blob: 43f757fcf30f6b1dc61cffd2c824e791db28c418 [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>
22#include <linux/bootmem.h>
23#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070024#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/module.h>
26#include <linux/suspend.h>
27#include <linux/pagevec.h>
28#include <linux/blkdev.h>
29#include <linux/slab.h>
David Rientjes5a3135c22007-10-16 23:25:53 -070030#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/notifier.h>
32#include <linux/topology.h>
33#include <linux/sysctl.h>
34#include <linux/cpu.h>
35#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070036#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/nodemask.h>
38#include <linux/vmalloc.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080039#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070040#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070041#include <linux/sort.h>
42#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070043#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080044#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070045#include <linux/page-isolation.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070048#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include "internal.h"
50
51/*
Christoph Lameter13808912007-10-16 01:25:27 -070052 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 */
Christoph Lameter13808912007-10-16 01:25:27 -070054nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
55 [N_POSSIBLE] = NODE_MASK_ALL,
56 [N_ONLINE] = { { [0] = 1UL } },
57#ifndef CONFIG_NUMA
58 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
59#ifdef CONFIG_HIGHMEM
60 [N_HIGH_MEMORY] = { { [0] = 1UL } },
61#endif
62 [N_CPU] = { { [0] = 1UL } },
63#endif /* NUMA */
64};
65EXPORT_SYMBOL(node_states);
66
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070067unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070068unsigned long totalreserve_pages __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070069long nr_swap_pages;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080070int percpu_pagelist_fraction;
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
Mel Gormand9c23402007-10-16 01:26:01 -070072#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
73int pageblock_order __read_mostly;
74#endif
75
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080076static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080077
Linus Torvalds1da177e2005-04-16 15:20:36 -070078/*
79 * results with 256, 32 in the lowmem_reserve sysctl:
80 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
81 * 1G machine -> (16M dma, 784M normal, 224M high)
82 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
83 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
84 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010085 *
86 * TBD: should special case ZONE_DMA32 machines here - in those we normally
87 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -070089int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -080090#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -070091 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -080092#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -070093#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -070094 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -070095#endif
Christoph Lametere53ef382006-09-25 23:31:14 -070096#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -070097 32,
Christoph Lametere53ef382006-09-25 23:31:14 -070098#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -070099 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700100};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
102EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
Helge Deller15ad7cd2006-12-06 20:40:36 -0800104static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800105#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700106 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800107#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700108#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700109 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700110#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700111 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700112#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700113 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700114#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700115 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700116};
117
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118int min_free_kbytes = 1024;
119
Yasunori Goto86356ab2006-06-23 02:03:09 -0700120unsigned long __meminitdata nr_kernel_pages;
121unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700122static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
Mel Gormanc7132162006-09-27 01:49:43 -0700124#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
125 /*
126 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
127 * ranges of memory (RAM) that may be registered with add_active_range().
128 * Ranges passed to add_active_range() will be merged if possible
129 * so the number of times add_active_range() can be called is
130 * related to the number of nodes and the number of holes
131 */
132 #ifdef CONFIG_MAX_ACTIVE_REGIONS
133 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
134 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
135 #else
136 #if MAX_NUMNODES >= 32
137 /* If there can be many nodes, allow up to 50 holes per node */
138 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
139 #else
140 /* By default, allow up to 256 distinct regions */
141 #define MAX_ACTIVE_REGIONS 256
142 #endif
143 #endif
144
Jan Beulich98011f52007-07-15 23:38:17 -0700145 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
146 static int __meminitdata nr_nodemap_entries;
147 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
148 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Mel Gormanfb014392006-09-27 01:49:59 -0700149#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
Jan Beulich98011f52007-07-15 23:38:17 -0700150 static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
151 static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
Mel Gormanfb014392006-09-27 01:49:59 -0700152#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gorman2a1e2742007-07-17 04:03:12 -0700153 unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700154 static unsigned long __initdata required_movablecore;
Paul Mundte2289292007-07-20 00:31:44 -0700155 unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gorman2a1e2742007-07-17 04:03:12 -0700156
157 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
158 int movable_zone;
159 EXPORT_SYMBOL(movable_zone);
Mel Gormanc7132162006-09-27 01:49:43 -0700160#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
161
Miklos Szeredi418508c2007-05-23 13:57:55 -0700162#if MAX_NUMNODES > 1
163int nr_node_ids __read_mostly = MAX_NUMNODES;
164EXPORT_SYMBOL(nr_node_ids);
165#endif
166
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700167int page_group_by_mobility_disabled __read_mostly;
168
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700169static void set_pageblock_migratetype(struct page *page, int migratetype)
170{
171 set_pageblock_flags_group(page, (unsigned long)migratetype,
172 PB_migrate, PB_migrate_end);
173}
174
Nick Piggin13e74442006-01-06 00:10:58 -0800175#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700176static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700178 int ret = 0;
179 unsigned seq;
180 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700181
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700182 do {
183 seq = zone_span_seqbegin(zone);
184 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
185 ret = 1;
186 else if (pfn < zone->zone_start_pfn)
187 ret = 1;
188 } while (zone_span_seqretry(zone, seq));
189
190 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700191}
192
193static int page_is_consistent(struct zone *zone, struct page *page)
194{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700195 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700196 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700198 return 0;
199
200 return 1;
201}
202/*
203 * Temporary debugging check for pages not lying within a given zone.
204 */
205static int bad_range(struct zone *zone, struct page *page)
206{
207 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700209 if (!page_is_consistent(zone, page))
210 return 1;
211
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 return 0;
213}
Nick Piggin13e74442006-01-06 00:10:58 -0800214#else
215static inline int bad_range(struct zone *zone, struct page *page)
216{
217 return 0;
218}
219#endif
220
Nick Piggin224abf92006-01-06 00:11:11 -0800221static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222{
Nick Piggin224abf92006-01-06 00:11:11 -0800223 printk(KERN_EMERG "Bad page state in process '%s'\n"
Hugh Dickins7365f3d2006-01-11 12:17:18 -0800224 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
225 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
226 KERN_EMERG "Backtrace:\n",
Nick Piggin224abf92006-01-06 00:11:11 -0800227 current->comm, page, (int)(2*sizeof(unsigned long)),
228 (unsigned long)page->flags, page->mapping,
229 page_mapcount(page), page_count(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 dump_stack();
Hugh Dickins334795e2005-06-21 17:15:08 -0700231 page->flags &= ~(1 << PG_lru |
232 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 1 << PG_active |
235 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700236 1 << PG_reclaim |
237 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 1 << PG_swapcache |
Nick Piggin676165a2006-04-10 11:21:48 +1000239 1 << PG_writeback |
240 1 << PG_buddy );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 set_page_count(page, 0);
242 reset_page_mapcount(page);
243 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700244 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245}
246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247/*
248 * Higher-order pages are called "compound pages". They are structured thusly:
249 *
250 * The first PAGE_SIZE page is called the "head page".
251 *
252 * The remaining PAGE_SIZE pages are called "tail pages".
253 *
254 * All pages have PG_compound set. All pages have their ->private pointing at
255 * the head page (even the head page has this).
256 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800257 * The first tail page's ->lru.next holds the address of the compound page's
258 * put_page() function. Its ->lru.prev holds the order of allocation.
259 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800261
262static void free_compound_page(struct page *page)
263{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700264 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800265}
266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267static void prep_compound_page(struct page *page, unsigned long order)
268{
269 int i;
270 int nr_pages = 1 << order;
271
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800272 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700273 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700274 __SetPageHead(page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700275 for (i = 1; i < nr_pages; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 struct page *p = page + i;
277
Christoph Lameterd85f3382007-05-06 14:49:39 -0700278 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700279 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 }
281}
282
283static void destroy_compound_page(struct page *page, unsigned long order)
284{
285 int i;
286 int nr_pages = 1 << order;
287
Christoph Lameterd85f3382007-05-06 14:49:39 -0700288 if (unlikely(compound_order(page) != order))
Nick Piggin224abf92006-01-06 00:11:11 -0800289 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
Christoph Lameter6d777952007-05-06 14:49:40 -0700291 if (unlikely(!PageHead(page)))
Christoph Lameterd85f3382007-05-06 14:49:39 -0700292 bad_page(page);
Christoph Lameter6d777952007-05-06 14:49:40 -0700293 __ClearPageHead(page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700294 for (i = 1; i < nr_pages; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 struct page *p = page + i;
296
Christoph Lameter6d777952007-05-06 14:49:40 -0700297 if (unlikely(!PageTail(p) |
Christoph Lameterd85f3382007-05-06 14:49:39 -0700298 (p->first_page != page)))
Nick Piggin224abf92006-01-06 00:11:11 -0800299 bad_page(page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700300 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 }
302}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
Nick Piggin17cf4402006-03-22 00:08:41 -0800304static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
305{
306 int i;
307
Nick Piggin725d7042006-09-25 23:30:55 -0700308 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
Andrew Morton6626c5d2006-03-22 00:08:42 -0800309 /*
310 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
311 * and __GFP_HIGHMEM from hard or soft interrupt context.
312 */
Nick Piggin725d7042006-09-25 23:30:55 -0700313 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800314 for (i = 0; i < (1 << order); i++)
315 clear_highpage(page + i);
316}
317
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700318static inline void set_page_order(struct page *page, int order)
319{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700320 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000321 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322}
323
324static inline void rmv_page_order(struct page *page)
325{
Nick Piggin676165a2006-04-10 11:21:48 +1000326 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700327 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328}
329
330/*
331 * Locate the struct page for both the matching buddy in our
332 * pair (buddy1) and the combined O(n+1) page they form (page).
333 *
334 * 1) Any buddy B1 will have an order O twin B2 which satisfies
335 * the following equation:
336 * B2 = B1 ^ (1 << O)
337 * For example, if the starting buddy (buddy2) is #8 its order
338 * 1 buddy is #10:
339 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
340 *
341 * 2) Any buddy B will have an order O+1 parent P which
342 * satisfies the following equation:
343 * P = B & ~(1 << O)
344 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200345 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 */
347static inline struct page *
348__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
349{
350 unsigned long buddy_idx = page_idx ^ (1 << order);
351
352 return page + (buddy_idx - page_idx);
353}
354
355static inline unsigned long
356__find_combined_index(unsigned long page_idx, unsigned int order)
357{
358 return (page_idx & ~(1 << order));
359}
360
361/*
362 * This function checks whether a page is free && is the buddy
363 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800364 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000365 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700366 * (c) a page and its buddy have the same order &&
367 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 *
Nick Piggin676165a2006-04-10 11:21:48 +1000369 * For recording whether a page is in the buddy system, we use PG_buddy.
370 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
371 *
372 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700374static inline int page_is_buddy(struct page *page, struct page *buddy,
375 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700377 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800378 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800379
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700380 if (page_zone_id(page) != page_zone_id(buddy))
381 return 0;
382
383 if (PageBuddy(buddy) && page_order(buddy) == order) {
384 BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700385 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000386 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700387 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388}
389
390/*
391 * Freeing function for a buddy system allocator.
392 *
393 * The concept of a buddy system is to maintain direct-mapped table
394 * (containing bit values) for memory blocks of various "orders".
395 * The bottom level table contains the map for the smallest allocatable
396 * units of memory (here, pages), and each level above it describes
397 * pairs of units from the levels below, hence, "buddies".
398 * At a high level, all that happens here is marking the table entry
399 * at the bottom level available, and propagating the changes upward
400 * as necessary, plus some accounting needed to play nicely with other
401 * parts of the VM system.
402 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000403 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700404 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 * So when we are allocating or freeing one, we can derive the state of the
406 * other. That is, if we allocate a small block, and both were
407 * free, the remainder of the region must be split into blocks.
408 * If a block is freed, and its buddy is also free, then this
409 * triggers coalescing into a block of larger size.
410 *
411 * -- wli
412 */
413
Nick Piggin48db57f2006-01-08 01:00:42 -0800414static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 struct zone *zone, unsigned int order)
416{
417 unsigned long page_idx;
418 int order_size = 1 << order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700419 int migratetype = get_pageblock_migratetype(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
Nick Piggin224abf92006-01-06 00:11:11 -0800421 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 destroy_compound_page(page, order);
423
424 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
425
Nick Piggin725d7042006-09-25 23:30:55 -0700426 VM_BUG_ON(page_idx & (order_size - 1));
427 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
Christoph Lameterd23ad422007-02-10 01:43:02 -0800429 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 while (order < MAX_ORDER-1) {
431 unsigned long combined_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 struct page *buddy;
433
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700435 if (!page_is_buddy(page, buddy, order))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700439 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800441 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 page = page + (combined_idx - page_idx);
443 page_idx = combined_idx;
444 order++;
445 }
446 set_page_order(page, order);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700447 list_add(&page->lru,
448 &zone->free_area[order].free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 zone->free_area[order].nr_free++;
450}
451
Nick Piggin224abf92006-01-06 00:11:11 -0800452static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
Nick Piggin92be2e332006-01-06 00:10:57 -0800454 if (unlikely(page_mapcount(page) |
455 (page->mapping != NULL) |
456 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 (page->flags & (
458 1 << PG_lru |
459 1 << PG_private |
460 1 << PG_locked |
461 1 << PG_active |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 1 << PG_slab |
463 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700464 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000465 1 << PG_reserved |
466 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800467 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700469 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800470 /*
471 * For now, we report if PG_reserved was found set, but do not
472 * clear it, and do not free the page. But we shall soon need
473 * to do more, for when the ZERO_PAGE count wraps negative.
