blob: cf268dfb5f5632c13ddde5ab7904fdf74af3264c [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/module.h>
29#include <linux/suspend.h>
30#include <linux/pagevec.h>
31#include <linux/blkdev.h>
32#include <linux/slab.h>
Dave Hansena238ab52011-05-24 17:12:16 -070033#include <linux/ratelimit.h>
David Rientjes5a3135c2007-10-16 23:25:53 -070034#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/notifier.h>
36#include <linux/topology.h>
37#include <linux/sysctl.h>
38#include <linux/cpu.h>
39#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070040#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/nodemask.h>
42#include <linux/vmalloc.h>
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -070043#include <linux/vmstat.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080044#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070045#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070046#include <linux/sort.h>
47#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070048#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080049#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070050#include <linux/page-isolation.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070051#include <linux/page_cgroup.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070052#include <linux/debugobjects.h>
Catalin Marinasdbb1f812009-06-11 13:23:19 +010053#include <linux/kmemleak.h>
Robert Jennings925cc712009-12-17 14:44:38 +000054#include <linux/memory.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>
Wu Fengguang718a3822010-03-10 15:20:43 -080057#include <linux/ftrace_event.h>
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -070058#include <linux/memcontrol.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070059#include <linux/prefetch.h>
Lisa Due0935212013-09-11 14:22:36 -070060#include <linux/mm_inline.h>
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +010061#include <linux/migrate.h>
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -080062#include <linux/page-debug-flags.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
64#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070065#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066#include "internal.h"
67
Lee Schermerhorn72812012010-05-26 14:44:56 -070068#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
69DEFINE_PER_CPU(int, numa_node);
70EXPORT_PER_CPU_SYMBOL(numa_node);
71#endif
72
Lee Schermerhorn7aac7892010-05-26 14:45:00 -070073#ifdef CONFIG_HAVE_MEMORYLESS_NODES
74/*
75 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
76 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
77 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
78 * defined in <linux/topology.h>.
79 */
80DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
81EXPORT_PER_CPU_SYMBOL(_numa_mem_);
82#endif
83
Linus Torvalds1da177e2005-04-16 15:20:36 -070084/*
Christoph Lameter13808912007-10-16 01:25:27 -070085 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 */
Christoph Lameter13808912007-10-16 01:25:27 -070087nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
88 [N_POSSIBLE] = NODE_MASK_ALL,
89 [N_ONLINE] = { { [0] = 1UL } },
90#ifndef CONFIG_NUMA
91 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
92#ifdef CONFIG_HIGHMEM
93 [N_HIGH_MEMORY] = { { [0] = 1UL } },
94#endif
95 [N_CPU] = { { [0] = 1UL } },
96#endif /* NUMA */
97};
98EXPORT_SYMBOL(node_states);
99
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700100unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -0700101unsigned long totalreserve_pages __read_mostly;
Johannes Weinerab8fabd2012-01-10 15:07:42 -0800102/*
103 * When calculating the number of globally allowed dirty pages, there
104 * is a certain number of per-zone reserves that should not be
105 * considered dirtyable memory. This is the sum of those reserves
106 * over all existing zones that contribute dirtyable memory.
107 */
108unsigned long dirty_balance_reserve __read_mostly;
109
Jack Cheung59f9f1c2011-11-29 16:52:49 -0800110#ifdef CONFIG_FIX_MOVABLE_ZONE
111unsigned long total_unmovable_pages __read_mostly;
112#endif
Rohit Seth8ad4b1f2006-01-08 01:00:40 -0800113int percpu_pagelist_fraction;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000114gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800116#ifdef CONFIG_PM_SLEEP
117/*
118 * The following functions are used by the suspend/hibernate code to temporarily
119 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
120 * while devices are suspended. To avoid races with the suspend/hibernate code,
121 * they should always be called with pm_mutex held (gfp_allowed_mask also should
122 * only be modified with pm_mutex held, unless the suspend/hibernate code is
123 * guaranteed not to run in parallel with that modification).
124 */
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100125
126static gfp_t saved_gfp_mask;
127
128void pm_restore_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800129{
130 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100131 if (saved_gfp_mask) {
132 gfp_allowed_mask = saved_gfp_mask;
133 saved_gfp_mask = 0;
134 }
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800135}
136
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100137void pm_restrict_gfp_mask(void)
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800138{
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800139 WARN_ON(!mutex_is_locked(&pm_mutex));
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100140 WARN_ON(saved_gfp_mask);
141 saved_gfp_mask = gfp_allowed_mask;
142 gfp_allowed_mask &= ~GFP_IOFS;
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800143}
Mel Gormanf90ac392012-01-10 15:07:15 -0800144
145bool pm_suspended_storage(void)
146{
147 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
148 return false;
149 return true;
150}
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800151#endif /* CONFIG_PM_SLEEP */
152
Mel Gormand9c23402007-10-16 01:26:01 -0700153#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
154int pageblock_order __read_mostly;
155#endif
156
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800157static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -0800158
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159/*
160 * results with 256, 32 in the lowmem_reserve sysctl:
161 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
162 * 1G machine -> (16M dma, 784M normal, 224M high)
163 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
164 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
165 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +0100166 *
167 * TBD: should special case ZONE_DMA32 machines here - in those we normally
168 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700170int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800171#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700172 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -0800173#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700174#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700175 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700176#endif
Christoph Lametere53ef382006-09-25 23:31:14 -0700177#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700178 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700179#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700180 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700181};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
183EXPORT_SYMBOL(totalram_pages);
Jack Cheung59f9f1c2011-11-29 16:52:49 -0800184#ifdef CONFIG_FIX_MOVABLE_ZONE
185EXPORT_SYMBOL(total_unmovable_pages);
186#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
Helge Deller15ad7cd2006-12-06 20:40:36 -0800188static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800189#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700190 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800191#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700192#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700193 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700194#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700195 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700196#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700197 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700198#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700199 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700200};
201
Rik van Riel3b321eb2011-09-01 19:26:50 +0000202/*
203 * Try to keep at least this much lowmem free. Do not allow normal
204 * allocations below this point, only high priority ones. Automatically
205 * tuned according to the amount of memory in the system.
206 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207int min_free_kbytes = 1024;
Arve Hjønnevågcde89492009-02-17 14:51:02 -0800208int min_free_order_shift = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
Rik van Riel3b321eb2011-09-01 19:26:50 +0000210/*
211 * Extra memory for the system to try freeing. Used to temporarily
212 * free memory, to make space for new workloads. Anyone can allocate
213 * down to the min watermarks controlled by min_free_kbytes above.
214 */
215int extra_free_kbytes = 0;
216
Jan Beulich2c85f512009-09-21 17:03:07 -0700217static unsigned long __meminitdata nr_kernel_pages;
218static unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700219static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
Tejun Heo0ee332c2011-12-08 10:22:09 -0800221#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
222static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
223static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
224static unsigned long __initdata required_kernelcore;
225static unsigned long __initdata required_movablecore;
226static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gormanc7132162006-09-27 01:49:43 -0700227
Tejun Heo0ee332c2011-12-08 10:22:09 -0800228/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
229int movable_zone;
230EXPORT_SYMBOL(movable_zone);
231#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -0700232
Miklos Szeredi418508c2007-05-23 13:57:55 -0700233#if MAX_NUMNODES > 1
234int nr_node_ids __read_mostly = MAX_NUMNODES;
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700235int nr_online_nodes __read_mostly = 1;
Miklos Szeredi418508c2007-05-23 13:57:55 -0700236EXPORT_SYMBOL(nr_node_ids);
Christoph Lameter62bc62a2009-06-16 15:32:15 -0700237EXPORT_SYMBOL(nr_online_nodes);
Miklos Szeredi418508c2007-05-23 13:57:55 -0700238#endif
239
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700240int page_group_by_mobility_disabled __read_mostly;
241
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700242static void set_pageblock_migratetype(struct page *page, int migratetype)
243{
Mel Gorman49255c62009-06-16 15:31:58 -0700244
245 if (unlikely(page_group_by_mobility_disabled))
246 migratetype = MIGRATE_UNMOVABLE;
247
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700248 set_pageblock_flags_group(page, (unsigned long)migratetype,
249 PB_migrate, PB_migrate_end);
250}
251
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700252bool oom_killer_disabled __read_mostly;
253
Nick Piggin13e74442006-01-06 00:10:58 -0800254#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700255static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700257 int ret = 0;
258 unsigned seq;
259 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700260
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700261 do {
262 seq = zone_span_seqbegin(zone);
263 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
264 ret = 1;
265 else if (pfn < zone->zone_start_pfn)
266 ret = 1;
267 } while (zone_span_seqretry(zone, seq));
268
269 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700270}
271
272static int page_is_consistent(struct zone *zone, struct page *page)
273{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700274 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700275 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700277 return 0;
278
279 return 1;
280}
281/*
282 * Temporary debugging check for pages not lying within a given zone.
283 */
284static int bad_range(struct zone *zone, struct page *page)
285{
286 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700288 if (!page_is_consistent(zone, page))
289 return 1;
290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 return 0;
292}
Nick Piggin13e74442006-01-06 00:10:58 -0800293#else
294static inline int bad_range(struct zone *zone, struct page *page)
295{
296 return 0;
297}
298#endif
299
Nick Piggin224abf92006-01-06 00:11:11 -0800300static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301{
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800302 static unsigned long resume;
303 static unsigned long nr_shown;
304 static unsigned long nr_unshown;
305
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200306 /* Don't complain about poisoned pages */
307 if (PageHWPoison(page)) {
Andrea Arcangelief2b4b92011-03-18 00:16:35 +0100308 reset_page_mapcount(page); /* remove PageBuddy */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200309 return;
310 }
311
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800312 /*
313 * Allow a burst of 60 reports, then keep quiet for that minute;
314 * or allow a steady drip of one report per second.
315 */
316 if (nr_shown == 60) {
317 if (time_before(jiffies, resume)) {
318 nr_unshown++;
319 goto out;
320 }
321 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800322 printk(KERN_ALERT
323 "BUG: Bad page state: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800324 nr_unshown);
325 nr_unshown = 0;
326 }
327 nr_shown = 0;
328 }
329 if (nr_shown++ == 0)
330 resume = jiffies + 60 * HZ;
331
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800332 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800333 current->comm, page_to_pfn(page));
Wu Fengguang718a3822010-03-10 15:20:43 -0800334 dump_page(page);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700335
Dave Jones4f318882011-10-31 17:07:24 -0700336 print_modules();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 dump_stack();
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800338out:
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800339 /* Leave bad fields for debug, except PageBuddy could make trouble */
Andrea Arcangelief2b4b92011-03-18 00:16:35 +0100340 reset_page_mapcount(page); /* remove PageBuddy */
Randy Dunlap9f158332005-09-13 01:25:16 -0700341 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342}
343
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344/*
345 * Higher-order pages are called "compound pages". They are structured thusly:
346 *
347 * The first PAGE_SIZE page is called the "head page".
348 *
349 * The remaining PAGE_SIZE pages are called "tail pages".
350 *
Wang Sheng-Hui6416b9f2011-11-17 10:53:50 +0100351 * All pages have PG_compound set. All tail pages have their ->first_page
352 * pointing at the head page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800354 * The first tail page's ->lru.next holds the address of the compound page's
355 * put_page() function. Its ->lru.prev holds the order of allocation.
356 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800358
359static void free_compound_page(struct page *page)
360{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700361 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800362}
363
Andi Kleen01ad1c02008-07-23 21:27:46 -0700364void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365{
366 int i;
367 int nr_pages = 1 << order;
368
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800369 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700370 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700371 __SetPageHead(page);
Andy Whitcroft18229df2008-11-06 12:53:27 -0800372 for (i = 1; i < nr_pages; i++) {
373 struct page *p = page + i;
Christoph Lameterd85f3382007-05-06 14:49:39 -0700374 __SetPageTail(p);
Youquan Song58a84aa2011-12-08 14:34:18 -0800375 set_page_count(p, 0);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700376 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 }
378}
379
Andrea Arcangeli59ff4212011-01-13 15:46:38 -0800380/* update __split_huge_page_refcount if you change this function */
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800381static int destroy_compound_page(struct page *page, unsigned long order)
Andy Whitcroft18229df2008-11-06 12:53:27 -0800382{
383 int i;
384 int nr_pages = 1 << order;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800385 int bad = 0;
Andy Whitcroft18229df2008-11-06 12:53:27 -0800386
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800387 if (unlikely(compound_order(page) != order) ||
388 unlikely(!PageHead(page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800389 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800390 bad++;
391 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
Christoph Lameter6d777952007-05-06 14:49:40 -0700393 __ClearPageHead(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800394
Andy Whitcroft18229df2008-11-06 12:53:27 -0800395 for (i = 1; i < nr_pages; i++) {
396 struct page *p = page + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
Alexey Zaytseve713a212009-01-10 02:47:57 +0300398 if (unlikely(!PageTail(p) || (p->first_page != page))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800399 bad_page(page);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800400 bad++;
401 }
Christoph Lameterd85f3382007-05-06 14:49:39 -0700402 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 }
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800404
405 return bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
Nick Piggin17cf4402006-03-22 00:08:41 -0800408static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
409{
410 int i;
411
Andrew Morton6626c5d2006-03-22 00:08:42 -0800412 /*
413 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
414 * and __GFP_HIGHMEM from hard or soft interrupt context.
415 */
Nick Piggin725d7042006-09-25 23:30:55 -0700416 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800417 for (i = 0; i < (1 << order); i++)
418 clear_highpage(page + i);
419}
420
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800421#ifdef CONFIG_DEBUG_PAGEALLOC
422unsigned int _debug_guardpage_minorder;
423
424static int __init debug_guardpage_minorder_setup(char *buf)
425{
426 unsigned long res;
427
428 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
429 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
430 return 0;
431 }
432 _debug_guardpage_minorder = res;
433 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
434 return 0;
435}
436__setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
437
438static inline void set_page_guard_flag(struct page *page)
439{
440 __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
441}
442
443static inline void clear_page_guard_flag(struct page *page)
444{
445 __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
446}
447#else
448static inline void set_page_guard_flag(struct page *page) { }
449static inline void clear_page_guard_flag(struct page *page) { }
450#endif
451
Andrew Morton6aa30012006-04-18 22:20:52 -0700452static inline void set_page_order(struct page *page, int order)
453{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700454 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000455 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456}
457
458static inline void rmv_page_order(struct page *page)
459{
Nick Piggin676165a2006-04-10 11:21:48 +1000460 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700461 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462}
463
464/*
465 * Locate the struct page for both the matching buddy in our
466 * pair (buddy1) and the combined O(n+1) page they form (page).
467 *
468 * 1) Any buddy B1 will have an order O twin B2 which satisfies
469 * the following equation:
470 * B2 = B1 ^ (1 << O)
471 * For example, if the starting buddy (buddy2) is #8 its order
472 * 1 buddy is #10:
473 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
474 *
475 * 2) Any buddy B will have an order O+1 parent P which
476 * satisfies the following equation:
477 * P = B & ~(1 << O)
478 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200479 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481static inline unsigned long
KyongHo Cho43506fa2011-01-13 15:47:24 -0800482__find_buddy_index(unsigned long page_idx, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483{
KyongHo Cho43506fa2011-01-13 15:47:24 -0800484 return page_idx ^ (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485}
486
487/*
488 * This function checks whether a page is free && is the buddy
489 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800490 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000491 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700492 * (c) a page and its buddy have the same order &&
493 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 *
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800495 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
496 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
Nick Piggin676165a2006-04-10 11:21:48 +1000497 *
498 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700500static inline int page_is_buddy(struct page *page, struct page *buddy,
501 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700503 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800504 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800505
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700506 if (page_zone_id(page) != page_zone_id(buddy))
507 return 0;
508
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800509 if (page_is_guard(buddy) && page_order(buddy) == order) {
510 VM_BUG_ON(page_count(buddy) != 0);
511 return 1;
512 }
513
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700514 if (PageBuddy(buddy) && page_order(buddy) == order) {
Nick Piggina3af9c32009-06-16 15:32:10 -0700515 VM_BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700516 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000517 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700518 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519}
520
521/*
522 * Freeing function for a buddy system allocator.
523 *
524 * The concept of a buddy system is to maintain direct-mapped table
525 * (containing bit values) for memory blocks of various "orders".
526 * The bottom level table contains the map for the smallest allocatable
527 * units of memory (here, pages), and each level above it describes
528 * pairs of units from the levels below, hence, "buddies".
529 * At a high level, all that happens here is marking the table entry
530 * at the bottom level available, and propagating the changes upward
531 * as necessary, plus some accounting needed to play nicely with other
532 * parts of the VM system.
533 * At each level, we keep a list of pages, which are heads of continuous
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800534 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700535 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 * So when we are allocating or freeing one, we can derive the state of the
Michal Nazarewiczadc95fd2012-01-11 15:16:11 +0100537 * other. That is, if we allocate a small block, and both were
538 * free, the remainder of the region must be split into blocks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 * If a block is freed, and its buddy is also free, then this
Michal Nazarewiczadc95fd2012-01-11 15:16:11 +0100540 * triggers coalescing into a block of larger size.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 *
542 * -- wli
543 */
544
Nick Piggin48db57f2006-01-08 01:00:42 -0800545static inline void __free_one_page(struct page *page,
Mel Gormaned0ae212009-06-16 15:32:07 -0700546 struct zone *zone, unsigned int order,
547 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
549 unsigned long page_idx;
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700550 unsigned long combined_idx;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800551 unsigned long uninitialized_var(buddy_idx);
Shashank Mittal53103492012-06-19 19:45:35 -0700552 struct page *buddy = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Nick Piggin224abf92006-01-06 00:11:11 -0800554 if (unlikely(PageCompound(page)))
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800555 if (unlikely(destroy_compound_page(page, order)))
556 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
Mel Gormaned0ae212009-06-16 15:32:07 -0700558 VM_BUG_ON(migratetype == -1);
559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
561
Mel Gormanf2260e62009-06-16 15:32:13 -0700562 VM_BUG_ON(page_idx & ((1 << order) - 1));
Nick Piggin725d7042006-09-25 23:30:55 -0700563 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 while (order < MAX_ORDER-1) {
KyongHo Cho43506fa2011-01-13 15:47:24 -0800566 buddy_idx = __find_buddy_index(page_idx, order);
567 buddy = page + (buddy_idx - page_idx);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700568 if (!page_is_buddy(page, buddy, order))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700569 break;
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800570 /*
571 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
572 * merge with it and move up one order.
573 */
574 if (page_is_guard(buddy)) {
575 clear_page_guard_flag(buddy);
576 set_page_private(page, 0);
Larry Bassel6d6e2c92012-12-14 14:21:05 -0800577 __mod_zone_freepage_state(zone, 1 << order,
578 migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800579 } else {
580 list_del(&buddy->lru);
581 zone->free_area[order].nr_free--;
582 rmv_page_order(buddy);
583 }
KyongHo Cho43506fa2011-01-13 15:47:24 -0800584 combined_idx = buddy_idx & page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 page = page + (combined_idx - page_idx);
586 page_idx = combined_idx;
587 order++;
588 }
589 set_page_order(page, order);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700590
591 /*
592 * If this is not the largest possible page, check if the buddy
593 * of the next-highest order is free. If it is, it's possible
594 * that pages are being freed that will coalesce soon. In case,
595 * that is happening, add the free page to the tail of the list
596 * so it's less likely to be used soon and more likely to be merged
597 * as a higher order page
598 */
Mel Gormanb7f50cf2010-10-26 14:21:11 -0700599 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700600 struct page *higher_page, *higher_buddy;
KyongHo Cho43506fa2011-01-13 15:47:24 -0800601 combined_idx = buddy_idx & page_idx;
602 higher_page = page + (combined_idx - page_idx);
603 buddy_idx = __find_buddy_index(combined_idx, order + 1);
604 higher_buddy = page + (buddy_idx - combined_idx);
Corrado Zoccolo6dda9d52010-05-24 14:31:54 -0700605 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
606 list_add_tail(&page->lru,
607 &zone->free_area[order].free_list[migratetype]);
608 goto out;
609 }
610 }
611
612 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
613out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 zone->free_area[order].nr_free++;
615}
616
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700617/*
618 * free_page_mlock() -- clean up attempts to free and mlocked() page.
619 * Page should not be on lru, so no need to fix that up.
620 * free_pages_check() will verify...
621 */
622static inline void free_page_mlock(struct page *page)
623{
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700624 __dec_zone_page_state(page, NR_MLOCK);
625 __count_vm_event(UNEVICTABLE_MLOCKFREED);
626}
KOSAKI Motohiro092cead2009-06-16 15:32:17 -0700627
Nick Piggin224abf92006-01-06 00:11:11 -0800628static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
Nick Piggin92be2e32006-01-06 00:10:57 -0800630 if (unlikely(page_mapcount(page) |
631 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700632 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700633 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
634 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800635 bad_page(page);
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800636 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800637 }
Hugh Dickins79f4b7b2009-01-06 14:40:05 -0800638 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
639 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
640 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641}
642
643/*
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700644 * Frees a number of pages from the PCP lists
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700646 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 *
648 * If the zone was previously in an "all pages pinned" state then look to
649 * see if this freeing clears that state.
650 *
651 * And clear the zone's pages_scanned counter, to hold off the "all pages are
652 * pinned" detection logic.
653 */
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700654static void free_pcppages_bulk(struct zone *zone, int count,
655 struct per_cpu_pages *pcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700657 int migratetype = 0;
Mel Gormana6f9edd2009-09-21 17:03:20 -0700658 int batch_free = 0;
Mel Gorman72853e22010-09-09 16:38:16 -0700659 int to_free = count;
Vinayak Menon898d5482014-10-07 11:16:52 +0530660 int free = 0;
661 int cma_free = 0;
Larry Bassel6d6e2c92012-12-14 14:21:05 -0800662 int mt = 0;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700663
Nick Pigginc54ad302006-01-06 00:10:56 -0800664 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700666
Mel Gorman72853e22010-09-09 16:38:16 -0700667 while (to_free) {
Nick Piggin48db57f2006-01-08 01:00:42 -0800668 struct page *page;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700669 struct list_head *list;
Nick Piggin48db57f2006-01-08 01:00:42 -0800670
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700671 /*
Mel Gormana6f9edd2009-09-21 17:03:20 -0700672 * Remove pages from lists in a round-robin fashion. A
673 * batch_free count is maintained that is incremented when an
674 * empty list is encountered. This is so more pages are freed
675 * off fuller lists instead of spinning excessively around empty
676 * lists
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700677 */
678 do {
Mel Gormana6f9edd2009-09-21 17:03:20 -0700679 batch_free++;
Mel Gorman5f8dcc22009-09-21 17:03:19 -0700680 if (++migratetype == MIGRATE_PCPTYPES)
681 migratetype = 0;
682 list = &pcp->lists[migratetype];
683 } while (list_empty(list));
684
Namhyung Kim1d168712011-03-22 16:32:45 -0700685 /* This is the only non-empty list. Free them all. */
686 if (batch_free == MIGRATE_PCPTYPES)
687 batch_free = to_free;
688
Mel Gormana6f9edd2009-09-21 17:03:20 -0700689 do {
690 page = list_entry(list->prev, struct page, lru);
Larry Bassel6d6e2c92012-12-14 14:21:05 -0800691 mt = get_pageblock_migratetype(page);
Vinayak Menon898d5482014-10-07 11:16:52 +0530692 if (likely(mt != MIGRATE_ISOLATE))
693 mt = page_private(page);
694
Mel Gormana6f9edd2009-09-21 17:03:20 -0700695 /* must delete as __free_one_page list manipulates */
696 list_del(&page->lru);
Hugh Dickinsa7016232010-01-29 17:46:34 +0000697 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
Vinayak Menon898d5482014-10-07 11:16:52 +0530698 __free_one_page(page, zone, 0, mt);
699 trace_mm_page_pcpu_drain(page, 0, mt);
700 if (likely(mt != MIGRATE_ISOLATE)) {
701 free++;
702 if (is_migrate_cma(mt))
703 cma_free++;
704 }
Mel Gorman72853e22010-09-09 16:38:16 -0700705 } while (--to_free && --batch_free && !list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 }
Vinayak Menon898d5482014-10-07 11:16:52 +0530707 __mod_zone_page_state(zone, NR_FREE_PAGES, free);
708 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, cma_free);
Nick Pigginc54ad302006-01-06 00:10:56 -0800709 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710}
711
Mel Gormaned0ae212009-06-16 15:32:07 -0700712static void free_one_page(struct zone *zone, struct page *page, int order,
713 int migratetype)
Nick Piggin48db57f2006-01-08 01:00:42 -0800714{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700715 spin_lock(&zone->lock);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700716 zone->pages_scanned = 0;
Mel Gormanf2260e62009-06-16 15:32:13 -0700717
Mel Gormaned0ae212009-06-16 15:32:07 -0700718 __free_one_page(page, zone, order, migratetype);
Larry Bassel6d6e2c92012-12-14 14:21:05 -0800719 if (unlikely(migratetype != MIGRATE_ISOLATE))
720 __mod_zone_freepage_state(zone, 1 << order, migratetype);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700721 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800722}
723
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700724static bool free_pages_prepare(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 int i;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800727 int bad = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -0800729 trace_mm_page_free(page, order);
Vegard Nossumb1eeab62008-11-25 16:55:53 +0100730 kmemcheck_free_shadow(page, order);
731
Andrea Arcangeli8dd60a32011-01-13 15:46:34 -0800732 if (PageAnon(page))
733 page->mapping = NULL;
734 for (i = 0; i < (1 << order); i++)
735 bad += free_pages_check(page + i);
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800736 if (bad)
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700737 return false;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800738
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700739 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700740 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700741 debug_check_no_obj_freed(page_address(page),
742 PAGE_SIZE << order);
743 }
Nick Piggindafb1362006-10-11 01:21:30 -0700744 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800745 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700746
KOSAKI Motohiroec95f532010-05-24 14:32:38 -0700747 return true;
748}
749
750static void __free_pages_ok(struct page *page, unsigned int order)
751{
752 unsigned long flags;
753 int wasMlocked = __TestClearPageMlocked(page);
754
755 if (!free_pages_prepare(page, order))
756 return;
757
Nick Pigginc54ad302006-01-06 00:10:56 -0800758 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +0200759 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -0700760 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700761 __count_vm_events(PGFREE, 1 << order);
Mel Gormaned0ae212009-06-16 15:32:07 -0700762 free_one_page(page_zone(page), page, order,
763 get_pageblock_migratetype(page));
Nick Pigginc54ad302006-01-06 00:10:56 -0800764 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765}
766
Laura Abbottf8382d82013-06-28 12:52:17 -0700767void __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800768{
Johannes Weinerc3993072012-01-10 15:08:10 -0800769 unsigned int nr_pages = 1 << order;
770 unsigned int loop;
David Howellsa226f6c2006-01-06 00:11:08 -0800771
Johannes Weinerc3993072012-01-10 15:08:10 -0800772 prefetchw(page);
773 for (loop = 0; loop < nr_pages; loop++) {
774 struct page *p = &page[loop];
David Howellsa226f6c2006-01-06 00:11:08 -0800775
Johannes Weinerc3993072012-01-10 15:08:10 -0800776 if (loop + 1 < nr_pages)
777 prefetchw(p + 1);
778 __ClearPageReserved(p);
779 set_page_count(p, 0);
David Howellsa226f6c2006-01-06 00:11:08 -0800780 }
Johannes Weinerc3993072012-01-10 15:08:10 -0800781
782 set_page_refcounted(page);
783 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800784}
785
Michal Nazarewiczd4158d22011-12-29 13:09:50 +0100786#ifdef CONFIG_CMA
Laura Abbott364dcd42012-11-27 10:17:24 -0800787bool is_cma_pageblock(struct page *page)
788{
789 return get_pageblock_migratetype(page) == MIGRATE_CMA;
790}
791
Michal Nazarewiczd4158d22011-12-29 13:09:50 +0100792/* Free whole pageblock and set it's migration type to MIGRATE_CMA. */
793void __init init_cma_reserved_pageblock(struct page *page)
794{
795 unsigned i = pageblock_nr_pages;
796 struct page *p = page;
797
798 do {
799 __ClearPageReserved(p);
800 set_page_count(p, 0);
801 } while (++p, --i);
802
803 set_page_refcounted(page);
804 set_pageblock_migratetype(page, MIGRATE_CMA);
805 __free_pages(page, pageblock_order);
806 totalram_pages += pageblock_nr_pages;
Marek Szyprowski178a68e2013-02-12 13:46:24 -0800807#ifdef CONFIG_HIGHMEM
808 if (PageHighMem(page))
809 totalhigh_pages += pageblock_nr_pages;
810#endif
Michal Nazarewiczd4158d22011-12-29 13:09:50 +0100811}
812#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
814/*
815 * The order of subdivision here is critical for the IO subsystem.
