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006195 return 0;
6196}
6197
Christoph Lameter96146342006-07-03 00:24:13 -07006198#ifdef CONFIG_NUMA
Joe Perchescccad5b2014-06-06 14:38:09 -07006199int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006200 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07006201{
6202 struct zone *zone;
6203 int rc;
6204
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006205 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07006206 if (rc)
6207 return rc;
6208
6209 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08006210 zone->min_unmapped_pages = (zone->managed_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07006211 sysctl_min_unmapped_ratio) / 100;
6212 return 0;
6213}
Christoph Lameter0ff38492006-09-25 23:31:52 -07006214
Joe Perchescccad5b2014-06-06 14:38:09 -07006215int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006216 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07006217{
6218 struct zone *zone;
6219 int rc;
6220
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006221 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07006222 if (rc)
6223 return rc;
6224
6225 for_each_zone(zone)
Jiang Liub40da042013-02-22 16:33:52 -08006226 zone->min_slab_pages = (zone->managed_pages *
Christoph Lameter0ff38492006-09-25 23:31:52 -07006227 sysctl_min_slab_ratio) / 100;
6228 return 0;
6229}
Christoph Lameter96146342006-07-03 00:24:13 -07006230#endif
6231
Linus Torvalds1da177e2005-04-16 15:20:36 -07006232/*
6233 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6234 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6235 * whenever sysctl_lowmem_reserve_ratio changes.
6236 *
6237 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07006238 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07006239 * if in function of the boot time zone sizes.
6240 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006241int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006242 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006243{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006244 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006245 setup_per_zone_lowmem_reserve();
6246 return 0;
6247}
6248
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006249/*
6250 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
Pintu Kumarb8af2942013-09-11 14:20:34 -07006251 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6252 * pagelist can have before it gets flushed back to buddy allocator.
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006253 */
Joe Perchescccad5b2014-06-06 14:38:09 -07006254int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07006255 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006256{
6257 struct zone *zone;
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006258 int old_percpu_pagelist_fraction;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006259 int ret;
6260
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006261 mutex_lock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006262 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6263
6264 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
6265 if (!write || ret < 0)
6266 goto out;
6267
6268 /* Sanity checking to avoid pcp imbalance */
6269 if (percpu_pagelist_fraction &&
6270 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6271 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6272 ret = -EINVAL;
6273 goto out;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006274 }
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006275
6276 /* No change? */
6277 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6278 goto out;
6279
6280 for_each_populated_zone(zone) {
6281 unsigned int cpu;
6282
6283 for_each_possible_cpu(cpu)
6284 pageset_set_high_and_batch(zone,
6285 per_cpu_ptr(zone->pageset, cpu));
6286 }
6287out:
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006288 mutex_unlock(&pcp_batch_high_lock);
David Rientjes7cd2b0a2014-06-23 13:22:04 -07006289 return ret;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08006290}
6291
Rasmus Villemoesa9919c72015-06-24 16:56:28 -07006292#ifdef CONFIG_NUMA
David S. Millerf034b5d2006-08-24 03:08:07 -07006293int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006294
Linus Torvalds1da177e2005-04-16 15:20:36 -07006295static int __init set_hashdist(char *str)
6296{
6297 if (!str)