474 */
475 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476}
477
478/*
479 * Frees a list of pages.
480 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700481 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 *
483 * If the zone was previously in an "all pages pinned" state then look to
484 * see if this freeing clears that state.
485 *
486 * And clear the zone's pages_scanned counter, to hold off the "all pages are
487 * pinned" detection logic.
488 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800489static void free_pages_bulk(struct zone *zone, int count,
490 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491{
Nick Pigginc54ad302006-01-06 00:10:56 -0800492 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700493 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800495 while (count--) {
496 struct page *page;
497
Nick Piggin725d7042006-09-25 23:30:55 -0700498 VM_BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800500 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800502 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800504 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505}
506
Nick Piggin48db57f2006-01-08 01:00:42 -0800507static void free_one_page(struct zone *zone, struct page *page, int order)
508{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700509 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700510 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700511 zone->pages_scanned = 0;
Paul Jackson0798e512006-12-06 20:31:38 -0800512 __free_one_page(page, zone, order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700513 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800514}
515
516static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517{
Nick Pigginc54ad302006-01-06 00:10:56 -0800518 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800520 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800523 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800524 if (reserved)
525 return;
526
Nick Piggin9858db52006-10-11 01:21:30 -0700527 if (!PageHighMem(page))
528 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Nick Piggindafb1362006-10-11 01:21:30 -0700529 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800530 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700531
Nick Pigginc54ad302006-01-06 00:10:56 -0800532 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700533 __count_vm_events(PGFREE, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800534 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800535 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536}
537
David Howellsa226f6c2006-01-06 00:11:08 -0800538/*
539 * permit the bootmem allocator to evade page validation on high-order frees
540 */
541void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
542{
543 if (order == 0) {
544 __ClearPageReserved(page);
545 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800546 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800547 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800548 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800549 int loop;
550
Nick Piggin545b1ea2006-03-22 00:08:07 -0800551 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800552 for (loop = 0; loop < BITS_PER_LONG; loop++) {
553 struct page *p = &page[loop];
554
Nick Piggin545b1ea2006-03-22 00:08:07 -0800555 if (loop + 1 < BITS_PER_LONG)
556 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800557 __ClearPageReserved(p);
558 set_page_count(p, 0);
559 }
560
Nick Piggin7835e982006-03-22 00:08:40 -0800561 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800562 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800563 }
564}
565
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
567/*
568 * The order of subdivision here is critical for the IO subsystem.
569 * Please do not alter this order without good reasons and regression
570 * testing. Specifically, as large blocks of memory are subdivided,
571 * the order in which smaller blocks are delivered depends on the order
572 * they're subdivided in this function. This is the primary factor
573 * influencing the order in which pages are delivered to the IO
574 * subsystem according to empirical testing, and this is also justified
575 * by considering the behavior of a buddy system containing a single
576 * large block of memory acted on by a series of small allocations.
577 * This behavior is a critical factor in sglist merging's success.
578 *
579 * -- wli
580 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800581static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700582 int low, int high, struct free_area *area,
583 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584{
585 unsigned long size = 1 << high;
586
587 while (high > low) {
588 area--;
589 high--;
590 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700591 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700592 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 area->nr_free++;
594 set_page_order(&page[size], high);
595 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596}
597
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598/*
599 * This page is about to be returned from the page allocator
600 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800601static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602{
Nick Piggin92be2e332006-01-06 00:10:57 -0800603 if (unlikely(page_mapcount(page) |
604 (page->mapping != NULL) |
605 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700606 (page->flags & (
607 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 1 << PG_private |
609 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 1 << PG_active |
611 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700612 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700614 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000615 1 << PG_reserved |
616 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800617 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
Hugh Dickins689bceb2005-11-21 21:32:20 -0800619 /*
620 * For now, we report if PG_reserved was found set, but do not
621 * clear it, and do not allocate the page: as a safety net.
622 */
623 if (PageReserved(page))
624 return 1;
625
Fengguang Wud77c2d72007-07-19 01:47:55 -0700626 page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 1 << PG_referenced | 1 << PG_arch_1 |
Nick Piggin5409bae2007-02-28 20:12:27 -0800628 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700629 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800630 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800631
632 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800634
635 if (gfp_flags & __GFP_ZERO)
636 prep_zero_page(page, order, gfp_flags);
637
638 if (order && (gfp_flags & __GFP_COMP))
639 prep_compound_page(page, order);
640
Hugh Dickins689bceb2005-11-21 21:32:20 -0800641 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642}
643
Mel Gorman56fd56b2007-10-16 01:25:58 -0700644/*
645 * Go through the free lists for the given migratetype and remove
646 * the smallest available page from the freelists
647 */
648static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
649 int migratetype)
650{
651 unsigned int current_order;
652 struct free_area * area;
653 struct page *page;
654
655 /* Find a page of the appropriate size in the preferred list */
656 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
657 area = &(zone->free_area[current_order]);
658 if (list_empty(&area->free_list[migratetype]))
659 continue;
660
661 page = list_entry(area->free_list[migratetype].next,
662 struct page, lru);
663 list_del(&page->lru);
664 rmv_page_order(page);
665 area->nr_free--;
666 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
667 expand(zone, page, order, current_order, area, migratetype);
668 return page;
669 }
670
671 return NULL;
672}
673
674
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700675/*
676 * This array describes the order lists are fallen back to when
677 * the free lists for the desirable migrate type are depleted
678 */
679static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700680 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
681 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
682 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
683 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700684};
685
Mel Gormanc361be52007-10-16 01:25:51 -0700686/*
687 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700688 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700689 * boundary. If alignment is required, use move_freepages_block()
690 */
691int move_freepages(struct zone *zone,
692 struct page *start_page, struct page *end_page,
693 int migratetype)
694{
695 struct page *page;
696 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700697 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700698
699#ifndef CONFIG_HOLES_IN_ZONE
700 /*
701 * page_zone is not safe to call in this context when
702 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
703 * anyway as we check zone boundaries in move_freepages_block().
704 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700705 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700706 */
707 BUG_ON(page_zone(start_page) != page_zone(end_page));
708#endif
709
710 for (page = start_page; page <= end_page;) {
711 if (!pfn_valid_within(page_to_pfn(page))) {
712 page++;
713 continue;
714 }
715
716 if (!PageBuddy(page)) {
717 page++;
718 continue;
719 }
720
721 order = page_order(page);
722 list_del(&page->lru);
723 list_add(&page->lru,
724 &zone->free_area[order].free_list[migratetype]);
725 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700726 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700727 }
728
Mel Gormand1003132007-10-16 01:26:00 -0700729 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700730}
731
732int move_freepages_block(struct zone *zone, struct page *page, int migratetype)
733{
734 unsigned long start_pfn, end_pfn;
735 struct page *start_page, *end_page;
736
737 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700738 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700739 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700740 end_page = start_page + pageblock_nr_pages - 1;
741 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700742
743 /* Do not cross zone boundaries */
744 if (start_pfn < zone->zone_start_pfn)
745 start_page = page;
746 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
747 return 0;
748
749 return move_freepages(zone, start_page, end_page, migratetype);
750}
751
Mel Gorman5adc5be2007-10-16 01:25:54 -0700752/* Return the page with the lowest PFN in the list */
753static struct page *min_page(struct list_head *list)
754{
755 unsigned long min_pfn = -1UL;
756 struct page *min_page = NULL, *page;;
757
758 list_for_each_entry(page, list, lru) {
759 unsigned long pfn = page_to_pfn(page);
760 if (pfn < min_pfn) {
761 min_pfn = pfn;
762 min_page = page;
763 }
764 }
765
766 return min_page;
767}
768
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700769/* Remove an element from the buddy allocator from the fallback list */
770static struct page *__rmqueue_fallback(struct zone *zone, int order,
771 int start_migratetype)
772{
773 struct free_area * area;
774 int current_order;
775 struct page *page;
776 int migratetype, i;
777
778 /* Find the largest possible block of pages in the other list */
779 for (current_order = MAX_ORDER-1; current_order >= order;
780 --current_order) {
781 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
782 migratetype = fallbacks[start_migratetype][i];
783
Mel Gorman56fd56b2007-10-16 01:25:58 -0700784 /* MIGRATE_RESERVE handled later if necessary */
785 if (migratetype == MIGRATE_RESERVE)
786 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700787
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700788 area = &(zone->free_area[current_order]);
789 if (list_empty(&area->free_list[migratetype]))
790 continue;
791
Mel Gorman5adc5be2007-10-16 01:25:54 -0700792 /* Bias kernel allocations towards low pfns */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700793 page = list_entry(area->free_list[migratetype].next,
794 struct page, lru);
Mel Gorman5adc5be2007-10-16 01:25:54 -0700795 if (unlikely(start_migratetype != MIGRATE_MOVABLE))
796 page = min_page(&area->free_list[migratetype]);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700797 area->nr_free--;
798
799 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700800 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700801 * pages to the preferred allocation list. If falling
802 * back for a reclaimable kernel allocation, be more
803 * agressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700804 */
Mel Gormand9c23402007-10-16 01:26:01 -0700805 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gorman46dafbc2007-10-16 01:25:55 -0700806 start_migratetype == MIGRATE_RECLAIMABLE) {
807 unsigned long pages;
808 pages = move_freepages_block(zone, page,
809 start_migratetype);
810
811 /* Claim the whole block if over half of it is free */
Mel Gormand9c23402007-10-16 01:26:01 -0700812 if (pages >= (1 << (pageblock_order-1)))
Mel Gorman46dafbc2007-10-16 01:25:55 -0700813 set_pageblock_migratetype(page,
814 start_migratetype);
815
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700816 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700817 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700818
819 /* Remove the page from the freelists */
820 list_del(&page->lru);
821 rmv_page_order(page);
822 __mod_zone_page_state(zone, NR_FREE_PAGES,
823 -(1UL << order));
824
Mel Gormand9c23402007-10-16 01:26:01 -0700825 if (current_order == pageblock_order)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700826 set_pageblock_migratetype(page,
827 start_migratetype);
828
829 expand(zone, page, order, current_order, area, migratetype);
830 return page;
831 }
832 }
833
Mel Gorman56fd56b2007-10-16 01:25:58 -0700834 /* Use MIGRATE_RESERVE rather than fail an allocation */
835 return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700836}
837
Mel Gorman56fd56b2007-10-16 01:25:58 -0700838/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 * Do the hard work of removing an element from the buddy allocator.
840 * Call me with the zone->lock already held.
841 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700842static struct page *__rmqueue(struct zone *zone, unsigned int order,
843 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 struct page *page;
846
Mel Gorman56fd56b2007-10-16 01:25:58 -0700847 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
Mel Gorman56fd56b2007-10-16 01:25:58 -0700849 if (unlikely(!page))
850 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700851
852 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853}
854
855/*
856 * Obtain a specified number of elements from the buddy allocator, all under
857 * a single hold of the lock, for efficiency. Add them to the supplied list.
858 * Returns the number of new pages which were placed at *list.
859 */
860static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700861 unsigned long count, struct list_head *list,
862 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865
Nick Pigginc54ad302006-01-06 00:10:56 -0800866 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700868 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800869 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 break;
Mel Gorman535131e62007-10-16 01:25:49 -0700871 list_add(&page->lru, list);
872 set_page_private(page, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800874 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800875 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876}
877
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700878#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800879/*
Christoph Lameter4037d452007-05-09 02:35:14 -0700880 * Called from the vmstat counter updater to drain pagesets of this
881 * currently executing processor on remote nodes after they have
882 * expired.