816 * Please do not alter this order without good reasons and regression
817 * testing. Specifically, as large blocks of memory are subdivided,
818 * the order in which smaller blocks are delivered depends on the order
819 * they're subdivided in this function. This is the primary factor
820 * influencing the order in which pages are delivered to the IO
821 * subsystem according to empirical testing, and this is also justified
822 * by considering the behavior of a buddy system containing a single
823 * large block of memory acted on by a series of small allocations.
824 * This behavior is a critical factor in sglist merging's success.
825 *
826 * -- wli
827 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800828static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700829 int low, int high, struct free_area *area,
830 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831{
832 unsigned long size = 1 << high;
833
834 while (high > low) {
835 area--;
836 high--;
837 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700838 VM_BUG_ON(bad_range(zone, &page[size]));
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800839
840#ifdef CONFIG_DEBUG_PAGEALLOC
841 if (high < debug_guardpage_minorder()) {
842 /*
843 * Mark as guard pages (or page), that will allow to
844 * merge back to allocator when buddy will be freed.
845 * Corresponding page table entries will not be touched,
846 * pages will stay not present in virtual address space
847 */
848 INIT_LIST_HEAD(&page[size].lru);
849 set_page_guard_flag(&page[size]);
850 set_page_private(&page[size], high);
851 /* Guard pages are not available for any usage */
Larry Bassel6d6e2c92012-12-14 14:21:05 -0800852 __mod_zone_freepage_state(zone, -(1 << high),
853 migratetype);
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -0800854 continue;
855 }
856#endif
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700857 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 area->nr_free++;
859 set_page_order(&page[size], high);
860 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861}
862
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863/*
864 * This page is about to be returned from the page allocator
865 */
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200866static inline int check_new_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867{
Nick Piggin92be2e32006-01-06 00:10:57 -0800868 if (unlikely(page_mapcount(page) |
869 (page->mapping != NULL) |
Nick Piggina3af9c32009-06-16 15:32:10 -0700870 (atomic_read(&page->_count) != 0) |
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -0700871 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
872 (mem_cgroup_bad_page_check(page)))) {
Nick Piggin224abf92006-01-06 00:11:11 -0800873 bad_page(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800874 return 1;
Hugh Dickins8cc3b392009-01-06 14:40:06 -0800875 }
Wu Fengguang2a7684a2009-09-16 11:50:12 +0200876 return 0;
877}
878
879static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
880{
881 int i;
882
883 for (i = 0; i < (1 << order); i++) {
884 struct page *p = page + i;
885 if (unlikely(check_new_page(p)))
886 return 1;
887 }
Hugh Dickins689bceb2005-11-21 21:32:20 -0800888
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700889 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800890 set_page_refcounted(page);
Nick Piggincc102502006-12-06 20:32:00 -0800891
892 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800894
895 if (gfp_flags & __GFP_ZERO)
896 prep_zero_page(page, order, gfp_flags);
897
898 if (order && (gfp_flags & __GFP_COMP))
899 prep_compound_page(page, order);
900
Hugh Dickins689bceb2005-11-21 21:32:20 -0800901 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902}
903
Mel Gorman56fd56b2007-10-16 01:25:58 -0700904/*
905 * Go through the free lists for the given migratetype and remove
906 * the smallest available page from the freelists
907 */
Mel Gorman728ec982009-06-16 15:32:04 -0700908static inline
909struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
Mel Gorman56fd56b2007-10-16 01:25:58 -0700910 int migratetype)
911{
912 unsigned int current_order;
913 struct free_area * area;
914 struct page *page;
915
916 /* Find a page of the appropriate size in the preferred list */
917 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
918 area = &(zone->free_area[current_order]);
919 if (list_empty(&area->free_list[migratetype]))
920 continue;
921
922 page = list_entry(area->free_list[migratetype].next,
923 struct page, lru);
924 list_del(&page->lru);
925 rmv_page_order(page);
926 area->nr_free--;
Mel Gorman56fd56b2007-10-16 01:25:58 -0700927 expand(zone, page, order, current_order, area, migratetype);
928 return page;
929 }
930
931 return NULL;
932}
933
934
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700935/*
936 * This array describes the order lists are fallen back to when
937 * the free lists for the desirable migrate type are depleted
938 */
Laura Abbott434f55a2013-02-18 07:17:06 -0800939static int fallbacks[MIGRATE_TYPES][4] = {
Michal Nazarewiczd4158d22011-12-29 13:09:50 +0100940 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
941 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
Michal Nazarewiczb5662d62012-11-20 16:37:50 +0100942#ifdef CONFIG_CMA
Laura Abbott434f55a2013-02-18 07:17:06 -0800943 [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
Michal Nazarewiczb5662d62012-11-20 16:37:50 +0100944 [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */
Laura Abbott434f55a2013-02-18 07:17:06 -0800945#else
946 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
Michal Nazarewiczd4158d22011-12-29 13:09:50 +0100947#endif
Michal Nazarewiczc6114bf2012-01-11 15:31:33 +0100948 [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */
949 [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700950};
951
Liam Mark3f1e5512013-03-27 12:34:51 -0700952int *get_migratetype_fallbacks(int mtype)
953{
954 return fallbacks[mtype];
955}
956
Mel Gormanc361be52007-10-16 01:25:51 -0700957/*
958 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700959 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700960 * boundary. If alignment is required, use move_freepages_block()
961 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700962static int move_freepages(struct zone *zone,
963 struct page *start_page, struct page *end_page,
964 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700965{
966 struct page *page;
967 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700968 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700969
970#ifndef CONFIG_HOLES_IN_ZONE
971 /*
972 * page_zone is not safe to call in this context when
973 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
974 * anyway as we check zone boundaries in move_freepages_block().
975 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700976 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700977 */
978 BUG_ON(page_zone(start_page) != page_zone(end_page));
979#endif
980
981 for (page = start_page; page <= end_page;) {
Adam Litke344c7902008-09-02 14:35:38 -0700982 /* Make sure we are not inadvertently changing nodes */
983 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
984
Mel Gormanc361be52007-10-16 01:25:51 -0700985 if (!pfn_valid_within(page_to_pfn(page))) {
986 page++;
987 continue;
988 }
989
990 if (!PageBuddy(page)) {
991 page++;
992 continue;
993 }
994
995 order = page_order(page);
Kirill A. Shutemov84be48d2011-03-22 16:33:41 -0700996 list_move(&page->lru,
997 &zone->free_area[order].free_list[migratetype]);
Mel Gormanc361be52007-10-16 01:25:51 -0700998 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700999 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -07001000 }
1001
Mel Gormand1003132007-10-16 01:26:00 -07001002 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -07001003}
1004
Adrian Bunkb69a7282008-07-23 21:28:12 -07001005static int move_freepages_block(struct zone *zone, struct page *page,
1006 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -07001007{
1008 unsigned long start_pfn, end_pfn;
1009 struct page *start_page, *end_page;
1010
1011 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -07001012 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -07001013 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -07001014 end_page = start_page + pageblock_nr_pages - 1;
1015 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -07001016
1017 /* Do not cross zone boundaries */
1018 if (start_pfn < zone->zone_start_pfn)
1019 start_page = page;
1020 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
1021 return 0;
1022
1023 return move_freepages(zone, start_page, end_page, migratetype);
1024}
1025
Mel Gorman2f66a682009-09-21 17:02:31 -07001026static void change_pageblock_range(struct page *pageblock_page,
1027 int start_order, int migratetype)
1028{
1029 int nr_pageblocks = 1 << (start_order - pageblock_order);
1030
1031 while (nr_pageblocks--) {
1032 set_pageblock_migratetype(pageblock_page, migratetype);
1033 pageblock_page += pageblock_nr_pages;
1034 }
1035}
1036
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001037/* Remove an element from the buddy allocator from the fallback list */
Mel Gorman0ac3a402009-06-16 15:32:06 -07001038static inline struct page *
1039__rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001040{
1041 struct free_area * area;
1042 int current_order;
1043 struct page *page;
1044 int migratetype, i;
1045
1046 /* Find the largest possible block of pages in the other list */
1047 for (current_order = MAX_ORDER-1; current_order >= order;
1048 --current_order) {
Michal Nazarewiczc6114bf2012-01-11 15:31:33 +01001049 for (i = 0;; i++) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001050 migratetype = fallbacks[start_migratetype][i];
1051
Mel Gorman56fd56b2007-10-16 01:25:58 -07001052 /* MIGRATE_RESERVE handled later if necessary */
1053 if (migratetype == MIGRATE_RESERVE)
Michal Nazarewiczc6114bf2012-01-11 15:31:33 +01001054 break;
Mel Gormane0104872007-10-16 01:25:53 -07001055
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001056 area = &(zone->free_area[current_order]);
1057 if (list_empty(&area->free_list[migratetype]))
1058 continue;
1059
1060 page = list_entry(area->free_list[migratetype].next,
1061 struct page, lru);
1062 area->nr_free--;
1063
1064 /*
Mel Gormanc361be52007-10-16 01:25:51 -07001065 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -07001066 * pages to the preferred allocation list. If falling
1067 * back for a reclaimable kernel allocation, be more
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001068 * aggressive about taking ownership of free pages
Laura Abbott434f55a2013-02-18 07:17:06 -08001069 *
1070 * On the other hand, never change migration
1071 * type of MIGRATE_CMA pageblocks nor move CMA
1072 * pages on different free lists. We don't
1073 * want unmovable pages to be allocated from
1074 * MIGRATE_CMA areas.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001075 */
Laura Abbott434f55a2013-02-18 07:17:06 -08001076 if (!is_migrate_cma(migratetype) &&
1077 (unlikely(current_order >= pageblock_order / 2) ||
1078 start_migratetype == MIGRATE_RECLAIMABLE ||
1079 page_group_by_mobility_disabled)) {
Michal Nazarewiczd4158d22011-12-29 13:09:50 +01001080 int pages;
Mel Gorman46dafbc2007-10-16 01:25:55 -07001081 pages = move_freepages_block(zone, page,
1082 start_migratetype);
1083
1084 /* Claim the whole block if over half of it is free */
Mel Gormandd5d2412009-09-05 11:17:11 -07001085 if (pages >= (1 << (pageblock_order-1)) ||
1086 page_group_by_mobility_disabled)
Mel Gorman46dafbc2007-10-16 01:25:55 -07001087 set_pageblock_migratetype(page,
1088 start_migratetype);
1089
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001090 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -07001091 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001092
1093 /* Remove the page from the freelists */
1094 list_del(&page->lru);
1095 rmv_page_order(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001096
Mel Gorman2f66a682009-09-21 17:02:31 -07001097 /* Take ownership for orders >= pageblock_order */
Laura Abbott434f55a2013-02-18 07:17:06 -08001098 if (current_order >= pageblock_order &&
1099 !is_migrate_cma(migratetype))
Mel Gorman2f66a682009-09-21 17:02:31 -07001100 change_pageblock_range(page, current_order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001101 start_migratetype);
1102
Michal Nazarewiczd4158d22011-12-29 13:09:50 +01001103 expand(zone, page, order, current_order, area,
Laura Abbott434f55a2013-02-18 07:17:06 -08001104 is_migrate_cma(migratetype)
1105 ? migratetype : start_migratetype);
Mel Gormane0fff1b2009-09-21 17:02:42 -07001106
1107 trace_mm_page_alloc_extfrag(page, order, current_order,
1108 start_migratetype, migratetype);
1109
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001110 return page;
1111 }
1112 }
1113
Mel Gorman728ec982009-06-16 15:32:04 -07001114 return NULL;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001115}
1116
Mel Gorman56fd56b2007-10-16 01:25:58 -07001117/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 * Do the hard work of removing an element from the buddy allocator.
1119 * Call me with the zone->lock already held.
1120 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001121static struct page *__rmqueue(struct zone *zone, unsigned int order,
1122 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 struct page *page;
1125
Laura Abbott434f55a2013-02-18 07:17:06 -08001126retry_reserve:
1127 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
Laura Abbott434f55a2013-02-18 07:17:06 -08001129 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
Michal Nazarewiczb5662d62012-11-20 16:37:50 +01001130 page = __rmqueue_fallback(zone, order, migratetype);
Michal Nazarewiczb5662d62012-11-20 16:37:50 +01001131
Laura Abbott434f55a2013-02-18 07:17:06 -08001132 /*
1133 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1134 * is used because __rmqueue_smallest is an inline function
1135 * and we want just one call site
1136 */
1137 if (!page) {
1138 migratetype = MIGRATE_RESERVE;
1139 goto retry_reserve;
1140 }
Mel Gorman728ec982009-06-16 15:32:04 -07001141 }
1142
Mel Gorman0d3d0622009-09-21 17:02:44 -07001143 trace_mm_page_alloc_zone_locked(page, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001144 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145}
1146
Heesub Shin771aaa62013-01-07 11:10:13 +09001147static struct page *__rmqueue_cma(struct zone *zone, unsigned int order,
1148 int migratetype)
1149{
1150 struct page *page = 0;
1151#ifdef CONFIG_CMA
1152 if (migratetype == MIGRATE_MOVABLE && !zone->cma_alloc)
1153 page = __rmqueue_smallest(zone, order, MIGRATE_CMA);
Laura Abbotte2958b82013-04-05 12:39:18 -07001154 if (!page)
Heesub Shin771aaa62013-01-07 11:10:13 +09001155#endif
1156retry_reserve :
1157 page = __rmqueue_smallest(zone, order, migratetype);
1158
1159
1160 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
1161 page = __rmqueue_fallback(zone, order, migratetype);
1162
1163 /*
1164 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1165 * is used because __rmqueue_smallest is an inline function
1166 * and we want just one call site
1167 */
1168 if (!page) {
1169 migratetype = MIGRATE_RESERVE;
1170 goto retry_reserve;
1171 }
1172 }
1173
1174 trace_mm_page_alloc_zone_locked(page, order, migratetype);
1175 return page;
1176}
1177
Michal Nazarewiczadc95fd2012-01-11 15:16:11 +01001178/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 * Obtain a specified number of elements from the buddy allocator, all under
1180 * a single hold of the lock, for efficiency. Add them to the supplied list.
1181 * Returns the number of new pages which were placed at *list.
1182 */
Michal Nazarewiczadc95fd2012-01-11 15:16:11 +01001183static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001184 unsigned long count, struct list_head *list,
Heesub Shin771aaa62013-01-07 11:10:13 +09001185 int migratetype, int cold, int cma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186{
Michal Nazarewiczd4158d22011-12-29 13:09:50 +01001187 int mt = migratetype, i;
Michal Nazarewiczadc95fd2012-01-11 15:16:11 +01001188
Nick Pigginc54ad302006-01-06 00:10:56 -08001189 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 for (i = 0; i < count; ++i) {
Heesub Shin771aaa62013-01-07 11:10:13 +09001191 struct page *page;
1192 if (cma)
1193 page = __rmqueue_cma(zone, order, migratetype);
1194 else
1195 page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001196 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -08001198
1199 /*
1200 * Split buddy pages returned by expand() are received here
1201 * in physical page order. The page is added to the callers and
1202 * list and the list head then moves forward. From the callers
1203 * perspective, the linked list is ordered by page number in
1204 * some conditions. This is useful for IO devices that can
1205 * merge IO requests if the physical pages are ordered
1206 * properly.
1207 */
Mel Gormane084b2d2009-07-29 15:02:04 -07001208 if (likely(cold == 0))
1209 list_add(&page->lru, list);
1210 else
1211 list_add_tail(&page->lru, list);
Michal Nazarewiczd4158d22011-12-29 13:09:50 +01001212 if (IS_ENABLED(CONFIG_CMA)) {
1213 mt = get_pageblock_migratetype(page);
1214 if (!is_migrate_cma(mt) && mt != MIGRATE_ISOLATE)
1215 mt = migratetype;
1216 }
1217 set_page_private(page, mt);
Mel Gorman81eabcb2007-12-17 16:20:05 -08001218 list = &page->lru;
Larry Bassel6d6e2c92012-12-14 14:21:05 -08001219 if (is_migrate_cma(mt))
1220 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1221 -(1 << order));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 }
Mel Gormanf2260e62009-06-16 15:32:13 -07001223 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
Nick Pigginc54ad302006-01-06 00:10:56 -08001224 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -08001225 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226}
1227
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001228#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001229/*
Christoph Lameter4037d452007-05-09 02:35:14 -07001230 * Called from the vmstat counter updater to drain pagesets of this
1231 * currently executing processor on remote nodes after they have
1232 * expired.
1233 *
Christoph Lameter879336c2006-03-22 00:09:08 -08001234 * Note that this function must be called with the thread pinned to
1235 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -08001236 */
Christoph Lameter4037d452007-05-09 02:35:14 -07001237void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001238{
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001239 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -07001240 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001241
Christoph Lameter4037d452007-05-09 02:35:14 -07001242 local_irq_save(flags);
1243 if (pcp->count >= pcp->batch)
1244 to_drain = pcp->batch;
1245 else
1246 to_drain = pcp->count;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001247 free_pcppages_bulk(zone, to_drain, pcp);
Christoph Lameter4037d452007-05-09 02:35:14 -07001248 pcp->count -= to_drain;
1249 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001250}
1251#endif
1252
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001253/*
1254 * Drain pages of the indicated processor.
1255 *
1256 * The processor must either be the current processor and the
1257 * thread pinned to the current processor or a processor that
1258 * is not online.
1259 */
1260static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261{
Nick Pigginc54ad302006-01-06 00:10:56 -08001262 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07001265 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001267 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001269 local_irq_save(flags);
1270 pset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001272 pcp = &pset->pcp;
David Rientjes2ff754f2011-01-25 15:07:23 -08001273 if (pcp->count) {
1274 free_pcppages_bulk(zone, pcp->count, pcp);
1275 pcp->count = 0;
1276 }
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001277 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 }
1279}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001281/*
1282 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1283 */
1284void drain_local_pages(void *arg)
1285{
1286 drain_pages(smp_processor_id());
1287}
1288
1289/*
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001290 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1291 *
1292 * Note that this code is protected against sending an IPI to an offline
1293 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1294 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1295 * nothing keeps CPUs from showing up after we populated the cpumask and
1296 * before the call to on_each_cpu_mask().
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001297 */
1298void drain_all_pages(void)
1299{
Gilad Ben-Yossef74046492012-03-28 14:42:45 -07001300 int cpu;
1301 struct per_cpu_pageset *pcp;
1302 struct zone *zone;
1303
1304 /*
1305 * Allocate in the BSS so we wont require allocation in
1306 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1307 */
1308 static cpumask_t cpus_with_pcps;
1309
1310 /*
1311 * We don't care about racing with CPU hotplug event
1312 * as offline notification will cause the notified
1313 * cpu to drain that CPU pcps and on_each_cpu_mask
1314 * disables preemption as part of its processing
1315 */
1316 for_each_online_cpu(cpu) {
1317 bool has_pcps = false;
1318 for_each_populated_zone(zone) {
1319 pcp = per_cpu_ptr(zone->pageset, cpu);
1320 if (pcp->pcp.count) {
1321 has_pcps = true;
1322 break;
1323 }
1324 }
1325 if (has_pcps)
1326 cpumask_set_cpu(cpu, &cpus_with_pcps);
1327 else
1328 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1329 }
1330 on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001331}
1332
Rafael J. Wysocki296699d2007-07-29 23:27:18 +02001333#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
1335void mark_free_pages(struct zone *zone)
1336{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001337 unsigned long pfn, max_zone_pfn;
1338 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001339 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 struct list_head *curr;
1341
1342 if (!zone->spanned_pages)
1343 return;
1344
1345 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001346
1347 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1348 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1349 if (pfn_valid(pfn)) {
1350 struct page *page = pfn_to_page(pfn);
1351
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001352 if (!swsusp_page_is_forbidden(page))
1353 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001354 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001356 for_each_migratetype_order(order, t) {
1357 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001358 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001360 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1361 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -07001362 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -07001363 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001364 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 spin_unlock_irqrestore(&zone->lock, flags);
1366}
Mel Gormane2c55dc2007-10-16 01:25:50 -07001367#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
1369/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 * Free a 0-order page
Li Hongfc916682010-03-05 13:41:54 -08001371 * cold == 1 ? free a cold page : free a hot page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 */
Li Hongfc916682010-03-05 13:41:54 -08001373void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374{
1375 struct zone *zone = page_zone(page);
1376 struct per_cpu_pages *pcp;
1377 unsigned long flags;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001378 int migratetype;
Johannes Weiner451ea252009-09-21 17:01:48 -07001379 int wasMlocked = __TestClearPageMlocked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
KOSAKI Motohiroec95f532010-05-24 14:32:38 -07001381 if (!free_pages_prepare(page, 0))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001382 return;
1383
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001384 migratetype = get_pageblock_migratetype(page);
1385 set_page_private(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 local_irq_save(flags);
Johannes Weinerc2773312009-06-19 19:30:56 +02001387 if (unlikely(wasMlocked))
Mel Gormanda456f12009-06-16 15:32:08 -07001388 free_page_mlock(page);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001389 __count_vm_event(PGFREE);
Mel Gormanda456f12009-06-16 15:32:08 -07001390
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001391 /*
1392 * We only track unmovable, reclaimable and movable on pcp lists.
1393 * Free ISOLATE pages back to the allocator because they are being
1394 * offlined but treat RESERVE as movable pages so we can get those
1395 * areas back if necessary. Otherwise, we may have to free
1396 * excessively into the page allocator
1397 */
1398 if (migratetype >= MIGRATE_PCPTYPES) {
Laura Abbottbe911be2013-02-12 13:30:04 -08001399 if (unlikely(migratetype == MIGRATE_ISOLATE) ||
1400 is_migrate_cma(migratetype)) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001401 free_one_page(zone, page, 0, migratetype);
1402 goto out;
1403 }
1404 migratetype = MIGRATE_MOVABLE;
1405 }
1406
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001407 pcp = &this_cpu_ptr(zone->pageset)->pcp;
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001408 if (cold)
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001409 list_add_tail(&page->lru, &pcp->lists[migratetype]);
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001410 else
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001411 list_add(&page->lru, &pcp->lists[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001413 if (pcp->count >= pcp->high) {
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001414 free_pcppages_bulk(zone, pcp->batch, pcp);
Nick Piggin48db57f2006-01-08 01:00:42 -08001415 pcp->count -= pcp->batch;
1416 }
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001417
1418out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420}
1421
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001422/*
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001423 * Free a list of 0-order pages
1424 */
1425void free_hot_cold_page_list(struct list_head *list, int cold)
1426{
1427 struct page *page, *next;
1428
1429 list_for_each_entry_safe(page, next, list, lru) {
Konstantin Khlebnikovb413d482012-01-10 15:07:09 -08001430 trace_mm_page_free_batched(page, cold);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -08001431 free_hot_cold_page(page, cold);
1432 }
1433}
1434
1435/*
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001436 * split_page takes a non-compound higher-order page, and splits it into
1437 * n (1<<order) sub-pages: page[0..n]
1438 * Each sub-page must be freed individually.
1439 *
1440 * Note: this is probably too low level an operation for use in drivers.
1441 * Please consult with lkml before using this in your driver.
1442 */
1443void split_page(struct page *page, unsigned int order)
1444{
1445 int i;
1446
Nick Piggin725d7042006-09-25 23:30:55 -07001447 VM_BUG_ON(PageCompound(page));
1448 VM_BUG_ON(!page_count(page));
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001449
1450#ifdef CONFIG_KMEMCHECK
1451 /*
1452 * Split shadow pages too, because free(page[0]) would
1453 * otherwise free the whole shadow.
1454 */
1455 if (kmemcheck_page_is_tracked(page))
1456 split_page(virt_to_page(page[0].shadow), order);
1457#endif
1458
Nick Piggin7835e982006-03-22 00:08:40 -08001459 for (i = 1; i < (1 << order); i++)
1460 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001461}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001462
Mel Gorman98149402013-01-11 14:32:16 -08001463static int __isolate_free_page(struct page *page, unsigned int order)
Mel Gorman748446b2010-05-24 14:32:27 -07001464{
Mel Gorman748446b2010-05-24 14:32:27 -07001465 unsigned long watermark;
1466 struct zone *zone;
Liam Mark4ca9a212012-11-27 18:49:58 -08001467 int mt;
Mel Gorman748446b2010-05-24 14:32:27 -07001468
1469 BUG_ON(!PageBuddy(page));
1470
1471 zone = page_zone(page);
Liam Mark4ca9a212012-11-27 18:49:58 -08001472 mt = get_pageblock_migratetype(page);
Mel Gorman748446b2010-05-24 14:32:27 -07001473
Laura Abbott102d27f2013-12-02 16:24:38 -08001474 if (mt != MIGRATE_ISOLATE) {
Marek Szyprowski993e9512012-12-11 16:02:57 -08001475 /* Obey watermarks as if the page was being allocated */
1476 watermark = low_wmark_pages(zone) + (1 << order);
Laura Abbott102d27f2013-12-02 16:24:38 -08001477 if (!is_migrate_cma(mt) &&
1478 !zone_watermark_ok(zone, 0, watermark, 0, 0))
Marek Szyprowski993e9512012-12-11 16:02:57 -08001479 return 0;
1480
Mel Gorman98149402013-01-11 14:32:16 -08001481 __mod_zone_freepage_state(zone, -(1UL << order), mt);
Marek Szyprowski993e9512012-12-11 16:02:57 -08001482 }
Mel Gorman748446b2010-05-24 14:32:27 -07001483
1484 /* Remove page from free list */
1485 list_del(&page->lru);
1486 zone->free_area[order].nr_free--;
1487 rmv_page_order(page);
Larry Bassel6d6e2c92012-12-14 14:21:05 -08001488
Mel Gorman98149402013-01-11 14:32:16 -08001489 /* Set the pageblock if the isolated page is at least a pageblock */
Mel Gorman748446b2010-05-24 14:32:27 -07001490 if (order >= pageblock_order - 1) {
1491 struct page *endpage = page + (1 << order) - 1;
Michal Nazarewiczd4158d22011-12-29 13:09:50 +01001492 for (; page < endpage; page += pageblock_nr_pages) {
Liam Mark4ca9a212012-11-27 18:49:58 -08001493 mt = get_pageblock_migratetype(page);
Michal Nazarewiczd4158d22011-12-29 13:09:50 +01001494 if (mt != MIGRATE_ISOLATE && !is_migrate_cma(mt))
1495 set_pageblock_migratetype(page,
1496 MIGRATE_MOVABLE);
1497 }
Mel Gorman748446b2010-05-24 14:32:27 -07001498 }
1499
Mel Gorman0a1104a2012-10-08 16:29:12 -07001500 return 1UL << order;
1501}
1502
1503/*
1504 * Similar to split_page except the page is already free. As this is only
1505 * being used for migration, the migratetype of the block also changes.
1506 * As this is called with interrupts disabled, the caller is responsible
1507 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1508 * are enabled.
1509 *
1510 * Note: this is probably too low level an operation for use in drivers.
1511 * Please consult with lkml before using this in your driver.
1512 */
1513int split_free_page(struct page *page)
1514{
1515 unsigned int order;
1516 int nr_pages;
1517
Mel Gorman0a1104a2012-10-08 16:29:12 -07001518 order = page_order(page);
1519
Mel Gorman98149402013-01-11 14:32:16 -08001520 nr_pages = __isolate_free_page(page, order);
Mel Gorman0a1104a2012-10-08 16:29:12 -07001521 if (!nr_pages)
1522 return 0;
1523
1524 /* Split into individual pages */
1525 set_page_refcounted(page);
1526 split_page(page, order);
1527 return nr_pages;
Mel Gorman748446b2010-05-24 14:32:27 -07001528}
1529
1530/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1532 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1533 * or two.
1534 */
Mel Gorman0a15c3e2009-06-16 15:32:05 -07001535static inline
1536struct page *buffered_rmqueue(struct zone *preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07001537 struct zone *zone, int order, gfp_t gfp_flags,
1538 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539{
1540 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001541 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 int cold = !!(gfp_flags & __GFP_COLD);
1543
Hugh Dickins689bceb2005-11-21 21:32:20 -08001544again:
Nick Piggin48db57f2006-01-08 01:00:42 -08001545 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001547 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 local_irq_save(flags);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09001550 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1551 list = &pcp->lists[migratetype];
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001552 if (list_empty(list)) {
Mel Gorman535131e2007-10-16 01:25:49 -07001553 pcp->count += rmqueue_bulk(zone, 0,
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001554 pcp->batch, list,
Heesub Shin771aaa62013-01-07 11:10:13 +09001555 migratetype, cold,
1556 gfp_flags & __GFP_CMA);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001557 if (unlikely(list_empty(list)))
Shaohua Li6fb332f2009-09-21 17:01:17 -07001558 goto failed;
Mel Gorman535131e2007-10-16 01:25:49 -07001559 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001560
Mel Gorman5f8dcc22009-09-21 17:03:19 -07001561 if (cold)
1562 page = list_entry(list->prev, struct page, lru);
1563 else
1564 page = list_entry(list->next, struct page, lru);
1565
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001566 list_del(&page->lru);
1567 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001568 } else {
Andrew Mortondab48da2009-06-16 15:32:37 -07001569 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1570 /*
1571 * __GFP_NOFAIL is not to be used in new code.