6298 return 0;
6299 hashdist = simple_strtoul(str, &str, 0);
6300 return 1;
6301}
6302__setup("hashdist=", set_hashdist);
6303#endif
6304
6305/*
6306 * allocate a large system hash table from bootmem
6307 * - it is assumed that the hash table must contain an exact power-of-2
6308 * quantity of entries
6309 * - limit is the number of hash buckets, not the total allocation size
6310 */
6311void *__init alloc_large_system_hash(const char *tablename,
6312 unsigned long bucketsize,
6313 unsigned long numentries,
6314 int scale,
6315 int flags,
6316 unsigned int *_hash_shift,
6317 unsigned int *_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00006318 unsigned long low_limit,
6319 unsigned long high_limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006320{
Tim Bird31fe62b2012-05-23 13:33:35 +00006321 unsigned long long max = high_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006322 unsigned long log2qty, size;
6323 void *table = NULL;
6324
6325 /* allow the kernel cmdline to have a say */
6326 if (!numentries) {
6327 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08006328 numentries = nr_kernel_pages;
Jerry Zhoua7e83312013-09-11 14:20:26 -07006329
6330 /* It isn't necessary when PAGE_SIZE >= 1MB */
6331 if (PAGE_SHIFT < 20)
6332 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006333
6334 /* limit to 1 bucket per 2^scale bytes of low memory */
6335 if (scale > PAGE_SHIFT)
6336 numentries >>= (scale - PAGE_SHIFT);
6337 else
6338 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08006339
6340 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07006341 if (unlikely(flags & HASH_SMALL)) {
6342 /* Makes no sense without HASH_EARLY */
6343 WARN_ON(!(flags & HASH_EARLY));
6344 if (!(numentries >> *_hash_shift)) {
6345 numentries = 1UL << *_hash_shift;
6346 BUG_ON(!numentries);
6347 }
6348 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08006349 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006350 }
John Hawkes6e692ed2006-03-25 03:08:02 -08006351 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006352
6353 /* limit allocation size to 1/16 total memory by default */
6354 if (max == 0) {
6355 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6356 do_div(max, bucketsize);
6357 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08006358 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006359
Tim Bird31fe62b2012-05-23 13:33:35 +00006360 if (numentries < low_limit)
6361 numentries = low_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006362 if (numentries > max)
6363 numentries = max;
6364
David Howellsf0d1b0b2006-12-08 02:37:49 -08006365 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006366
6367 do {
6368 size = bucketsize << log2qty;
6369 if (flags & HASH_EARLY)
Santosh Shilimkar67828322014-01-21 15:50:25 -08006370 table = memblock_virt_alloc_nopanic(size, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006371 else if (hashdist)
6372 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6373 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07006374 /*
6375 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07006376 * some pages at the end of hash table which
6377 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07006378 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006379 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07006380 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01006381 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6382 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006383 }
6384 } while (!table && size > PAGE_SIZE && --log2qty);
6385
6386 if (!table)
6387 panic("Failed to allocate %s hash table\n", tablename);
6388
Robin Holtf241e6602010-10-07 12:59:26 -07006389 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07006390 tablename,
Robin Holtf241e6602010-10-07 12:59:26 -07006391 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08006392 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006393 size);
6394
6395 if (_hash_shift)
6396 *_hash_shift = log2qty;
6397 if (_hash_mask)
6398 *_hash_mask = (1 << log2qty) - 1;
6399
6400 return table;
6401}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08006402
Mel Gorman835c1342007-10-16 01:25:47 -07006403/* Return a pointer to the bitmap storing bits affecting a block of pages */
6404static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6405 unsigned long pfn)