883 *
Christoph Lameter879336c2006-03-22 00:09:08 -0800884 * Note that this function must be called with the thread pinned to
885 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800886 */
Christoph Lameter4037d452007-05-09 02:35:14 -0700887void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700888{
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700889 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -0700890 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700891
Christoph Lameter4037d452007-05-09 02:35:14 -0700892 local_irq_save(flags);
893 if (pcp->count >= pcp->batch)
894 to_drain = pcp->batch;
895 else
896 to_drain = pcp->count;
897 free_pages_bulk(zone, to_drain, &pcp->list, 0);
898 pcp->count -= to_drain;
899 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700900}
901#endif
902
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903static void __drain_pages(unsigned int cpu)
904{
Nick Pigginc54ad302006-01-06 00:10:56 -0800905 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 struct zone *zone;
907 int i;
908
909 for_each_zone(zone) {
910 struct per_cpu_pageset *pset;
911
Christoph Lameterf2e12bb2007-01-05 16:37:02 -0800912 if (!populated_zone(zone))
913 continue;
914
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700915 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
917 struct per_cpu_pages *pcp;
918
919 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800920 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800921 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
922 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800923 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 }
925 }
926}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
Rafael J. Wysocki296699d2007-07-29 23:27:18 +0200928#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
930void mark_free_pages(struct zone *zone)
931{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700932 unsigned long pfn, max_zone_pfn;
933 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700934 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 struct list_head *curr;
936
937 if (!zone->spanned_pages)
938 return;
939
940 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700941
942 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
943 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
944 if (pfn_valid(pfn)) {
945 struct page *page = pfn_to_page(pfn);
946
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700947 if (!swsusp_page_is_forbidden(page))
948 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700949 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700951 for_each_migratetype_order(order, t) {
952 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700953 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700955 pfn = page_to_pfn(list_entry(curr, struct page, lru));
956 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700957 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700958 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700959 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 spin_unlock_irqrestore(&zone->lock, flags);
961}
Mel Gormane2c55dc2007-10-16 01:25:50 -0700962#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
964/*
965 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
966 */
967void drain_local_pages(void)
968{
969 unsigned long flags;
970
971 local_irq_save(flags);
972 __drain_pages(smp_processor_id());
973 local_irq_restore(flags);
974}
Mel Gormane2c55dc2007-10-16 01:25:50 -0700975
976void smp_drain_local_pages(void *arg)
977{
978 drain_local_pages();
979}
980
981/*
982 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
983 */
984void drain_all_local_pages(void)
985{
986 unsigned long flags;
987
988 local_irq_save(flags);
989 __drain_pages(smp_processor_id());
990 local_irq_restore(flags);
991
992 smp_call_function(smp_drain_local_pages, NULL, 0, 1);
993}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995/*
996 * Free a 0-order page
997 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998static void fastcall free_hot_cold_page(struct page *page, int cold)
999{
1000 struct zone *zone = page_zone(page);
1001 struct per_cpu_pages *pcp;
1002 unsigned long flags;
1003
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 if (PageAnon(page))
1005 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -08001006 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -08001007 return;
1008
Nick Piggin9858db52006-10-11 01:21:30 -07001009 if (!PageHighMem(page))
1010 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Nick Piggindafb1362006-10-11 01:21:30 -07001011 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -08001012 kernel_map_pages(page, 1, 0);
1013
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001014 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001016 __count_vm_event(PGFREE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 list_add(&page->lru, &pcp->list);
Mel Gorman535131e62007-10-16 01:25:49 -07001018 set_page_private(page, get_pageblock_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001020 if (pcp->count >= pcp->high) {
1021 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1022 pcp->count -= pcp->batch;
1023 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 local_irq_restore(flags);
1025 put_cpu();
1026}
1027
1028void fastcall free_hot_page(struct page *page)
1029{
1030 free_hot_cold_page(page, 0);
1031}
1032
1033void fastcall free_cold_page(struct page *page)
1034{
1035 free_hot_cold_page(page, 1);
1036}
1037
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001038/*
1039 * split_page takes a non-compound higher-order page, and splits it into
1040 * n (1<<order) sub-pages: page[0..n]
1041 * Each sub-page must be freed individually.
1042 *
1043 * Note: this is probably too low level an operation for use in drivers.
1044 * Please consult with lkml before using this in your driver.
1045 */
1046void split_page(struct page *page, unsigned int order)
1047{
1048 int i;
1049
Nick Piggin725d7042006-09-25 23:30:55 -07001050 VM_BUG_ON(PageCompound(page));
1051 VM_BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -08001052 for (i = 1; i < (1 << order); i++)
1053 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001054}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001055
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056/*
1057 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1058 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1059 * or two.
1060 */
Nick Piggina74609f2006-01-06 00:11:20 -08001061static struct page *buffered_rmqueue(struct zonelist *zonelist,
1062 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063{
1064 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001065 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -08001067 int cpu;
Mel Gorman64c5e132007-10-16 01:25:59 -07001068 int migratetype = allocflags_to_migratetype(gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069
Hugh Dickins689bceb2005-11-21 21:32:20 -08001070again:
Nick Piggina74609f2006-01-06 00:11:20 -08001071 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -08001072 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 struct per_cpu_pages *pcp;
1074
Nick Piggina74609f2006-01-06 00:11:20 -08001075 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001077 if (!pcp->count) {
nkalmala941c7102006-11-02 22:07:04 -08001078 pcp->count = rmqueue_bulk(zone, 0,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001079 pcp->batch, &pcp->list, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001080 if (unlikely(!pcp->count))
1081 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 }
Mel Gorman535131e62007-10-16 01:25:49 -07001083
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001084 /* Find a page of the appropriate migrate type */
1085 list_for_each_entry(page, &pcp->list, lru)
1086 if (page_private(page) == migratetype)
1087 break;
1088
1089 /* Allocate more to the pcp list if necessary */
1090 if (unlikely(&page->lru == &pcp->list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001091 pcp->count += rmqueue_bulk(zone, 0,
1092 pcp->batch, &pcp->list, migratetype);
1093 page = list_entry(pcp->list.next, struct page, lru);
Mel Gorman535131e62007-10-16 01:25:49 -07001094 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001095
1096 list_del(&page->lru);
1097 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001098 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001100 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001101 spin_unlock(&zone->lock);
1102 if (!page)
1103 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 }
1105
Christoph Lameterf8891e52006-06-30 01:55:45 -07001106 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Christoph Lameterca889e62006-06-30 01:55:44 -07001107 zone_statistics(zonelist, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001108 local_irq_restore(flags);
1109 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110
Nick Piggin725d7042006-09-25 23:30:55 -07001111 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001112 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001113 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001115
1116failed:
1117 local_irq_restore(flags);
1118 put_cpu();
1119 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120}
1121
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001122#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -08001123#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1124#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1125#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1126#define ALLOC_HARDER 0x10 /* try to alloc harder */
1127#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1128#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001129
Akinobu Mita933e3122006-12-08 02:39:45 -08001130#ifdef CONFIG_FAIL_PAGE_ALLOC
1131
1132static struct fail_page_alloc_attr {
1133 struct fault_attr attr;
1134
1135 u32 ignore_gfp_highmem;
1136 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001137 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001138
1139#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1140
1141 struct dentry *ignore_gfp_highmem_file;
1142 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001143 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001144
1145#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1146
1147} fail_page_alloc = {
1148 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001149 .ignore_gfp_wait = 1,
1150 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001151 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001152};
1153
1154static int __init setup_fail_page_alloc(char *str)
1155{
1156 return setup_fault_attr(&fail_page_alloc.attr, str);
1157}
1158__setup("fail_page_alloc=", setup_fail_page_alloc);
1159
1160static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1161{
Akinobu Mita54114992007-07-15 23:40:23 -07001162 if (order < fail_page_alloc.min_order)
1163 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001164 if (gfp_mask & __GFP_NOFAIL)
1165 return 0;
1166 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1167 return 0;
1168 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1169 return 0;
1170
1171 return should_fail(&fail_page_alloc.attr, 1 << order);
1172}
1173
1174#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1175
1176static int __init fail_page_alloc_debugfs(void)
1177{
1178 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1179 struct dentry *dir;
1180 int err;
1181
1182 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1183 "fail_page_alloc");
1184 if (err)
1185 return err;
1186 dir = fail_page_alloc.attr.dentries.dir;
1187
1188 fail_page_alloc.ignore_gfp_wait_file =
1189 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1190 &fail_page_alloc.ignore_gfp_wait);
1191
1192 fail_page_alloc.ignore_gfp_highmem_file =
1193 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1194 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001195 fail_page_alloc.min_order_file =
1196 debugfs_create_u32("min-order", mode, dir,
1197 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001198
1199 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001200 !fail_page_alloc.ignore_gfp_highmem_file ||
1201 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001202 err = -ENOMEM;
1203 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1204 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001205 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001206 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1207 }
1208
1209 return err;
1210}
1211
1212late_initcall(fail_page_alloc_debugfs);
1213
1214#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1215
1216#else /* CONFIG_FAIL_PAGE_ALLOC */
1217
1218static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1219{
1220 return 0;
1221}
1222
1223#endif /* CONFIG_FAIL_PAGE_ALLOC */
1224
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225/*
1226 * Return 1 if free pages are above 'mark'. This takes into account the order
1227 * of the allocation.
1228 */
1229int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001230 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231{
1232 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001233 long min = mark;
1234 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 int o;
1236
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001237 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001239 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 min -= min / 4;
1241
1242 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1243 return 0;
1244 for (o = 0; o < order; o++) {
1245 /* At the next order, this order's pages become unavailable */
1246 free_pages -= z->free_area[o].nr_free << o;
1247
1248 /* Require fewer higher order pages to be free */
1249 min >>= 1;
1250
1251 if (free_pages <= min)
1252 return 0;
1253 }
1254 return 1;
1255}
1256
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001257#ifdef CONFIG_NUMA
1258/*
1259 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1260 * skip over zones that are not allowed by the cpuset, or that have
1261 * been recently (in last second) found to be nearly full. See further
1262 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1263 * that have to skip over alot of full or unallowed zones.
1264 *
1265 * If the zonelist cache is present in the passed in zonelist, then
1266 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001267 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001268 *
1269 * If the zonelist cache is not available for this zonelist, does
1270 * nothing and returns NULL.
1271 *
1272 * If the fullzones BITMAP in the zonelist cache is stale (more than
1273 * a second since last zap'd) then we zap it out (clear its bits.)
1274 *
1275 * We hold off even calling zlc_setup, until after we've checked the
1276 * first zone in the zonelist, on the theory that most allocations will
1277 * be satisfied from that first zone, so best to examine that zone as
1278 * quickly as we can.
1279 */
1280static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1281{
1282 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1283 nodemask_t *allowednodes; /* zonelist_cache approximation */
1284
1285 zlc = zonelist->zlcache_ptr;
1286 if (!zlc)
1287 return NULL;
1288
1289 if (jiffies - zlc->last_full_zap > 1 * HZ) {
1290 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1291 zlc->last_full_zap = jiffies;
1292 }
1293
1294 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1295 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001296 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001297 return allowednodes;
1298}
1299
1300/*
1301 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1302 * if it is worth looking at further for free memory:
1303 * 1) Check that the zone isn't thought to be full (doesn't have its
1304 * bit set in the zonelist_cache fullzones BITMAP).
1305 * 2) Check that the zones node (obtained from the zonelist_cache
1306 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1307 * Return true (non-zero) if zone is worth looking at further, or
1308 * else return false (zero) if it is not.
1309 *
1310 * This check -ignores- the distinction between various watermarks,
1311 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1312 * found to be full for any variation of these watermarks, it will
1313 * be considered full for up to one second by all requests, unless
1314 * we are so low on memory on all allowed nodes that we are forced
1315 * into the second scan of the zonelist.
1316 *
1317 * In the second scan we ignore this zonelist cache and exactly
1318 * apply the watermarks to all zones, even it is slower to do so.
1319 * We are low on memory in the second scan, and should leave no stone
1320 * unturned looking for a free page.
1321 */
1322static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1323 nodemask_t *allowednodes)
1324{
1325 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1326 int i; /* index of *z in zonelist zones */
1327 int n; /* node that zone *z is on */
1328
1329 zlc = zonelist->zlcache_ptr;
1330 if (!zlc)
1331 return 1;
1332
1333 i = z - zonelist->zones;
1334 n = zlc->z_to_n[i];
1335
1336 /* This zone is worth trying if it is allowed but not full */
1337 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1338}
1339
1340/*
1341 * Given 'z' scanning a zonelist, set the corresponding bit in
1342 * zlc->fullzones, so that subsequent attempts to allocate a page
1343 * from that zone don't waste time re-examining it.
1344 */
1345static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1346{
1347 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1348 int i; /* index of *z in zonelist zones */
1349
1350 zlc = zonelist->zlcache_ptr;
1351 if (!zlc)
1352 return;
1353
1354 i = z - zonelist->zones;
1355
1356 set_bit(i, zlc->fullzones);
1357}
1358
1359#else /* CONFIG_NUMA */
1360
1361static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1362{
1363 return NULL;
1364}
1365
1366static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1367 nodemask_t *allowednodes)
1368{
1369 return 1;
1370}
1371
1372static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1373{
1374}
1375#endif /* CONFIG_NUMA */
1376
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001377/*
Paul Jackson0798e512006-12-06 20:31:38 -08001378 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001379 * a page.
1380 */
1381static struct page *
1382get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1383 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -07001384{
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001385 struct zone **z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001386 struct page *page = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001387 int classzone_idx = zone_idx(zonelist->zones[0]);
Christoph Lameter1192d522006-09-25 23:31:45 -07001388 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001389 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1390 int zlc_active = 0; /* set if using zonelist_cache */
1391 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gormanb377fd32007-08-22 14:02:05 -07001392 enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001393
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001394zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001395 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001396 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001397 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1398 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001399 z = zonelist->zones;
1400
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001401 do {
Mel Gormanb377fd32007-08-22 14:02:05 -07001402 /*
1403 * In NUMA, this could be a policy zonelist which contains
1404 * zones that may not be allowed by the current gfp_mask.