1572 *
1573 * All __GFP_NOFAIL callers should be fixed so that they
1574 * properly detect and handle allocation failures.
1575 *
1576 * We most definitely don't want callers attempting to
Linus Torvalds4923abf2009-06-24 12:16:49 -07001577 * allocate greater than order-1 page units with
Andrew Mortondab48da2009-06-16 15:32:37 -07001578 * __GFP_NOFAIL.
1579 */
Linus Torvalds4923abf2009-06-24 12:16:49 -07001580 WARN_ON_ONCE(order > 1);
Andrew Mortondab48da2009-06-16 15:32:37 -07001581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 spin_lock_irqsave(&zone->lock, flags);
Heesub Shin771aaa62013-01-07 11:10:13 +09001583 if (gfp_flags & __GFP_CMA)
1584 page = __rmqueue_cma(zone, order, migratetype);
1585 else
1586 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001587 spin_unlock(&zone->lock);
1588 if (!page)
1589 goto failed;
Larry Bassel6d6e2c92012-12-14 14:21:05 -08001590 __mod_zone_freepage_state(zone, -(1 << order),
1591 get_pageblock_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 }
1593
Christoph Lameterf8891e52006-06-30 01:55:45 -07001594 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Andi Kleen78afd562011-03-22 16:33:12 -07001595 zone_statistics(preferred_zone, zone, gfp_flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001596 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597
Nick Piggin725d7042006-09-25 23:30:55 -07001598 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001599 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001600 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001602
1603failed:
1604 local_irq_restore(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001605 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606}
1607
Akinobu Mita933e3122006-12-08 02:39:45 -08001608#ifdef CONFIG_FAIL_PAGE_ALLOC
1609
Akinobu Mitab2588c42011-07-26 16:09:03 -07001610static struct {
Akinobu Mita933e3122006-12-08 02:39:45 -08001611 struct fault_attr attr;
1612
1613 u32 ignore_gfp_highmem;
1614 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001615 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001616} fail_page_alloc = {
1617 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001618 .ignore_gfp_wait = 1,
1619 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001620 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001621};
1622
1623static int __init setup_fail_page_alloc(char *str)
1624{
1625 return setup_fault_attr(&fail_page_alloc.attr, str);
1626}
1627__setup("fail_page_alloc=", setup_fail_page_alloc);
1628
1629static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1630{
Akinobu Mita54114992007-07-15 23:40:23 -07001631 if (order < fail_page_alloc.min_order)
1632 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001633 if (gfp_mask & __GFP_NOFAIL)
1634 return 0;
1635 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1636 return 0;
1637 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1638 return 0;
1639
1640 return should_fail(&fail_page_alloc.attr, 1 << order);
1641}
1642
1643#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1644
1645static int __init fail_page_alloc_debugfs(void)
1646{
Al Virof4ae40a2011-07-24 04:33:43 -04001647 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
Akinobu Mita933e3122006-12-08 02:39:45 -08001648 struct dentry *dir;
Akinobu Mita933e3122006-12-08 02:39:45 -08001649
Akinobu Mitadd48c082011-08-03 16:21:01 -07001650 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1651 &fail_page_alloc.attr);
1652 if (IS_ERR(dir))
1653 return PTR_ERR(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001654
Akinobu Mitab2588c42011-07-26 16:09:03 -07001655 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1656 &fail_page_alloc.ignore_gfp_wait))
1657 goto fail;
1658 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1659 &fail_page_alloc.ignore_gfp_highmem))
1660 goto fail;
1661 if (!debugfs_create_u32("min-order", mode, dir,
1662 &fail_page_alloc.min_order))
1663 goto fail;
Akinobu Mita933e3122006-12-08 02:39:45 -08001664
Akinobu Mitab2588c42011-07-26 16:09:03 -07001665 return 0;
1666fail:
Akinobu Mitadd48c082011-08-03 16:21:01 -07001667 debugfs_remove_recursive(dir);
Akinobu Mita933e3122006-12-08 02:39:45 -08001668
Akinobu Mitab2588c42011-07-26 16:09:03 -07001669 return -ENOMEM;
Akinobu Mita933e3122006-12-08 02:39:45 -08001670}
1671
1672late_initcall(fail_page_alloc_debugfs);
1673
1674#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1675
1676#else /* CONFIG_FAIL_PAGE_ALLOC */
1677
1678static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1679{
1680 return 0;
1681}
1682
1683#endif /* CONFIG_FAIL_PAGE_ALLOC */
1684
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685/*
Mel Gorman88f5acf2011-01-13 15:45:41 -08001686 * Return true if free pages are above 'mark'. This takes into account the order
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 * of the allocation.
1688 */
Mel Gorman88f5acf2011-01-13 15:45:41 -08001689static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1690 int classzone_idx, int alloc_flags, long free_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691{
Jack Cheung9f41da82011-11-28 16:41:28 -08001692 /* free_pages may go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001693 long min = mark;
Jack Cheung9f41da82011-11-28 16:41:28 -08001694 long lowmem_reserve = z->lowmem_reserve[classzone_idx];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 int o;
Tomasz Stanislawski18d4e2d2013-05-09 09:50:46 +02001696 long free_cma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697
Michal Hockodf0a6da2012-01-10 15:08:02 -08001698 free_pages -= (1 << order) - 1;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001699 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001701 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 min -= min / 4;
Bartlomiej Zolnierkiewicz19644312012-10-08 16:32:05 -07001703#ifdef CONFIG_CMA
1704 /* If allocation can't use CMA areas don't use free CMA pages */
1705 if (!(alloc_flags & ALLOC_CMA))
Tomasz Stanislawski18d4e2d2013-05-09 09:50:46 +02001706 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
Bartlomiej Zolnierkiewicz19644312012-10-08 16:32:05 -07001707#endif
Tomasz Stanislawski18d4e2d2013-05-09 09:50:46 +02001708
1709 if (free_pages - free_cma <= min + lowmem_reserve)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001710 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 for (o = 0; o < order; o++) {
1712 /* At the next order, this order's pages become unavailable */
1713 free_pages -= z->free_area[o].nr_free << o;
1714
1715 /* Require fewer higher order pages to be free */
Arve Hjønnevågcde89492009-02-17 14:51:02 -08001716 min >>= min_free_order_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
1718 if (free_pages <= min)
Mel Gorman88f5acf2011-01-13 15:45:41 -08001719 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 }
Mel Gorman88f5acf2011-01-13 15:45:41 -08001721 return true;
1722}
1723
1724bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1725 int classzone_idx, int alloc_flags)
1726{
1727 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1728 zone_page_state(z, NR_FREE_PAGES));
1729}
1730
1731bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1732 int classzone_idx, int alloc_flags)
1733{
1734 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1735
1736 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1737 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1738
1739 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1740 free_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741}
1742
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001743#ifdef CONFIG_NUMA
1744/*
1745 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1746 * skip over zones that are not allowed by the cpuset, or that have
1747 * been recently (in last second) found to be nearly full. See further
1748 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001749 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001750 *
1751 * If the zonelist cache is present in the passed in zonelist, then
1752 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001753 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001754 *
1755 * If the zonelist cache is not available for this zonelist, does
1756 * nothing and returns NULL.
1757 *
1758 * If the fullzones BITMAP in the zonelist cache is stale (more than
1759 * a second since last zap'd) then we zap it out (clear its bits.)
1760 *
1761 * We hold off even calling zlc_setup, until after we've checked the
1762 * first zone in the zonelist, on the theory that most allocations will
1763 * be satisfied from that first zone, so best to examine that zone as
1764 * quickly as we can.
1765 */
1766static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1767{
1768 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1769 nodemask_t *allowednodes; /* zonelist_cache approximation */
1770
1771 zlc = zonelist->zlcache_ptr;
1772 if (!zlc)
1773 return NULL;
1774
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001775 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001776 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1777 zlc->last_full_zap = jiffies;
1778 }
1779
1780 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1781 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001782 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001783 return allowednodes;
1784}
1785
1786/*
1787 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1788 * if it is worth looking at further for free memory:
1789 * 1) Check that the zone isn't thought to be full (doesn't have its
1790 * bit set in the zonelist_cache fullzones BITMAP).
1791 * 2) Check that the zones node (obtained from the zonelist_cache
1792 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1793 * Return true (non-zero) if zone is worth looking at further, or
1794 * else return false (zero) if it is not.
1795 *
1796 * This check -ignores- the distinction between various watermarks,
1797 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1798 * found to be full for any variation of these watermarks, it will
1799 * be considered full for up to one second by all requests, unless
1800 * we are so low on memory on all allowed nodes that we are forced
1801 * into the second scan of the zonelist.
1802 *
1803 * In the second scan we ignore this zonelist cache and exactly
1804 * apply the watermarks to all zones, even it is slower to do so.
1805 * We are low on memory in the second scan, and should leave no stone
1806 * unturned looking for a free page.
1807 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001808static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001809 nodemask_t *allowednodes)
1810{
1811 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1812 int i; /* index of *z in zonelist zones */
1813 int n; /* node that zone *z is on */
1814
1815 zlc = zonelist->zlcache_ptr;
1816 if (!zlc)
1817 return 1;
1818
Mel Gormandd1a2392008-04-28 02:12:17 -07001819 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001820 n = zlc->z_to_n[i];
1821
1822 /* This zone is worth trying if it is allowed but not full */
1823 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1824}
1825
1826/*
1827 * Given 'z' scanning a zonelist, set the corresponding bit in
1828 * zlc->fullzones, so that subsequent attempts to allocate a page
1829 * from that zone don't waste time re-examining it.
1830 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001831static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001832{
1833 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1834 int i; /* index of *z in zonelist zones */
1835
1836 zlc = zonelist->zlcache_ptr;
1837 if (!zlc)
1838 return;
1839
Mel Gormandd1a2392008-04-28 02:12:17 -07001840 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001841
1842 set_bit(i, zlc->fullzones);
1843}
1844
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001845/*
1846 * clear all zones full, called after direct reclaim makes progress so that
1847 * a zone that was recently full is not skipped over for up to a second
1848 */
1849static void zlc_clear_zones_full(struct zonelist *zonelist)
1850{
1851 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1852
1853 zlc = zonelist->zlcache_ptr;
1854 if (!zlc)
1855 return;
1856
1857 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1858}
1859
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001860#else /* CONFIG_NUMA */
1861
1862static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1863{
1864 return NULL;
1865}
1866
Mel Gormandd1a2392008-04-28 02:12:17 -07001867static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001868 nodemask_t *allowednodes)
1869{
1870 return 1;
1871}
1872
Mel Gormandd1a2392008-04-28 02:12:17 -07001873static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001874{
1875}
Mel Gorman76d3fbf2011-07-25 17:12:30 -07001876
1877static void zlc_clear_zones_full(struct zonelist *zonelist)
1878{
1879}
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001880#endif /* CONFIG_NUMA */
1881
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001882/*
Paul Jackson0798e512006-12-06 20:31:38 -08001883 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001884 * a page.
1885 */
1886static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001887get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman5117f452009-06-16 15:31:59 -07001888 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07001889 struct zone *preferred_zone, int migratetype)
Martin Hicks753ee722005-06-21 17:14:41 -07001890{
Mel Gormandd1a2392008-04-28 02:12:17 -07001891 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001892 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001893 int classzone_idx;
Mel Gorman5117f452009-06-16 15:31:59 -07001894 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001895 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1896 int zlc_active = 0; /* set if using zonelist_cache */
1897 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001898
Mel Gorman19770b32008-04-28 02:12:18 -07001899 classzone_idx = zone_idx(preferred_zone);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001900zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001901 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001902 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001903 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1904 */
Mel Gorman19770b32008-04-28 02:12:18 -07001905 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1906 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001907 if (NUMA_BUILD && zlc_active &&
1908 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1909 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001910 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001911 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Mel Gormancd38b112011-07-25 17:12:29 -07001912 continue;
Johannes Weinera756cf52012-01-10 15:07:49 -08001913 /*
1914 * When allocating a page cache page for writing, we
1915 * want to get it from a zone that is within its dirty
1916 * limit, such that no single zone holds more than its
1917 * proportional share of globally allowed dirty pages.
1918 * The dirty limits take into account the zone's
1919 * lowmem reserves and high watermark so that kswapd
1920 * should be able to balance it without having to
1921 * write pages from its LRU list.
1922 *
1923 * This may look like it could increase pressure on
1924 * lower zones by failing allocations in higher zones
1925 * before they are full. But the pages that do spill
1926 * over are limited as the lower zones are protected
1927 * by this very same mechanism. It should not become
1928 * a practical burden to them.
1929 *
1930 * XXX: For now, allow allocations to potentially
1931 * exceed the per-zone dirty limit in the slowpath
1932 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1933 * which is important when on a NUMA setup the allowed
1934 * zones are together not big enough to reach the
1935 * global limit. The proper fix for these situations
1936 * will require awareness of zones in the
1937 * dirty-throttling and the flusher threads.
1938 */
1939 if ((alloc_flags & ALLOC_WMARK_LOW) &&
1940 (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone))
1941 goto this_zone_full;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001942
Mel Gorman41858962009-06-16 15:32:12 -07001943 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001944 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001945 unsigned long mark;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001946 int ret;
1947
Mel Gorman41858962009-06-16 15:32:12 -07001948 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
Mel Gormanfa5e0842009-06-16 15:33:22 -07001949 if (zone_watermark_ok(zone, order, mark,
1950 classzone_idx, alloc_flags))
1951 goto try_this_zone;
1952
Mel Gormancd38b112011-07-25 17:12:29 -07001953 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
1954 /*
1955 * we do zlc_setup if there are multiple nodes
1956 * and before considering the first zone allowed
1957 * by the cpuset.
1958 */
1959 allowednodes = zlc_setup(zonelist, alloc_flags);
1960 zlc_active = 1;
1961 did_zlc_setup = 1;
1962 }
1963
Mel Gormanfa5e0842009-06-16 15:33:22 -07001964 if (zone_reclaim_mode == 0)
1965 goto this_zone_full;
1966
Mel Gormancd38b112011-07-25 17:12:29 -07001967 /*
1968 * As we may have just activated ZLC, check if the first
1969 * eligible zone has failed zone_reclaim recently.
1970 */
1971 if (NUMA_BUILD && zlc_active &&
1972 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1973 continue;
1974
Mel Gormanfa5e0842009-06-16 15:33:22 -07001975 ret = zone_reclaim(zone, gfp_mask, order);
1976 switch (ret) {
1977 case ZONE_RECLAIM_NOSCAN:
1978 /* did not scan */
Mel Gormancd38b112011-07-25 17:12:29 -07001979 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001980 case ZONE_RECLAIM_FULL:
1981 /* scanned but unreclaimable */
Mel Gormancd38b112011-07-25 17:12:29 -07001982 continue;
Mel Gormanfa5e0842009-06-16 15:33:22 -07001983 default:
1984 /* did we reclaim enough */
1985 if (!zone_watermark_ok(zone, order, mark,
1986 classzone_idx, alloc_flags))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001987 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001988 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001989 }
1990
Mel Gormanfa5e0842009-06-16 15:33:22 -07001991try_this_zone:
Mel Gorman3dd28262009-06-16 15:32:00 -07001992 page = buffered_rmqueue(preferred_zone, zone, order,
1993 gfp_mask, migratetype);
Paul Jackson0798e512006-12-06 20:31:38 -08001994 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001995 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001996this_zone_full:
1997 if (NUMA_BUILD)
1998 zlc_mark_zone_full(zonelist, z);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001999 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002000
2001 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
2002 /* Disable zlc cache for second zonelist scan */
2003 zlc_active = 0;
2004 goto zonelist_scan;
2005 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002006 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07002007}
2008
David Rientjes29423e72011-03-22 16:30:47 -07002009/*
2010 * Large machines with many possible nodes should not always dump per-node
2011 * meminfo in irq context.
2012 */
2013static inline bool should_suppress_show_mem(void)
2014{
2015 bool ret = false;
2016
2017#if NODES_SHIFT > 8
2018 ret = in_interrupt();
2019#endif
2020 return ret;
2021}
2022
Dave Hansena238ab52011-05-24 17:12:16 -07002023static DEFINE_RATELIMIT_STATE(nopage_rs,
2024 DEFAULT_RATELIMIT_INTERVAL,
2025 DEFAULT_RATELIMIT_BURST);
2026
2027void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2028{
Dave Hansena238ab52011-05-24 17:12:16 -07002029 unsigned int filter = SHOW_MEM_FILTER_NODES;
2030
Stanislaw Gruszkac0a32fc2012-01-10 15:07:28 -08002031 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2032 debug_guardpage_minorder() > 0)
Dave Hansena238ab52011-05-24 17:12:16 -07002033 return;
2034
2035 /*
David Rientjesdb155a22013-04-29 15:06:11 -07002036 * Walking all memory to count page types is very expensive and should
2037 * be inhibited in non-blockable contexts.
2038 */
2039 if (!(gfp_mask & __GFP_WAIT))
2040 filter |= SHOW_MEM_FILTER_PAGE_COUNT;
2041
2042 /*
Dave Hansena238ab52011-05-24 17:12:16 -07002043 * This documents exceptions given to allocations in certain
2044 * contexts that are allowed to allocate outside current's set
2045 * of allowed nodes.
2046 */
2047 if (!(gfp_mask & __GFP_NOMEMALLOC))
2048 if (test_thread_flag(TIF_MEMDIE) ||
2049 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2050 filter &= ~SHOW_MEM_FILTER_NODES;
2051 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2052 filter &= ~SHOW_MEM_FILTER_NODES;
2053
2054 if (fmt) {
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002055 struct va_format vaf;
2056 va_list args;
2057
Dave Hansena238ab52011-05-24 17:12:16 -07002058 va_start(args, fmt);
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002059
2060 vaf.fmt = fmt;
2061 vaf.va = &args;
2062
2063 pr_warn("%pV", &vaf);
2064
Dave Hansena238ab52011-05-24 17:12:16 -07002065 va_end(args);
2066 }
2067
Joe Perches3ee9a4f2011-10-31 17:08:35 -07002068 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2069 current->comm, order, gfp_mask);
Dave Hansena238ab52011-05-24 17:12:16 -07002070
2071 dump_stack();
2072 if (!should_suppress_show_mem())
2073 show_mem(filter);
2074}
2075
Mel Gorman11e33f62009-06-16 15:31:57 -07002076static inline int
2077should_alloc_retry(gfp_t gfp_mask, unsigned int order,
Mel Gormanf90ac392012-01-10 15:07:15 -08002078 unsigned long did_some_progress,
Mel Gorman11e33f62009-06-16 15:31:57 -07002079 unsigned long pages_reclaimed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080{
Mel Gorman11e33f62009-06-16 15:31:57 -07002081 /* Do not loop if specifically requested */
2082 if (gfp_mask & __GFP_NORETRY)
2083 return 0;
2084
Mel Gormanf90ac392012-01-10 15:07:15 -08002085 /* Always retry if specifically requested */
2086 if (gfp_mask & __GFP_NOFAIL)
2087 return 1;
2088
2089 /*
2090 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2091 * making forward progress without invoking OOM. Suspend also disables
2092 * storage devices so kswapd will not help. Bail if we are suspending.
2093 */
2094 if (!did_some_progress && pm_suspended_storage())
2095 return 0;
2096
Mel Gorman11e33f62009-06-16 15:31:57 -07002097 /*
2098 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2099 * means __GFP_NOFAIL, but that may not be true in other
2100 * implementations.
2101 */
2102 if (order <= PAGE_ALLOC_COSTLY_ORDER)
2103 return 1;
2104
2105 /*
2106 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2107 * specified, then we retry until we no longer reclaim any pages
2108 * (above), or we've reclaimed an order of pages at least as
2109 * large as the allocation's order. In both cases, if the
2110 * allocation still fails, we stop retrying.
2111 */
2112 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2113 return 1;
2114
Mel Gorman11e33f62009-06-16 15:31:57 -07002115 return 0;
2116}
2117
2118static inline struct page *
2119__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2120 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002121 nodemask_t *nodemask, struct zone *preferred_zone,
2122 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002123{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125
Mel Gorman11e33f62009-06-16 15:31:57 -07002126 /* Acquire the OOM killer lock for the zones in zonelist */
Minchan Kimff321fe2010-08-09 17:18:57 -07002127 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002128 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 return NULL;
2130 }
Jens Axboe6b1de912005-11-17 21:35:02 +01002131
Mel Gorman11e33f62009-06-16 15:31:57 -07002132 /*
2133 * Go through the zonelist yet one more time, keep very high watermark
2134 * here, this is only to catch a parallel oom killing, we must fail if
2135 * we're still under heavy pressure.
2136 */
2137 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
2138 order, zonelist, high_zoneidx,
Mel Gorman5117f452009-06-16 15:31:59 -07002139 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
Mel Gorman3dd28262009-06-16 15:32:00 -07002140 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002141 if (page)
Mel Gorman11e33f62009-06-16 15:31:57 -07002142 goto out;
2143
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002144 if (!(gfp_mask & __GFP_NOFAIL)) {
2145 /* The OOM killer will not help higher order allocs */
2146 if (order > PAGE_ALLOC_COSTLY_ORDER)
2147 goto out;
David Rientjes03668b32010-08-09 17:18:54 -07002148 /* The OOM killer does not needlessly kill tasks for lowmem */
2149 if (high_zoneidx < ZONE_NORMAL)
2150 goto out;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -08002151 /*
2152 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2153 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2154 * The caller should handle page allocation failure by itself if
2155 * it specifies __GFP_THISNODE.
2156 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2157 */
2158 if (gfp_mask & __GFP_THISNODE)
2159 goto out;
2160 }
Mel Gorman11e33f62009-06-16 15:31:57 -07002161 /* Exhausted what can be done so it's blamo time */
David Rientjes08ab9b12012-03-21 16:34:04 -07002162 out_of_memory(zonelist, gfp_mask, order, nodemask, false);
Mel Gorman11e33f62009-06-16 15:31:57 -07002163
2164out:
2165 clear_zonelist_oom(zonelist, gfp_mask);
2166 return page;
2167}
2168
Mel Gorman56de7262010-05-24 14:32:30 -07002169#ifdef CONFIG_COMPACTION
2170/* Try memory compaction for high-order allocations before reclaim */
2171static struct page *
2172__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2173 struct zonelist *zonelist, enum zone_type high_zoneidx,
2174 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002175 int migratetype, bool sync_migration,
Mel Gorman27d1af82012-08-21 16:16:17 -07002176 bool *contended_compaction, bool *deferred_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002177 unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002178{
Mel Gorman66199712012-01-12 17:19:41 -08002179 if (!order)
Mel Gorman56de7262010-05-24 14:32:30 -07002180 return NULL;
2181
Rik van Rielaff62242012-03-21 16:33:52 -07002182 if (compaction_deferred(preferred_zone, order)) {
Mel Gorman66199712012-01-12 17:19:41 -08002183 *deferred_compaction = true;
2184 return NULL;
2185 }
2186
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002187 current->flags |= PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002188 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
Mel Gorman27d1af82012-08-21 16:16:17 -07002189 nodemask, sync_migration,
Mel Gorman98149402013-01-11 14:32:16 -08002190 contended_compaction);
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002191 current->flags &= ~PF_MEMALLOC;
Mel Gorman56de7262010-05-24 14:32:30 -07002192
Mel Gorman0a1104a2012-10-08 16:29:12 -07002193 if (*did_some_progress != COMPACT_SKIPPED) {
Mel Gorman98149402013-01-11 14:32:16 -08002194 struct page *page;
2195
Mel Gorman56de7262010-05-24 14:32:30 -07002196 /* Page migration frees to the PCP lists but we want merging */
2197 drain_pages(get_cpu());
2198 put_cpu();
2199
2200 page = get_page_from_freelist(gfp_mask, nodemask,
2201 order, zonelist, high_zoneidx,
2202 alloc_flags, preferred_zone,
2203 migratetype);
2204 if (page) {
Mel Gorman9317a2b2012-10-08 16:32:47 -07002205 preferred_zone->compact_blockskip_flush = false;
Mel Gorman4f92e252010-05-24 14:32:32 -07002206 preferred_zone->compact_considered = 0;
2207 preferred_zone->compact_defer_shift = 0;
Rik van Rielaff62242012-03-21 16:33:52 -07002208 if (order >= preferred_zone->compact_order_failed)
2209 preferred_zone->compact_order_failed = order + 1;
Mel Gorman56de7262010-05-24 14:32:30 -07002210 count_vm_event(COMPACTSUCCESS);
2211 return page;
2212 }
2213
2214 /*
2215 * It's bad if compaction run occurs and fails.
2216 * The most likely reason is that pages exist,
2217 * but not enough to satisfy watermarks.
2218 */
2219 count_vm_event(COMPACTFAIL);
Mel Gorman66199712012-01-12 17:19:41 -08002220
2221 /*
2222 * As async compaction considers a subset of pageblocks, only
2223 * defer if the failure was a sync compaction failure.
2224 */
2225 if (sync_migration)
Rik van Rielaff62242012-03-21 16:33:52 -07002226 defer_compaction(preferred_zone, order);
Mel Gorman56de7262010-05-24 14:32:30 -07002227
2228 cond_resched();
2229 }
2230
2231 return NULL;
2232}
2233#else
2234static inline struct page *
2235__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2236 struct zonelist *zonelist, enum zone_type high_zoneidx,
2237 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002238 int migratetype, bool sync_migration,
Mel Gorman27d1af82012-08-21 16:16:17 -07002239 bool *contended_compaction, bool *deferred_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002240 unsigned long *did_some_progress)
Mel Gorman56de7262010-05-24 14:32:30 -07002241{
2242 return NULL;
2243}
2244#endif /* CONFIG_COMPACTION */
2245
Marek Szyprowski1ed36512012-01-25 12:09:52 +01002246/* Perform direct synchronous page reclaim */
2247static int
2248__perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist,
2249 nodemask_t *nodemask)
Mel Gorman11e33f62009-06-16 15:31:57 -07002250{
Mel Gorman11e33f62009-06-16 15:31:57 -07002251 struct reclaim_state reclaim_state;
Marek Szyprowski1ed36512012-01-25 12:09:52 +01002252 int progress;
Mel Gorman11e33f62009-06-16 15:31:57 -07002253
2254 cond_resched();
2255
2256 /* We now go into synchronous reclaim */
2257 cpuset_memory_pressure_bump();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002258 current->flags |= PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002259 lockdep_set_current_reclaim_state(gfp_mask);
2260 reclaim_state.reclaimed_slab = 0;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002261 current->reclaim_state = &reclaim_state;
Mel Gorman11e33f62009-06-16 15:31:57 -07002262
Marek Szyprowski1ed36512012-01-25 12:09:52 +01002263 progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
Mel Gorman11e33f62009-06-16 15:31:57 -07002264
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002265 current->reclaim_state = NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002266 lockdep_clear_current_reclaim_state();
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002267 current->flags &= ~PF_MEMALLOC;
Mel Gorman11e33f62009-06-16 15:31:57 -07002268
2269 cond_resched();
2270
Marek Szyprowski1ed36512012-01-25 12:09:52 +01002271 return progress;
2272}
2273
2274/* The really slow allocator path where we enter direct reclaim */
2275static inline struct page *
2276__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2277 struct zonelist *zonelist, enum zone_type high_zoneidx,
2278 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2279 int migratetype, unsigned long *did_some_progress)
2280{
2281 struct page *page = NULL;
2282 bool drained = false;
2283
2284 *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist,
2285 nodemask);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002286 if (unlikely(!(*did_some_progress)))
2287 return NULL;
Mel Gorman11e33f62009-06-16 15:31:57 -07002288
Mel Gorman76d3fbf2011-07-25 17:12:30 -07002289 /* After successful reclaim, reconsider all zones for allocation */
2290 if (NUMA_BUILD)
2291 zlc_clear_zones_full(zonelist);
2292
Mel Gorman9ee493c2010-09-09 16:38:18 -07002293retry:
2294 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002295 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002296 alloc_flags, preferred_zone,
2297 migratetype);
Mel Gorman9ee493c2010-09-09 16:38:18 -07002298
2299 /*
2300 * If an allocation failed after direct reclaim, it could be because
2301 * pages are pinned on the per-cpu lists. Drain them and try again
2302 */
2303 if (!page && !drained) {
2304 drain_all_pages();
2305 drained = true;
2306 goto retry;
2307 }
2308
Mel Gorman11e33f62009-06-16 15:31:57 -07002309 return page;
2310}
2311
Mel Gorman11e33f62009-06-16 15:31:57 -07002312/*
2313 * This is called in the allocator slow-path if the allocation request is of
2314 * sufficient urgency to ignore watermarks and take other desperate measures
2315 */
2316static inline struct page *
2317__alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2318 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002319 nodemask_t *nodemask, struct zone *preferred_zone,
2320 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002321{
2322 struct page *page;
2323
2324 do {
2325 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002326 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
Mel Gorman3dd28262009-06-16 15:32:00 -07002327 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002328
2329 if (!page && gfp_mask & __GFP_NOFAIL)
Mel Gorman0e093d992010-10-26 14:21:45 -07002330 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Mel Gorman11e33f62009-06-16 15:31:57 -07002331 } while (!page && (gfp_mask & __GFP_NOFAIL));
2332
2333 return page;
2334}
2335
2336static inline
2337void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
Mel Gorman99504742011-01-13 15:46:20 -08002338 enum zone_type high_zoneidx,
2339 enum zone_type classzone_idx)
Mel Gorman11e33f62009-06-16 15:31:57 -07002340{
2341 struct zoneref *z;
2342 struct zone *zone;
2343
2344 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
Mel Gorman99504742011-01-13 15:46:20 -08002345 wakeup_kswapd(zone, order, classzone_idx);
Mel Gorman11e33f62009-06-16 15:31:57 -07002346}
2347
Peter Zijlstra341ce062009-06-16 15:32:02 -07002348static inline int
2349gfp_to_alloc_flags(gfp_t gfp_mask)
2350{
Peter Zijlstra341ce062009-06-16 15:32:02 -07002351 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2352 const gfp_t wait = gfp_mask & __GFP_WAIT;
2353
Mel Gormana56f57f2009-06-16 15:32:02 -07002354 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
Namhyung Kime6223a32010-10-26 14:21:59 -07002355 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
Mel Gormana56f57f2009-06-16 15:32:02 -07002356
Peter Zijlstra341ce062009-06-16 15:32:02 -07002357 /*
2358 * The caller may dip into page reserves a bit more if the caller
2359 * cannot run direct reclaim, or if the caller has realtime scheduling
2360 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2361 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2362 */
Namhyung Kime6223a32010-10-26 14:21:59 -07002363 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
Peter Zijlstra341ce062009-06-16 15:32:02 -07002364
2365 if (!wait) {
Andrea Arcangeli5c3240d2011-01-13 15:46:49 -08002366 /*
2367 * Not worth trying to allocate harder for
2368 * __GFP_NOMEMALLOC even if it can't schedule.