6406{
6407#ifdef CONFIG_SPARSEMEM
6408 return __pfn_to_section(pfn)->pageblock_flags;
6409#else
6410 return zone->pageblock_flags;
6411#endif /* CONFIG_SPARSEMEM */
6412}
Andrew Morton6220ec72006-10-19 23:29:05 -07006413
Mel Gorman835c1342007-10-16 01:25:47 -07006414static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6415{
6416#ifdef CONFIG_SPARSEMEM
6417 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07006418 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006419#else
Laura Abbottc060f942013-01-11 14:31:51 -08006420 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07006421 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07006422#endif /* CONFIG_SPARSEMEM */
6423}
6424
6425/**
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006426 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
Mel Gorman835c1342007-10-16 01:25:47 -07006427 * @page: The page within the block of interest
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006428 * @pfn: The target page frame number
6429 * @end_bitidx: The last bit of interest to retrieve
6430 * @mask: mask of bits that the caller is interested in
6431 *
6432 * Return: pageblock_bits flags
Mel Gorman835c1342007-10-16 01:25:47 -07006433 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006434unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006435 unsigned long end_bitidx,
6436 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006437{
6438 struct zone *zone;
6439 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006440 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006441 unsigned long word;
Mel Gorman835c1342007-10-16 01:25:47 -07006442
6443 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006444 bitmap = get_pageblock_bitmap(zone, pfn);
6445 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006446 word_bitidx = bitidx / BITS_PER_LONG;
6447 bitidx &= (BITS_PER_LONG-1);
Mel Gorman835c1342007-10-16 01:25:47 -07006448
Mel Gormane58469b2014-06-04 16:10:16 -07006449 word = bitmap[word_bitidx];
6450 bitidx += end_bitidx;
6451 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
Mel Gorman835c1342007-10-16 01:25:47 -07006452}
6453
6454/**
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006455 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
Mel Gorman835c1342007-10-16 01:25:47 -07006456 * @page: The page within the block of interest
Mel Gorman835c1342007-10-16 01:25:47 -07006457 * @flags: The flags to set
Randy Dunlap1aab4d72014-07-27 14:15:33 -07006458 * @pfn: The target page frame number
6459 * @end_bitidx: The last bit of interest
6460 * @mask: mask of bits that the caller is interested in
Mel Gorman835c1342007-10-16 01:25:47 -07006461 */
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006462void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6463 unsigned long pfn,
Mel Gormane58469b2014-06-04 16:10:16 -07006464 unsigned long end_bitidx,
6465 unsigned long mask)
Mel Gorman835c1342007-10-16 01:25:47 -07006466{
6467 struct zone *zone;
6468 unsigned long *bitmap;
Mel Gormandc4b0ca2014-06-04 16:10:17 -07006469 unsigned long bitidx, word_bitidx;
Mel Gormane58469b2014-06-04 16:10:16 -07006470 unsigned long old_word, word;
6471
6472 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
Mel Gorman835c1342007-10-16 01:25:47 -07006473
6474 zone = page_zone(page);
Mel Gorman835c1342007-10-16 01:25:47 -07006475 bitmap = get_pageblock_bitmap(zone, pfn);
6476 bitidx = pfn_to_bitidx(zone, pfn);
Mel Gormane58469b2014-06-04 16:10:16 -07006477 word_bitidx = bitidx / BITS_PER_LONG;
6478 bitidx &= (BITS_PER_LONG-1);
6479
Sasha Levin309381fea2014-01-23 15:52:54 -08006480 VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
Mel Gorman835c1342007-10-16 01:25:47 -07006481
Mel Gormane58469b2014-06-04 16:10:16 -07006482 bitidx += end_bitidx;
6483 mask <<= (BITS_PER_LONG - bitidx - 1);
6484 flags <<= (BITS_PER_LONG - bitidx - 1);
6485
Jason Low4db0c3c2015-04-15 16:14:08 -07006486 word = READ_ONCE(bitmap[word_bitidx]);
Mel Gormane58469b2014-06-04 16:10:16 -07006487 for (;;) {
6488 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6489 if (word == old_word)
6490 break;
6491 word = old_word;
6492 }
Mel Gorman835c1342007-10-16 01:25:47 -07006493}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006494
6495/*
Minchan Kim80934512012-07-31 16:43:01 -07006496 * This function checks whether pageblock includes unmovable pages or not.