1405 * Check the zone is allowed by the current flags
1406 */
1407 if (unlikely(alloc_should_filter_zonelist(zonelist))) {
1408 if (highest_zoneidx == -1)
1409 highest_zoneidx = gfp_zone(gfp_mask);
1410 if (zone_idx(*z) > highest_zoneidx)
1411 continue;
1412 }
1413
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001414 if (NUMA_BUILD && zlc_active &&
1415 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1416 continue;
Christoph Lameter1192d522006-09-25 23:31:45 -07001417 zone = *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001418 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001419 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001420 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001421
1422 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001423 unsigned long mark;
1424 if (alloc_flags & ALLOC_WMARK_MIN)
Christoph Lameter1192d522006-09-25 23:31:45 -07001425 mark = zone->pages_min;
Nick Piggin31488902005-11-28 13:44:03 -08001426 else if (alloc_flags & ALLOC_WMARK_LOW)
Christoph Lameter1192d522006-09-25 23:31:45 -07001427 mark = zone->pages_low;
Nick Piggin31488902005-11-28 13:44:03 -08001428 else
Christoph Lameter1192d522006-09-25 23:31:45 -07001429 mark = zone->pages_high;
Paul Jackson0798e512006-12-06 20:31:38 -08001430 if (!zone_watermark_ok(zone, order, mark,
1431 classzone_idx, alloc_flags)) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001432 if (!zone_reclaim_mode ||
Christoph Lameter1192d522006-09-25 23:31:45 -07001433 !zone_reclaim(zone, gfp_mask, order))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001434 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001435 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001436 }
1437
Christoph Lameter1192d522006-09-25 23:31:45 -07001438 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
Paul Jackson0798e512006-12-06 20:31:38 -08001439 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001440 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001441this_zone_full:
1442 if (NUMA_BUILD)
1443 zlc_mark_zone_full(zonelist, z);
1444try_next_zone:
1445 if (NUMA_BUILD && !did_zlc_setup) {
1446 /* we do zlc_setup after the first zone is tried */
1447 allowednodes = zlc_setup(zonelist, alloc_flags);
1448 zlc_active = 1;
1449 did_zlc_setup = 1;
1450 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001451 } while (*(++z) != NULL);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001452
1453 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1454 /* Disable zlc cache for second zonelist scan */
1455 zlc_active = 0;
1456 goto zonelist_scan;
1457 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001458 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001459}
1460
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461/*
1462 * This is the 'heart' of the zoned buddy allocator.
1463 */
1464struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +01001465__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 struct zonelist *zonelist)
1467{
Al Viro260b2362005-10-21 03:22:44 -04001468 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001469 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 struct page *page;
1471 struct reclaim_state reclaim_state;
1472 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001474 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 int did_some_progress;
1476
1477 might_sleep_if(wait);
1478
Akinobu Mita933e3122006-12-08 02:39:45 -08001479 if (should_fail_alloc_page(gfp_mask, order))
1480 return NULL;
1481
Jens Axboe6b1de912005-11-17 21:35:02 +01001482restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001483 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001485 if (unlikely(*z == NULL)) {
Christoph Lameter523b9452007-10-16 01:25:37 -07001486 /*
1487 * Happens if we have an empty zonelist as a result of
1488 * GFP_THISNODE being used on a memoryless node
1489 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 return NULL;
1491 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001492
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001493 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001494 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001495 if (page)
1496 goto got_pg;
1497
Christoph Lameter952f3b52006-12-06 20:33:26 -08001498 /*
1499 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1500 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1501 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1502 * using a larger set of nodes after it has established that the
1503 * allowed per node queues are empty and that nodes are
1504 * over allocated.
1505 */
1506 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1507 goto nopage;
1508
Paul Jackson0798e512006-12-06 20:31:38 -08001509 for (z = zonelist->zones; *z; z++)
Chris Wright43b0bc02006-06-25 05:47:55 -07001510 wakeup_kswapd(*z, order);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001511
Paul Jackson9bf22292005-09-06 15:18:12 -07001512 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001513 * OK, we're below the kswapd watermark and have kicked background
1514 * reclaim. Now things get more complex, so set up alloc_flags according
1515 * to how we want to proceed.
1516 *
1517 * The caller may dip into page reserves a bit more if the caller
1518 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -08001519 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1520 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -07001521 */
Nick Piggin31488902005-11-28 13:44:03 -08001522 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001523 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1524 alloc_flags |= ALLOC_HARDER;
1525 if (gfp_mask & __GFP_HIGH)
1526 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -07001527 if (wait)
1528 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
1530 /*
1531 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001532 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 *
1534 * This is the last chance, in general, before the goto nopage.
1535 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -07001536 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001538 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1539 if (page)
1540 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541
1542 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -07001543
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001544rebalance:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001545 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1546 && !in_interrupt()) {
1547 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -08001548nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001549 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001550 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -08001551 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001552 if (page)
1553 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -08001554 if (gfp_mask & __GFP_NOFAIL) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001555 congestion_wait(WRITE, HZ/50);
Kirill Korotaev885036d2005-11-13 16:06:41 -08001556 goto nofail_alloc;
1557 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 }
1559 goto nopage;
1560 }
1561
1562 /* Atomic allocations - we can't balance anything */
1563 if (!wait)
1564 goto nopage;
1565
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 cond_resched();
1567
1568 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001569 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 p->flags |= PF_MEMALLOC;
1571 reclaim_state.reclaimed_slab = 0;
1572 p->reclaim_state = &reclaim_state;
1573
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001574 did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
1576 p->reclaim_state = NULL;
1577 p->flags &= ~PF_MEMALLOC;
1578
1579 cond_resched();
1580
Mel Gormane2c55dc2007-10-16 01:25:50 -07001581 if (order != 0)
1582 drain_all_local_pages();
1583
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001585 page = get_page_from_freelist(gfp_mask, order,
1586 zonelist, alloc_flags);
1587 if (page)
1588 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
David Rientjesff0ceb92007-10-16 23:25:56 -07001590 if (!try_set_zone_oom(zonelist)) {
1591 schedule_timeout_uninterruptible(1);
1592 goto restart;
1593 }
1594
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 /*
1596 * Go through the zonelist yet one more time, keep
1597 * very high watermark here, this is only to catch
1598 * a parallel oom killing, we must fail if we're still
1599 * under heavy pressure.
1600 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001601 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001602 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
David Rientjesff0ceb92007-10-16 23:25:56 -07001603 if (page) {
1604 clear_zonelist_oom(zonelist);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001605 goto got_pg;
David Rientjesff0ceb92007-10-16 23:25:56 -07001606 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607
Mel Gormana8bbf722007-07-31 00:37:30 -07001608 /* The OOM killer will not help higher order allocs so fail */
David Rientjesff0ceb92007-10-16 23:25:56 -07001609 if (order > PAGE_ALLOC_COSTLY_ORDER) {
1610 clear_zonelist_oom(zonelist);
Mel Gormana8bbf722007-07-31 00:37:30 -07001611 goto nopage;
David Rientjesff0ceb92007-10-16 23:25:56 -07001612 }
Mel Gormana8bbf722007-07-31 00:37:30 -07001613
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001614 out_of_memory(zonelist, gfp_mask, order);
David Rientjesff0ceb92007-10-16 23:25:56 -07001615 clear_zonelist_oom(zonelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 goto restart;
1617 }
1618
1619 /*
1620 * Don't let big-order allocations loop unless the caller explicitly
1621 * requests that. Wait for some write requests to complete then retry.
1622 *
1623 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1624 * <= 3, but that may not be true in other implementations.
1625 */
1626 do_retry = 0;
1627 if (!(gfp_mask & __GFP_NORETRY)) {
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001628 if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
1629 (gfp_mask & __GFP_REPEAT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 do_retry = 1;
1631 if (gfp_mask & __GFP_NOFAIL)
1632 do_retry = 1;
1633 }
1634 if (do_retry) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001635 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 goto rebalance;
1637 }
1638
1639nopage:
1640 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1641 printk(KERN_WARNING "%s: page allocation failure."
1642 " order:%d, mode:0x%x\n",
1643 p->comm, order, gfp_mask);
1644 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001645 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 return page;
1649}
1650
1651EXPORT_SYMBOL(__alloc_pages);
1652
1653/*
1654 * Common helper functions.
1655 */
Al Virodd0fc662005-10-07 07:46:04 +01001656fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657{
1658 struct page * page;
1659 page = alloc_pages(gfp_mask, order);
1660 if (!page)
1661 return 0;
1662 return (unsigned long) page_address(page);
1663}
1664
1665EXPORT_SYMBOL(__get_free_pages);
1666
Al Virodd0fc662005-10-07 07:46:04 +01001667fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668{
1669 struct page * page;
1670
1671 /*
1672 * get_zeroed_page() returns a 32-bit address, which cannot represent
1673 * a highmem page
1674 */
Nick Piggin725d7042006-09-25 23:30:55 -07001675 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
1677 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1678 if (page)
1679 return (unsigned long) page_address(page);
1680 return 0;
1681}
1682
1683EXPORT_SYMBOL(get_zeroed_page);
1684
1685void __pagevec_free(struct pagevec *pvec)
1686{
1687 int i = pagevec_count(pvec);
1688
1689 while (--i >= 0)
1690 free_hot_cold_page(pvec->pages[i], pvec->cold);
1691}
1692
1693fastcall void __free_pages(struct page *page, unsigned int order)
1694{
Nick Pigginb5810032005-10-29 18:16:12 -07001695 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 if (order == 0)
1697 free_hot_page(page);
1698 else
1699 __free_pages_ok(page, order);
1700 }
1701}
1702
1703EXPORT_SYMBOL(__free_pages);
1704
1705fastcall void free_pages(unsigned long addr, unsigned int order)
1706{
1707 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001708 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 __free_pages(virt_to_page((void *)addr), order);
1710 }
1711}
1712
1713EXPORT_SYMBOL(free_pages);
1714
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715static unsigned int nr_free_zone_pages(int offset)
1716{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001717 /* Just pick one node, since fallback list is circular */
1718 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 unsigned int sum = 0;
1720
Martin J. Blighe310fd42005-07-29 22:59:18 -07001721 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1722 struct zone **zonep = zonelist->zones;
1723 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724
Martin J. Blighe310fd42005-07-29 22:59:18 -07001725 for (zone = *zonep++; zone; zone = *zonep++) {
1726 unsigned long size = zone->present_pages;
1727 unsigned long high = zone->pages_high;
1728 if (size > high)
1729 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 }
1731
1732 return sum;
1733}
1734
1735/*
1736 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1737 */
1738unsigned int nr_free_buffer_pages(void)
1739{
Al Viroaf4ca452005-10-21 02:55:38 -04001740 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741}
Meelap Shahc2f1a552007-07-17 04:04:39 -07001742EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
1744/*
1745 * Amount of free RAM allocatable within all zones
1746 */
1747unsigned int nr_free_pagecache_pages(void)
1748{
Mel Gorman2a1e2742007-07-17 04:03:12 -07001749 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001751
1752static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001754 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08001755 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758void si_meminfo(struct sysinfo *val)
1759{
1760 val->totalram = totalram_pages;
1761 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001762 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 val->totalhigh = totalhigh_pages;
1765 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 val->mem_unit = PAGE_SIZE;
1767}
1768
1769EXPORT_SYMBOL(si_meminfo);
1770
1771#ifdef CONFIG_NUMA
1772void si_meminfo_node(struct sysinfo *val, int nid)
1773{
1774 pg_data_t *pgdat = NODE_DATA(nid);
1775
1776 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001777 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001778#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001780 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1781 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001782#else
1783 val->totalhigh = 0;
1784 val->freehigh = 0;
1785#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 val->mem_unit = PAGE_SIZE;
1787}
1788#endif
1789
1790#define K(x) ((x) << (PAGE_SHIFT-10))
1791
1792/*
1793 * Show free area list (used inside shift_scroll-lock stuff)
1794 * We also calculate the percentage fragmentation. We do this by counting the
1795 * memory on each free list with the exception of the first item on the list.