2369 */
2370 if (!(gfp_mask & __GFP_NOMEMALLOC))
2371 alloc_flags |= ALLOC_HARDER;
Peter Zijlstra341ce062009-06-16 15:32:02 -07002372 /*
2373 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2374 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2375 */
2376 alloc_flags &= ~ALLOC_CPUSET;
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002377 } else if (unlikely(rt_task(current)) && !in_interrupt())
Peter Zijlstra341ce062009-06-16 15:32:02 -07002378 alloc_flags |= ALLOC_HARDER;
2379
2380 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2381 if (!in_interrupt() &&
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002382 ((current->flags & PF_MEMALLOC) ||
Peter Zijlstra341ce062009-06-16 15:32:02 -07002383 unlikely(test_thread_flag(TIF_MEMDIE))))
2384 alloc_flags |= ALLOC_NO_WATERMARKS;
2385 }
Bartlomiej Zolnierkiewicz19644312012-10-08 16:32:05 -07002386#ifdef CONFIG_CMA
2387 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2388 alloc_flags |= ALLOC_CMA;
2389#endif
Peter Zijlstra341ce062009-06-16 15:32:02 -07002390 return alloc_flags;
2391}
2392
Mel Gorman11e33f62009-06-16 15:31:57 -07002393static inline struct page *
2394__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2395 struct zonelist *zonelist, enum zone_type high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002396 nodemask_t *nodemask, struct zone *preferred_zone,
2397 int migratetype)
Mel Gorman11e33f62009-06-16 15:31:57 -07002398{
2399 const gfp_t wait = gfp_mask & __GFP_WAIT;
2400 struct page *page = NULL;
2401 int alloc_flags;
2402 unsigned long pages_reclaimed = 0;
2403 unsigned long did_some_progress;
Mel Gorman77f1fe62011-01-13 15:45:57 -08002404 bool sync_migration = false;
Mel Gorman66199712012-01-12 17:19:41 -08002405 bool deferred_compaction = false;
Mel Gorman27d1af82012-08-21 16:16:17 -07002406 bool contended_compaction = false;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002407
Christoph Lameter952f3b52006-12-06 20:33:26 -08002408 /*
Mel Gorman72807a72009-06-16 15:32:18 -07002409 * In the slowpath, we sanity check order to avoid ever trying to
2410 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2411 * be using allocators in order of preference for an area that is
2412 * too large.
2413 */
Mel Gorman1fc28b72009-07-29 15:04:08 -07002414 if (order >= MAX_ORDER) {
2415 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
Mel Gorman72807a72009-06-16 15:32:18 -07002416 return NULL;
Mel Gorman1fc28b72009-07-29 15:04:08 -07002417 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002418
Christoph Lameter952f3b52006-12-06 20:33:26 -08002419 /*
2420 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2421 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2422 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2423 * using a larger set of nodes after it has established that the
2424 * allowed per node queues are empty and that nodes are
2425 * over allocated.
2426 */
2427 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2428 goto nopage;
2429
Mel Gormancc4a6852009-11-11 14:26:14 -08002430restart:
Andrea Arcangeli32dba982011-01-13 15:46:49 -08002431 if (!(gfp_mask & __GFP_NO_KSWAPD))
2432 wake_all_kswapd(order, zonelist, high_zoneidx,
Mel Gorman99504742011-01-13 15:46:20 -08002433 zone_idx(preferred_zone));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002434
Paul Jackson9bf22292005-09-06 15:18:12 -07002435 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002436 * OK, we're below the kswapd watermark and have kicked background
2437 * reclaim. Now things get more complex, so set up alloc_flags according
2438 * to how we want to proceed.
Paul Jackson9bf22292005-09-06 15:18:12 -07002439 */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002440 alloc_flags = gfp_to_alloc_flags(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
David Rientjesf33261d2011-01-25 15:07:20 -08002442 /*
2443 * Find the true preferred zone if the allocation is unconstrained by
2444 * cpusets.
2445 */
2446 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2447 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2448 &preferred_zone);
2449
Andrew Barrycfa54a02011-05-24 17:12:52 -07002450rebalance:
Peter Zijlstra341ce062009-06-16 15:32:02 -07002451 /* This is the last chance, in general, before the goto nopage. */
Mel Gorman19770b32008-04-28 02:12:18 -07002452 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Peter Zijlstra341ce062009-06-16 15:32:02 -07002453 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2454 preferred_zone, migratetype);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002455 if (page)
2456 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457
Mel Gorman11e33f62009-06-16 15:31:57 -07002458 /* Allocate without watermarks if the context allows */
Peter Zijlstra341ce062009-06-16 15:32:02 -07002459 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2460 page = __alloc_pages_high_priority(gfp_mask, order,
2461 zonelist, high_zoneidx, nodemask,
2462 preferred_zone, migratetype);
2463 if (page)
2464 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 }
2466
2467 /* Atomic allocations - we can't balance anything */
2468 if (!wait)
2469 goto nopage;
2470
Peter Zijlstra341ce062009-06-16 15:32:02 -07002471 /* Avoid recursion of direct reclaim */
Andrew Mortonc06b1fc2011-01-13 15:47:32 -08002472 if (current->flags & PF_MEMALLOC)
Peter Zijlstra341ce062009-06-16 15:32:02 -07002473 goto nopage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474
David Rientjes6583bb62009-07-29 15:02:06 -07002475 /* Avoid allocations with no watermarks from looping endlessly */
2476 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2477 goto nopage;
2478
Mel Gorman77f1fe62011-01-13 15:45:57 -08002479 /*
2480 * Try direct compaction. The first pass is asynchronous. Subsequent
2481 * attempts after direct reclaim are synchronous
2482 */
Mel Gorman56de7262010-05-24 14:32:30 -07002483 page = __alloc_pages_direct_compact(gfp_mask, order,
2484 zonelist, high_zoneidx,
2485 nodemask,
2486 alloc_flags, preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002487 migratetype, sync_migration,
Mel Gorman27d1af82012-08-21 16:16:17 -07002488 &contended_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002489 &deferred_compaction,
2490 &did_some_progress);
Mel Gorman56de7262010-05-24 14:32:30 -07002491 if (page)
2492 goto got_pg;
Andrea Arcangelic6a140b2011-05-24 17:11:38 -07002493 sync_migration = true;
Mel Gorman56de7262010-05-24 14:32:30 -07002494
Mel Gorman66199712012-01-12 17:19:41 -08002495 /*
2496 * If compaction is deferred for high-order allocations, it is because
2497 * sync compaction recently failed. In this is the case and the caller
Mel Gorman27d1af82012-08-21 16:16:17 -07002498 * requested a movable allocation that does not heavily disrupt the
2499 * system then fail the allocation instead of entering direct reclaim.
Mel Gorman66199712012-01-12 17:19:41 -08002500 */
Mel Gorman27d1af82012-08-21 16:16:17 -07002501 if ((deferred_compaction || contended_compaction) &&
2502 (gfp_mask & __GFP_NO_KSWAPD))
Mel Gorman66199712012-01-12 17:19:41 -08002503 goto nopage;
2504
Mel Gorman11e33f62009-06-16 15:31:57 -07002505 /* Try direct reclaim and then allocating */
2506 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2507 zonelist, high_zoneidx,
2508 nodemask,
Mel Gorman5117f452009-06-16 15:31:59 -07002509 alloc_flags, preferred_zone,
Mel Gorman3dd28262009-06-16 15:32:00 -07002510 migratetype, &did_some_progress);
Mel Gorman11e33f62009-06-16 15:31:57 -07002511 if (page)
2512 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 /*
Mel Gorman11e33f62009-06-16 15:31:57 -07002515 * If we failed to make any progress reclaiming, then we are
2516 * running out of options and have to consider going OOM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 */
Mel Gorman11e33f62009-06-16 15:31:57 -07002518 if (!did_some_progress) {
2519 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -07002520 if (oom_killer_disabled)
2521 goto nopage;
David Rientjes29fd66d2012-03-28 14:42:41 -07002522 /* Coredumps can quickly deplete all memory reserves */
2523 if ((current->flags & PF_DUMPCORE) &&
2524 !(gfp_mask & __GFP_NOFAIL))
2525 goto nopage;
Mel Gorman11e33f62009-06-16 15:31:57 -07002526 page = __alloc_pages_may_oom(gfp_mask, order,
2527 zonelist, high_zoneidx,
Mel Gorman3dd28262009-06-16 15:32:00 -07002528 nodemask, preferred_zone,
2529 migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002530 if (page)
2531 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532
David Rientjes03668b32010-08-09 17:18:54 -07002533 if (!(gfp_mask & __GFP_NOFAIL)) {
2534 /*
2535 * The oom killer is not called for high-order
2536 * allocations that may fail, so if no progress
2537 * is being made, there are no other options and
2538 * retrying is unlikely to help.
2539 */
2540 if (order > PAGE_ALLOC_COSTLY_ORDER)
2541 goto nopage;
2542 /*
2543 * The oom killer is not called for lowmem
2544 * allocations to prevent needlessly killing
2545 * innocent tasks.
2546 */
2547 if (high_zoneidx < ZONE_NORMAL)
2548 goto nopage;
2549 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 goto restart;
2552 }
2553 }
2554
Mel Gorman11e33f62009-06-16 15:31:57 -07002555 /* Check if we should retry the allocation */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07002556 pages_reclaimed += did_some_progress;
Mel Gormanf90ac392012-01-10 15:07:15 -08002557 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2558 pages_reclaimed)) {
Mel Gorman11e33f62009-06-16 15:31:57 -07002559 /* Wait for some write requests to complete then retry */
Mel Gorman0e093d992010-10-26 14:21:45 -07002560 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 goto rebalance;
Mel Gorman3e7d3442011-01-13 15:45:56 -08002562 } else {
2563 /*
2564 * High-order allocations do not necessarily loop after
2565 * direct reclaim and reclaim/compaction depends on compaction
2566 * being called after reclaim so call directly if necessary
2567 */
2568 page = __alloc_pages_direct_compact(gfp_mask, order,
2569 zonelist, high_zoneidx,
2570 nodemask,
2571 alloc_flags, preferred_zone,
Mel Gorman66199712012-01-12 17:19:41 -08002572 migratetype, sync_migration,
Mel Gorman27d1af82012-08-21 16:16:17 -07002573 &contended_compaction,
Mel Gorman66199712012-01-12 17:19:41 -08002574 &deferred_compaction,
2575 &did_some_progress);
Mel Gorman3e7d3442011-01-13 15:45:56 -08002576 if (page)
2577 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 }
2579
2580nopage:
Dave Hansena238ab52011-05-24 17:12:16 -07002581 warn_alloc_failed(gfp_mask, order, NULL);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002582 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583got_pg:
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002584 if (kmemcheck_enabled)
2585 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 return page;
Mel Gorman11e33f62009-06-16 15:31:57 -07002587
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588}
Mel Gorman11e33f62009-06-16 15:31:57 -07002589
2590/*
2591 * This is the 'heart' of the zoned buddy allocator.
2592 */
2593struct page *
2594__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2595 struct zonelist *zonelist, nodemask_t *nodemask)
2596{
2597 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gorman5117f452009-06-16 15:31:59 -07002598 struct zone *preferred_zone;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002599 struct page *page = NULL;
Mel Gorman3dd28262009-06-16 15:32:00 -07002600 int migratetype = allocflags_to_migratetype(gfp_mask);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002601 unsigned int cpuset_mems_cookie;
Bartlomiej Zolnierkiewicz19644312012-10-08 16:32:05 -07002602 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET;
Mel Gorman11e33f62009-06-16 15:31:57 -07002603
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10002604 gfp_mask &= gfp_allowed_mask;
2605
Mel Gorman11e33f62009-06-16 15:31:57 -07002606 lockdep_trace_alloc(gfp_mask);
2607
2608 might_sleep_if(gfp_mask & __GFP_WAIT);
2609
2610 if (should_fail_alloc_page(gfp_mask, order))
2611 return NULL;
2612
2613 /*
2614 * Check the zones suitable for the gfp_mask contain at least one
2615 * valid zone. It's possible to have an empty zonelist as a result
2616 * of GFP_THISNODE and a memoryless node
2617 */
2618 if (unlikely(!zonelist->_zonerefs->zone))
2619 return NULL;
2620
Mel Gormancc9a6c82012-03-21 16:34:11 -07002621retry_cpuset:
2622 cpuset_mems_cookie = get_mems_allowed();
2623
Mel Gorman5117f452009-06-16 15:31:59 -07002624 /* The preferred zone is used for statistics later */
David Rientjesf33261d2011-01-25 15:07:20 -08002625 first_zones_zonelist(zonelist, high_zoneidx,
2626 nodemask ? : &cpuset_current_mems_allowed,
2627 &preferred_zone);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002628 if (!preferred_zone)
2629 goto out;
Mel Gorman5117f452009-06-16 15:31:59 -07002630
Bartlomiej Zolnierkiewicz19644312012-10-08 16:32:05 -07002631#ifdef CONFIG_CMA
2632 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2633 alloc_flags |= ALLOC_CMA;
2634#endif
Mel Gorman5117f452009-06-16 15:31:59 -07002635 /* First allocation attempt */
Mel Gorman11e33f62009-06-16 15:31:57 -07002636 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Bartlomiej Zolnierkiewicz19644312012-10-08 16:32:05 -07002637 zonelist, high_zoneidx, alloc_flags,
Mel Gorman3dd28262009-06-16 15:32:00 -07002638 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002639 if (unlikely(!page))
2640 page = __alloc_pages_slowpath(gfp_mask, order,
Mel Gorman5117f452009-06-16 15:31:59 -07002641 zonelist, high_zoneidx, nodemask,
Mel Gorman3dd28262009-06-16 15:32:00 -07002642 preferred_zone, migratetype);
Mel Gorman11e33f62009-06-16 15:31:57 -07002643
Mel Gorman4b4f2782009-09-21 17:02:41 -07002644 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002645
2646out:
2647 /*
2648 * When updating a task's mems_allowed, it is possible to race with
2649 * parallel threads in such a way that an allocation can fail while
2650 * the mask is being updated. If a page allocation is about to fail,
2651 * check if the cpuset changed during allocation and if so, retry.
2652 */
2653 if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
2654 goto retry_cpuset;
2655
Mel Gorman11e33f62009-06-16 15:31:57 -07002656 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657}
Mel Gormand2391712009-06-16 15:31:52 -07002658EXPORT_SYMBOL(__alloc_pages_nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659
2660/*
2661 * Common helper functions.
2662 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08002663unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664{
Akinobu Mita945a1112009-09-21 17:01:47 -07002665 struct page *page;
2666
2667 /*
2668 * __get_free_pages() returns a 32-bit address, which cannot represent
2669 * a highmem page
2670 */
2671 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2672
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 page = alloc_pages(gfp_mask, order);
2674 if (!page)
2675 return 0;
2676 return (unsigned long) page_address(page);
2677}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678EXPORT_SYMBOL(__get_free_pages);
2679
Harvey Harrison920c7a52008-02-04 22:29:26 -08002680unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681{
Akinobu Mita945a1112009-09-21 17:01:47 -07002682 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684EXPORT_SYMBOL(get_zeroed_page);
2685
Harvey Harrison920c7a52008-02-04 22:29:26 -08002686void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687{
Nick Pigginb5810032005-10-29 18:16:12 -07002688 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 if (order == 0)
Li Hongfc916682010-03-05 13:41:54 -08002690 free_hot_cold_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 else
2692 __free_pages_ok(page, order);
2693 }
2694}
2695
2696EXPORT_SYMBOL(__free_pages);
2697
Harvey Harrison920c7a52008-02-04 22:29:26 -08002698void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699{
2700 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07002701 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 __free_pages(virt_to_page((void *)addr), order);
2703 }
2704}
2705
2706EXPORT_SYMBOL(free_pages);
2707
Andi Kleenee85c2e2011-05-11 15:13:34 -07002708static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2709{
2710 if (addr) {
2711 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2712 unsigned long used = addr + PAGE_ALIGN(size);
2713
2714 split_page(virt_to_page((void *)addr), order);
2715 while (used < alloc_end) {
2716 free_page(used);
2717 used += PAGE_SIZE;
2718 }
2719 }
2720 return (void *)addr;
2721}
2722
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002723/**
2724 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2725 * @size: the number of bytes to allocate
2726 * @gfp_mask: GFP flags for the allocation
2727 *
2728 * This function is similar to alloc_pages(), except that it allocates the
2729 * minimum number of pages to satisfy the request. alloc_pages() can only
2730 * allocate memory in power-of-two pages.
2731 *
2732 * This function is also limited by MAX_ORDER.
2733 *
2734 * Memory allocated by this function must be released by free_pages_exact().
2735 */
2736void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2737{
2738 unsigned int order = get_order(size);
2739 unsigned long addr;
2740
2741 addr = __get_free_pages(gfp_mask, order);
Andi Kleenee85c2e2011-05-11 15:13:34 -07002742 return make_alloc_exact(addr, order, size);
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002743}
2744EXPORT_SYMBOL(alloc_pages_exact);
2745
2746/**
Andi Kleenee85c2e2011-05-11 15:13:34 -07002747 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2748 * pages on a node.
Randy Dunlapb5e6ab52011-05-16 13:16:54 -07002749 * @nid: the preferred node ID where memory should be allocated
Andi Kleenee85c2e2011-05-11 15:13:34 -07002750 * @size: the number of bytes to allocate
2751 * @gfp_mask: GFP flags for the allocation
2752 *
2753 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2754 * back.
2755 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2756 * but is not exact.
2757 */
2758void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2759{
2760 unsigned order = get_order(size);
2761 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2762 if (!p)
2763 return NULL;
2764 return make_alloc_exact((unsigned long)page_address(p), order, size);
2765}
2766EXPORT_SYMBOL(alloc_pages_exact_nid);
2767
2768/**
Timur Tabi2be0ffe2008-07-23 21:28:11 -07002769 * free_pages_exact - release memory allocated via alloc_pages_exact()
2770 * @virt: the value returned by alloc_pages_exact.
2771 * @size: size of allocation, same value as passed to alloc_pages_exact().
2772 *
2773 * Release the memory allocated by a previous call to alloc_pages_exact.
2774 */
2775void free_pages_exact(void *virt, size_t size)
2776{
2777 unsigned long addr = (unsigned long)virt;
2778 unsigned long end = addr + PAGE_ALIGN(size);
2779
2780 while (addr < end) {
2781 free_page(addr);
2782 addr += PAGE_SIZE;
2783 }
2784}
2785EXPORT_SYMBOL(free_pages_exact);
2786
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787static unsigned int nr_free_zone_pages(int offset)
2788{
Mel Gormandd1a2392008-04-28 02:12:17 -07002789 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002790 struct zone *zone;
2791
Martin J. Blighe310fd42005-07-29 22:59:18 -07002792 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 unsigned int sum = 0;
2794
Mel Gorman0e884602008-04-28 02:12:14 -07002795 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796
Mel Gorman54a6eb52008-04-28 02:12:16 -07002797 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07002798 unsigned long size = zone->present_pages;
Mel Gorman41858962009-06-16 15:32:12 -07002799 unsigned long high = high_wmark_pages(zone);
Martin J. Blighe310fd42005-07-29 22:59:18 -07002800 if (size > high)
2801 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802 }
2803
2804 return sum;
2805}
2806
2807/*
2808 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2809 */
2810unsigned int nr_free_buffer_pages(void)
2811{
Al Viroaf4ca452005-10-21 02:55:38 -04002812 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813}
Meelap Shahc2f1a552007-07-17 04:04:39 -07002814EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815
2816/*
2817 * Amount of free RAM allocatable within all zones
2818 */
2819unsigned int nr_free_pagecache_pages(void)
2820{
Mel Gorman2a1e2742007-07-17 04:03:12 -07002821 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002823
2824static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07002826 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08002827 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830void si_meminfo(struct sysinfo *val)
2831{
2832 val->totalram = totalram_pages;
2833 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002834 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 val->totalhigh = totalhigh_pages;
2837 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 val->mem_unit = PAGE_SIZE;
2839}
2840
2841EXPORT_SYMBOL(si_meminfo);
2842
2843#ifdef CONFIG_NUMA
2844void si_meminfo_node(struct sysinfo *val, int nid)
2845{
2846 pg_data_t *pgdat = NODE_DATA(nid);
2847
2848 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002849 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002850#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08002852 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2853 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002854#else
2855 val->totalhigh = 0;
2856 val->freehigh = 0;
2857#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 val->mem_unit = PAGE_SIZE;
2859}
2860#endif
2861
David Rientjesddd588b2011-03-22 16:30:46 -07002862/*
David Rientjes7bf02ea2011-05-24 17:11:16 -07002863 * Determine whether the node should be displayed or not, depending on whether
2864 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
David Rientjesddd588b2011-03-22 16:30:46 -07002865 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002866bool skip_free_areas_node(unsigned int flags, int nid)
David Rientjesddd588b2011-03-22 16:30:46 -07002867{
2868 bool ret = false;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002869 unsigned int cpuset_mems_cookie;
David Rientjesddd588b2011-03-22 16:30:46 -07002870
2871 if (!(flags & SHOW_MEM_FILTER_NODES))
2872 goto out;
2873
Mel Gormancc9a6c82012-03-21 16:34:11 -07002874 do {
2875 cpuset_mems_cookie = get_mems_allowed();
2876 ret = !node_isset(nid, cpuset_current_mems_allowed);
2877 } while (!put_mems_allowed(cpuset_mems_cookie));
David Rientjesddd588b2011-03-22 16:30:46 -07002878out:
2879 return ret;
2880}
2881
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882#define K(x) ((x) << (PAGE_SHIFT-10))
2883
Rabin Vincent394b8082012-12-11 16:00:24 -08002884static void show_migration_types(unsigned char type)
2885{
2886 static const char types[MIGRATE_TYPES] = {
2887 [MIGRATE_UNMOVABLE] = 'U',
2888 [MIGRATE_RECLAIMABLE] = 'E',
2889 [MIGRATE_MOVABLE] = 'M',
2890 [MIGRATE_RESERVE] = 'R',
2891#ifdef CONFIG_CMA
2892 [MIGRATE_CMA] = 'C',
2893#endif
2894 [MIGRATE_ISOLATE] = 'I',
2895 };
2896 char tmp[MIGRATE_TYPES + 1];
2897 char *p = tmp;
2898 int i;
2899
2900 for (i = 0; i < MIGRATE_TYPES; i++) {
2901 if (type & (1 << i))
2902 *p++ = types[i];
2903 }
2904
2905 *p = '\0';
2906 printk("(%s) ", tmp);
2907}
2908
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909/*
2910 * Show free area list (used inside shift_scroll-lock stuff)
2911 * We also calculate the percentage fragmentation. We do this by counting the
2912 * memory on each free list with the exception of the first item on the list.