6497 * If @count is not zero, it is okay to include less @count unmovable pages
6498 *
Pintu Kumarb8af2942013-09-11 14:20:34 -07006499 * PageLRU check without isolation or lru_lock could race so that
Minchan Kim80934512012-07-31 16:43:01 -07006500 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6501 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006502 */
Wen Congyangb023f462012-12-11 16:00:45 -08006503bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6504 bool skip_hwpoisoned_pages)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006505{
6506 unsigned long pfn, iter, found;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006507 int mt;
6508
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006509 /*
6510 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim80934512012-07-31 16:43:01 -07006511 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006512 */
6513 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim80934512012-07-31 16:43:01 -07006514 return false;
Michal Nazarewicz47118af2011-12-29 13:09:50 +01006515 mt = get_pageblock_migratetype(page);
6516 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim80934512012-07-31 16:43:01 -07006517 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006518
6519 pfn = page_to_pfn(page);
6520 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6521 unsigned long check = pfn + iter;
6522
Namhyung Kim29723fc2011-02-25 14:44:25 -08006523 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006524 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08006525
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006526 page = pfn_to_page(check);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07006527
6528 /*
6529 * Hugepages are not in LRU lists, but they're movable.
6530 * We need not scan over tail pages bacause we don't
6531 * handle each tail page individually in migration.
6532 */
6533 if (PageHuge(page)) {
6534 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6535 continue;
6536 }
6537
Minchan Kim97d255c2012-07-31 16:42:59 -07006538 /*
6539 * We can't use page_count without pin a page
6540 * because another CPU can free compound page.
6541 * This check already skips compound tails of THP
6542 * because their page->_count is zero at all time.
6543 */
6544 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006545 if (PageBuddy(page))
6546 iter += (1 << page_order(page)) - 1;
6547 continue;
6548 }
Minchan Kim97d255c2012-07-31 16:42:59 -07006549
Wen Congyangb023f462012-12-11 16:00:45 -08006550 /*
6551 * The HWPoisoned page may be not in buddy system, and
6552 * page_count() is not 0.
6553 */
6554 if (skip_hwpoisoned_pages && PageHWPoison(page))
6555 continue;
6556
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006557 if (!PageLRU(page))
6558 found++;
6559 /*
Johannes Weiner6b4f7792014-12-12 16:56:13 -08006560 * If there are RECLAIMABLE pages, we need to check
6561 * it. But now, memory offline itself doesn't call
6562 * shrink_node_slabs() and it still to be fixed.
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006563 */
6564 /*
6565 * If the page is not RAM, page_count()should be 0.
6566 * we don't need more check. This is an _used_ not-movable page.
6567 *
6568 * The problematic thing here is PG_reserved pages. PG_reserved
6569 * is set to both of a memory hole page and a _used_ kernel
6570 * page at boot.
6571 */
6572 if (found > count)
Minchan Kim80934512012-07-31 16:43:01 -07006573 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006574 }
Minchan Kim80934512012-07-31 16:43:01 -07006575 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07006576}
6577
6578bool is_pageblock_removable_nolock(struct page *page)
6579{
Michal Hocko656a0702012-01-20 14:33:58 -08006580 struct zone *zone;
6581 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08006582
6583 /*
6584 * We have to be careful here because we are iterating over memory
6585 * sections which are not zone aware so we might end up outside of
6586 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08006587 * We have to take care about the node as well. If the node is offline
6588 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08006589 */
Michal Hocko656a0702012-01-20 14:33:58 -08006590 if (!node_online(page_to_nid(page)))
6591 return false;
6592
6593 zone = page_zone(page);
6594 pfn = page_to_pfn(page);
Cody P Schafer108bcc92013-02-22 16:35:23 -08006595 if (!zone_spans_pfn(zone, pfn))
Michal Hocko687875f2012-01-20 14:33:55 -08006596 return false;
6597
Wen Congyangb023f462012-12-11 16:00:45 -08006598 return !has_unmovable_pages(zone, page, 0, true);
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07006599}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006600
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006601#ifdef CONFIG_CMA
6602
6603static unsigned long pfn_max_align_down(unsigned long pfn)
6604{
6605 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6606 pageblock_nr_pages) - 1);
6607}
6608