1796 */
1797void show_free_areas(void)
1798{
Jes Sorensenc7241912006-09-27 01:50:05 -07001799 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 struct zone *zone;
1801
1802 for_each_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07001803 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07001805
1806 show_node(zone);
1807 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
Dave Jones6b482c62005-11-10 15:45:56 -05001809 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 struct per_cpu_pageset *pageset;
1811
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001812 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813
Jes Sorensenc7241912006-09-27 01:50:05 -07001814 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1815 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1816 cpu, pageset->pcp[0].high,
1817 pageset->pcp[0].batch, pageset->pcp[0].count,
1818 pageset->pcp[1].high, pageset->pcp[1].batch,
1819 pageset->pcp[1].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 }
1821 }
1822
Andrew Mortona25700a2007-02-08 14:20:40 -08001823 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
Christoph Lameterd23ad422007-02-10 01:43:02 -08001824 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
Christoph Lameter65e458d42007-02-10 01:43:05 -08001825 global_page_state(NR_ACTIVE),
1826 global_page_state(NR_INACTIVE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07001827 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07001828 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001829 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08001830 global_page_state(NR_FREE_PAGES),
Christoph Lameter972d1a72006-09-25 23:31:51 -07001831 global_page_state(NR_SLAB_RECLAIMABLE) +
1832 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001833 global_page_state(NR_FILE_MAPPED),
Andrew Mortona25700a2007-02-08 14:20:40 -08001834 global_page_state(NR_PAGETABLE),
1835 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836
1837 for_each_zone(zone) {
1838 int i;
1839
Jes Sorensenc7241912006-09-27 01:50:05 -07001840 if (!populated_zone(zone))
1841 continue;
1842
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 show_node(zone);
1844 printk("%s"
1845 " free:%lukB"
1846 " min:%lukB"
1847 " low:%lukB"
1848 " high:%lukB"
1849 " active:%lukB"
1850 " inactive:%lukB"
1851 " present:%lukB"
1852 " pages_scanned:%lu"
1853 " all_unreclaimable? %s"
1854 "\n",
1855 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08001856 K(zone_page_state(zone, NR_FREE_PAGES)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 K(zone->pages_min),
1858 K(zone->pages_low),
1859 K(zone->pages_high),
Christoph Lameterc8785382007-02-10 01:43:01 -08001860 K(zone_page_state(zone, NR_ACTIVE)),
1861 K(zone_page_state(zone, NR_INACTIVE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 K(zone->present_pages),
1863 zone->pages_scanned,
David Rientjese815af92007-10-16 23:25:54 -07001864 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 );
1866 printk("lowmem_reserve[]:");
1867 for (i = 0; i < MAX_NR_ZONES; i++)
1868 printk(" %lu", zone->lowmem_reserve[i]);
1869 printk("\n");
1870 }
1871
1872 for_each_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001873 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874
Jes Sorensenc7241912006-09-27 01:50:05 -07001875 if (!populated_zone(zone))
1876 continue;
1877
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 show_node(zone);
1879 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
1881 spin_lock_irqsave(&zone->lock, flags);
1882 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001883 nr[order] = zone->free_area[order].nr_free;
1884 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 }
1886 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001887 for (order = 0; order < MAX_ORDER; order++)
1888 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 printk("= %lukB\n", K(total));
1890 }
1891
1892 show_swap_cache_info();
1893}
1894
1895/*
1896 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001897 *
1898 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001900static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
1901 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902{
Christoph Lameter1a932052006-01-06 00:11:16 -08001903 struct zone *zone;
1904
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001905 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001906 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001907
1908 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001909 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08001910 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001911 if (populated_zone(zone)) {
Christoph Lameter070f8032006-01-06 00:11:19 -08001912 zonelist->zones[nr_zones++] = zone;
1913 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08001915
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001916 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08001917 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918}
1919
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001920
1921/*
1922 * zonelist_order:
1923 * 0 = automatic detection of better ordering.
1924 * 1 = order by ([node] distance, -zonetype)
1925 * 2 = order by (-zonetype, [node] distance)
1926 *
1927 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1928 * the same zonelist. So only NUMA can configure this param.
1929 */
1930#define ZONELIST_ORDER_DEFAULT 0
1931#define ZONELIST_ORDER_NODE 1
1932#define ZONELIST_ORDER_ZONE 2
1933
1934/* zonelist order in the kernel.
1935 * set_zonelist_order() will set this to NODE or ZONE.
1936 */
1937static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
1938static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
1939
1940
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001942/* The value user specified ....changed by config */
1943static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1944/* string for sysctl */
1945#define NUMA_ZONELIST_ORDER_LEN 16
1946char numa_zonelist_order[16] = "default";
1947
1948/*
1949 * interface for configure zonelist ordering.
1950 * command line option "numa_zonelist_order"
1951 * = "[dD]efault - default, automatic configuration.
1952 * = "[nN]ode - order by node locality, then by zone within node
1953 * = "[zZ]one - order by zone, then by locality within zone
1954 */
1955
1956static int __parse_numa_zonelist_order(char *s)
1957{
1958 if (*s == 'd' || *s == 'D') {
1959 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1960 } else if (*s == 'n' || *s == 'N') {
1961 user_zonelist_order = ZONELIST_ORDER_NODE;
1962 } else if (*s == 'z' || *s == 'Z') {
1963 user_zonelist_order = ZONELIST_ORDER_ZONE;
1964 } else {
1965 printk(KERN_WARNING
1966 "Ignoring invalid numa_zonelist_order value: "
1967 "%s\n", s);
1968 return -EINVAL;
1969 }
1970 return 0;
1971}
1972
1973static __init int setup_numa_zonelist_order(char *s)
1974{
1975 if (s)
1976 return __parse_numa_zonelist_order(s);
1977 return 0;
1978}
1979early_param("numa_zonelist_order", setup_numa_zonelist_order);
1980
1981/*
1982 * sysctl handler for numa_zonelist_order
1983 */
1984int numa_zonelist_order_handler(ctl_table *table, int write,
1985 struct file *file, void __user *buffer, size_t *length,
1986 loff_t *ppos)
1987{
1988 char saved_string[NUMA_ZONELIST_ORDER_LEN];
1989 int ret;
1990
1991 if (write)
1992 strncpy(saved_string, (char*)table->data,
1993 NUMA_ZONELIST_ORDER_LEN);
1994 ret = proc_dostring(table, write, file, buffer, length, ppos);
1995 if (ret)
1996 return ret;
1997 if (write) {
1998 int oldval = user_zonelist_order;
1999 if (__parse_numa_zonelist_order((char*)table->data)) {
2000 /*
2001 * bogus value. restore saved string
2002 */
2003 strncpy((char*)table->data, saved_string,
2004 NUMA_ZONELIST_ORDER_LEN);
2005 user_zonelist_order = oldval;
2006 } else if (oldval != user_zonelist_order)
2007 build_all_zonelists();
2008 }
2009 return 0;
2010}
2011
2012
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013#define MAX_NODE_LOAD (num_online_nodes())
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002014static int node_load[MAX_NUMNODES];
2015
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002017 * 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 -07002018 * @node: node whose fallback list we're appending
2019 * @used_node_mask: nodemask_t of already used nodes
2020 *
2021 * We use a number of factors to determine which is the next node that should
2022 * appear on a given node's fallback list. The node should not have appeared
2023 * already in @node's fallback list, and it should be the next closest node
2024 * according to the distance array (which contains arbitrary distance values
2025 * from each node to each node in the system), and should also prefer nodes
2026 * with no CPUs, since presumably they'll have very little allocation pressure
2027 * on them otherwise.
2028 * It returns -1 if no node is found.
2029 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002030static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002032 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 int min_val = INT_MAX;
2034 int best_node = -1;
2035
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002036 /* Use the local node if we haven't already */
2037 if (!node_isset(node, *used_node_mask)) {
2038 node_set(node, *used_node_mask);
2039 return node;
2040 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002042 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002043 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
2045 /* Don't want a node to appear more than once */
2046 if (node_isset(n, *used_node_mask))
2047 continue;
2048
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 /* Use the distance array to find the distance */
2050 val = node_distance(node, n);
2051
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002052 /* Penalize nodes under us ("prefer the next node") */
2053 val += (n < node);
2054
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 /* Give preference to headless and unused nodes */
2056 tmp = node_to_cpumask(n);
2057 if (!cpus_empty(tmp))
2058 val += PENALTY_FOR_NODE_WITH_CPUS;
2059
2060 /* Slight preference for less loaded node */
2061 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2062 val += node_load[n];
2063
2064 if (val < min_val) {
2065 min_val = val;
2066 best_node = n;
2067 }
2068 }
2069
2070 if (best_node >= 0)
2071 node_set(best_node, *used_node_mask);
2072
2073 return best_node;
2074}
2075
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002076
2077/*
2078 * Build zonelists ordered by node and zones within node.
2079 * This results in maximum locality--normal zone overflows into local
2080 * DMA zone, if any--but risks exhausting DMA zone.
2081 */
2082static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083{
Christoph Lameter19655d32006-09-25 23:31:19 -07002084 enum zone_type i;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002085 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002087
2088 for (i = 0; i < MAX_NR_ZONES; i++) {
2089 zonelist = pgdat->node_zonelists + i;
2090 for (j = 0; zonelist->zones[j] != NULL; j++)
2091 ;
2092 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2093 zonelist->zones[j] = NULL;
2094 }
2095}
2096
2097/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002098 * Build gfp_thisnode zonelists
2099 */
2100static void build_thisnode_zonelists(pg_data_t *pgdat)
2101{
2102 enum zone_type i;
2103 int j;
2104 struct zonelist *zonelist;
2105
2106 for (i = 0; i < MAX_NR_ZONES; i++) {
2107 zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i;
2108 j = build_zonelists_node(pgdat, zonelist, 0, i);
2109 zonelist->zones[j] = NULL;
2110 }
2111}
2112
2113/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002114 * Build zonelists ordered by zone and nodes within zones.
2115 * This results in conserving DMA zone[s] until all Normal memory is
2116 * exhausted, but results in overflowing to remote node while memory
2117 * may still exist in local DMA zone.
2118 */
2119static int node_order[MAX_NUMNODES];
2120
2121static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2122{
2123 enum zone_type i;
2124 int pos, j, node;
2125 int zone_type; /* needs to be signed */
2126 struct zone *z;
2127 struct zonelist *zonelist;
2128
2129 for (i = 0; i < MAX_NR_ZONES; i++) {
2130 zonelist = pgdat->node_zonelists + i;
2131 pos = 0;
2132 for (zone_type = i; zone_type >= 0; zone_type--) {
2133 for (j = 0; j < nr_nodes; j++) {
2134 node = node_order[j];
2135 z = &NODE_DATA(node)->node_zones[zone_type];
2136 if (populated_zone(z)) {
2137 zonelist->zones[pos++] = z;
2138 check_highest_zone(zone_type);
2139 }
2140 }
2141 }
2142 zonelist->zones[pos] = NULL;
2143 }
2144}
2145
2146static int default_zonelist_order(void)
2147{
2148 int nid, zone_type;
2149 unsigned long low_kmem_size,total_size;
2150 struct zone *z;
2151 int average_size;
2152 /*
2153 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2154 * If they are really small and used heavily, the system can fall
2155 * into OOM very easily.
2156 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2157 */
2158 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2159 low_kmem_size = 0;
2160 total_size = 0;
2161 for_each_online_node(nid) {
2162 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2163 z = &NODE_DATA(nid)->node_zones[zone_type];
2164 if (populated_zone(z)) {
2165 if (zone_type < ZONE_NORMAL)
2166 low_kmem_size += z->present_pages;
2167 total_size += z->present_pages;
2168 }
2169 }
2170 }
2171 if (!low_kmem_size || /* there are no DMA area. */
2172 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2173 return ZONELIST_ORDER_NODE;
2174 /*
2175 * look into each node's config.
2176 * If there is a node whose DMA/DMA32 memory is very big area on
2177 * local memory, NODE_ORDER may be suitable.
2178 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002179 average_size = total_size /
2180 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002181 for_each_online_node(nid) {
2182 low_kmem_size = 0;
2183 total_size = 0;
2184 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2185 z = &NODE_DATA(nid)->node_zones[zone_type];
2186 if (populated_zone(z)) {
2187 if (zone_type < ZONE_NORMAL)
2188 low_kmem_size += z->present_pages;
2189 total_size += z->present_pages;
2190 }
2191 }
2192 if (low_kmem_size &&
2193 total_size > average_size && /* ignore small node */
2194 low_kmem_size > total_size * 70/100)
2195 return ZONELIST_ORDER_NODE;
2196 }
2197 return ZONELIST_ORDER_ZONE;
2198}
2199
2200static void set_zonelist_order(void)
2201{
2202 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2203 current_zonelist_order = default_zonelist_order();
2204 else
2205 current_zonelist_order = user_zonelist_order;
2206}
2207
2208static void build_zonelists(pg_data_t *pgdat)
2209{
2210 int j, node, load;
2211 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002213 int local_node, prev_node;
2214 struct zonelist *zonelist;
2215 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216
2217 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002218 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 zonelist = pgdat->node_zonelists + i;
2220 zonelist->zones[0] = NULL;
2221 }
2222
2223 /* NUMA-aware ordering of nodes */
2224 local_node = pgdat->node_id;
2225 load = num_online_nodes();
2226 prev_node = local_node;
2227 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002228
2229 memset(node_load, 0, sizeof(node_load));
2230 memset(node_order, 0, sizeof(node_order));
2231 j = 0;
2232
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002234 int distance = node_distance(local_node, node);
2235
2236 /*
2237 * If another node is sufficiently far away then it is better
2238 * to reclaim pages in a zone before going off node.
2239 */
2240 if (distance > RECLAIM_DISTANCE)
2241 zone_reclaim_mode = 1;
2242
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 /*
2244 * We don't want to pressure a particular node.
2245 * So adding penalty to the first node in same
2246 * distance group to make it round-robin.
2247 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002248 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002249 node_load[node] = load;
2250
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 prev_node = node;
2252 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002253 if (order == ZONELIST_ORDER_NODE)
2254 build_zonelists_in_node_order(pgdat, node);
2255 else
2256 node_order[j++] = node; /* remember order */
2257 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002259 if (order == ZONELIST_ORDER_ZONE) {
2260 /* calculate node order -- i.e., DMA last! */
2261 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002263
2264 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265}
2266
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002267/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002268static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002269{
2270 int i;
2271
2272 for (i = 0; i < MAX_NR_ZONES; i++) {
2273 struct zonelist *zonelist;
2274 struct zonelist_cache *zlc;
2275 struct zone **z;
2276
2277 zonelist = pgdat->node_zonelists + i;
2278 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2279 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2280 for (z = zonelist->zones; *z; z++)
2281 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
2282 }
2283}
2284
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002285
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286#else /* CONFIG_NUMA */
2287
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002288static void set_zonelist_order(void)
2289{
2290 current_zonelist_order = ZONELIST_ORDER_ZONE;
2291}
2292
2293static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294{
Christoph Lameter19655d32006-09-25 23:31:19 -07002295 int node, local_node;
2296 enum zone_type i,j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297
2298 local_node = pgdat->node_id;
Christoph Lameter19655d32006-09-25 23:31:19 -07002299 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 struct zonelist *zonelist;
2301
2302 zonelist = pgdat->node_zonelists + i;
2303
Christoph Lameter19655d32006-09-25 23:31:19 -07002304 j = build_zonelists_node(pgdat, zonelist, 0, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 /*
2306 * Now we build the zonelist so that it contains the zones
2307 * of all the other nodes.