David Rientjesddd588b2011-03-22 16:30:46 -07002913 * Suppresses nodes that are not allowed by current's cpuset if
2914 * SHOW_MEM_FILTER_NODES is passed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 */
David Rientjes7bf02ea2011-05-24 17:11:16 -07002916void show_free_areas(unsigned int filter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917{
Jes Sorensenc7241912006-09-27 01:50:05 -07002918 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919 struct zone *zone;
2920
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002921 for_each_populated_zone(zone) {
David Rientjes7bf02ea2011-05-24 17:11:16 -07002922 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002923 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07002924 show_node(zone);
2925 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926
Dave Jones6b482c62005-11-10 15:45:56 -05002927 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 struct per_cpu_pageset *pageset;
2929
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09002930 pageset = per_cpu_ptr(zone->pageset, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002932 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2933 cpu, pageset->pcp.high,
2934 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935 }
2936 }
2937
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002938 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2939 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002940 " unevictable:%lu"
Andrew Mortonb76146e2009-10-26 16:49:52 -07002941 " dirty:%lu writeback:%lu unstable:%lu\n"
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002942 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
Larry Bassel6d6e2c92012-12-14 14:21:05 -08002943 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
2944 " free_cma:%lu\n",
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002945 global_page_state(NR_ACTIVE_ANON),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002946 global_page_state(NR_INACTIVE_ANON),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002947 global_page_state(NR_ISOLATED_ANON),
2948 global_page_state(NR_ACTIVE_FILE),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002949 global_page_state(NR_INACTIVE_FILE),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002950 global_page_state(NR_ISOLATED_FILE),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002951 global_page_state(NR_UNEVICTABLE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07002952 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07002953 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07002954 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08002955 global_page_state(NR_FREE_PAGES),
KOSAKI Motohiro3701b032009-09-21 17:01:29 -07002956 global_page_state(NR_SLAB_RECLAIMABLE),
2957 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07002958 global_page_state(NR_FILE_MAPPED),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002959 global_page_state(NR_SHMEM),
Andrew Mortona25700a2007-02-08 14:20:40 -08002960 global_page_state(NR_PAGETABLE),
Larry Bassel6d6e2c92012-12-14 14:21:05 -08002961 global_page_state(NR_BOUNCE),
2962 global_page_state(NR_FREE_CMA_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002964 for_each_populated_zone(zone) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965 int i;
2966
David Rientjes7bf02ea2011-05-24 17:11:16 -07002967 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07002968 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 show_node(zone);
2970 printk("%s"
2971 " free:%lukB"
2972 " min:%lukB"
2973 " low:%lukB"
2974 " high:%lukB"
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002975 " active_anon:%lukB"
2976 " inactive_anon:%lukB"
2977 " active_file:%lukB"
2978 " inactive_file:%lukB"
Lee Schermerhorn7b854122008-10-18 20:26:40 -07002979 " unevictable:%lukB"
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07002980 " isolated(anon):%lukB"
2981 " isolated(file):%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982 " present:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002983 " mlocked:%lukB"
2984 " dirty:%lukB"
2985 " writeback:%lukB"
2986 " mapped:%lukB"
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07002987 " shmem:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002988 " slab_reclaimable:%lukB"
2989 " slab_unreclaimable:%lukB"
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07002990 " kernel_stack:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002991 " pagetables:%lukB"
2992 " unstable:%lukB"
2993 " bounce:%lukB"
Larry Bassel6d6e2c92012-12-14 14:21:05 -08002994 " free_cma:%lukB"
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07002995 " writeback_tmp:%lukB"
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 " pages_scanned:%lu"
2997 " all_unreclaimable? %s"
2998 "\n",
2999 zone->name,
Mel Gorman88f5acf2011-01-13 15:45:41 -08003000 K(zone_page_state(zone, NR_FREE_PAGES)),
Mel Gorman41858962009-06-16 15:32:12 -07003001 K(min_wmark_pages(zone)),
3002 K(low_wmark_pages(zone)),
3003 K(high_wmark_pages(zone)),
Rik van Riel4f98a2f2008-10-18 20:26:32 -07003004 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3005 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3006 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3007 K(zone_page_state(zone, NR_INACTIVE_FILE)),
Lee Schermerhorn7b854122008-10-18 20:26:40 -07003008 K(zone_page_state(zone, NR_UNEVICTABLE)),
KOSAKI Motohiroa7312862009-09-21 17:01:37 -07003009 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3010 K(zone_page_state(zone, NR_ISOLATED_FILE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 K(zone->present_pages),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003012 K(zone_page_state(zone, NR_MLOCK)),
3013 K(zone_page_state(zone, NR_FILE_DIRTY)),
3014 K(zone_page_state(zone, NR_WRITEBACK)),
3015 K(zone_page_state(zone, NR_FILE_MAPPED)),
KOSAKI Motohiro4b021082009-09-21 17:01:33 -07003016 K(zone_page_state(zone, NR_SHMEM)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003017 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3018 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
KOSAKI Motohiroc6a7f572009-09-21 17:01:32 -07003019 zone_page_state(zone, NR_KERNEL_STACK) *
3020 THREAD_SIZE / 1024,
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003021 K(zone_page_state(zone, NR_PAGETABLE)),
3022 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3023 K(zone_page_state(zone, NR_BOUNCE)),
Larry Bassel6d6e2c92012-12-14 14:21:05 -08003024 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
KOSAKI Motohiro4a0aa732009-09-21 17:01:30 -07003025 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026 zone->pages_scanned,
Lisa Due0935212013-09-11 14:22:36 -07003027 (!zone_reclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 );
3029 printk("lowmem_reserve[]:");
3030 for (i = 0; i < MAX_NR_ZONES; i++)
3031 printk(" %lu", zone->lowmem_reserve[i]);
3032 printk("\n");
3033 }
3034
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07003035 for_each_populated_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003036 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Rabin Vincent394b8082012-12-11 16:00:24 -08003037 unsigned char types[MAX_ORDER];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038
David Rientjes7bf02ea2011-05-24 17:11:16 -07003039 if (skip_free_areas_node(filter, zone_to_nid(zone)))
David Rientjesddd588b2011-03-22 16:30:46 -07003040 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 show_node(zone);
3042 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043
3044 spin_lock_irqsave(&zone->lock, flags);
3045 for (order = 0; order < MAX_ORDER; order++) {
Rabin Vincent394b8082012-12-11 16:00:24 -08003046 struct free_area *area = &zone->free_area[order];
3047 int type;
3048
3049 nr[order] = area->nr_free;
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003050 total += nr[order] << order;
Rabin Vincent394b8082012-12-11 16:00:24 -08003051
3052 types[order] = 0;
3053 for (type = 0; type < MIGRATE_TYPES; type++) {
3054 if (!list_empty(&area->free_list[type]))
3055 types[order] |= 1 << type;
3056 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057 }
3058 spin_unlock_irqrestore(&zone->lock, flags);
Rabin Vincent394b8082012-12-11 16:00:24 -08003059 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07003060 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Rabin Vincent394b8082012-12-11 16:00:24 -08003061 if (nr[order])
3062 show_migration_types(types[order]);
3063 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 printk("= %lukB\n", K(total));
3065 }
3066
Larry Woodmane6f36022008-02-04 22:29:30 -08003067 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3068
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 show_swap_cache_info();
3070}
3071
Mel Gorman19770b32008-04-28 02:12:18 -07003072static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3073{
3074 zoneref->zone = zone;
3075 zoneref->zone_idx = zone_idx(zone);
3076}
3077
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078/*
3079 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08003080 *
3081 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003083static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
3084 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085{
Christoph Lameter1a932052006-01-06 00:11:16 -08003086 struct zone *zone;
3087
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003088 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003089 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08003090
3091 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003092 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08003093 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08003094 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003095 zoneref_set_zone(zone,
3096 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08003097 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08003099
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003100 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08003101 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102}
3103
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003104
3105/*
3106 * zonelist_order:
3107 * 0 = automatic detection of better ordering.
3108 * 1 = order by ([node] distance, -zonetype)
3109 * 2 = order by (-zonetype, [node] distance)
3110 *
3111 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3112 * the same zonelist. So only NUMA can configure this param.
3113 */
3114#define ZONELIST_ORDER_DEFAULT 0
3115#define ZONELIST_ORDER_NODE 1
3116#define ZONELIST_ORDER_ZONE 2
3117
3118/* zonelist order in the kernel.
3119 * set_zonelist_order() will set this to NODE or ZONE.
3120 */
3121static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3122static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3123
3124
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003126/* The value user specified ....changed by config */
3127static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3128/* string for sysctl */
3129#define NUMA_ZONELIST_ORDER_LEN 16
3130char numa_zonelist_order[16] = "default";
3131
3132/*
3133 * interface for configure zonelist ordering.
3134 * command line option "numa_zonelist_order"
3135 * = "[dD]efault - default, automatic configuration.
3136 * = "[nN]ode - order by node locality, then by zone within node
3137 * = "[zZ]one - order by zone, then by locality within zone
3138 */
3139
3140static int __parse_numa_zonelist_order(char *s)
3141{
3142 if (*s == 'd' || *s == 'D') {
3143 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3144 } else if (*s == 'n' || *s == 'N') {
3145 user_zonelist_order = ZONELIST_ORDER_NODE;
3146 } else if (*s == 'z' || *s == 'Z') {
3147 user_zonelist_order = ZONELIST_ORDER_ZONE;
3148 } else {
3149 printk(KERN_WARNING
3150 "Ignoring invalid numa_zonelist_order value: "
3151 "%s\n", s);
3152 return -EINVAL;
3153 }
3154 return 0;
3155}
3156
3157static __init int setup_numa_zonelist_order(char *s)
3158{
Volodymyr G. Lukiianykecb256f2011-01-13 15:46:26 -08003159 int ret;
3160
3161 if (!s)
3162 return 0;
3163
3164 ret = __parse_numa_zonelist_order(s);
3165 if (ret == 0)
3166 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3167
3168 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003169}
3170early_param("numa_zonelist_order", setup_numa_zonelist_order);
3171
3172/*
3173 * sysctl handler for numa_zonelist_order
3174 */
3175int numa_zonelist_order_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003176 void __user *buffer, size_t *length,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003177 loff_t *ppos)
3178{
3179 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3180 int ret;
Andi Kleen443c6f12009-12-23 21:00:47 +01003181 static DEFINE_MUTEX(zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003182
Andi Kleen443c6f12009-12-23 21:00:47 +01003183 mutex_lock(&zl_order_mutex);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003184 if (write)
Andi Kleen443c6f12009-12-23 21:00:47 +01003185 strcpy(saved_string, (char*)table->data);
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07003186 ret = proc_dostring(table, write, buffer, length, ppos);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003187 if (ret)
Andi Kleen443c6f12009-12-23 21:00:47 +01003188 goto out;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003189 if (write) {
3190 int oldval = user_zonelist_order;
3191 if (__parse_numa_zonelist_order((char*)table->data)) {
3192 /*
3193 * bogus value. restore saved string
3194 */
3195 strncpy((char*)table->data, saved_string,
3196 NUMA_ZONELIST_ORDER_LEN);
3197 user_zonelist_order = oldval;
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003198 } else if (oldval != user_zonelist_order) {
3199 mutex_lock(&zonelists_mutex);
Haicheng Li1f522502010-05-24 14:32:51 -07003200 build_all_zonelists(NULL);
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003201 mutex_unlock(&zonelists_mutex);
3202 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003203 }
Andi Kleen443c6f12009-12-23 21:00:47 +01003204out:
3205 mutex_unlock(&zl_order_mutex);
3206 return ret;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003207}
3208
3209
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003210#define MAX_NODE_LOAD (nr_online_nodes)
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003211static int node_load[MAX_NUMNODES];
3212
Linus Torvalds1da177e2005-04-16 15:20:36 -07003213/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07003214 * 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 -07003215 * @node: node whose fallback list we're appending
3216 * @used_node_mask: nodemask_t of already used nodes
3217 *
3218 * We use a number of factors to determine which is the next node that should
3219 * appear on a given node's fallback list. The node should not have appeared
3220 * already in @node's fallback list, and it should be the next closest node
3221 * according to the distance array (which contains arbitrary distance values
3222 * from each node to each node in the system), and should also prefer nodes
3223 * with no CPUs, since presumably they'll have very little allocation pressure
3224 * on them otherwise.
3225 * It returns -1 if no node is found.
3226 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003227static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003228{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003229 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230 int min_val = INT_MAX;
3231 int best_node = -1;
Rusty Russella70f7302009-03-13 14:49:46 +10303232 const struct cpumask *tmp = cpumask_of_node(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003234 /* Use the local node if we haven't already */
3235 if (!node_isset(node, *used_node_mask)) {
3236 node_set(node, *used_node_mask);
3237 return node;
3238 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003240 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241
3242 /* Don't want a node to appear more than once */
3243 if (node_isset(n, *used_node_mask))
3244 continue;
3245
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246 /* Use the distance array to find the distance */
3247 val = node_distance(node, n);
3248
Linus Torvalds4cf808e2006-02-17 20:38:21 +01003249 /* Penalize nodes under us ("prefer the next node") */
3250 val += (n < node);
3251
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252 /* Give preference to headless and unused nodes */
Rusty Russella70f7302009-03-13 14:49:46 +10303253 tmp = cpumask_of_node(n);
3254 if (!cpumask_empty(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255 val += PENALTY_FOR_NODE_WITH_CPUS;
3256
3257 /* Slight preference for less loaded node */
3258 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3259 val += node_load[n];
3260
3261 if (val < min_val) {
3262 min_val = val;
3263 best_node = n;
3264 }
3265 }
3266
3267 if (best_node >= 0)
3268 node_set(best_node, *used_node_mask);
3269
3270 return best_node;
3271}
3272
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003273
3274/*
3275 * Build zonelists ordered by node and zones within node.
3276 * This results in maximum locality--normal zone overflows into local
3277 * DMA zone, if any--but risks exhausting DMA zone.
3278 */
3279static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003281 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003282 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003283
Mel Gorman54a6eb52008-04-28 02:12:16 -07003284 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07003285 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07003286 ;
3287 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3288 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07003289 zonelist->_zonerefs[j].zone = NULL;
3290 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003291}
3292
3293/*
Christoph Lameter523b9452007-10-16 01:25:37 -07003294 * Build gfp_thisnode zonelists
3295 */
3296static void build_thisnode_zonelists(pg_data_t *pgdat)
3297{
Christoph Lameter523b9452007-10-16 01:25:37 -07003298 int j;
3299 struct zonelist *zonelist;
3300
Mel Gorman54a6eb52008-04-28 02:12:16 -07003301 zonelist = &pgdat->node_zonelists[1];
3302 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07003303 zonelist->_zonerefs[j].zone = NULL;
3304 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07003305}
3306
3307/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003308 * Build zonelists ordered by zone and nodes within zones.
3309 * This results in conserving DMA zone[s] until all Normal memory is
3310 * exhausted, but results in overflowing to remote node while memory
3311 * may still exist in local DMA zone.
3312 */
3313static int node_order[MAX_NUMNODES];
3314
3315static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3316{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003317 int pos, j, node;
3318 int zone_type; /* needs to be signed */
3319 struct zone *z;
3320 struct zonelist *zonelist;
3321
Mel Gorman54a6eb52008-04-28 02:12:16 -07003322 zonelist = &pgdat->node_zonelists[0];
3323 pos = 0;
3324 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3325 for (j = 0; j < nr_nodes; j++) {
3326 node = node_order[j];
3327 z = &NODE_DATA(node)->node_zones[zone_type];
3328 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07003329 zoneref_set_zone(z,
3330 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07003331 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003332 }
3333 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003334 }
Mel Gormandd1a2392008-04-28 02:12:17 -07003335 zonelist->_zonerefs[pos].zone = NULL;
3336 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003337}
3338
3339static int default_zonelist_order(void)
3340{
3341 int nid, zone_type;
3342 unsigned long low_kmem_size,total_size;
3343 struct zone *z;
3344 int average_size;
3345 /*
Thomas Weber88393162010-03-16 11:47:56 +01003346 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003347 * If they are really small and used heavily, the system can fall
3348 * into OOM very easily.
David Rientjese325c902010-05-24 14:32:13 -07003349 * This function detect ZONE_DMA/DMA32 size and configures zone order.
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003350 */
3351 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3352 low_kmem_size = 0;
3353 total_size = 0;
3354 for_each_online_node(nid) {
3355 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3356 z = &NODE_DATA(nid)->node_zones[zone_type];
3357 if (populated_zone(z)) {
3358 if (zone_type < ZONE_NORMAL)
3359 low_kmem_size += z->present_pages;
3360 total_size += z->present_pages;
David Rientjese325c902010-05-24 14:32:13 -07003361 } else if (zone_type == ZONE_NORMAL) {
3362 /*
3363 * If any node has only lowmem, then node order
3364 * is preferred to allow kernel allocations
3365 * locally; otherwise, they can easily infringe
3366 * on other nodes when there is an abundance of
3367 * lowmem available to allocate from.
3368 */
3369 return ZONELIST_ORDER_NODE;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003370 }
3371 }
3372 }
3373 if (!low_kmem_size || /* there are no DMA area. */
3374 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3375 return ZONELIST_ORDER_NODE;
3376 /*
3377 * look into each node's config.
3378 * If there is a node whose DMA/DMA32 memory is very big area on
3379 * local memory, NODE_ORDER may be suitable.
3380 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003381 average_size = total_size /
3382 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003383 for_each_online_node(nid) {
3384 low_kmem_size = 0;
3385 total_size = 0;
3386 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3387 z = &NODE_DATA(nid)->node_zones[zone_type];
3388 if (populated_zone(z)) {
3389 if (zone_type < ZONE_NORMAL)
3390 low_kmem_size += z->present_pages;
3391 total_size += z->present_pages;
3392 }
3393 }
3394 if (low_kmem_size &&
3395 total_size > average_size && /* ignore small node */
3396 low_kmem_size > total_size * 70/100)
3397 return ZONELIST_ORDER_NODE;
3398 }
3399 return ZONELIST_ORDER_ZONE;
3400}
3401
3402static void set_zonelist_order(void)
3403{
3404 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3405 current_zonelist_order = default_zonelist_order();
3406 else
3407 current_zonelist_order = user_zonelist_order;
3408}
3409
3410static void build_zonelists(pg_data_t *pgdat)
3411{
3412 int j, node, load;
3413 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003415 int local_node, prev_node;
3416 struct zonelist *zonelist;
3417 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418
3419 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07003420 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07003422 zonelist->_zonerefs[0].zone = NULL;
3423 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424 }
3425
3426 /* NUMA-aware ordering of nodes */
3427 local_node = pgdat->node_id;
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003428 load = nr_online_nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429 prev_node = local_node;
3430 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003431
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003432 memset(node_order, 0, sizeof(node_order));
3433 j = 0;
3434
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08003436 int distance = node_distance(local_node, node);
3437
3438 /*
3439 * If another node is sufficiently far away then it is better
3440 * to reclaim pages in a zone before going off node.
3441 */
3442 if (distance > RECLAIM_DISTANCE)
3443 zone_reclaim_mode = 1;
3444
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445 /*
3446 * We don't want to pressure a particular node.
3447 * So adding penalty to the first node in same
3448 * distance group to make it round-robin.
3449 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08003450 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003451 node_load[node] = load;
3452
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 prev_node = node;
3454 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003455 if (order == ZONELIST_ORDER_NODE)
3456 build_zonelists_in_node_order(pgdat, node);
3457 else
3458 node_order[j++] = node; /* remember order */
3459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003461 if (order == ZONELIST_ORDER_ZONE) {
3462 /* calculate node order -- i.e., DMA last! */
3463 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003464 }
Christoph Lameter523b9452007-10-16 01:25:37 -07003465
3466 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003467}
3468
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003469/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003470static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003471{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003472 struct zonelist *zonelist;
3473 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07003474 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003475
Mel Gorman54a6eb52008-04-28 02:12:16 -07003476 zonelist = &pgdat->node_zonelists[0];
3477 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3478 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07003479 for (z = zonelist->_zonerefs; z->zone; z++)
3480 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003481}
3482
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003483#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3484/*
3485 * Return node id of node used for "local" allocations.
3486 * I.e., first node id of first zone in arg node's generic zonelist.
3487 * Used for initializing percpu 'numa_mem', which is used primarily
3488 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3489 */
3490int local_memory_node(int node)
3491{
3492 struct zone *zone;
3493
3494 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3495 gfp_zone(GFP_KERNEL),
3496 NULL,
3497 &zone);
3498 return zone->node;
3499}
3500#endif
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003501
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502#else /* CONFIG_NUMA */
3503
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003504static void set_zonelist_order(void)
3505{
3506 current_zonelist_order = ZONELIST_ORDER_ZONE;
3507}
3508
3509static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510{
Christoph Lameter19655d32006-09-25 23:31:19 -07003511 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003512 enum zone_type j;
3513 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003514
3515 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516
Mel Gorman54a6eb52008-04-28 02:12:16 -07003517 zonelist = &pgdat->node_zonelists[0];
3518 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519
Mel Gorman54a6eb52008-04-28 02:12:16 -07003520 /*
3521 * Now we build the zonelist so that it contains the zones
3522 * of all the other nodes.
3523 * We don't want to pressure a particular node, so when
3524 * building the zones for node N, we make sure that the
3525 * zones coming right after the local ones are those from
3526 * node N+1 (modulo N)
3527 */
3528 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3529 if (!node_online(node))
3530 continue;
3531 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3532 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003533 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07003534 for (node = 0; node < local_node; node++) {
3535 if (!node_online(node))
3536 continue;
3537 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3538 MAX_NR_ZONES - 1);
3539 }
3540
Mel Gormandd1a2392008-04-28 02:12:17 -07003541 zonelist->_zonerefs[j].zone = NULL;
3542 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003543}
3544
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003545/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003546static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003547{
Mel Gorman54a6eb52008-04-28 02:12:16 -07003548 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003549}
3550
Linus Torvalds1da177e2005-04-16 15:20:36 -07003551#endif /* CONFIG_NUMA */
3552
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003553/*
3554 * Boot pageset table. One per cpu which is going to be used for all
3555 * zones and all nodes. The parameters will be set in such a way
3556 * that an item put on a list will immediately be handed over to
3557 * the buddy list. This is safe since pageset manipulation is done
3558 * with interrupts disabled.
3559 *
3560 * The boot_pagesets must be kept even after bootup is complete for
3561 * unused processors and/or zones. They do play a role for bootstrapping
3562 * hotplugged processors.
3563 *
3564 * zoneinfo_show() and maybe other functions do
3565 * not check if the processor is online before following the pageset pointer.
3566 * Other parts of the kernel may not check if the zone is available.
3567 */
3568static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3569static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
Haicheng Li1f522502010-05-24 14:32:51 -07003570static void setup_zone_pageset(struct zone *zone);
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003571
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003572/*
3573 * Global mutex to protect against size modification of zonelists
3574 * as well as to serialize pageset setup for the new populated zone.
3575 */
3576DEFINE_MUTEX(zonelists_mutex);
3577
Rusty Russell9b1a4d32008-07-28 12:16:30 -05003578/* return values int ....just for stop_machine() */
Haicheng Li1f522502010-05-24 14:32:51 -07003579static __init_refok int __build_all_zonelists(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580{
Yasunori Goto68113782006-06-23 02:03:11 -07003581 int nid;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003582 int cpu;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003583
Bo Liu7f9cfb32009-08-18 14:11:19 -07003584#ifdef CONFIG_NUMA
3585 memset(node_load, 0, sizeof(node_load));
3586#endif
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003587 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07003588 pg_data_t *pgdat = NODE_DATA(nid);
3589
3590 build_zonelists(pgdat);
3591 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08003592 }
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003593
3594 /*
3595 * Initialize the boot_pagesets that are going to be used
3596 * for bootstrapping processors. The real pagesets for
3597 * each zone will be allocated later when the per cpu
3598 * allocator is available.
3599 *
3600 * boot_pagesets are used also for bootstrapping offline
3601 * cpus if the system is already booted because the pagesets
3602 * are needed to initialize allocators on a specific cpu too.
3603 * F.e. the percpu allocator needs the page allocator which
3604 * needs the percpu allocator in order to allocate its pagesets
3605 * (a chicken-egg dilemma).
3606 */
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003607 for_each_possible_cpu(cpu) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09003608 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3609
Lee Schermerhorn7aac7892010-05-26 14:45:00 -07003610#ifdef CONFIG_HAVE_MEMORYLESS_NODES
3611 /*
3612 * We now know the "local memory node" for each node--
3613 * i.e., the node of the first zone in the generic zonelist.
3614 * Set up numa_mem percpu variable for on-line cpus. During
3615 * boot, only the boot cpu should be on-line; we'll init the
3616 * secondary cpus' numa_mem as they come on-line. During
3617 * node/memory hotplug, we'll fixup all on-line cpus.
3618 */
3619 if (cpu_online(cpu))
3620 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3621#endif
3622 }
3623
Yasunori Goto68113782006-06-23 02:03:11 -07003624 return 0;
3625}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003626
Haicheng Li4eaf3f62010-05-24 14:32:52 -07003627/*
3628 * Called with zonelists_mutex held always
3629 * unless system_state == SYSTEM_BOOTING.
3630 */
Paul Mundt9f6ae442011-04-14 15:21:57 -07003631void __ref build_all_zonelists(void *data)
Yasunori Goto68113782006-06-23 02:03:11 -07003632{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003633 set_zonelist_order();
3634
Yasunori Goto68113782006-06-23 02:03:11 -07003635 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07003636 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07003637 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07003638 cpuset_init_current_mems_allowed();
3639 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02003640 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07003641 of zonelist */
KAMEZAWA Hiroyukie9959f02010-11-24 12:57:09 -08003642#ifdef CONFIG_MEMORY_HOTPLUG
3643 if (data)
3644 setup_zone_pageset((struct zone *)data);
3645#endif
3646 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07003647 /* cpuset refresh routine should be here */
3648 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07003649 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003650 /*
3651 * Disable grouping by mobility if the number of pages in the
3652 * system is too low to allow the mechanism to work. It would be
3653 * more accurate, but expensive to check per-zone. This check is
3654 * made on memory-hotadd so a system can start with mobility
3655 * disabled and enable it later
3656 */
Mel Gormand9c23402007-10-16 01:26:01 -07003657 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003658 page_group_by_mobility_disabled = 1;
3659 else
3660 page_group_by_mobility_disabled = 0;
3661
3662 printk("Built %i zonelists in %s order, mobility grouping %s. "
3663 "Total pages: %ld\n",
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003664 nr_online_nodes,
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003665 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07003666 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07003667 vm_total_pages);
3668#ifdef CONFIG_NUMA
3669 printk("Policy zone: %s\n", zone_names[policy_zone]);
3670#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003671}
3672
3673/*
3674 * Helper functions to size the waitqueue hash table.
3675 * Essentially these want to choose hash table sizes sufficiently
3676 * large so that collisions trying to wait on pages are rare.
3677 * But in fact, the number of active page waitqueues on typical
3678 * systems is ridiculously low, less than 200. So this is even
3679 * conservative, even though it seems large.
3680 *
3681 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3682 * waitqueues, i.e. the size of the waitq table given the number of pages.
3683 */
3684#define PAGES_PER_WAITQUEUE 256
3685
Yasunori Gotocca448f2006-06-23 02:03:10 -07003686#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07003687static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688{
3689 unsigned long size = 1;
3690
3691 pages /= PAGES_PER_WAITQUEUE;
3692
3693 while (size < pages)
3694 size <<= 1;
3695
3696 /*
3697 * Once we have dozens or even hundreds of threads sleeping
3698 * on IO we've got bigger problems than wait queue collision.
3699 * Limit the size of the wait table to a reasonable size.
3700 */
3701 size = min(size, 4096UL);
3702
3703 return max(size, 4UL);
3704}
Yasunori Gotocca448f2006-06-23 02:03:10 -07003705#else
3706/*
3707 * A zone's size might be changed by hot-add, so it is not possible to determine
3708 * a suitable size for its wait_table. So we use the maximum size now.
3709 *
3710 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3711 *
3712 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3713 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3714 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3715 *
3716 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3717 * or more by the traditional way. (See above). It equals:
3718 *
3719 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3720 * ia64(16K page size) : = ( 8G + 4M)byte.
3721 * powerpc (64K page size) : = (32G +16M)byte.
3722 */
3723static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3724{
3725 return 4096UL;
3726}
3727#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003728
3729/*
3730 * This is an integer logarithm so that shifts can be used later
3731 * to extract the more random high bits from the multiplicative
3732 * hash function before the remainder is taken.
3733 */
3734static inline unsigned long wait_table_bits(unsigned long size)
3735{
3736 return ffz(~size);
3737}
3738
3739#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3740
Mel Gorman56fd56b2007-10-16 01:25:58 -07003741/*
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003742 * Check if a pageblock contains reserved pages
3743 */
3744static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3745{
3746 unsigned long pfn;
3747
3748 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3749 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3750 return 1;
3751 }
3752 return 0;
3753}
3754
3755/*
Mel Gormand9c23402007-10-16 01:26:01 -07003756 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman41858962009-06-16 15:32:12 -07003757 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3758 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
Mel Gorman56fd56b2007-10-16 01:25:58 -07003759 * higher will lead to a bigger reserve which will get freed as contiguous
3760 * blocks as reclaim kicks in
3761 */
3762static void setup_zone_migrate_reserve(struct zone *zone)
3763{
Arve Hjønnevåg6d3163c2011-05-24 17:12:24 -07003764 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003765 struct page *page;
Mel Gorman78986a62009-09-21 17:03:02 -07003766 unsigned long block_migratetype;
3767 int reserve;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003768
Michal Hockod02156382011-12-08 14:34:27 -08003769 /*
3770 * Get the start pfn, end pfn and the number of blocks to reserve
3771 * We have to be careful to be aligned to pageblock_nr_pages to
3772 * make sure that we always check pfn_valid for the first page in
3773 * the block.
3774 */
Mel Gorman56fd56b2007-10-16 01:25:58 -07003775 start_pfn = zone->zone_start_pfn;
3776 end_pfn = start_pfn + zone->spanned_pages;
Michal Hockod02156382011-12-08 14:34:27 -08003777 start_pfn = roundup(start_pfn, pageblock_nr_pages);
Mel Gorman41858962009-06-16 15:32:12 -07003778 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
Mel Gormand9c23402007-10-16 01:26:01 -07003779 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003780
Mel Gorman78986a62009-09-21 17:03:02 -07003781 /*
3782 * Reserve blocks are generally in place to help high-order atomic
3783 * allocations that are short-lived. A min_free_kbytes value that
3784 * would result in more than 2 reserve blocks for atomic allocations
3785 * is assumed to be in place to help anti-fragmentation for the
3786 * future allocation of hugepages at runtime.
3787 */
3788 reserve = min(2, reserve);
3789
Mel Gormand9c23402007-10-16 01:26:01 -07003790 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07003791 if (!pfn_valid(pfn))
3792 continue;
3793 page = pfn_to_page(pfn);
3794
Adam Litke344c7902008-09-02 14:35:38 -07003795 /* Watch out for overlapping nodes */
3796 if (page_to_nid(page) != zone_to_nid(zone))
3797 continue;
3798
Mel Gorman56fd56b2007-10-16 01:25:58 -07003799 block_migratetype = get_pageblock_migratetype(page);
3800
Mel Gorman938929f2012-01-10 15:07:14 -08003801 /* Only test what is necessary when the reserves are not met */
3802 if (reserve > 0) {
3803 /*
3804 * Blocks with reserved pages will never free, skip
3805 * them.
3806 */
3807 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
3808 if (pageblock_is_reserved(pfn, block_end_pfn))
3809 continue;
Mel Gorman56fd56b2007-10-16 01:25:58 -07003810
Mel Gorman938929f2012-01-10 15:07:14 -08003811 /* If this block is reserved, account for it */
3812 if (block_migratetype == MIGRATE_RESERVE) {
3813 reserve--;
3814 continue;
3815 }
3816
3817 /* Suitable for reserving if this block is movable */
3818 if (block_migratetype == MIGRATE_MOVABLE) {
3819 set_pageblock_migratetype(page,
3820 MIGRATE_RESERVE);
3821 move_freepages_block(zone, page,
3822 MIGRATE_RESERVE);
3823 reserve--;
3824 continue;
3825 }
Mel Gorman56fd56b2007-10-16 01:25:58 -07003826 }
3827
3828 /*
3829 * If the reserve is met and this is a previous reserved block,
3830 * take it back
3831 */
3832 if (block_migratetype == MIGRATE_RESERVE) {
3833 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3834 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3835 }
3836 }
3837}
Mel Gormanac0e5b72007-10-16 01:25:58 -07003838
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839/*
3840 * Initially all pages are reserved - free ones are freed
3841 * up by free_all_bootmem() once the early boot process is
3842 * done. Non-atomic initialization, single-pass.