6609static unsigned long pfn_max_align_up(unsigned long pfn)
6610{
6611 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6612 pageblock_nr_pages));
6613}
6614
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006615/* [start, end) must belong to a single zone. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006616static int __alloc_contig_migrate_range(struct compact_control *cc,
6617 unsigned long start, unsigned long end)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006618{
6619 /* This function is based on compact_zone() from compaction.c. */
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006620 unsigned long nr_reclaimed;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006621 unsigned long pfn = start;
6622 unsigned int tries = 0;
6623 int ret = 0;
6624
Marek Szyprowskibe49a6e2012-12-12 13:51:19 -08006625 migrate_prep();
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006626
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006627 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006628 if (fatal_signal_pending(current)) {
6629 ret = -EINTR;
6630 break;
6631 }
6632
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006633 if (list_empty(&cc->migratepages)) {
6634 cc->nr_migratepages = 0;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07006635 pfn = isolate_migratepages_range(cc, pfn, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006636 if (!pfn) {
6637 ret = -EINTR;
6638 break;
6639 }
6640 tries = 0;
6641 } else if (++tries == 5) {
6642 ret = ret < 0 ? ret : -EBUSY;
6643 break;
6644 }
6645
Minchan Kimbeb51ea2012-10-08 16:33:51 -07006646 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6647 &cc->migratepages);
6648 cc->nr_migratepages -= nr_reclaimed;
Minchan Kim02c6de82012-10-08 16:31:55 -07006649
Hugh Dickins9c620e22013-02-22 16:35:14 -08006650 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
David Rientjese0b9dae2014-06-04 16:08:28 -07006651 NULL, 0, cc->mode, MR_CMA);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006652 }
Srinivas Pandruvada2a6f5122013-02-22 16:32:09 -08006653 if (ret < 0) {
6654 putback_movable_pages(&cc->migratepages);
6655 return ret;
6656 }
6657 return 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006658}
6659
6660/**
6661 * alloc_contig_range() -- tries to allocate given range of pages
6662 * @start: start PFN to allocate
6663 * @end: one-past-the-last PFN to allocate
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006664 * @migratetype: migratetype of the underlaying pageblocks (either
6665 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6666 * in range must have the same migratetype and it must
6667 * be either of the two.
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006668 *
6669 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6670 * aligned, however it's the caller's responsibility to guarantee that
6671 * we are the only thread that changes migrate type of pageblocks the
6672 * pages fall in.
6673 *
6674 * The PFN range must belong to a single zone.
6675 *
6676 * Returns zero on success or negative error code. On success all
6677 * pages which PFN is in [start, end) are allocated for the caller and
6678 * need to be freed with free_contig_range().
6679 */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006680int alloc_contig_range(unsigned long start, unsigned long end,
6681 unsigned migratetype)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006682{
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006683 unsigned long outer_start, outer_end;
Kirill A. Shutemovd00181b2015-11-06 16:29:57 -08006684 unsigned int order;
6685 int ret = 0;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006686
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006687 struct compact_control cc = {
6688 .nr_migratepages = 0,
6689 .order = -1,
6690 .zone = page_zone(pfn_to_page(start)),
David Rientjese0b9dae2014-06-04 16:08:28 -07006691 .mode = MIGRATE_SYNC,
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006692 .ignore_skip_hint = true,
6693 };
6694 INIT_LIST_HEAD(&cc.migratepages);
6695
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006696 /*
6697 * What we do here is we mark all pageblocks in range as
6698 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6699 * have different sizes, and due to the way page allocator
6700 * work, we align the range to biggest of the two pages so
6701 * that page allocator won't try to merge buddies from
6702 * different pageblocks and change MIGRATE_ISOLATE to some
6703 * other migration type.
6704 *
6705 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6706 * migrate the pages from an unaligned range (ie. pages that
6707 * we are interested in). This will put all the pages in
6708 * range back to page allocator as MIGRATE_ISOLATE.
6709 *
6710 * When this is done, we take the pages in range from page
6711 * allocator removing them from the buddy system. This way
6712 * page allocator will never consider using them.