2308 * We don't want to pressure a particular node, so when
2309 * building the zones for node N, we make sure that the
2310 * zones coming right after the local ones are those from
2311 * node N+1 (modulo N)
2312 */
2313 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2314 if (!node_online(node))
2315 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07002316 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 }
2318 for (node = 0; node < local_node; node++) {
2319 if (!node_online(node))
2320 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07002321 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 }
2323
2324 zonelist->zones[j] = NULL;
2325 }
2326}
2327
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002328/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002329static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002330{
2331 int i;
2332
2333 for (i = 0; i < MAX_NR_ZONES; i++)
2334 pgdat->node_zonelists[i].zlcache_ptr = NULL;
2335}
2336
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337#endif /* CONFIG_NUMA */
2338
Yasunori Goto68113782006-06-23 02:03:11 -07002339/* return values int ....just for stop_machine_run() */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002340static int __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341{
Yasunori Goto68113782006-06-23 02:03:11 -07002342 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002343
2344 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07002345 pg_data_t *pgdat = NODE_DATA(nid);
2346
2347 build_zonelists(pgdat);
2348 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002349 }
Yasunori Goto68113782006-06-23 02:03:11 -07002350 return 0;
2351}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002353void build_all_zonelists(void)
Yasunori Goto68113782006-06-23 02:03:11 -07002354{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002355 set_zonelist_order();
2356
Yasunori Goto68113782006-06-23 02:03:11 -07002357 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07002358 __build_all_zonelists(NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07002359 cpuset_init_current_mems_allowed();
2360 } else {
2361 /* we have to stop all cpus to guaranntee there is no user
2362 of zonelist */
2363 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
2364 /* cpuset refresh routine should be here */
2365 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07002366 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002367 /*
2368 * Disable grouping by mobility if the number of pages in the
2369 * system is too low to allow the mechanism to work. It would be
2370 * more accurate, but expensive to check per-zone. This check is
2371 * made on memory-hotadd so a system can start with mobility
2372 * disabled and enable it later
2373 */
Mel Gormand9c23402007-10-16 01:26:01 -07002374 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002375 page_group_by_mobility_disabled = 1;
2376 else
2377 page_group_by_mobility_disabled = 0;
2378
2379 printk("Built %i zonelists in %s order, mobility grouping %s. "
2380 "Total pages: %ld\n",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002381 num_online_nodes(),
2382 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002383 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002384 vm_total_pages);
2385#ifdef CONFIG_NUMA
2386 printk("Policy zone: %s\n", zone_names[policy_zone]);
2387#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388}
2389
2390/*
2391 * Helper functions to size the waitqueue hash table.
2392 * Essentially these want to choose hash table sizes sufficiently
2393 * large so that collisions trying to wait on pages are rare.
2394 * But in fact, the number of active page waitqueues on typical
2395 * systems is ridiculously low, less than 200. So this is even
2396 * conservative, even though it seems large.
2397 *
2398 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2399 * waitqueues, i.e. the size of the waitq table given the number of pages.
2400 */
2401#define PAGES_PER_WAITQUEUE 256
2402
Yasunori Gotocca448f2006-06-23 02:03:10 -07002403#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07002404static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405{
2406 unsigned long size = 1;
2407
2408 pages /= PAGES_PER_WAITQUEUE;
2409
2410 while (size < pages)
2411 size <<= 1;
2412
2413 /*
2414 * Once we have dozens or even hundreds of threads sleeping
2415 * on IO we've got bigger problems than wait queue collision.
2416 * Limit the size of the wait table to a reasonable size.
2417 */
2418 size = min(size, 4096UL);
2419
2420 return max(size, 4UL);
2421}
Yasunori Gotocca448f2006-06-23 02:03:10 -07002422#else
2423/*
2424 * A zone's size might be changed by hot-add, so it is not possible to determine
2425 * a suitable size for its wait_table. So we use the maximum size now.
2426 *
2427 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2428 *
2429 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2430 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2431 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2432 *
2433 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2434 * or more by the traditional way. (See above). It equals:
2435 *
2436 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2437 * ia64(16K page size) : = ( 8G + 4M)byte.
2438 * powerpc (64K page size) : = (32G +16M)byte.
2439 */
2440static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2441{
2442 return 4096UL;
2443}
2444#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445
2446/*
2447 * This is an integer logarithm so that shifts can be used later
2448 * to extract the more random high bits from the multiplicative
2449 * hash function before the remainder is taken.
2450 */
2451static inline unsigned long wait_table_bits(unsigned long size)
2452{
2453 return ffz(~size);
2454}
2455
2456#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2457
Mel Gorman56fd56b2007-10-16 01:25:58 -07002458/*
Mel Gormand9c23402007-10-16 01:26:01 -07002459 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman56fd56b2007-10-16 01:25:58 -07002460 * of blocks reserved is based on zone->pages_min. The memory within the
2461 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2462 * higher will lead to a bigger reserve which will get freed as contiguous
2463 * blocks as reclaim kicks in
2464 */
2465static void setup_zone_migrate_reserve(struct zone *zone)
2466{
2467 unsigned long start_pfn, pfn, end_pfn;
2468 struct page *page;
2469 unsigned long reserve, block_migratetype;
2470
2471 /* Get the start pfn, end pfn and the number of blocks to reserve */
2472 start_pfn = zone->zone_start_pfn;
2473 end_pfn = start_pfn + zone->spanned_pages;
Mel Gormand9c23402007-10-16 01:26:01 -07002474 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2475 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002476
Mel Gormand9c23402007-10-16 01:26:01 -07002477 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07002478 if (!pfn_valid(pfn))
2479 continue;
2480 page = pfn_to_page(pfn);
2481
2482 /* Blocks with reserved pages will never free, skip them. */
2483 if (PageReserved(page))
2484 continue;
2485
2486 block_migratetype = get_pageblock_migratetype(page);
2487
2488 /* If this block is reserved, account for it */
2489 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2490 reserve--;
2491 continue;
2492 }
2493
2494 /* Suitable for reserving if this block is movable */
2495 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2496 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2497 move_freepages_block(zone, page, MIGRATE_RESERVE);
2498 reserve--;
2499 continue;
2500 }
2501
2502 /*
2503 * If the reserve is met and this is a previous reserved block,
2504 * take it back
2505 */
2506 if (block_migratetype == MIGRATE_RESERVE) {
2507 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2508 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2509 }
2510 }
2511}
Mel Gormanac0e5b72007-10-16 01:25:58 -07002512
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513/*
2514 * Initially all pages are reserved - free ones are freed
2515 * up by free_all_bootmem() once the early boot process is
2516 * done. Non-atomic initialization, single-pass.
2517 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002518void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08002519 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07002522 unsigned long end_pfn = start_pfn + size;
2523 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524
Greg Ungerercbe8dd42006-01-12 01:05:24 -08002525 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08002526 /*
2527 * There can be holes in boot-time mem_map[]s
2528 * handed to this function. They do not
2529 * exist on hotplugged memory.
2530 */
2531 if (context == MEMMAP_EARLY) {
2532 if (!early_pfn_valid(pfn))
2533 continue;
2534 if (!early_pfn_in_nid(pfn, nid))
2535 continue;
2536 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002537 page = pfn_to_page(pfn);
2538 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08002539 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 reset_page_mapcount(page);
2541 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002542
2543 /*
2544 * Mark the block movable so that blocks are reserved for
2545 * movable at startup. This will force kernel allocations
2546 * to reserve their blocks rather than leaking throughout
2547 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07002548 * kernel allocations are made. Later some blocks near
2549 * the start are marked MIGRATE_RESERVE by
2550 * setup_zone_migrate_reserve()
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002551 */
Mel Gormand9c23402007-10-16 01:26:01 -07002552 if ((pfn & (pageblock_nr_pages-1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07002553 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002554
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 INIT_LIST_HEAD(&page->lru);
2556#ifdef WANT_PAGE_VIRTUAL
2557 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2558 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07002559 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 }
2562}
2563
Paul Mundt6ea6e682007-07-15 23:38:20 -07002564static void __meminit zone_init_free_lists(struct pglist_data *pgdat,
2565 struct zone *zone, unsigned long size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002567 int order, t;
2568 for_each_migratetype_order(order, t) {
2569 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 zone->free_area[order].nr_free = 0;
2571 }
2572}
2573
2574#ifndef __HAVE_ARCH_MEMMAP_INIT
2575#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08002576 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577#endif
2578
Paul Mundtd09c6b82007-06-14 15:13:16 +09002579static int __devinit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002580{
2581 int batch;
2582
2583 /*
2584 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07002585 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002586 *
2587 * OK, so we don't know how big the cache is. So guess.
2588 */
2589 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07002590 if (batch * PAGE_SIZE > 512 * 1024)
2591 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002592 batch /= 4; /* We effectively *= 4 below */
2593 if (batch < 1)
2594 batch = 1;
2595
2596 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002597 * Clamp the batch to a 2^n - 1 value. Having a power
2598 * of 2 value was found to be more likely to have
2599 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002600 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002601 * For example if 2 tasks are alternately allocating
2602 * batches of pages, one task can end up with a lot
2603 * of pages of one half of the possible page colors
2604 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002605 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002606 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07002607
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002608 return batch;
2609}
2610
Christoph Lameter2caaad42005-06-21 17:15:00 -07002611inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2612{
2613 struct per_cpu_pages *pcp;
2614
Magnus Damm1c6fe942005-10-26 01:58:59 -07002615 memset(p, 0, sizeof(*p));
2616
Christoph Lameter2caaad42005-06-21 17:15:00 -07002617 pcp = &p->pcp[0]; /* hot */
2618 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002619 pcp->high = 6 * batch;
2620 pcp->batch = max(1UL, 1 * batch);
2621 INIT_LIST_HEAD(&pcp->list);
2622
2623 pcp = &p->pcp[1]; /* cold*/
2624 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002625 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07002626 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07002627 INIT_LIST_HEAD(&pcp->list);
2628}
2629
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002630/*
2631 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2632 * to the value high for the pageset p.
2633 */
2634
2635static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2636 unsigned long high)
2637{
2638 struct per_cpu_pages *pcp;
2639
2640 pcp = &p->pcp[0]; /* hot list */
2641 pcp->high = high;
2642 pcp->batch = max(1UL, high/4);
2643 if ((high/4) > (PAGE_SHIFT * 8))
2644 pcp->batch = PAGE_SHIFT * 8;
2645}
2646
2647
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002648#ifdef CONFIG_NUMA
2649/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07002650 * Boot pageset table. One per cpu which is going to be used for all
2651 * zones and all nodes. The parameters will be set in such a way
2652 * that an item put on a list will immediately be handed over to
2653 * the buddy list. This is safe since pageset manipulation is done
2654 * with interrupts disabled.
2655 *
2656 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07002657 *
2658 * The boot_pagesets must be kept even after bootup is complete for
2659 * unused processors and/or zones. They do play a role for bootstrapping
2660 * hotplugged processors.
2661 *
2662 * zoneinfo_show() and maybe other functions do
2663 * not check if the processor is online before following the pageset pointer.
2664 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07002665 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08002666static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07002667
2668/*
2669 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002670 * per cpu pageset array in struct zone.
2671 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08002672static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002673{
2674 struct zone *zone, *dzone;
Christoph Lameter37c07082007-10-16 01:25:36 -07002675 int node = cpu_to_node(cpu);
2676
2677 node_set_state(node, N_CPU); /* this node has a cpu */
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002678
2679 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002680
Christoph Lameter66a55032006-09-27 01:50:09 -07002681 if (!populated_zone(zone))
2682 continue;
2683
Nick Piggin23316bc2006-01-08 01:00:41 -08002684 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lameter37c07082007-10-16 01:25:36 -07002685 GFP_KERNEL, node);
Nick Piggin23316bc2006-01-08 01:00:41 -08002686 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002687 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002688
Nick Piggin23316bc2006-01-08 01:00:41 -08002689 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002690
2691 if (percpu_pagelist_fraction)
2692 setup_pagelist_highmark(zone_pcp(zone, cpu),
2693 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002694 }
2695
2696 return 0;
2697bad:
2698 for_each_zone(dzone) {
Andrew Morton64191682007-08-30 23:56:17 -07002699 if (!populated_zone(dzone))
2700 continue;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002701 if (dzone == zone)
2702 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08002703 kfree(zone_pcp(dzone, cpu));
2704 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002705 }
2706 return -ENOMEM;
2707}
2708
2709static inline void free_zone_pagesets(int cpu)
2710{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002711 struct zone *zone;
2712
2713 for_each_zone(zone) {
2714 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2715
David Rientjesf3ef9ea2006-09-25 16:24:57 -07002716 /* Free per_cpu_pageset if it is slab allocated */
2717 if (pset != &boot_pageset[cpu])
2718 kfree(pset);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002719 zone_pcp(zone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002720 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002721}
2722
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07002723static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002724 unsigned long action,
2725 void *hcpu)
2726{
2727 int cpu = (long)hcpu;
2728 int ret = NOTIFY_OK;
2729
2730 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08002731 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002732 case CPU_UP_PREPARE_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002733 if (process_zones(cpu))
2734 ret = NOTIFY_BAD;
2735 break;
2736 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002737 case CPU_UP_CANCELED_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002738 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002739 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002740 free_zone_pagesets(cpu);
2741 break;
2742 default:
2743 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002744 }
2745 return ret;
2746}
2747
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07002748static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002749 { &pageset_cpuup_callback, NULL, 0 };
2750
Al Viro78d99552005-12-15 09:18:25 +00002751void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002752{
2753 int err;
2754
2755 /* Initialize per_cpu_pageset for cpu 0.