3843 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01003844void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa02007-01-10 23:15:30 -08003845 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003846{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003847 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07003848 unsigned long end_pfn = start_pfn + size;
3849 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003850 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003851
Hugh Dickins22b31ee2009-01-06 14:40:09 -08003852 if (highest_memmap_pfn < end_pfn - 1)
3853 highest_memmap_pfn = end_pfn - 1;
3854
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003855 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08003856 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa02007-01-10 23:15:30 -08003857 /*
3858 * There can be holes in boot-time mem_map[]s
3859 * handed to this function. They do not
3860 * exist on hotplugged memory.
3861 */
3862 if (context == MEMMAP_EARLY) {
3863 if (!early_pfn_valid(pfn))
3864 continue;
3865 if (!early_pfn_in_nid(pfn, nid))
3866 continue;
3867 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003868 page = pfn_to_page(pfn);
3869 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07003870 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08003871 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872 reset_page_mapcount(page);
3873 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003874 /*
3875 * Mark the block movable so that blocks are reserved for
3876 * movable at startup. This will force kernel allocations
3877 * to reserve their blocks rather than leaking throughout
3878 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07003879 * kernel allocations are made. Later some blocks near
3880 * the start are marked MIGRATE_RESERVE by
3881 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003882 *
3883 * bitmap is created for zone's valid pfn range. but memmap
3884 * can be created for invalid pages (for alignment)
3885 * check here not to call set_pageblock_migratetype() against
3886 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003887 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07003888 if ((z->zone_start_pfn <= pfn)
3889 && (pfn < z->zone_start_pfn + z->spanned_pages)
3890 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07003891 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003892
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893 INIT_LIST_HEAD(&page->lru);
3894#ifdef WANT_PAGE_VIRTUAL
3895 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3896 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07003897 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003898#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899 }
3900}
3901
Andi Kleen1e548de2008-02-04 22:29:26 -08003902static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07003904 int order, t;
3905 for_each_migratetype_order(order, t) {
3906 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003907 zone->free_area[order].nr_free = 0;
3908 }
3909}
3910
3911#ifndef __HAVE_ARCH_MEMMAP_INIT
3912#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa02007-01-10 23:15:30 -08003913 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003914#endif
3915
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08003916static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003917{
David Howells3a6be872009-05-06 16:03:03 -07003918#ifdef CONFIG_MMU
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003919 int batch;
3920
3921 /*
3922 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07003923 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003924 *
3925 * OK, so we don't know how big the cache is. So guess.
3926 */
3927 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07003928 if (batch * PAGE_SIZE > 512 * 1024)
3929 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003930 batch /= 4; /* We effectively *= 4 below */
3931 if (batch < 1)
3932 batch = 1;
3933
3934 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003935 * Clamp the batch to a 2^n - 1 value. Having a power
3936 * of 2 value was found to be more likely to have
3937 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003938 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11003939 * For example if 2 tasks are alternately allocating
3940 * batches of pages, one task can end up with a lot
3941 * of pages of one half of the possible page colors
3942 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003943 */
David Howells91552032009-05-06 16:03:02 -07003944 batch = rounddown_pow_of_two(batch + batch/2) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07003945
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003946 return batch;
David Howells3a6be872009-05-06 16:03:03 -07003947
3948#else
3949 /* The deferral and batching of frees should be suppressed under NOMMU
3950 * conditions.
3951 *
3952 * The problem is that NOMMU needs to be able to allocate large chunks
3953 * of contiguous memory as there's no hardware page translation to
3954 * assemble apparent contiguous memory from discontiguous pages.
3955 *
3956 * Queueing large contiguous runs of pages for batching, however,
3957 * causes the pages to actually be freed in smaller chunks. As there
3958 * can be a significant delay between the individual batches being
3959 * recycled, this leads to the once large chunks of space being
3960 * fragmented and becoming unavailable for high-order allocations.
3961 */
3962 return 0;
3963#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07003964}
3965
Adrian Bunkb69a7282008-07-23 21:28:12 -07003966static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07003967{
3968 struct per_cpu_pages *pcp;
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003969 int migratetype;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003970
Magnus Damm1c6fe942005-10-26 01:58:59 -07003971 memset(p, 0, sizeof(*p));
3972
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003973 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003974 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07003975 pcp->high = 6 * batch;
3976 pcp->batch = max(1UL, 1 * batch);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07003977 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3978 INIT_LIST_HEAD(&pcp->lists[migratetype]);
Christoph Lameter2caaad42005-06-21 17:15:00 -07003979}
3980
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003981/*
3982 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3983 * to the value high for the pageset p.
3984 */
3985
3986static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3987 unsigned long high)
3988{
3989 struct per_cpu_pages *pcp;
3990
Christoph Lameter3dfa5722008-02-04 22:29:19 -08003991 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003992 pcp->high = high;
3993 pcp->batch = max(1UL, high/4);
3994 if ((high/4) > (PAGE_SHIFT * 8))
3995 pcp->batch = PAGE_SHIFT * 8;
3996}
3997
Nikanth Karthikesan58c2ee42011-03-15 10:59:02 +05303998static void setup_zone_pageset(struct zone *zone)
Wu Fengguang319774e2010-05-24 14:32:49 -07003999{
4000 int cpu;
4001
4002 zone->pageset = alloc_percpu(struct per_cpu_pageset);
4003
4004 for_each_possible_cpu(cpu) {
4005 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4006
4007 setup_pageset(pcp, zone_batchsize(zone));
4008
4009 if (percpu_pagelist_fraction)
4010 setup_pagelist_highmark(pcp,
4011 (zone->present_pages /
4012 percpu_pagelist_fraction));
4013 }
4014}
4015
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004016/*
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004017 * Allocate per cpu pagesets and initialize them.
4018 * Before this call only boot pagesets were available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07004019 */
Al Viro78d99552005-12-15 09:18:25 +00004020void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004021{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004022 struct zone *zone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004023
Wu Fengguang319774e2010-05-24 14:32:49 -07004024 for_each_populated_zone(zone)
4025 setup_zone_pageset(zone);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07004026}
4027
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004028static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07004029int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07004030{
4031 int i;
4032 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004033 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07004034
4035 /*
4036 * The per-page waitqueue mechanism uses hashed waitqueues
4037 * per zone.
4038 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07004039 zone->wait_table_hash_nr_entries =
4040 wait_table_hash_nr_entries(zone_size_pages);
4041 zone->wait_table_bits =
4042 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004043 alloc_size = zone->wait_table_hash_nr_entries
4044 * sizeof(wait_queue_head_t);
4045
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07004046 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07004047 zone->wait_table = (wait_queue_head_t *)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004048 alloc_bootmem_node_nopanic(pgdat, alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004049 } else {
4050 /*
4051 * This case means that a zone whose size was 0 gets new memory
4052 * via memory hot-add.
4053 * But it may be the case that a new node was hot-added. In
4054 * this case vmalloc() will not be able to use this new node's
4055 * memory - this wait_table must be initialized to use this new
4056 * node itself as well.
4057 * To use this new node's memory, further consideration will be
4058 * necessary.
4059 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07004060 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004061 }
4062 if (!zone->wait_table)
4063 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07004064
Yasunori Goto02b694d2006-06-23 02:03:08 -07004065 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07004066 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07004067
4068 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004069}
4070
Shaohua Li112067f2009-09-21 17:01:16 -07004071static int __zone_pcp_update(void *data)
4072{
4073 struct zone *zone = data;
4074 int cpu;
4075 unsigned long batch = zone_batchsize(zone), flags;
4076
Thomas Gleixner2d30a1f2010-03-10 15:20:40 -08004077 for_each_possible_cpu(cpu) {
Shaohua Li112067f2009-09-21 17:01:16 -07004078 struct per_cpu_pageset *pset;
4079 struct per_cpu_pages *pcp;
4080
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004081 pset = per_cpu_ptr(zone->pageset, cpu);
Shaohua Li112067f2009-09-21 17:01:16 -07004082 pcp = &pset->pcp;
4083
4084 local_irq_save(flags);
Mel Gorman5f8dcc22009-09-21 17:03:19 -07004085 free_pcppages_bulk(zone, pcp->count, pcp);
Shaohua Li112067f2009-09-21 17:01:16 -07004086 setup_pageset(pset, batch);
4087 local_irq_restore(flags);
4088 }
4089 return 0;
4090}
4091
4092void zone_pcp_update(struct zone *zone)
4093{
4094 stop_machine(__zone_pcp_update, zone, NULL);
4095}
4096
Matt Tolentinoc09b4242006-01-17 07:03:44 +01004097static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07004098{
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004099 /*
4100 * per cpu subsystem is not up at this point. The following code
4101 * relies on the ability of the linker to provide the
4102 * offset of a (static) per cpu variable into the per cpu area.
4103 */
4104 zone->pageset = &boot_pageset;
Dave Hansened8ece22005-10-29 18:16:50 -07004105
Anton Blanchardf5335c02006-03-25 03:06:49 -08004106 if (zone->present_pages)
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09004107 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4108 zone->name, zone->present_pages,
4109 zone_batchsize(zone));
Dave Hansened8ece22005-10-29 18:16:50 -07004110}
4111
Yasunori Goto718127c2006-06-23 02:03:10 -07004112__meminit int init_currently_empty_zone(struct zone *zone,
4113 unsigned long zone_start_pfn,
Dave Hansena2f3aa02007-01-10 23:15:30 -08004114 unsigned long size,
4115 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07004116{
4117 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07004118 int ret;
4119 ret = zone_wait_table_init(zone, size);
4120 if (ret)
4121 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07004122 pgdat->nr_zones = zone_idx(zone) + 1;
4123
Dave Hansened8ece22005-10-29 18:16:50 -07004124 zone->zone_start_pfn = zone_start_pfn;
4125
Mel Gorman708614e2008-07-23 21:26:51 -07004126 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4127 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4128 pgdat->node_id,
4129 (unsigned long)zone_idx(zone),
4130 zone_start_pfn, (zone_start_pfn + size));
4131
Andi Kleen1e548de2008-02-04 22:29:26 -08004132 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07004133
4134 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07004135}
4136
Tejun Heo0ee332c2011-12-08 10:22:09 -08004137#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004138#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4139/*
4140 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4141 * Architectures may implement their own version but if add_active_range()
4142 * was used and there are no special requirements, this is a convenient
4143 * alternative
4144 */
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004145int __meminit __early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004146{
Tejun Heoc13291a2011-07-12 10:46:30 +02004147 unsigned long start_pfn, end_pfn;
4148 int i, nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004149
Tejun Heoc13291a2011-07-12 10:46:30 +02004150 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004151 if (start_pfn <= pfn && pfn < end_pfn)
Tejun Heoc13291a2011-07-12 10:46:30 +02004152 return nid;
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004153 /* This is a memory hole */
4154 return -1;
Mel Gormanc7132162006-09-27 01:49:43 -07004155}
4156#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4157
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004158int __meminit early_pfn_to_nid(unsigned long pfn)
4159{
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004160 int nid;
4161
4162 nid = __early_pfn_to_nid(pfn);
4163 if (nid >= 0)
4164 return nid;
4165 /* just returns 0 */
4166 return 0;
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004167}
4168
KAMEZAWA Hiroyukicc2559b2009-02-18 14:48:33 -08004169#ifdef CONFIG_NODES_SPAN_OTHER_NODES
4170bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4171{
4172 int nid;
4173
4174 nid = __early_pfn_to_nid(pfn);
4175 if (nid >= 0 && nid != node)
4176 return false;
4177 return true;
4178}
4179#endif
KAMEZAWA Hiroyukif2dbcfa2009-02-18 14:48:32 -08004180
Mel Gormanc7132162006-09-27 01:49:43 -07004181/**
4182 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004183 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4184 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07004185 *
4186 * If an architecture guarantees that all ranges registered with
4187 * add_active_ranges() contain no holes and may be freed, this
4188 * this function may be used instead of calling free_bootmem() manually.
4189 */
Tejun Heoc13291a2011-07-12 10:46:30 +02004190void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07004191{
Tejun Heoc13291a2011-07-12 10:46:30 +02004192 unsigned long start_pfn, end_pfn;
4193 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004194
Tejun Heoc13291a2011-07-12 10:46:30 +02004195 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4196 start_pfn = min(start_pfn, max_low_pfn);
4197 end_pfn = min(end_pfn, max_low_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004198
Tejun Heoc13291a2011-07-12 10:46:30 +02004199 if (start_pfn < end_pfn)
4200 free_bootmem_node(NODE_DATA(this_nid),
4201 PFN_PHYS(start_pfn),
4202 (end_pfn - start_pfn) << PAGE_SHIFT);
Mel Gormanc7132162006-09-27 01:49:43 -07004203 }
4204}
4205
4206/**
4207 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004208 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07004209 *
4210 * If an architecture guarantees that all ranges registered with
4211 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004212 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07004213 */
4214void __init sparse_memory_present_with_active_regions(int nid)
4215{
Tejun Heoc13291a2011-07-12 10:46:30 +02004216 unsigned long start_pfn, end_pfn;
4217 int i, this_nid;
Mel Gormanc7132162006-09-27 01:49:43 -07004218
Tejun Heoc13291a2011-07-12 10:46:30 +02004219 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4220 memory_present(this_nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004221}
4222
4223/**
4224 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004225 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4226 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4227 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07004228 *
4229 * It returns the start and end page frame of a node based on information
4230 * provided by an arch calling add_active_range(). If called for a node
4231 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004232 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07004233 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004234void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004235 unsigned long *start_pfn, unsigned long *end_pfn)
4236{
Tejun Heoc13291a2011-07-12 10:46:30 +02004237 unsigned long this_start_pfn, this_end_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004238 int i;
Tejun Heoc13291a2011-07-12 10:46:30 +02004239
Mel Gormanc7132162006-09-27 01:49:43 -07004240 *start_pfn = -1UL;
4241 *end_pfn = 0;
4242
Tejun Heoc13291a2011-07-12 10:46:30 +02004243 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4244 *start_pfn = min(*start_pfn, this_start_pfn);
4245 *end_pfn = max(*end_pfn, this_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004246 }
4247
Christoph Lameter633c0662007-10-16 01:25:37 -07004248 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07004249 *start_pfn = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07004250}
4251
4252/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07004253 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4254 * assumption is made that zones within a node are ordered in monotonic
4255 * increasing memory addresses so that the "highest" populated zone is used
4256 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004257static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004258{
4259 int zone_index;
4260 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4261 if (zone_index == ZONE_MOVABLE)
4262 continue;
4263
4264 if (arch_zone_highest_possible_pfn[zone_index] >
4265 arch_zone_lowest_possible_pfn[zone_index])
4266 break;
4267 }
4268
4269 VM_BUG_ON(zone_index == -1);
4270 movable_zone = zone_index;
4271}
4272
4273/*
4274 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004275 * because it is sized independent of architecture. Unlike the other zones,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004276 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4277 * in each node depending on the size of each node and how evenly kernelcore
4278 * is distributed. This helper function adjusts the zone ranges
4279 * provided by the architecture for a given node by using the end of the
4280 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4281 * zones within a node are in order of monotonic increases memory addresses
4282 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004283static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07004284 unsigned long zone_type,
4285 unsigned long node_start_pfn,
4286 unsigned long node_end_pfn,
4287 unsigned long *zone_start_pfn,
4288 unsigned long *zone_end_pfn)
4289{
4290 /* Only adjust if ZONE_MOVABLE is on this node */
4291 if (zone_movable_pfn[nid]) {
4292 /* Size ZONE_MOVABLE */
4293 if (zone_type == ZONE_MOVABLE) {
4294 *zone_start_pfn = zone_movable_pfn[nid];
4295 *zone_end_pfn = min(node_end_pfn,
4296 arch_zone_highest_possible_pfn[movable_zone]);
4297
4298 /* Adjust for ZONE_MOVABLE starting within this range */
4299 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4300 *zone_end_pfn > zone_movable_pfn[nid]) {
4301 *zone_end_pfn = zone_movable_pfn[nid];
4302
4303 /* Check if this whole range is within ZONE_MOVABLE */
4304 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4305 *zone_start_pfn = *zone_end_pfn;
4306 }
4307}
4308
4309/*
Mel Gormanc7132162006-09-27 01:49:43 -07004310 * Return the number of pages a zone spans in a node, including holes
4311 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4312 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004313static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004314 unsigned long zone_type,
4315 unsigned long *ignored)
4316{
4317 unsigned long node_start_pfn, node_end_pfn;
4318 unsigned long zone_start_pfn, zone_end_pfn;
4319
4320 /* Get the start and end of the node and zone */
4321 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4322 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4323 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07004324 adjust_zone_range_for_zone_movable(nid, zone_type,
4325 node_start_pfn, node_end_pfn,
4326 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004327
4328 /* Check that this node has pages within the zone's required range */
4329 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4330 return 0;
4331
4332 /* Move the zone boundaries inside the node if necessary */
4333 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4334 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4335
4336 /* Return the spanned pages */
4337 return zone_end_pfn - zone_start_pfn;
4338}
4339
4340/*
4341 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004342 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07004343 */
Yinghai Lu32996252009-12-15 17:59:02 -08004344unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004345 unsigned long range_start_pfn,
4346 unsigned long range_end_pfn)
4347{
Tejun Heo96e907d2011-07-12 10:46:29 +02004348 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4349 unsigned long start_pfn, end_pfn;
4350 int i;
Mel Gormanc7132162006-09-27 01:49:43 -07004351
Tejun Heo96e907d2011-07-12 10:46:29 +02004352 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4353 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4354 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4355 nr_absent -= end_pfn - start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004356 }
Tejun Heo96e907d2011-07-12 10:46:29 +02004357 return nr_absent;
Mel Gormanc7132162006-09-27 01:49:43 -07004358}
4359
4360/**
4361 * absent_pages_in_range - Return number of page frames in holes within a range
4362 * @start_pfn: The start PFN to start searching for holes
4363 * @end_pfn: The end PFN to stop searching for holes
4364 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004365 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07004366 */
4367unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4368 unsigned long end_pfn)
4369{
4370 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4371}
4372
4373/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004374static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004375 unsigned long zone_type,
4376 unsigned long *ignored)
4377{
Tejun Heo96e907d2011-07-12 10:46:29 +02004378 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4379 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman9c7cd682006-09-27 01:49:58 -07004380 unsigned long node_start_pfn, node_end_pfn;
4381 unsigned long zone_start_pfn, zone_end_pfn;
4382
4383 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
Tejun Heo96e907d2011-07-12 10:46:29 +02004384 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4385 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004386
Mel Gorman2a1e2742007-07-17 04:03:12 -07004387 adjust_zone_range_for_zone_movable(nid, zone_type,
4388 node_start_pfn, node_end_pfn,
4389 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07004390 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004391}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004392
Tejun Heo0ee332c2011-12-08 10:22:09 -08004393#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Paul Mundt6ea6e682007-07-15 23:38:20 -07004394static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004395 unsigned long zone_type,
4396 unsigned long *zones_size)
4397{
4398 return zones_size[zone_type];
4399}
4400
Paul Mundt6ea6e682007-07-15 23:38:20 -07004401static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07004402 unsigned long zone_type,
4403 unsigned long *zholes_size)
4404{
4405 if (!zholes_size)
4406 return 0;
4407
4408 return zholes_size[zone_type];
4409}
Mel Gorman0e0b8642006-09-27 01:49:56 -07004410
Tejun Heo0ee332c2011-12-08 10:22:09 -08004411#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004412
Yasunori Gotoa3142c82007-05-08 00:23:07 -07004413static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07004414 unsigned long *zones_size, unsigned long *zholes_size)
4415{
4416 unsigned long realtotalpages, totalpages = 0;
4417 enum zone_type i;
4418
4419 for (i = 0; i < MAX_NR_ZONES; i++)
4420 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4421 zones_size);
4422 pgdat->node_spanned_pages = totalpages;
4423
4424 realtotalpages = totalpages;
4425 for (i = 0; i < MAX_NR_ZONES; i++)
4426 realtotalpages -=
4427 zone_absent_pages_in_node(pgdat->node_id, i,
4428 zholes_size);
4429 pgdat->node_present_pages = realtotalpages;
4430 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4431 realtotalpages);
4432}
4433
Mel Gorman835c1342007-10-16 01:25:47 -07004434#ifndef CONFIG_SPARSEMEM
4435/*
4436 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07004437 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4438 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07004439 * round what is now in bits to nearest long in bits, then return it in
4440 * bytes.
4441 */
4442static unsigned long __init usemap_size(unsigned long zonesize)
4443{
4444 unsigned long usemapsize;
4445
Mel Gormand9c23402007-10-16 01:26:01 -07004446 usemapsize = roundup(zonesize, pageblock_nr_pages);
4447 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07004448 usemapsize *= NR_PAGEBLOCK_BITS;
4449 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4450
4451 return usemapsize / 8;
4452}
4453
4454static void __init setup_usemap(struct pglist_data *pgdat,
4455 struct zone *zone, unsigned long zonesize)
4456{
4457 unsigned long usemapsize = usemap_size(zonesize);
4458 zone->pageblock_flags = NULL;
Julia Lawall58a01a42009-01-06 14:39:28 -08004459 if (usemapsize)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004460 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4461 usemapsize);
Mel Gorman835c1342007-10-16 01:25:47 -07004462}
4463#else
Jesper Juhlfa9f90b2010-11-28 21:39:34 +01004464static inline void setup_usemap(struct pglist_data *pgdat,
Mel Gorman835c1342007-10-16 01:25:47 -07004465 struct zone *zone, unsigned long zonesize) {}
4466#endif /* CONFIG_SPARSEMEM */
4467
Mel Gormand9c23402007-10-16 01:26:01 -07004468#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08004469
4470/* Return a sensible default order for the pageblock size. */
4471static inline int pageblock_default_order(void)
4472{
4473 if (HPAGE_SHIFT > PAGE_SHIFT)
4474 return HUGETLB_PAGE_ORDER;
4475
4476 return MAX_ORDER-1;
4477}
4478
Mel Gormand9c23402007-10-16 01:26:01 -07004479/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4480static inline void __init set_pageblock_order(unsigned int order)
4481{
4482 /* Check that pageblock_nr_pages has not already been setup */
4483 if (pageblock_order)
4484 return;
4485
4486 /*
4487 * Assume the largest contiguous order of interest is a huge page.
4488 * This value may be variable depending on boot parameters on IA64
4489 */
4490 pageblock_order = order;
4491}
4492#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4493
Mel Gormanba72cb82007-11-28 16:21:13 -08004494/*
4495 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4496 * and pageblock_default_order() are unused as pageblock_order is set
4497 * at compile-time. See include/linux/pageblock-flags.h for the values of
4498 * pageblock_order based on the kernel config
4499 */
4500static inline int pageblock_default_order(unsigned int order)
4501{
4502 return MAX_ORDER-1;
4503}
Mel Gormand9c23402007-10-16 01:26:01 -07004504#define set_pageblock_order(x) do {} while (0)
4505
4506#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4507
Linus Torvalds1da177e2005-04-16 15:20:36 -07004508/*
4509 * Set up the zone data structures:
4510 * - mark all pages reserved
4511 * - mark all memory queues empty
4512 * - clear the memory bitmaps
4513 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08004514static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004515 unsigned long *zones_size, unsigned long *zholes_size)
4516{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07004517 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07004518 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004519 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07004520 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004521
Dave Hansen208d54e2005-10-29 18:16:52 -07004522 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004523 pgdat->nr_zones = 0;
4524 init_waitqueue_head(&pgdat->kswapd_wait);
4525 pgdat->kswapd_max_order = 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004526 pgdat_page_cgroup_init(pgdat);
Michal Nazarewiczadc95fd2012-01-11 15:16:11 +01004527
Linus Torvalds1da177e2005-04-16 15:20:36 -07004528 for (j = 0; j < MAX_NR_ZONES; j++) {
4529 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004530 unsigned long size, realsize, memmap_pages;
Hugh Dickins41113042012-01-12 17:20:01 -08004531 enum lru_list lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004532
Mel Gormanc7132162006-09-27 01:49:43 -07004533 size = zone_spanned_pages_in_node(nid, j, zones_size);
4534 realsize = size - zone_absent_pages_in_node(nid, j,
4535 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004536
Mel Gorman0e0b8642006-09-27 01:49:56 -07004537 /*
4538 * Adjust realsize so that it accounts for how much memory
4539 * is used by this zone for memmap. This affects the watermark
4540 * and per-cpu initialisations
4541 */
Johannes Weinerf7232152008-05-23 13:04:21 -07004542 memmap_pages =
4543 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07004544 if (realsize >= memmap_pages) {
4545 realsize -= memmap_pages;
Yinghai Lu5594c8c2009-01-06 14:39:14 -08004546 if (memmap_pages)
4547 printk(KERN_DEBUG
4548 " %s zone: %lu pages used for memmap\n",
4549 zone_names[j], memmap_pages);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004550 } else
4551 printk(KERN_WARNING
4552 " %s zone: %lu pages exceeds realsize %lu\n",
4553 zone_names[j], memmap_pages, realsize);
4554
Christoph Lameter62672762007-02-10 01:43:07 -08004555 /* Account for reserved pages */
4556 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07004557 realsize -= dma_reserve;
Yinghai Lud903ef92008-10-18 20:27:06 -07004558 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08004559 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07004560 }
4561
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07004562 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004563 nr_kernel_pages += realsize;
4564 nr_all_pages += realsize;
4565
4566 zone->spanned_pages = size;
4567 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07004568#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07004569 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07004570 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07004571 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07004572 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07004573#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004574 zone->name = zone_names[j];
4575 spin_lock_init(&zone->lock);
4576 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07004577 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004579
Dave Hansened8ece22005-10-29 18:16:50 -07004580 zone_pcp_init(zone);
Hugh Dickins41113042012-01-12 17:20:01 -08004581 for_each_lru(lru)
4582 INIT_LIST_HEAD(&zone->lruvec.lists[lru]);
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08004583 zone->reclaim_stat.recent_rotated[0] = 0;
4584 zone->reclaim_stat.recent_rotated[1] = 0;
4585 zone->reclaim_stat.recent_scanned[0] = 0;
4586 zone->reclaim_stat.recent_scanned[1] = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07004587 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07004588 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004589 if (!size)
4590 continue;
4591
Mel Gormanba72cb82007-11-28 16:21:13 -08004592 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07004593 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa02007-01-10 23:15:30 -08004594 ret = init_currently_empty_zone(zone, zone_start_pfn,
4595 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07004596 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07004597 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004598 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004599 }
4600}
4601
Sam Ravnborg577a32f2007-05-17 23:29:25 +02004602static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004603{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004604 /* Skip empty nodes */
4605 if (!pgdat->node_spanned_pages)
4606 return;
4607
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004608#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07004609 /* ia64 gets its own node_mem_map, before this, without bootmem */
4610 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07004611 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004612 struct page *map;
4613
Bob Piccoe984bb42006-05-20 15:00:31 -07004614 /*
4615 * The zone's endpoints aren't required to be MAX_ORDER
4616 * aligned but the node_mem_map endpoints must be in order
4617 * for the buddy allocator to function correctly.
4618 */
4619 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4620 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4621 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4622 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07004623 map = alloc_remap(pgdat->node_id, size);
4624 if (!map)
Yinghai Lu8f389a992011-05-11 15:13:32 -07004625 map = alloc_bootmem_node_nopanic(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07004626 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004627 }
Roman Zippel12d810c2007-05-31 00:40:54 -07004628#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004629 /*
4630 * With no DISCONTIG, the global mem_map is just set as node 0's
4631 */
Mel Gormanc7132162006-09-27 01:49:43 -07004632 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004633 mem_map = NODE_DATA(0)->node_mem_map;
Tejun Heo0ee332c2011-12-08 10:22:09 -08004634#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Mel Gormanc7132162006-09-27 01:49:43 -07004635 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08004636 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Tejun Heo0ee332c2011-12-08 10:22:09 -08004637#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07004638 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004639#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004640#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004641}
4642
Johannes Weiner9109fb72008-07-23 21:27:20 -07004643void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4644 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004645{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004646 pg_data_t *pgdat = NODE_DATA(nid);
4647
Linus Torvalds1da177e2005-04-16 15:20:36 -07004648 pgdat->node_id = nid;
4649 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004650 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004651
4652 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07004653#ifdef CONFIG_FLAT_NODE_MEM_MAP
4654 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4655 nid, (unsigned long)pgdat,
4656 (unsigned long)pgdat->node_mem_map);
4657#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004658
4659 free_area_init_core(pgdat, zones_size, zholes_size);
4660}
4661
Tejun Heo0ee332c2011-12-08 10:22:09 -08004662#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07004663
4664#if MAX_NUMNODES > 1
4665/*
4666 * Figure out the number of possible node ids.