6713 *
6714 * This lets us mark the pageblocks back as
6715 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6716 * aligned range but not in the unaligned, original range are
6717 * put back to page allocator so that buddy can use them.
6718 */
6719
6720 ret = start_isolate_page_range(pfn_max_align_down(start),
Wen Congyangb023f462012-12-11 16:00:45 -08006721 pfn_max_align_up(end), migratetype,
6722 false);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006723 if (ret)
Bob Liu86a595f2012-10-25 13:37:56 -07006724 return ret;
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006725
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006726 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006727 if (ret)
6728 goto done;
6729
6730 /*
6731 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6732 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6733 * more, all pages in [start, end) are free in page allocator.
6734 * What we are going to do is to allocate all pages from
6735 * [start, end) (that is remove them from page allocator).
6736 *
6737 * The only problem is that pages at the beginning and at the
6738 * end of interesting range may be not aligned with pages that
6739 * page allocator holds, ie. they can be part of higher order
6740 * pages. Because of this, we reserve the bigger range and
6741 * once this is done free the pages we are not interested in.
6742 *
6743 * We don't have to hold zone->lock here because the pages are
6744 * isolated thus they won't get removed from buddy.
6745 */
6746
6747 lru_add_drain_all();
Vlastimil Babka510f5502014-12-10 15:43:07 -08006748 drain_all_pages(cc.zone);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006749
6750 order = 0;
6751 outer_start = start;
6752 while (!PageBuddy(pfn_to_page(outer_start))) {
6753 if (++order >= MAX_ORDER) {
6754 ret = -EBUSY;
6755 goto done;
6756 }
6757 outer_start &= ~0UL << order;
6758 }
6759
6760 /* Make sure the range is really isolated. */
Wen Congyangb023f462012-12-11 16:00:45 -08006761 if (test_pages_isolated(outer_start, end, false)) {
Michal Nazarewiczdae803e2014-11-13 15:19:27 -08006762 pr_info("%s: [%lx, %lx) PFNs busy\n",
6763 __func__, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006764 ret = -EBUSY;
6765 goto done;
6766 }
6767
Marek Szyprowski49f223a2012-01-25 12:49:24 +01006768 /* Grab isolated pages from freelists. */
Mel Gormanbb13ffe2012-10-08 16:32:41 -07006769 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006770 if (!outer_end) {
6771 ret = -EBUSY;
6772 goto done;
6773 }
6774
6775 /* Free head and tail (if any) */
6776 if (start != outer_start)
6777 free_contig_range(outer_start, start - outer_start);
6778 if (end != outer_end)
6779 free_contig_range(end, outer_end - end);
6780
6781done:
6782 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +02006783 pfn_max_align_up(end), migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006784 return ret;
6785}
6786
6787void free_contig_range(unsigned long pfn, unsigned nr_pages)
6788{
Marek Szyprowskibcc2b022012-12-20 15:05:18 -08006789 unsigned int count = 0;
6790
6791 for (; nr_pages--; pfn++) {
6792 struct page *page = pfn_to_page(pfn);
6793
6794 count += page_count(page) != 1;
6795 __free_page(page);
6796 }
6797 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +01006798}
6799#endif
6800
Jiang Liu4ed7e022012-07-31 16:43:35 -07006801#ifdef CONFIG_MEMORY_HOTPLUG
Cody P Schafer0a647f32013-07-03 15:01:33 -07006802/*
6803 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6804 * page high values need to be recalulated.