2756 * A cpuup callback will do this for every cpu
2757 * as it comes online
2758 */
2759 err = process_zones(smp_processor_id());
2760 BUG_ON(err);
2761 register_cpu_notifier(&pageset_notifier);
2762}
2763
2764#endif
2765
Sam Ravnborg577a32f2007-05-17 23:29:25 +02002766static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07002767int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07002768{
2769 int i;
2770 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002771 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07002772
2773 /*
2774 * The per-page waitqueue mechanism uses hashed waitqueues
2775 * per zone.
2776 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07002777 zone->wait_table_hash_nr_entries =
2778 wait_table_hash_nr_entries(zone_size_pages);
2779 zone->wait_table_bits =
2780 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002781 alloc_size = zone->wait_table_hash_nr_entries
2782 * sizeof(wait_queue_head_t);
2783
2784 if (system_state == SYSTEM_BOOTING) {
2785 zone->wait_table = (wait_queue_head_t *)
2786 alloc_bootmem_node(pgdat, alloc_size);
2787 } else {
2788 /*
2789 * This case means that a zone whose size was 0 gets new memory
2790 * via memory hot-add.
2791 * But it may be the case that a new node was hot-added. In
2792 * this case vmalloc() will not be able to use this new node's
2793 * memory - this wait_table must be initialized to use this new
2794 * node itself as well.
2795 * To use this new node's memory, further consideration will be
2796 * necessary.
2797 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07002798 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002799 }
2800 if (!zone->wait_table)
2801 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07002802
Yasunori Goto02b694d2006-06-23 02:03:08 -07002803 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07002804 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002805
2806 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002807}
2808
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002809static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07002810{
2811 int cpu;
2812 unsigned long batch = zone_batchsize(zone);
2813
2814 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2815#ifdef CONFIG_NUMA
2816 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002817 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002818 setup_pageset(&boot_pageset[cpu],0);
2819#else
2820 setup_pageset(zone_pcp(zone,cpu), batch);
2821#endif
2822 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08002823 if (zone->present_pages)
2824 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2825 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07002826}
2827
Yasunori Goto718127c2006-06-23 02:03:10 -07002828__meminit int init_currently_empty_zone(struct zone *zone,
2829 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08002830 unsigned long size,
2831 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07002832{
2833 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002834 int ret;
2835 ret = zone_wait_table_init(zone, size);
2836 if (ret)
2837 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07002838 pgdat->nr_zones = zone_idx(zone) + 1;
2839
Dave Hansened8ece22005-10-29 18:16:50 -07002840 zone->zone_start_pfn = zone_start_pfn;
2841
2842 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2843
2844 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
Yasunori Goto718127c2006-06-23 02:03:10 -07002845
2846 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002847}
2848
Mel Gormanc7132162006-09-27 01:49:43 -07002849#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2850/*
2851 * Basic iterator support. Return the first range of PFNs for a node
2852 * Note: nid == MAX_NUMNODES returns first region regardless of node
2853 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002854static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07002855{
2856 int i;
2857
2858 for (i = 0; i < nr_nodemap_entries; i++)
2859 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2860 return i;
2861
2862 return -1;
2863}
2864
2865/*
2866 * Basic iterator support. Return the next active range of PFNs for a node
2867 * Note: nid == MAX_NUMNODES returns next region regardles of node
2868 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002869static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07002870{
2871 for (index = index + 1; index < nr_nodemap_entries; index++)
2872 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2873 return index;
2874
2875 return -1;
2876}
2877
2878#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2879/*
2880 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2881 * Architectures may implement their own version but if add_active_range()
2882 * was used and there are no special requirements, this is a convenient
2883 * alternative
2884 */
Stephen Rothwell6f076f52007-05-10 03:15:27 -07002885int __meminit early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07002886{
2887 int i;
2888
2889 for (i = 0; i < nr_nodemap_entries; i++) {
2890 unsigned long start_pfn = early_node_map[i].start_pfn;
2891 unsigned long end_pfn = early_node_map[i].end_pfn;
2892
2893 if (start_pfn <= pfn && pfn < end_pfn)
2894 return early_node_map[i].nid;
2895 }
2896
2897 return 0;
2898}
2899#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2900
2901/* Basic iterator support to walk early_node_map[] */
2902#define for_each_active_range_index_in_nid(i, nid) \
2903 for (i = first_active_region_index_in_nid(nid); i != -1; \
2904 i = next_active_region_index_in_nid(i, nid))
2905
2906/**
2907 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002908 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2909 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07002910 *
2911 * If an architecture guarantees that all ranges registered with
2912 * add_active_ranges() contain no holes and may be freed, this
2913 * this function may be used instead of calling free_bootmem() manually.
2914 */
2915void __init free_bootmem_with_active_regions(int nid,
2916 unsigned long max_low_pfn)
2917{
2918 int i;
2919
2920 for_each_active_range_index_in_nid(i, nid) {
2921 unsigned long size_pages = 0;
2922 unsigned long end_pfn = early_node_map[i].end_pfn;
2923
2924 if (early_node_map[i].start_pfn >= max_low_pfn)
2925 continue;
2926
2927 if (end_pfn > max_low_pfn)
2928 end_pfn = max_low_pfn;
2929
2930 size_pages = end_pfn - early_node_map[i].start_pfn;
2931 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2932 PFN_PHYS(early_node_map[i].start_pfn),
2933 size_pages << PAGE_SHIFT);
2934 }
2935}
2936
2937/**
2938 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002939 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07002940 *
2941 * If an architecture guarantees that all ranges registered with
2942 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002943 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07002944 */
2945void __init sparse_memory_present_with_active_regions(int nid)
2946{
2947 int i;
2948
2949 for_each_active_range_index_in_nid(i, nid)
2950 memory_present(early_node_map[i].nid,
2951 early_node_map[i].start_pfn,
2952 early_node_map[i].end_pfn);
2953}
2954
2955/**
Mel Gormanfb014392006-09-27 01:49:59 -07002956 * push_node_boundaries - Push node boundaries to at least the requested boundary
2957 * @nid: The nid of the node to push the boundary for
2958 * @start_pfn: The start pfn of the node
2959 * @end_pfn: The end pfn of the node
2960 *
2961 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2962 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2963 * be hotplugged even though no physical memory exists. This function allows
2964 * an arch to push out the node boundaries so mem_map is allocated that can
2965 * be used later.
2966 */
2967#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2968void __init push_node_boundaries(unsigned int nid,
2969 unsigned long start_pfn, unsigned long end_pfn)
2970{
2971 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2972 nid, start_pfn, end_pfn);
2973
2974 /* Initialise the boundary for this node if necessary */
2975 if (node_boundary_end_pfn[nid] == 0)
2976 node_boundary_start_pfn[nid] = -1UL;
2977
2978 /* Update the boundaries */
2979 if (node_boundary_start_pfn[nid] > start_pfn)
2980 node_boundary_start_pfn[nid] = start_pfn;
2981 if (node_boundary_end_pfn[nid] < end_pfn)
2982 node_boundary_end_pfn[nid] = end_pfn;
2983}
2984
2985/* If necessary, push the node boundary out for reserve hotadd */
Jan Beulich98011f52007-07-15 23:38:17 -07002986static void __meminit account_node_boundary(unsigned int nid,
Mel Gormanfb014392006-09-27 01:49:59 -07002987 unsigned long *start_pfn, unsigned long *end_pfn)
2988{
2989 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2990 nid, *start_pfn, *end_pfn);
2991
2992 /* Return if boundary information has not been provided */
2993 if (node_boundary_end_pfn[nid] == 0)
2994 return;
2995
2996 /* Check the boundaries and update if necessary */
2997 if (node_boundary_start_pfn[nid] < *start_pfn)
2998 *start_pfn = node_boundary_start_pfn[nid];
2999 if (node_boundary_end_pfn[nid] > *end_pfn)
3000 *end_pfn = node_boundary_end_pfn[nid];
3001}
3002#else
3003void __init push_node_boundaries(unsigned int nid,
3004 unsigned long start_pfn, unsigned long end_pfn) {}
3005
Jan Beulich98011f52007-07-15 23:38:17 -07003006static void __meminit account_node_boundary(unsigned int nid,
Mel Gormanfb014392006-09-27 01:49:59 -07003007 unsigned long *start_pfn, unsigned long *end_pfn) {}
3008#endif
3009
3010
3011/**
Mel Gormanc7132162006-09-27 01:49:43 -07003012 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003013 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3014 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3015 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003016 *
3017 * It returns the start and end page frame of a node based on information
3018 * provided by an arch calling add_active_range(). If called for a node
3019 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003020 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003021 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003022void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003023 unsigned long *start_pfn, unsigned long *end_pfn)
3024{
3025 int i;
3026 *start_pfn = -1UL;
3027 *end_pfn = 0;
3028
3029 for_each_active_range_index_in_nid(i, nid) {
3030 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3031 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3032 }
3033
Christoph Lameter633c0662007-10-16 01:25:37 -07003034 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003035 *start_pfn = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07003036
3037 /* Push the node boundaries out if requested */
3038 account_node_boundary(nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003039}
3040
3041/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003042 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3043 * assumption is made that zones within a node are ordered in monotonic
3044 * increasing memory addresses so that the "highest" populated zone is used
3045 */
3046void __init find_usable_zone_for_movable(void)
3047{
3048 int zone_index;
3049 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3050 if (zone_index == ZONE_MOVABLE)
3051 continue;
3052
3053 if (arch_zone_highest_possible_pfn[zone_index] >
3054 arch_zone_lowest_possible_pfn[zone_index])
3055 break;
3056 }
3057
3058 VM_BUG_ON(zone_index == -1);
3059 movable_zone = zone_index;
3060}
3061
3062/*
3063 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3064 * because it is sized independant of architecture. Unlike the other zones,
3065 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3066 * in each node depending on the size of each node and how evenly kernelcore
3067 * is distributed. This helper function adjusts the zone ranges
3068 * provided by the architecture for a given node by using the end of the
3069 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3070 * zones within a node are in order of monotonic increases memory addresses
3071 */
3072void __meminit adjust_zone_range_for_zone_movable(int nid,
3073 unsigned long zone_type,
3074 unsigned long node_start_pfn,
3075 unsigned long node_end_pfn,
3076 unsigned long *zone_start_pfn,
3077 unsigned long *zone_end_pfn)
3078{
3079 /* Only adjust if ZONE_MOVABLE is on this node */
3080 if (zone_movable_pfn[nid]) {
3081 /* Size ZONE_MOVABLE */
3082 if (zone_type == ZONE_MOVABLE) {
3083 *zone_start_pfn = zone_movable_pfn[nid];
3084 *zone_end_pfn = min(node_end_pfn,
3085 arch_zone_highest_possible_pfn[movable_zone]);
3086
3087 /* Adjust for ZONE_MOVABLE starting within this range */
3088 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3089 *zone_end_pfn > zone_movable_pfn[nid]) {
3090 *zone_end_pfn = zone_movable_pfn[nid];
3091
3092 /* Check if this whole range is within ZONE_MOVABLE */
3093 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3094 *zone_start_pfn = *zone_end_pfn;
3095 }
3096}
3097
3098/*
Mel Gormanc7132162006-09-27 01:49:43 -07003099 * Return the number of pages a zone spans in a node, including holes
3100 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3101 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003102static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003103 unsigned long zone_type,
3104 unsigned long *ignored)
3105{
3106 unsigned long node_start_pfn, node_end_pfn;
3107 unsigned long zone_start_pfn, zone_end_pfn;
3108
3109 /* Get the start and end of the node and zone */
3110 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3111 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3112 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003113 adjust_zone_range_for_zone_movable(nid, zone_type,
3114 node_start_pfn, node_end_pfn,
3115 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003116
3117 /* Check that this node has pages within the zone's required range */
3118 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3119 return 0;
3120
3121 /* Move the zone boundaries inside the node if necessary */
3122 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3123 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3124
3125 /* Return the spanned pages */
3126 return zone_end_pfn - zone_start_pfn;
3127}
3128
3129/*
3130 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003131 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003132 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003133unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003134 unsigned long range_start_pfn,
3135 unsigned long range_end_pfn)
3136{
3137 int i = 0;
3138 unsigned long prev_end_pfn = 0, hole_pages = 0;
3139 unsigned long start_pfn;
3140
3141 /* Find the end_pfn of the first active range of pfns in the node */
3142 i = first_active_region_index_in_nid(nid);
3143 if (i == -1)
3144 return 0;
3145
Mel Gormanb5445f92007-07-26 10:41:18 -07003146 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3147
Mel Gorman9c7cd682006-09-27 01:49:58 -07003148 /* Account for ranges before physical memory on this node */
3149 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003150 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003151
3152 /* Find all holes for the zone within the node */
3153 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3154
3155 /* No need to continue if prev_end_pfn is outside the zone */
3156 if (prev_end_pfn >= range_end_pfn)
3157 break;
3158
3159 /* Make sure the end of the zone is not within the hole */
3160 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3161 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3162
3163 /* Update the hole size cound and move on */
3164 if (start_pfn > range_start_pfn) {
3165 BUG_ON(prev_end_pfn > start_pfn);
3166 hole_pages += start_pfn - prev_end_pfn;
3167 }
3168 prev_end_pfn = early_node_map[i].end_pfn;
3169 }
3170
Mel Gorman9c7cd682006-09-27 01:49:58 -07003171 /* Account for ranges past physical memory on this node */
3172 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003173 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003174 max(range_start_pfn, prev_end_pfn);
3175
Mel Gormanc7132162006-09-27 01:49:43 -07003176 return hole_pages;
3177}
3178
3179/**
3180 * absent_pages_in_range - Return number of page frames in holes within a range
3181 * @start_pfn: The start PFN to start searching for holes
3182 * @end_pfn: The end PFN to stop searching for holes
3183 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003184 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003185 */
3186unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3187 unsigned long end_pfn)
3188{
3189 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3190}
3191
3192/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003193static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003194 unsigned long zone_type,
3195 unsigned long *ignored)
3196{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003197 unsigned long node_start_pfn, node_end_pfn;
3198 unsigned long zone_start_pfn, zone_end_pfn;
3199
3200 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3201 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3202 node_start_pfn);
3203 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3204 node_end_pfn);
3205
Mel Gorman2a1e2742007-07-17 04:03:12 -07003206 adjust_zone_range_for_zone_movable(nid, zone_type,
3207 node_start_pfn, node_end_pfn,
3208 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07003209 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003210}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003211
Mel Gormanc7132162006-09-27 01:49:43 -07003212#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07003213static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003214 unsigned long zone_type,
3215 unsigned long *zones_size)
3216{
3217 return zones_size[zone_type];
3218}
3219
Paul Mundt6ea6e682007-07-15 23:38:20 -07003220static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003221 unsigned long zone_type,
3222 unsigned long *zholes_size)
3223{
3224 if (!zholes_size)
3225 return 0;
3226
3227 return zholes_size[zone_type];
3228}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003229
Mel Gormanc7132162006-09-27 01:49:43 -07003230#endif
3231
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003232static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07003233 unsigned long *zones_size, unsigned long *zholes_size)
3234{
3235 unsigned long realtotalpages, totalpages = 0;
3236 enum zone_type i;
3237
3238 for (i = 0; i < MAX_NR_ZONES; i++)
3239 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3240 zones_size);
3241 pgdat->node_spanned_pages = totalpages;
3242
3243 realtotalpages = totalpages;
3244 for (i = 0; i < MAX_NR_ZONES; i++)
3245 realtotalpages -=
3246 zone_absent_pages_in_node(pgdat->node_id, i,
3247 zholes_size);
3248 pgdat->node_present_pages = realtotalpages;
3249 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3250 realtotalpages);
3251}
3252
Mel Gorman835c1342007-10-16 01:25:47 -07003253#ifndef CONFIG_SPARSEMEM
3254/*
3255 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07003256 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3257 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07003258 * round what is now in bits to nearest long in bits, then return it in
3259 * bytes.