4667 */
4668static void __init setup_nr_node_ids(void)
4669{
4670 unsigned int node;
4671 unsigned int highest = 0;
4672
4673 for_each_node_mask(node, node_possible_map)
4674 highest = node;
4675 nr_node_ids = highest + 1;
4676}
4677#else
4678static inline void setup_nr_node_ids(void)
4679{
4680}
4681#endif
4682
Mel Gormanc7132162006-09-27 01:49:43 -07004683/**
Tejun Heo1e019792011-07-12 09:45:34 +02004684 * node_map_pfn_alignment - determine the maximum internode alignment
4685 *
4686 * This function should be called after node map is populated and sorted.
4687 * It calculates the maximum power of two alignment which can distinguish
4688 * all the nodes.
4689 *
4690 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4691 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4692 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4693 * shifted, 1GiB is enough and this function will indicate so.
4694 *
4695 * This is used to test whether pfn -> nid mapping of the chosen memory
4696 * model has fine enough granularity to avoid incorrect mapping for the
4697 * populated node map.
4698 *
4699 * Returns the determined alignment in pfn's. 0 if there is no alignment
4700 * requirement (single node).
4701 */
4702unsigned long __init node_map_pfn_alignment(void)
4703{
4704 unsigned long accl_mask = 0, last_end = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02004705 unsigned long start, end, mask;
Tejun Heo1e019792011-07-12 09:45:34 +02004706 int last_nid = -1;
Tejun Heoc13291a2011-07-12 10:46:30 +02004707 int i, nid;
Tejun Heo1e019792011-07-12 09:45:34 +02004708
Tejun Heoc13291a2011-07-12 10:46:30 +02004709 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
Tejun Heo1e019792011-07-12 09:45:34 +02004710 if (!start || last_nid < 0 || last_nid == nid) {
4711 last_nid = nid;
4712 last_end = end;
4713 continue;
4714 }
4715
4716 /*
4717 * Start with a mask granular enough to pin-point to the
4718 * start pfn and tick off bits one-by-one until it becomes
4719 * too coarse to separate the current node from the last.
4720 */
4721 mask = ~((1 << __ffs(start)) - 1);
4722 while (mask && last_end <= (start & (mask << 1)))
4723 mask <<= 1;
4724
4725 /* accumulate all internode masks */
4726 accl_mask |= mask;
4727 }
4728
4729 /* convert mask to number of pages */
4730 return ~accl_mask + 1;
4731}
4732
Mel Gormana6af2bc2007-02-10 01:42:57 -08004733/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07004734static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07004735{
Mel Gormana6af2bc2007-02-10 01:42:57 -08004736 unsigned long min_pfn = ULONG_MAX;
Tejun Heoc13291a2011-07-12 10:46:30 +02004737 unsigned long start_pfn;
4738 int i;
Mel Gorman1abbfb42006-11-23 12:01:41 +00004739
Tejun Heoc13291a2011-07-12 10:46:30 +02004740 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
4741 min_pfn = min(min_pfn, start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07004742
Mel Gormana6af2bc2007-02-10 01:42:57 -08004743 if (min_pfn == ULONG_MAX) {
4744 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07004745 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08004746 return 0;
4747 }
4748
4749 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07004750}
4751
4752/**
4753 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4754 *
4755 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004756 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07004757 */
4758unsigned long __init find_min_pfn_with_active_regions(void)
4759{
4760 return find_min_pfn_for_node(MAX_NUMNODES);
4761}
4762
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004763/*
4764 * early_calculate_totalpages()
4765 * Sum pages in active regions for movable zone.
4766 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4767 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07004768static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07004769{
Mel Gorman7e63efe2007-07-17 04:03:15 -07004770 unsigned long totalpages = 0;
Tejun Heoc13291a2011-07-12 10:46:30 +02004771 unsigned long start_pfn, end_pfn;
4772 int i, nid;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004773
Tejun Heoc13291a2011-07-12 10:46:30 +02004774 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
4775 unsigned long pages = end_pfn - start_pfn;
4776
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004777 totalpages += pages;
4778 if (pages)
Tejun Heoc13291a2011-07-12 10:46:30 +02004779 node_set_state(nid, N_HIGH_MEMORY);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004780 }
4781 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07004782}
4783
Mel Gorman2a1e2742007-07-17 04:03:12 -07004784/*
4785 * Find the PFN the Movable zone begins in each node. Kernel memory
4786 * is spread evenly between nodes as long as the nodes have enough
4787 * memory. When they don't, some nodes will have more kernelcore than
4788 * others
4789 */
Kautuk Consulb224ef82012-03-21 16:34:15 -07004790static void __init find_zone_movable_pfns_for_nodes(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004791{
4792 int i, nid;
4793 unsigned long usable_startpfn;
4794 unsigned long kernelcore_node, kernelcore_remaining;
Yinghai Lu66918dc2009-06-30 11:41:37 -07004795 /* save the state before borrow the nodemask */
4796 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004797 unsigned long totalpages = early_calculate_totalpages();
4798 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004799
Larry Bassele6436862011-10-14 10:59:07 -07004800#ifdef CONFIG_FIX_MOVABLE_ZONE
4801 required_movablecore = movable_reserved_size >> PAGE_SHIFT;
4802#endif
Mel Gorman7e63efe2007-07-17 04:03:15 -07004803 /*
4804 * If movablecore was specified, calculate what size of
4805 * kernelcore that corresponds so that memory usable for
4806 * any allocation type is evenly spread. If both kernelcore
4807 * and movablecore are specified, then the value of kernelcore
4808 * will be used for required_kernelcore if it's greater than
4809 * what movablecore would have allowed.
4810 */
4811 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07004812 unsigned long corepages;
4813
4814 /*
4815 * Round-up so that ZONE_MOVABLE is at least as large as what
4816 * was requested by the user
4817 */
4818 required_movablecore =
4819 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4820 corepages = totalpages - required_movablecore;
4821
4822 required_kernelcore = max(required_kernelcore, corepages);
4823 }
4824
Mel Gorman2a1e2742007-07-17 04:03:12 -07004825 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4826 if (!required_kernelcore)
Yinghai Lu66918dc2009-06-30 11:41:37 -07004827 goto out;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004828
4829 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4830 find_usable_zone_for_movable();
4831 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4832
4833restart:
4834 /* Spread kernelcore memory as evenly as possible throughout nodes */
4835 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004836 for_each_node_state(nid, N_HIGH_MEMORY) {
Tejun Heoc13291a2011-07-12 10:46:30 +02004837 unsigned long start_pfn, end_pfn;
4838
Mel Gorman2a1e2742007-07-17 04:03:12 -07004839 /*
4840 * Recalculate kernelcore_node if the division per node
4841 * now exceeds what is necessary to satisfy the requested
4842 * amount of memory for the kernel
4843 */
4844 if (required_kernelcore < kernelcore_node)
4845 kernelcore_node = required_kernelcore / usable_nodes;
4846
4847 /*
4848 * As the map is walked, we track how much memory is usable
4849 * by the kernel using kernelcore_remaining. When it is
4850 * 0, the rest of the node is usable by ZONE_MOVABLE
4851 */
4852 kernelcore_remaining = kernelcore_node;
4853
4854 /* Go through each range of PFNs within this node */
Tejun Heoc13291a2011-07-12 10:46:30 +02004855 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004856 unsigned long size_pages;
4857
Tejun Heoc13291a2011-07-12 10:46:30 +02004858 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07004859 if (start_pfn >= end_pfn)
4860 continue;
4861
4862 /* Account for what is only usable for kernelcore */
4863 if (start_pfn < usable_startpfn) {
4864 unsigned long kernel_pages;
4865 kernel_pages = min(end_pfn, usable_startpfn)
4866 - start_pfn;
4867
4868 kernelcore_remaining -= min(kernel_pages,
4869 kernelcore_remaining);
4870 required_kernelcore -= min(kernel_pages,
4871 required_kernelcore);
4872
4873 /* Continue if range is now fully accounted */
4874 if (end_pfn <= usable_startpfn) {
4875
4876 /*
4877 * Push zone_movable_pfn to the end so
4878 * that if we have to rebalance
4879 * kernelcore across nodes, we will
4880 * not double account here
4881 */
4882 zone_movable_pfn[nid] = end_pfn;
4883 continue;
4884 }
4885 start_pfn = usable_startpfn;
4886 }
4887
4888 /*
4889 * The usable PFN range for ZONE_MOVABLE is from
4890 * start_pfn->end_pfn. Calculate size_pages as the
4891 * number of pages used as kernelcore
4892 */
4893 size_pages = end_pfn - start_pfn;
4894 if (size_pages > kernelcore_remaining)
4895 size_pages = kernelcore_remaining;
4896 zone_movable_pfn[nid] = start_pfn + size_pages;
4897
4898 /*
4899 * Some kernelcore has been met, update counts and
4900 * break if the kernelcore for this node has been
4901 * satisified
4902 */
4903 required_kernelcore -= min(required_kernelcore,
4904 size_pages);
4905 kernelcore_remaining -= size_pages;
4906 if (!kernelcore_remaining)
4907 break;
4908 }
4909 }
4910
4911 /*
4912 * If there is still required_kernelcore, we do another pass with one
4913 * less node in the count. This will push zone_movable_pfn[nid] further
4914 * along on the nodes that still have memory until kernelcore is
4915 * satisified
4916 */
4917 usable_nodes--;
4918 if (usable_nodes && required_kernelcore > usable_nodes)
4919 goto restart;
4920
4921 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4922 for (nid = 0; nid < MAX_NUMNODES; nid++)
4923 zone_movable_pfn[nid] =
4924 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
Yinghai Lu66918dc2009-06-30 11:41:37 -07004925
4926out:
4927 /* restore the node_state */
4928 node_states[N_HIGH_MEMORY] = saved_node_state;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004929}
4930
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004931/* Any regular memory on that node ? */
4932static void check_for_regular_memory(pg_data_t *pgdat)
4933{
4934#ifdef CONFIG_HIGHMEM
4935 enum zone_type zone_type;
4936
4937 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4938 struct zone *zone = &pgdat->node_zones[zone_type];
Bob Liud0048b02012-01-12 17:19:07 -08004939 if (zone->present_pages) {
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004940 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
Bob Liud0048b02012-01-12 17:19:07 -08004941 break;
4942 }
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004943 }
4944#endif
4945}
4946
Mel Gormanc7132162006-09-27 01:49:43 -07004947/**
4948 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004949 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07004950 *
4951 * This will call free_area_init_node() for each active node in the system.
4952 * Using the page ranges provided by add_active_range(), the size of each
4953 * zone in each node and their holes is calculated. If the maximum PFN
4954 * between two adjacent zones match, it is assumed that the zone is empty.
4955 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4956 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4957 * starts where the previous one ended. For example, ZONE_DMA32 starts
4958 * at arch_max_dma_pfn.
4959 */
4960void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4961{
Tejun Heoc13291a2011-07-12 10:46:30 +02004962 unsigned long start_pfn, end_pfn;
4963 int i, nid;
Mel Gormana6af2bc2007-02-10 01:42:57 -08004964
Mel Gormanc7132162006-09-27 01:49:43 -07004965 /* Record where the zone boundaries are */
4966 memset(arch_zone_lowest_possible_pfn, 0,
4967 sizeof(arch_zone_lowest_possible_pfn));
4968 memset(arch_zone_highest_possible_pfn, 0,
4969 sizeof(arch_zone_highest_possible_pfn));
4970 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4971 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4972 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07004973 if (i == ZONE_MOVABLE)
4974 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07004975 arch_zone_lowest_possible_pfn[i] =
4976 arch_zone_highest_possible_pfn[i-1];
4977 arch_zone_highest_possible_pfn[i] =
4978 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4979 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07004980 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4981 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4982
4983 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4984 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
Kautuk Consulb224ef82012-03-21 16:34:15 -07004985 find_zone_movable_pfns_for_nodes();
Mel Gormanc7132162006-09-27 01:49:43 -07004986
Mel Gormanc7132162006-09-27 01:49:43 -07004987 /* Print out the zone ranges */
4988 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07004989 for (i = 0; i < MAX_NR_ZONES; i++) {
4990 if (i == ZONE_MOVABLE)
4991 continue;
David Rientjes72f0ba02010-03-05 13:42:14 -08004992 printk(" %-8s ", zone_names[i]);
4993 if (arch_zone_lowest_possible_pfn[i] ==
4994 arch_zone_highest_possible_pfn[i])
4995 printk("empty\n");
4996 else
4997 printk("%0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07004998 arch_zone_lowest_possible_pfn[i],
4999 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07005000 }
5001
5002 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5003 printk("Movable zone start PFN for each node\n");
5004 for (i = 0; i < MAX_NUMNODES; i++) {
5005 if (zone_movable_pfn[i])
5006 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
5007 }
Mel Gormanc7132162006-09-27 01:49:43 -07005008
5009 /* Print out the early_node_map[] */
Tejun Heoc13291a2011-07-12 10:46:30 +02005010 printk("Early memory PFN ranges\n");
5011 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
5012 printk(" %3d: %0#10lx -> %0#10lx\n", nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07005013
5014 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07005015 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08005016 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07005017 for_each_online_node(nid) {
5018 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07005019 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07005020 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07005021
5022 /* Any memory on that node */
5023 if (pgdat->node_present_pages)
5024 node_set_state(nid, N_HIGH_MEMORY);
5025 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07005026 }
5027}
Mel Gorman2a1e2742007-07-17 04:03:12 -07005028
Mel Gorman7e63efe2007-07-17 04:03:15 -07005029static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07005030{
5031 unsigned long long coremem;
5032 if (!p)
5033 return -EINVAL;
5034
5035 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005036 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07005037
Mel Gorman7e63efe2007-07-17 04:03:15 -07005038 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07005039 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5040
5041 return 0;
5042}
Mel Gormaned7ed362007-07-17 04:03:14 -07005043
Mel Gorman7e63efe2007-07-17 04:03:15 -07005044/*
5045 * kernelcore=size sets the amount of memory for use for allocations that
5046 * cannot be reclaimed or migrated.
5047 */
5048static int __init cmdline_parse_kernelcore(char *p)
5049{
5050 return cmdline_parse_core(p, &required_kernelcore);
5051}
5052
5053/*
5054 * movablecore=size sets the amount of memory for use for allocations that
5055 * can be reclaimed or migrated.
5056 */
5057static int __init cmdline_parse_movablecore(char *p)
5058{
5059 return cmdline_parse_core(p, &required_movablecore);
5060}
5061
Mel Gormaned7ed362007-07-17 04:03:14 -07005062early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07005063early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07005064
Tejun Heo0ee332c2011-12-08 10:22:09 -08005065#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
Mel Gormanc7132162006-09-27 01:49:43 -07005066
Mel Gorman0e0b8642006-09-27 01:49:56 -07005067/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005068 * set_dma_reserve - set the specified number of pages reserved in the first zone
5069 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07005070 *
5071 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5072 * In the DMA zone, a significant percentage may be consumed by kernel image
5073 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07005074 * function may optionally be used to account for unfreeable pages in the
5075 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5076 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07005077 */
5078void __init set_dma_reserve(unsigned long new_dma_reserve)
5079{
5080 dma_reserve = new_dma_reserve;
5081}
5082
Linus Torvalds1da177e2005-04-16 15:20:36 -07005083void __init free_area_init(unsigned long *zones_size)
5084{
Johannes Weiner9109fb72008-07-23 21:27:20 -07005085 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005086 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5087}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005088
Linus Torvalds1da177e2005-04-16 15:20:36 -07005089static int page_alloc_cpu_notify(struct notifier_block *self,
5090 unsigned long action, void *hcpu)
5091{
5092 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005093
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07005094 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -07005095 lru_add_drain_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005096 drain_pages(cpu);
5097
5098 /*
5099 * Spill the event counters of the dead processor
5100 * into the current processors event counters.
5101 * This artificially elevates the count of the current
5102 * processor.
5103 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07005104 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005105
5106 /*
5107 * Zero the differential counters of the dead processor
5108 * so that the vm statistics are consistent.
5109 *
5110 * This is only okay since the processor is dead and cannot
5111 * race with what we are doing.
5112 */
Christoph Lameter2244b952006-06-30 01:55:33 -07005113 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005114 }
5115 return NOTIFY_OK;
5116}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005117
5118void __init page_alloc_init(void)
5119{
5120 hotcpu_notifier(page_alloc_cpu_notify, 0);
5121}
5122
5123/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005124 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5125 * or min_free_kbytes changes.
5126 */
5127static void calculate_totalreserve_pages(void)
5128{
5129 struct pglist_data *pgdat;
5130 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005131 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005132
5133 for_each_online_pgdat(pgdat) {
5134 for (i = 0; i < MAX_NR_ZONES; i++) {
5135 struct zone *zone = pgdat->node_zones + i;
5136 unsigned long max = 0;
5137
5138 /* Find valid and maximum lowmem_reserve in the zone */
5139 for (j = i; j < MAX_NR_ZONES; j++) {
5140 if (zone->lowmem_reserve[j] > max)
5141 max = zone->lowmem_reserve[j];
5142 }
5143
Mel Gorman41858962009-06-16 15:32:12 -07005144 /* we treat the high watermark as reserved pages. */
5145 max += high_wmark_pages(zone);
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005146
5147 if (max > zone->present_pages)
5148 max = zone->present_pages;
5149 reserve_pages += max;
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005150 /*
5151 * Lowmem reserves are not available to
5152 * GFP_HIGHUSER page cache allocations and
5153 * kswapd tries to balance zones to their high
5154 * watermark. As a result, neither should be
5155 * regarded as dirtyable memory, to prevent a
5156 * situation where reclaim has to clean pages
5157 * in order to balance the zones.
5158 */
5159 zone->dirty_balance_reserve = max;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005160 }
5161 }
Johannes Weinerab8fabd2012-01-10 15:07:42 -08005162 dirty_balance_reserve = reserve_pages;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005163 totalreserve_pages = reserve_pages;
5164}
5165
5166/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07005167 * setup_per_zone_lowmem_reserve - called whenever
5168 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5169 * has a correct pages reserved value, so an adequate number of
5170 * pages are left in the zone after a successful __alloc_pages().
5171 */
5172static void setup_per_zone_lowmem_reserve(void)
5173{
5174 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005175 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005176
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08005177 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005178 for (j = 0; j < MAX_NR_ZONES; j++) {
5179 struct zone *zone = pgdat->node_zones + j;
5180 unsigned long present_pages = zone->present_pages;
5181
5182 zone->lowmem_reserve[j] = 0;
5183
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005184 idx = j;
5185 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005186 struct zone *lower_zone;
5187
Christoph Lameter2f6726e2006-09-25 23:31:18 -07005188 idx--;
5189
Linus Torvalds1da177e2005-04-16 15:20:36 -07005190 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5191 sysctl_lowmem_reserve_ratio[idx] = 1;
5192
5193 lower_zone = pgdat->node_zones + idx;
5194 lower_zone->lowmem_reserve[j] = present_pages /
5195 sysctl_lowmem_reserve_ratio[idx];
5196 present_pages += lower_zone->present_pages;
5197 }
5198 }
5199 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005200
5201 /* update totalreserve_pages */
5202 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005203}
5204
Mel Gormane12aade2011-04-25 21:36:42 +00005205static void __setup_per_zone_wmarks(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005206{
5207 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
Rik van Riel3b321eb2011-09-01 19:26:50 +00005208 unsigned long pages_low = extra_free_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005209 unsigned long lowmem_pages = 0;
5210 struct zone *zone;
5211 unsigned long flags;
5212
5213 /* Calculate total number of !ZONE_HIGHMEM pages */
5214 for_each_zone(zone) {
5215 if (!is_highmem(zone))
5216 lowmem_pages += zone->present_pages;
5217 }
5218
5219 for_each_zone(zone) {
Rik van Riel3b321eb2011-09-01 19:26:50 +00005220 u64 min, low;
Andrew Mortonac924c62006-05-15 09:43:59 -07005221
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005222 spin_lock_irqsave(&zone->lock, flags);
Rik van Riel3b321eb2011-09-01 19:26:50 +00005223 min = (u64)pages_min * zone->present_pages;
5224 do_div(min, lowmem_pages);
5225 low = (u64)pages_low * zone->present_pages;
5226 do_div(low, vm_total_pages);
5227
Linus Torvalds1da177e2005-04-16 15:20:36 -07005228 if (is_highmem(zone)) {
5229 /*
Nick Piggin669ed172005-11-13 16:06:45 -08005230 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5231 * need highmem pages, so cap pages_min to a small
5232 * value here.
5233 *
Mel Gorman41858962009-06-16 15:32:12 -07005234 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
Nick Piggin669ed172005-11-13 16:06:45 -08005235 * deltas controls asynch page reclaim, and so should
5236 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005237 */
5238 int min_pages;
5239
5240 min_pages = zone->present_pages / 1024;
5241 if (min_pages < SWAP_CLUSTER_MAX)
5242 min_pages = SWAP_CLUSTER_MAX;
5243 if (min_pages > 128)
5244 min_pages = 128;
Mel Gorman41858962009-06-16 15:32:12 -07005245 zone->watermark[WMARK_MIN] = min_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005246 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08005247 /*
5248 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07005249 * proportionate to the zone's size.
5250 */
Rik van Riel3b321eb2011-09-01 19:26:50 +00005251 zone->watermark[WMARK_MIN] = min;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005252 }
5253
Rik van Riel3b321eb2011-09-01 19:26:50 +00005254 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) +
5255 low + (min >> 2);
5256 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) +
5257 low + (min >> 1);
Marek Szyprowskif1f63882012-01-25 12:49:24 +01005258
Mel Gorman56fd56b2007-10-16 01:25:58 -07005259 setup_zone_migrate_reserve(zone);
Gerald Schaefer1125b4e2008-10-18 20:27:11 -07005260 spin_unlock_irqrestore(&zone->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005261 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07005262
5263 /* update totalreserve_pages */
5264 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005265}
5266
Mel Gormane12aade2011-04-25 21:36:42 +00005267/**
5268 * setup_per_zone_wmarks - called when min_free_kbytes changes
5269 * or when memory is hot-{added|removed}
5270 *
5271 * Ensures that the watermark[min,low,high] values for each zone are set
5272 * correctly with respect to min_free_kbytes.
5273 */
5274void setup_per_zone_wmarks(void)
5275{
5276 mutex_lock(&zonelists_mutex);
5277 __setup_per_zone_wmarks();
5278 mutex_unlock(&zonelists_mutex);
5279}
5280
Randy Dunlap55a44622009-09-21 17:01:20 -07005281/*
Rik van Riel556adec2008-10-18 20:26:34 -07005282 * The inactive anon list should be small enough that the VM never has to
5283 * do too much work, but large enough that each inactive page has a chance
5284 * to be referenced again before it is swapped out.
5285 *
5286 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5287 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5288 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5289 * the anonymous pages are kept on the inactive list.
5290 *
5291 * total target max
5292 * memory ratio inactive anon
5293 * -------------------------------------
5294 * 10MB 1 5MB
5295 * 100MB 1 50MB
5296 * 1GB 3 250MB
5297 * 10GB 10 0.9GB
5298 * 100GB 31 3GB
5299 * 1TB 101 10GB
5300 * 10TB 320 32GB
5301 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005302static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
Minchan Kim96cb4df2009-06-16 15:32:49 -07005303{
5304 unsigned int gb, ratio;
5305
5306 /* Zone size in gigabytes */
5307 gb = zone->present_pages >> (30 - PAGE_SHIFT);
5308 if (gb)
5309 ratio = int_sqrt(10 * gb);
5310 else
5311 ratio = 1;
5312
5313 zone->inactive_ratio = ratio;
5314}
5315
KOSAKI Motohiro839a4fc2011-05-24 17:11:31 -07005316static void __meminit setup_per_zone_inactive_ratio(void)
Rik van Riel556adec2008-10-18 20:26:34 -07005317{
5318 struct zone *zone;
5319
Minchan Kim96cb4df2009-06-16 15:32:49 -07005320 for_each_zone(zone)
5321 calculate_zone_inactive_ratio(zone);
Rik van Riel556adec2008-10-18 20:26:34 -07005322}
5323
Linus Torvalds1da177e2005-04-16 15:20:36 -07005324/*
5325 * Initialise min_free_kbytes.
5326 *
5327 * For small machines we want it small (128k min). For large machines
5328 * we want it large (64MB max). But it is not linear, because network
5329 * bandwidth does not increase linearly with machine size. We use
5330 *
5331 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5332 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5333 *
5334 * which yields
5335 *
5336 * 16MB: 512k
5337 * 32MB: 724k
5338 * 64MB: 1024k
5339 * 128MB: 1448k
5340 * 256MB: 2048k
5341 * 512MB: 2896k
5342 * 1024MB: 4096k
5343 * 2048MB: 5792k
5344 * 4096MB: 8192k
5345 * 8192MB: 11584k
5346 * 16384MB: 16384k
5347 */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07005348int __meminit init_per_zone_wmark_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005349{
5350 unsigned long lowmem_kbytes;
5351
5352 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5353
5354 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5355 if (min_free_kbytes < 128)
5356 min_free_kbytes = 128;
5357 if (min_free_kbytes > 65536)
5358 min_free_kbytes = 65536;
Minchan Kimbc75d332009-06-16 15:32:48 -07005359 setup_per_zone_wmarks();
KOSAKI Motohiroa6cccdc2011-05-24 17:11:33 -07005360 refresh_zone_stat_thresholds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005361 setup_per_zone_lowmem_reserve();
Rik van Riel556adec2008-10-18 20:26:34 -07005362 setup_per_zone_inactive_ratio();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005363 return 0;
5364}
Minchan Kimbc75d332009-06-16 15:32:48 -07005365module_init(init_per_zone_wmark_min)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005366
5367/*
5368 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5369 * that we can call two helper functions whenever min_free_kbytes
Rik van Riel3b321eb2011-09-01 19:26:50 +00005370 * or extra_free_kbytes changes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005371 */
5372int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005373 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005374{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005375 proc_dointvec(table, write, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07005376 if (write)
Minchan Kimbc75d332009-06-16 15:32:48 -07005377 setup_per_zone_wmarks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005378 return 0;
5379}
5380
Christoph Lameter96146342006-07-03 00:24:13 -07005381#ifdef CONFIG_NUMA
5382int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005383 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter96146342006-07-03 00:24:13 -07005384{
5385 struct zone *zone;
5386 int rc;
5387
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005388 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter96146342006-07-03 00:24:13 -07005389 if (rc)
5390 return rc;
5391
5392 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07005393 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07005394 sysctl_min_unmapped_ratio) / 100;
5395 return 0;
5396}
Christoph Lameter0ff38492006-09-25 23:31:52 -07005397
5398int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005399 void __user *buffer, size_t *length, loff_t *ppos)
Christoph Lameter0ff38492006-09-25 23:31:52 -07005400{
5401 struct zone *zone;
5402 int rc;
5403
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005404 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
Christoph Lameter0ff38492006-09-25 23:31:52 -07005405 if (rc)
5406 return rc;
5407
5408 for_each_zone(zone)
5409 zone->min_slab_pages = (zone->present_pages *
5410 sysctl_min_slab_ratio) / 100;
5411 return 0;
5412}
Christoph Lameter96146342006-07-03 00:24:13 -07005413#endif
5414
Linus Torvalds1da177e2005-04-16 15:20:36 -07005415/*
5416 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5417 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5418 * whenever sysctl_lowmem_reserve_ratio changes.
5419 *
5420 * The reserve ratio obviously has absolutely no relation with the
Mel Gorman41858962009-06-16 15:32:12 -07005421 * minimum watermarks. The lowmem reserve ratio can only make sense
Linus Torvalds1da177e2005-04-16 15:20:36 -07005422 * if in function of the boot time zone sizes.
5423 */
5424int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005425 void __user *buffer, size_t *length, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005426{
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005427 proc_dointvec_minmax(table, write, buffer, length, ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005428 setup_per_zone_lowmem_reserve();
5429 return 0;
5430}
5431
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005432/*
5433 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5434 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5435 * can have before it gets flushed back to buddy allocator.