6805 */
Jiang Liu4ed7e022012-07-31 16:43:35 -07006806void __meminit zone_pcp_update(struct zone *zone)
6807{
Cody P Schafer0a647f32013-07-03 15:01:33 -07006808 unsigned cpu;
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006809 mutex_lock(&pcp_batch_high_lock);
Cody P Schafer0a647f32013-07-03 15:01:33 -07006810 for_each_possible_cpu(cpu)
Cody P Schafer169f6c12013-07-03 15:01:41 -07006811 pageset_set_high_and_batch(zone,
6812 per_cpu_ptr(zone->pageset, cpu));
Cody P Schaferc8e251f2013-07-03 15:01:29 -07006813 mutex_unlock(&pcp_batch_high_lock);
Jiang Liu4ed7e022012-07-31 16:43:35 -07006814}
6815#endif
6816
Jiang Liu340175b2012-07-31 16:43:32 -07006817void zone_pcp_reset(struct zone *zone)
6818{
6819 unsigned long flags;
Minchan Kim5a883812012-10-08 16:33:39 -07006820 int cpu;
6821 struct per_cpu_pageset *pset;
Jiang Liu340175b2012-07-31 16:43:32 -07006822
6823 /* avoid races with drain_pages() */
6824 local_irq_save(flags);
6825 if (zone->pageset != &boot_pageset) {
Minchan Kim5a883812012-10-08 16:33:39 -07006826 for_each_online_cpu(cpu) {
6827 pset = per_cpu_ptr(zone->pageset, cpu);
6828 drain_zonestat(zone, pset);
6829 }
Jiang Liu340175b2012-07-31 16:43:32 -07006830 free_percpu(zone->pageset);
6831 zone->pageset = &boot_pageset;
6832 }
6833 local_irq_restore(flags);
6834}
6835
Wen Congyang6dcd73d2012-12-11 16:01:01 -08006836#ifdef CONFIG_MEMORY_HOTREMOVE
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006837/*
6838 * All pages in the range must be isolated before calling this.
6839 */
6840void
6841__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6842{
6843 struct page *page;
6844 struct zone *zone;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006845 unsigned int order, i;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006846 unsigned long pfn;
6847 unsigned long flags;
6848 /* find the first valid pfn */
6849 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6850 if (pfn_valid(pfn))
6851 break;
6852 if (pfn == end_pfn)
6853 return;
6854 zone = page_zone(pfn_to_page(pfn));
6855 spin_lock_irqsave(&zone->lock, flags);
6856 pfn = start_pfn;
6857 while (pfn < end_pfn) {
6858 if (!pfn_valid(pfn)) {
6859 pfn++;
6860 continue;
6861 }
6862 page = pfn_to_page(pfn);
Wen Congyangb023f462012-12-11 16:00:45 -08006863 /*
6864 * The HWPoisoned page may be not in buddy system, and
6865 * page_count() is not 0.
6866 */
6867 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6868 pfn++;
6869 SetPageReserved(page);
6870 continue;
6871 }
6872
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006873 BUG_ON(page_count(page));
6874 BUG_ON(!PageBuddy(page));
6875 order = page_order(page);
6876#ifdef CONFIG_DEBUG_VM
6877 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6878 pfn, 1 << order, end_pfn);
6879#endif
6880 list_del(&page->lru);
6881 rmv_page_order(page);
6882 zone->free_area[order].nr_free--;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006883 for (i = 0; i < (1 << order); i++)
6884 SetPageReserved((page+i));
6885 pfn += (1 << order);
6886 }
6887 spin_unlock_irqrestore(&zone->lock, flags);
6888}
6889#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006890
6891#ifdef CONFIG_MEMORY_FAILURE
6892bool is_free_buddy_page(struct page *page)
6893{
6894 struct zone *zone = page_zone(page);
6895 unsigned long pfn = page_to_pfn(page);
6896 unsigned long flags;
Mel Gorman7aeb09f2014-06-04 16:10:21 -07006897 unsigned int order;
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006898
6899 spin_lock_irqsave(&zone->lock, flags);
6900 for (order = 0; order < MAX_ORDER; order++) {
6901 struct page *page_head = page - (pfn & ((1 << order) - 1));
6902
6903 if (PageBuddy(page_head) && page_order(page_head) >= order)
6904 break;
6905 }
6906 spin_unlock_irqrestore(&zone->lock, flags);
6907
6908 return order < MAX_ORDER;
6909}
6910#endif