5436 */
5437
5438int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005439 void __user *buffer, size_t *length, loff_t *ppos)
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005440{
5441 struct zone *zone;
5442 unsigned int cpu;
5443 int ret;
5444
Alexey Dobriyan8d65af72009-09-23 15:57:19 -07005445 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
Sasha Levin93278812012-05-10 13:01:44 -07005446 if (!write || (ret < 0))
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005447 return ret;
Dimitri Sivanich364df0e2009-06-23 12:37:04 -07005448 for_each_populated_zone(zone) {
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005449 for_each_possible_cpu(cpu) {
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005450 unsigned long high;
5451 high = zone->present_pages / percpu_pagelist_fraction;
Christoph Lameter99dcc3e2010-01-05 15:34:51 +09005452 setup_pagelist_highmark(
5453 per_cpu_ptr(zone->pageset, cpu), high);
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08005454 }
5455 }
5456 return 0;
5457}
5458
David S. Millerf034b5d2006-08-24 03:08:07 -07005459int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005460
5461#ifdef CONFIG_NUMA
5462static int __init set_hashdist(char *str)
5463{
5464 if (!str)
5465 return 0;
5466 hashdist = simple_strtoul(str, &str, 0);
5467 return 1;
5468}
5469__setup("hashdist=", set_hashdist);
5470#endif
5471
5472/*
5473 * allocate a large system hash table from bootmem
5474 * - it is assumed that the hash table must contain an exact power-of-2
5475 * quantity of entries
5476 * - limit is the number of hash buckets, not the total allocation size
5477 */
5478void *__init alloc_large_system_hash(const char *tablename,
5479 unsigned long bucketsize,
5480 unsigned long numentries,
5481 int scale,
5482 int flags,
5483 unsigned int *_hash_shift,
5484 unsigned int *_hash_mask,
5485 unsigned long limit)
5486{
5487 unsigned long long max = limit;
5488 unsigned long log2qty, size;
5489 void *table = NULL;
5490
5491 /* allow the kernel cmdline to have a say */
5492 if (!numentries) {
5493 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08005494 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005495 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5496 numentries >>= 20 - PAGE_SHIFT;
5497 numentries <<= 20 - PAGE_SHIFT;
5498
5499 /* limit to 1 bucket per 2^scale bytes of low memory */
5500 if (scale > PAGE_SHIFT)
5501 numentries >>= (scale - PAGE_SHIFT);
5502 else
5503 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08005504
5505 /* Make sure we've got at least a 0-order allocation.. */
Jan Beulich2c85f512009-09-21 17:03:07 -07005506 if (unlikely(flags & HASH_SMALL)) {
5507 /* Makes no sense without HASH_EARLY */
5508 WARN_ON(!(flags & HASH_EARLY));
5509 if (!(numentries >> *_hash_shift)) {
5510 numentries = 1UL << *_hash_shift;
5511 BUG_ON(!numentries);
5512 }
5513 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
Paul Mundt9ab37b82007-01-05 16:36:30 -08005514 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005515 }
John Hawkes6e692ed2006-03-25 03:08:02 -08005516 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005517
5518 /* limit allocation size to 1/16 total memory by default */
5519 if (max == 0) {
5520 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5521 do_div(max, bucketsize);
5522 }
Dimitri Sivanich074b8512012-02-08 12:39:07 -08005523 max = min(max, 0x80000000ULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005524
5525 if (numentries > max)
5526 numentries = max;
5527
David Howellsf0d1b0b2006-12-08 02:37:49 -08005528 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005529
5530 do {
5531 size = bucketsize << log2qty;
5532 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07005533 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005534 else if (hashdist)
5535 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5536 else {
Eric Dumazet1037b832007-07-15 23:38:05 -07005537 /*
5538 * If bucketsize is not a power-of-two, we may free
Mel Gormana1dd2682009-06-16 15:32:19 -07005539 * some pages at the end of hash table which
5540 * alloc_pages_exact() automatically does
Eric Dumazet1037b832007-07-15 23:38:05 -07005541 */
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005542 if (get_order(size) < MAX_ORDER) {
Mel Gormana1dd2682009-06-16 15:32:19 -07005543 table = alloc_pages_exact(size, GFP_ATOMIC);
Catalin Marinas264ef8a2009-07-07 10:33:01 +01005544 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5545 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005546 }
5547 } while (!table && size > PAGE_SIZE && --log2qty);
5548
5549 if (!table)
5550 panic("Failed to allocate %s hash table\n", tablename);
5551
Robin Holtf241e662010-10-07 12:59:26 -07005552 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005553 tablename,
Robin Holtf241e662010-10-07 12:59:26 -07005554 (1UL << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08005555 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005556 size);
5557
5558 if (_hash_shift)
5559 *_hash_shift = log2qty;
5560 if (_hash_mask)
5561 *_hash_mask = (1 << log2qty) - 1;
5562
5563 return table;
5564}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08005565
Mel Gorman835c1342007-10-16 01:25:47 -07005566/* Return a pointer to the bitmap storing bits affecting a block of pages */
5567static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5568 unsigned long pfn)
5569{
5570#ifdef CONFIG_SPARSEMEM
5571 return __pfn_to_section(pfn)->pageblock_flags;
5572#else
5573 return zone->pageblock_flags;
5574#endif /* CONFIG_SPARSEMEM */
5575}
Andrew Morton6220ec72006-10-19 23:29:05 -07005576
Mel Gorman835c1342007-10-16 01:25:47 -07005577static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5578{
5579#ifdef CONFIG_SPARSEMEM
5580 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07005581 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005582#else
Laura Abbott5bca7222012-11-30 14:07:01 -08005583 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
Mel Gormand9c23402007-10-16 01:26:01 -07005584 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07005585#endif /* CONFIG_SPARSEMEM */
5586}
5587
5588/**
Mel Gormand9c23402007-10-16 01:26:01 -07005589 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
Mel Gorman835c1342007-10-16 01:25:47 -07005590 * @page: The page within the block of interest
5591 * @start_bitidx: The first bit of interest to retrieve
5592 * @end_bitidx: The last bit of interest
5593 * returns pageblock_bits flags
5594 */
5595unsigned long get_pageblock_flags_group(struct page *page,
5596 int start_bitidx, int end_bitidx)
5597{
5598 struct zone *zone;
5599 unsigned long *bitmap;
5600 unsigned long pfn, bitidx;
5601 unsigned long flags = 0;
5602 unsigned long value = 1;
5603
5604 zone = page_zone(page);
5605 pfn = page_to_pfn(page);
5606 bitmap = get_pageblock_bitmap(zone, pfn);
5607 bitidx = pfn_to_bitidx(zone, pfn);
5608
5609 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5610 if (test_bit(bitidx + start_bitidx, bitmap))
5611 flags |= value;
5612
5613 return flags;
5614}
5615
5616/**
Mel Gormand9c23402007-10-16 01:26:01 -07005617 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
Mel Gorman835c1342007-10-16 01:25:47 -07005618 * @page: The page within the block of interest
5619 * @start_bitidx: The first bit of interest
5620 * @end_bitidx: The last bit of interest
5621 * @flags: The flags to set
5622 */
5623void set_pageblock_flags_group(struct page *page, unsigned long flags,
5624 int start_bitidx, int end_bitidx)
5625{
5626 struct zone *zone;
5627 unsigned long *bitmap;
5628 unsigned long pfn, bitidx;
5629 unsigned long value = 1;
5630
5631 zone = page_zone(page);
5632 pfn = page_to_pfn(page);
5633 bitmap = get_pageblock_bitmap(zone, pfn);
5634 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07005635 VM_BUG_ON(pfn < zone->zone_start_pfn);
5636 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07005637
5638 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5639 if (flags & value)
5640 __set_bit(bitidx + start_bitidx, bitmap);
5641 else
5642 __clear_bit(bitidx + start_bitidx, bitmap);
5643}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005644
5645/*
Minchan Kim13f662c2012-07-31 16:43:01 -07005646 * This function checks whether pageblock includes unmovable pages or not.
5647 * If @count is not zero, it is okay to include less @count unmovable pages
5648 *
5649 * PageLRU check wihtout isolation or lru_lock could race so that
5650 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
5651 * expect this function should be exact.
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005652 */
Minchan Kim13f662c2012-07-31 16:43:01 -07005653static bool
5654__has_unmovable_pages(struct zone *zone, struct page *page, int count)
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005655{
5656 unsigned long pfn, iter, found;
Michal Nazarewiczd4158d22011-12-29 13:09:50 +01005657 int mt;
5658
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005659 /*
5660 * For avoiding noise data, lru_add_drain_all() should be called
Minchan Kim13f662c2012-07-31 16:43:01 -07005661 * If ZONE_MOVABLE, the zone never contains unmovable pages
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005662 */
5663 if (zone_idx(zone) == ZONE_MOVABLE)
Minchan Kim13f662c2012-07-31 16:43:01 -07005664 return false;
Michal Nazarewiczd4158d22011-12-29 13:09:50 +01005665 mt = get_pageblock_migratetype(page);
5666 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
Minchan Kim13f662c2012-07-31 16:43:01 -07005667 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005668
5669 pfn = page_to_pfn(page);
5670 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5671 unsigned long check = pfn + iter;
5672
Namhyung Kim29723fc2011-02-25 14:44:25 -08005673 if (!pfn_valid_within(check))
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005674 continue;
Namhyung Kim29723fc2011-02-25 14:44:25 -08005675
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005676 page = pfn_to_page(check);
Minchan Kim259ca6a2012-07-31 16:42:59 -07005677 /*
5678 * We can't use page_count without pin a page
5679 * because another CPU can free compound page.
5680 * This check already skips compound tails of THP
5681 * because their page->_count is zero at all time.
5682 */
5683 if (!atomic_read(&page->_count)) {
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005684 if (PageBuddy(page))
5685 iter += (1 << page_order(page)) - 1;
5686 continue;
5687 }
Minchan Kim259ca6a2012-07-31 16:42:59 -07005688
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005689 if (!PageLRU(page))
5690 found++;
5691 /*
5692 * If there are RECLAIMABLE pages, we need to check it.
5693 * But now, memory offline itself doesn't call shrink_slab()
5694 * and it still to be fixed.
5695 */
5696 /*
5697 * If the page is not RAM, page_count()should be 0.
5698 * we don't need more check. This is an _used_ not-movable page.
5699 *
5700 * The problematic thing here is PG_reserved pages. PG_reserved
5701 * is set to both of a memory hole page and a _used_ kernel
5702 * page at boot.
5703 */
5704 if (found > count)
Minchan Kim13f662c2012-07-31 16:43:01 -07005705 return true;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005706 }
Minchan Kim13f662c2012-07-31 16:43:01 -07005707 return false;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005708}
5709
5710bool is_pageblock_removable_nolock(struct page *page)
5711{
Michal Hocko656a0702012-01-20 14:33:58 -08005712 struct zone *zone;
5713 unsigned long pfn;
Michal Hocko687875f2012-01-20 14:33:55 -08005714
5715 /*
5716 * We have to be careful here because we are iterating over memory
5717 * sections which are not zone aware so we might end up outside of
5718 * the zone but still within the section.
Michal Hocko656a0702012-01-20 14:33:58 -08005719 * We have to take care about the node as well. If the node is offline
5720 * its NODE_DATA will be NULL - see page_zone.
Michal Hocko687875f2012-01-20 14:33:55 -08005721 */
Michal Hocko656a0702012-01-20 14:33:58 -08005722 if (!node_online(page_to_nid(page)))
5723 return false;
5724
5725 zone = page_zone(page);
5726 pfn = page_to_pfn(page);
5727 if (zone->zone_start_pfn > pfn ||
Michal Hocko687875f2012-01-20 14:33:55 -08005728 zone->zone_start_pfn + zone->spanned_pages <= pfn)
5729 return false;
5730
Minchan Kim13f662c2012-07-31 16:43:01 -07005731 return !__has_unmovable_pages(zone, page, 0);
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005732}
5733
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005734int set_migratetype_isolate(struct page *page)
5735{
5736 struct zone *zone;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005737 unsigned long flags, pfn;
Robert Jennings925cc712009-12-17 14:44:38 +00005738 struct memory_isolate_notify arg;
5739 int notifier_ret;
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005740 int ret = -EBUSY;
5741
5742 zone = page_zone(page);
Robert Jennings925cc712009-12-17 14:44:38 +00005743
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005744 spin_lock_irqsave(&zone->lock, flags);
Robert Jennings925cc712009-12-17 14:44:38 +00005745
5746 pfn = page_to_pfn(page);
5747 arg.start_pfn = pfn;
5748 arg.nr_pages = pageblock_nr_pages;
5749 arg.pages_found = 0;
5750
5751 /*
5752 * It may be possible to isolate a pageblock even if the
5753 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5754 * notifier chain is used by balloon drivers to return the
5755 * number of pages in a range that are held by the balloon
5756 * driver to shrink memory. If all the pages are accounted for
5757 * by balloons, are free, or on the LRU, isolation can continue.
5758 * Later, for example, when memory hotplug notifier runs, these
5759 * pages reported as "can be isolated" should be isolated(freed)
5760 * by the balloon driver through the memory notifier chain.
5761 */
5762 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5763 notifier_ret = notifier_to_errno(notifier_ret);
KAMEZAWA Hiroyuki4b204772010-10-26 14:21:29 -07005764 if (notifier_ret)
Robert Jennings925cc712009-12-17 14:44:38 +00005765 goto out;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005766 /*
5767 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5768 * We just check MOVABLE pages.
5769 */
Minchan Kim13f662c2012-07-31 16:43:01 -07005770 if (!__has_unmovable_pages(zone, page, arg.pages_found))
Robert Jennings925cc712009-12-17 14:44:38 +00005771 ret = 0;
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005772 /*
Minchan Kim13f662c2012-07-31 16:43:01 -07005773 * Unmovable means "not-on-lru" pages. If Unmovable pages are
5774 * larger than removable-by-driver pages reported by notifier,
5775 * we'll fail.
KAMEZAWA Hiroyuki49ac8252010-10-26 14:21:30 -07005776 */
5777
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005778out:
Robert Jennings925cc712009-12-17 14:44:38 +00005779 if (!ret) {
Larry Bassel6d6e2c92012-12-14 14:21:05 -08005780 unsigned long nr_pages;
5781 int migratetype = get_pageblock_migratetype(page);
5782
Robert Jennings925cc712009-12-17 14:44:38 +00005783 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
Larry Bassel6d6e2c92012-12-14 14:21:05 -08005784 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE);
5785
5786 __mod_zone_freepage_state(zone, -nr_pages, migratetype);
Robert Jennings925cc712009-12-17 14:44:38 +00005787 }
5788
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005789 spin_unlock_irqrestore(&zone->lock, flags);
5790 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08005791 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005792 return ret;
5793}
5794
Michal Nazarewiczc80cd922012-04-03 15:06:15 +02005795void unset_migratetype_isolate(struct page *page, unsigned migratetype)
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005796{
5797 struct zone *zone;
Larry Bassel6d6e2c92012-12-14 14:21:05 -08005798 unsigned long flags, nr_pages;
5799
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005800 zone = page_zone(page);
5801 spin_lock_irqsave(&zone->lock, flags);
5802 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5803 goto out;
Larry Bassel6d6e2c92012-12-14 14:21:05 -08005804 nr_pages = move_freepages_block(zone, page, migratetype);
5805 __mod_zone_freepage_state(zone, nr_pages, migratetype);
Michal Nazarewiczc80cd922012-04-03 15:06:15 +02005806 set_pageblock_migratetype(page, migratetype);
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07005807out:
5808 spin_unlock_irqrestore(&zone->lock, flags);
5809}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07005810
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005811#ifdef CONFIG_CMA
5812
5813static unsigned long pfn_max_align_down(unsigned long pfn)
5814{
5815 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
5816 pageblock_nr_pages) - 1);
5817}
5818
5819static unsigned long pfn_max_align_up(unsigned long pfn)
5820{
5821 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
5822 pageblock_nr_pages));
5823}
5824
5825static struct page *
5826__alloc_contig_migrate_alloc(struct page *page, unsigned long private,
5827 int **resultp)
5828{
Rabin Vincentfbfcd6f2012-07-05 15:52:23 +05305829 gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
5830
5831 if (PageHighMem(page))
5832 gfp_mask |= __GFP_HIGHMEM;
5833
5834 return alloc_page(gfp_mask);
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005835}
5836
5837/* [start, end) must belong to a single zone. */
Mel Gorman6b506642012-10-08 16:32:41 -07005838static int __alloc_contig_migrate_range(struct compact_control *cc,
5839 unsigned long start, unsigned long end)
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005840{
5841 /* This function is based on compact_zone() from compaction.c. */
Minchan Kim726cfb42012-10-08 16:33:51 -07005842 unsigned long nr_reclaimed;
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005843 unsigned long pfn = start;
5844 unsigned int tries = 0;
5845 int ret = 0;
5846
woojoong.lee539118f2012-12-03 17:58:43 -08005847 migrate_prep();
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005848
Mel Gorman6b506642012-10-08 16:32:41 -07005849 while (pfn < end || !list_empty(&cc->migratepages)) {
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005850 if (fatal_signal_pending(current)) {
5851 ret = -EINTR;
5852 break;
5853 }
5854
Mel Gorman6b506642012-10-08 16:32:41 -07005855 if (list_empty(&cc->migratepages)) {
5856 cc->nr_migratepages = 0;
5857 pfn = isolate_migratepages_range(cc->zone, cc,
Minchan Kimd05b4522012-10-08 16:33:48 -07005858 pfn, end, true);
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005859 if (!pfn) {
5860 ret = -EINTR;
5861 break;
5862 }
5863 tries = 0;
5864 } else if (++tries == 5) {
5865 ret = ret < 0 ? ret : -EBUSY;
5866 break;
5867 }
5868
Mel Gorman6b506642012-10-08 16:32:41 -07005869 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, &cc->migratepages);
Minchan Kimaab50f12012-10-08 16:31:55 -07005870
Mel Gorman6b506642012-10-08 16:32:41 -07005871 cc->nr_migratepages -= nr_reclaimed;
5872
5873 ret = migrate_pages(&cc->migratepages,
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005874 __alloc_contig_migrate_alloc,
Minchan Kim132fb792012-05-11 09:37:13 +02005875 0, false, MIGRATE_SYNC);
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005876 }
5877
Mel Gorman6b506642012-10-08 16:32:41 -07005878 putback_lru_pages(&cc->migratepages);
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005879 return ret > 0 ? 0 : ret;
5880}
5881
5882/**
5883 * alloc_contig_range() -- tries to allocate given range of pages
5884 * @start: start PFN to allocate
5885 * @end: one-past-the-last PFN to allocate
Michal Nazarewiczc80cd922012-04-03 15:06:15 +02005886 * @migratetype: migratetype of the underlaying pageblocks (either
5887 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
5888 * in range must have the same migratetype and it must
5889 * be either of the two.
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005890 *
5891 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
5892 * aligned, however it's the caller's responsibility to guarantee that
5893 * we are the only thread that changes migrate type of pageblocks the
5894 * pages fall in.
5895 *
5896 * The PFN range must belong to a single zone.
5897 *
5898 * Returns zero on success or negative error code. On success all
5899 * pages which PFN is in [start, end) are allocated for the caller and
5900 * need to be freed with free_contig_range().
5901 */
Michal Nazarewiczc80cd922012-04-03 15:06:15 +02005902int alloc_contig_range(unsigned long start, unsigned long end,
5903 unsigned migratetype)
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005904{
5905 struct zone *zone = page_zone(pfn_to_page(start));
5906 unsigned long outer_start, outer_end;
5907 int ret = 0, order;
5908
Mel Gorman6b506642012-10-08 16:32:41 -07005909 struct compact_control cc = {
5910 .nr_migratepages = 0,
5911 .order = -1,
5912 .zone = page_zone(pfn_to_page(start)),
5913 .sync = true,
5914 .ignore_skip_hint = true,
5915 };
5916 INIT_LIST_HEAD(&cc.migratepages);
5917
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005918 /*
5919 * What we do here is we mark all pageblocks in range as
5920 * MIGRATE_ISOLATE. Because pageblock and max order pages may
5921 * have different sizes, and due to the way page allocator
5922 * work, we align the range to biggest of the two pages so
5923 * that page allocator won't try to merge buddies from
5924 * different pageblocks and change MIGRATE_ISOLATE to some
5925 * other migration type.
5926 *
5927 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
5928 * migrate the pages from an unaligned range (ie. pages that
5929 * we are interested in). This will put all the pages in
5930 * range back to page allocator as MIGRATE_ISOLATE.
5931 *
5932 * When this is done, we take the pages in range from page
5933 * allocator removing them from the buddy system. This way
5934 * page allocator will never consider using them.
5935 *
5936 * This lets us mark the pageblocks back as
5937 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
5938 * aligned range but not in the unaligned, original range are
5939 * put back to page allocator so that buddy can use them.
5940 */
5941
5942 ret = start_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewiczc80cd922012-04-03 15:06:15 +02005943 pfn_max_align_up(end), migratetype);
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005944 if (ret)
5945 goto done;
5946
Heesub Shin771aaa62013-01-07 11:10:13 +09005947 zone->cma_alloc = 1;
5948
Mel Gorman6b506642012-10-08 16:32:41 -07005949 ret = __alloc_contig_migrate_range(&cc, start, end);
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005950 if (ret)
5951 goto done;
5952
5953 /*
5954 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
5955 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
5956 * more, all pages in [start, end) are free in page allocator.
5957 * What we are going to do is to allocate all pages from
5958 * [start, end) (that is remove them from page allocator).
5959 *
5960 * The only problem is that pages at the beginning and at the
5961 * end of interesting range may be not aligned with pages that
5962 * page allocator holds, ie. they can be part of higher order
5963 * pages. Because of this, we reserve the bigger range and
5964 * once this is done free the pages we are not interested in.
5965 *
5966 * We don't have to hold zone->lock here because the pages are
5967 * isolated thus they won't get removed from buddy.
5968 */
5969
5970 lru_add_drain_all();
5971 drain_all_pages();
5972
5973 order = 0;
5974 outer_start = start;
5975 while (!PageBuddy(pfn_to_page(outer_start))) {
5976 if (++order >= MAX_ORDER) {
5977 ret = -EBUSY;
5978 goto done;
5979 }
5980 outer_start &= ~0UL << order;
5981 }
5982
5983 /* Make sure the range is really isolated. */
5984 if (test_pages_isolated(outer_start, end)) {
5985 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
5986 outer_start, end);
5987 ret = -EBUSY;
5988 goto done;
5989 }
5990
Marek Szyprowskif1f63882012-01-25 12:49:24 +01005991
5992 /* Grab isolated pages from freelists. */
Mel Gorman6b506642012-10-08 16:32:41 -07005993 outer_end = isolate_freepages_range(&cc, outer_start, end);
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01005994 if (!outer_end) {
5995 ret = -EBUSY;
5996 goto done;
5997 }
5998
5999 /* Free head and tail (if any) */
6000 if (start != outer_start)
6001 free_contig_range(outer_start, start - outer_start);
6002 if (end != outer_end)
6003 free_contig_range(end, outer_end - end);
6004
6005done:
6006 undo_isolate_page_range(pfn_max_align_down(start),
Michal Nazarewiczc80cd922012-04-03 15:06:15 +02006007 pfn_max_align_up(end), migratetype);
Heesub Shin771aaa62013-01-07 11:10:13 +09006008 zone->cma_alloc = 0;
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01006009 return ret;
6010}
6011
6012void free_contig_range(unsigned long pfn, unsigned nr_pages)
6013{
Marek Szyprowskidad0d8b2012-12-20 15:05:18 -08006014 unsigned int count = 0;
6015
6016 for (; nr_pages--; pfn++) {
6017 struct page *page = pfn_to_page(pfn);
6018
6019 count += page_count(page) != 1;
6020 __free_page(page);
6021 }
6022 WARN(count != 0, "%d pages are still in use!\n", count);
Michal Nazarewicz4c1ff372011-12-29 13:09:50 +01006023}
6024#endif
6025
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07006026#ifdef CONFIG_MEMORY_HOTREMOVE
6027/*
6028 * All pages in the range must be isolated before calling this.
6029 */
6030void
6031__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6032{
6033 struct page *page;
6034 struct zone *zone;
6035 int order, i;
6036 unsigned long pfn;
6037 unsigned long flags;
6038 /* find the first valid pfn */
6039 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6040 if (pfn_valid(pfn))
6041 break;
6042 if (pfn == end_pfn)
6043 return;
6044 zone = page_zone(pfn_to_page(pfn));
6045 spin_lock_irqsave(&zone->lock, flags);
6046 pfn = start_pfn;
6047 while (pfn < end_pfn) {
6048 if (!pfn_valid(pfn)) {
6049 pfn++;
6050 continue;
6051 }
6052 page = pfn_to_page(pfn);
6053 BUG_ON(page_count(page));
6054 BUG_ON(!PageBuddy(page));
6055 order = page_order(page);
6056#ifdef CONFIG_DEBUG_VM
6057 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6058 pfn, 1 << order, end_pfn);
6059#endif
6060 list_del(&page->lru);
6061 rmv_page_order(page);
6062 zone->free_area[order].nr_free--;
6063 __mod_zone_page_state(zone, NR_FREE_PAGES,
6064 - (1UL << order));
6065 for (i = 0; i < (1 << order); i++)
6066 SetPageReserved((page+i));
6067 pfn += (1 << order);
6068 }
6069 spin_unlock_irqrestore(&zone->lock, flags);
6070}
6071#endif
Wu Fengguang8d22ba12009-12-16 12:19:58 +01006072
6073#ifdef CONFIG_MEMORY_FAILURE
6074bool is_free_buddy_page(struct page *page)
6075{
6076 struct zone *zone = page_zone(page);
6077 unsigned long pfn = page_to_pfn(page);
6078 unsigned long flags;
6079 int order;
6080
6081 spin_lock_irqsave(&zone->lock, flags);
6082 for (order = 0; order < MAX_ORDER; order++) {
6083 struct page *page_head = page - (pfn & ((1 << order) - 1));
6084
6085 if (PageBuddy(page_head) && page_order(page_head) >= order)
6086 break;
6087 }
6088 spin_unlock_irqrestore(&zone->lock, flags);
6089
6090 return order < MAX_ORDER;
6091}
6092#endif
Wu Fengguang718a3822010-03-10 15:20:43 -08006093
6094static struct trace_print_flags pageflag_names[] = {
6095 {1UL << PG_locked, "locked" },
6096 {1UL << PG_error, "error" },
6097 {1UL << PG_referenced, "referenced" },
6098 {1UL << PG_uptodate, "uptodate" },
6099 {1UL << PG_dirty, "dirty" },
6100 {1UL << PG_lru, "lru" },
6101 {1UL << PG_active, "active" },
6102 {1UL << PG_slab, "slab" },
6103 {1UL << PG_owner_priv_1, "owner_priv_1" },
6104 {1UL << PG_arch_1, "arch_1" },
6105 {1UL << PG_reserved, "reserved" },
6106 {1UL << PG_private, "private" },
6107 {1UL << PG_private_2, "private_2" },
6108 {1UL << PG_writeback, "writeback" },
6109#ifdef CONFIG_PAGEFLAGS_EXTENDED
6110 {1UL << PG_head, "head" },
6111 {1UL << PG_tail, "tail" },
6112#else
6113 {1UL << PG_compound, "compound" },
6114#endif
6115 {1UL << PG_swapcache, "swapcache" },
6116 {1UL << PG_mappedtodisk, "mappedtodisk" },
6117 {1UL << PG_reclaim, "reclaim" },
Wu Fengguang718a3822010-03-10 15:20:43 -08006118 {1UL << PG_swapbacked, "swapbacked" },
6119 {1UL << PG_unevictable, "unevictable" },
6120#ifdef CONFIG_MMU
6121 {1UL << PG_mlocked, "mlocked" },
6122#endif
6123#ifdef CONFIG_ARCH_USES_PG_UNCACHED
6124 {1UL << PG_uncached, "uncached" },
6125#endif
6126#ifdef CONFIG_MEMORY_FAILURE
6127 {1UL << PG_hwpoison, "hwpoison" },
6128#endif
Lee Susman0cc1a0c2013-05-05 17:31:17 +03006129 {1UL << PG_readahead, "PG_readahead" },
Wu Fengguang718a3822010-03-10 15:20:43 -08006130 {-1UL, NULL },
6131};
6132
6133static void dump_page_flags(unsigned long flags)
6134{
6135 const char *delim = "";
6136 unsigned long mask;
6137 int i;
6138
6139 printk(KERN_ALERT "page flags: %#lx(", flags);
6140
6141 /* remove zone id */
6142 flags &= (1UL << NR_PAGEFLAGS) - 1;
6143
6144 for (i = 0; pageflag_names[i].name && flags; i++) {
6145
6146 mask = pageflag_names[i].mask;
6147 if ((flags & mask) != mask)
6148 continue;
6149
6150 flags &= ~mask;
6151 printk("%s%s", delim, pageflag_names[i].name);
6152 delim = "|";
6153 }
6154
6155 /* check for left over flags */
6156 if (flags)
6157 printk("%s%#lx", delim, flags);
6158
6159 printk(")\n");
6160}
6161
6162void dump_page(struct page *page)
6163{
6164 printk(KERN_ALERT
6165 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
Andrea Arcangeli4e9f64c2011-01-13 15:46:29 -08006166 page, atomic_read(&page->_count), page_mapcount(page),
Wu Fengguang718a3822010-03-10 15:20:43 -08006167 page->mapping, page->index);
6168 dump_page_flags(page->flags);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07006169 mem_cgroup_print_bad_page(page);
Wu Fengguang718a3822010-03-10 15:20:43 -08006170}