3260 */
3261static unsigned long __init usemap_size(unsigned long zonesize)
3262{
3263 unsigned long usemapsize;
3264
Mel Gormand9c23402007-10-16 01:26:01 -07003265 usemapsize = roundup(zonesize, pageblock_nr_pages);
3266 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07003267 usemapsize *= NR_PAGEBLOCK_BITS;
3268 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3269
3270 return usemapsize / 8;
3271}
3272
3273static void __init setup_usemap(struct pglist_data *pgdat,
3274 struct zone *zone, unsigned long zonesize)
3275{
3276 unsigned long usemapsize = usemap_size(zonesize);
3277 zone->pageblock_flags = NULL;
3278 if (usemapsize) {
3279 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3280 memset(zone->pageblock_flags, 0, usemapsize);
3281 }
3282}
3283#else
3284static void inline setup_usemap(struct pglist_data *pgdat,
3285 struct zone *zone, unsigned long zonesize) {}
3286#endif /* CONFIG_SPARSEMEM */
3287
Mel Gormand9c23402007-10-16 01:26:01 -07003288#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3289/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3290static inline void __init set_pageblock_order(unsigned int order)
3291{
3292 /* Check that pageblock_nr_pages has not already been setup */
3293 if (pageblock_order)
3294 return;
3295
3296 /*
3297 * Assume the largest contiguous order of interest is a huge page.
3298 * This value may be variable depending on boot parameters on IA64
3299 */
3300 pageblock_order = order;
3301}
3302#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3303
3304/* Defined this way to avoid accidently referencing HUGETLB_PAGE_ORDER */
3305#define set_pageblock_order(x) do {} while (0)
3306
3307#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3308
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309/*
3310 * Set up the zone data structures:
3311 * - mark all pages reserved
3312 * - mark all memory queues empty
3313 * - clear the memory bitmaps
3314 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07003315static void __meminit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316 unsigned long *zones_size, unsigned long *zholes_size)
3317{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07003318 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07003319 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07003321 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322
Dave Hansen208d54e2005-10-29 18:16:52 -07003323 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 pgdat->nr_zones = 0;
3325 init_waitqueue_head(&pgdat->kswapd_wait);
3326 pgdat->kswapd_max_order = 0;
3327
3328 for (j = 0; j < MAX_NR_ZONES; j++) {
3329 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003330 unsigned long size, realsize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003331
Mel Gormanc7132162006-09-27 01:49:43 -07003332 size = zone_spanned_pages_in_node(nid, j, zones_size);
3333 realsize = size - zone_absent_pages_in_node(nid, j,
3334 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335
Mel Gorman0e0b8642006-09-27 01:49:56 -07003336 /*
3337 * Adjust realsize so that it accounts for how much memory
3338 * is used by this zone for memmap. This affects the watermark
3339 * and per-cpu initialisations
3340 */
3341 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
3342 if (realsize >= memmap_pages) {
3343 realsize -= memmap_pages;
3344 printk(KERN_DEBUG
3345 " %s zone: %lu pages used for memmap\n",
3346 zone_names[j], memmap_pages);
3347 } else
3348 printk(KERN_WARNING
3349 " %s zone: %lu pages exceeds realsize %lu\n",
3350 zone_names[j], memmap_pages, realsize);
3351
Christoph Lameter62672762007-02-10 01:43:07 -08003352 /* Account for reserved pages */
3353 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07003354 realsize -= dma_reserve;
Christoph Lameter62672762007-02-10 01:43:07 -08003355 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3356 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003357 }
3358
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003359 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 nr_kernel_pages += realsize;
3361 nr_all_pages += realsize;
3362
3363 zone->spanned_pages = size;
3364 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07003365#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07003366 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07003367 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07003368 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07003369 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07003370#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371 zone->name = zone_names[j];
3372 spin_lock_init(&zone->lock);
3373 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07003374 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376
Martin Bligh3bb1a8522006-10-28 10:38:24 -07003377 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003378
Dave Hansened8ece22005-10-29 18:16:50 -07003379 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380 INIT_LIST_HEAD(&zone->active_list);
3381 INIT_LIST_HEAD(&zone->inactive_list);
3382 zone->nr_scan_active = 0;
3383 zone->nr_scan_inactive = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07003384 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07003385 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386 if (!size)
3387 continue;
3388
Mel Gormand9c23402007-10-16 01:26:01 -07003389 set_pageblock_order(HUGETLB_PAGE_ORDER);
Mel Gorman835c1342007-10-16 01:25:47 -07003390 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08003391 ret = init_currently_empty_zone(zone, zone_start_pfn,
3392 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07003393 BUG_ON(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395 }
3396}
3397
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003398static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 /* Skip empty nodes */
3401 if (!pgdat->node_spanned_pages)
3402 return;
3403
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003404#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405 /* ia64 gets its own node_mem_map, before this, without bootmem */
3406 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07003407 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003408 struct page *map;
3409
Bob Piccoe984bb42006-05-20 15:00:31 -07003410 /*
3411 * The zone's endpoints aren't required to be MAX_ORDER
3412 * aligned but the node_mem_map endpoints must be in order
3413 * for the buddy allocator to function correctly.
3414 */
3415 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3416 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3417 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3418 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07003419 map = alloc_remap(pgdat->node_id, size);
3420 if (!map)
3421 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07003422 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423 }
Roman Zippel12d810c2007-05-31 00:40:54 -07003424#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003425 /*
3426 * With no DISCONTIG, the global mem_map is just set as node 0's
3427 */
Mel Gormanc7132162006-09-27 01:49:43 -07003428 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07003430#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3431 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3432 mem_map -= pgdat->node_start_pfn;
3433#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3434 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003436#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003437}
3438
Yasunori Goto86356ab2006-06-23 02:03:09 -07003439void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440 unsigned long *zones_size, unsigned long node_start_pfn,
3441 unsigned long *zholes_size)
3442{
3443 pgdat->node_id = nid;
3444 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003445 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446
3447 alloc_node_mem_map(pgdat);
3448
3449 free_area_init_core(pgdat, zones_size, zholes_size);
3450}
3451
Mel Gormanc7132162006-09-27 01:49:43 -07003452#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07003453
3454#if MAX_NUMNODES > 1
3455/*
3456 * Figure out the number of possible node ids.
3457 */
3458static void __init setup_nr_node_ids(void)
3459{
3460 unsigned int node;
3461 unsigned int highest = 0;
3462
3463 for_each_node_mask(node, node_possible_map)
3464 highest = node;
3465 nr_node_ids = highest + 1;
3466}
3467#else
3468static inline void setup_nr_node_ids(void)
3469{
3470}
3471#endif
3472
Mel Gormanc7132162006-09-27 01:49:43 -07003473/**
3474 * add_active_range - Register a range of PFNs backed by physical memory
3475 * @nid: The node ID the range resides on
3476 * @start_pfn: The start PFN of the available physical memory
3477 * @end_pfn: The end PFN of the available physical memory
3478 *
3479 * These ranges are stored in an early_node_map[] and later used by
3480 * free_area_init_nodes() to calculate zone sizes and holes. If the
3481 * range spans a memory hole, it is up to the architecture to ensure
3482 * the memory is not freed by the bootmem allocator. If possible
3483 * the range being registered will be merged with existing ranges.
3484 */
3485void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3486 unsigned long end_pfn)
3487{
3488 int i;
3489
3490 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
3491 "%d entries of %d used\n",
3492 nid, start_pfn, end_pfn,
3493 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3494
3495 /* Merge with existing active regions if possible */
3496 for (i = 0; i < nr_nodemap_entries; i++) {
3497 if (early_node_map[i].nid != nid)
3498 continue;
3499
3500 /* Skip if an existing region covers this new one */
3501 if (start_pfn >= early_node_map[i].start_pfn &&
3502 end_pfn <= early_node_map[i].end_pfn)
3503 return;
3504
3505 /* Merge forward if suitable */
3506 if (start_pfn <= early_node_map[i].end_pfn &&
3507 end_pfn > early_node_map[i].end_pfn) {
3508 early_node_map[i].end_pfn = end_pfn;
3509 return;
3510 }
3511
3512 /* Merge backward if suitable */
3513 if (start_pfn < early_node_map[i].end_pfn &&
3514 end_pfn >= early_node_map[i].start_pfn) {
3515 early_node_map[i].start_pfn = start_pfn;
3516 return;
3517 }
3518 }
3519
3520 /* Check that early_node_map is large enough */
3521 if (i >= MAX_ACTIVE_REGIONS) {
3522 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3523 MAX_ACTIVE_REGIONS);
3524 return;
3525 }
3526
3527 early_node_map[i].nid = nid;
3528 early_node_map[i].start_pfn = start_pfn;
3529 early_node_map[i].end_pfn = end_pfn;
3530 nr_nodemap_entries = i + 1;
3531}
3532
3533/**
3534 * shrink_active_range - Shrink an existing registered range of PFNs
3535 * @nid: The node id the range is on that should be shrunk
3536 * @old_end_pfn: The old end PFN of the range
3537 * @new_end_pfn: The new PFN of the range
3538 *
3539 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3540 * The map is kept at the end physical page range that has already been
3541 * registered with add_active_range(). This function allows an arch to shrink
3542 * an existing registered range.
3543 */
3544void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
3545 unsigned long new_end_pfn)
3546{
3547 int i;
3548
3549 /* Find the old active region end and shrink */
3550 for_each_active_range_index_in_nid(i, nid)
3551 if (early_node_map[i].end_pfn == old_end_pfn) {
3552 early_node_map[i].end_pfn = new_end_pfn;
3553 break;
3554 }
3555}
3556
3557/**
3558 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003559 *
Mel Gormanc7132162006-09-27 01:49:43 -07003560 * During discovery, it may be found that a table like SRAT is invalid
3561 * and an alternative discovery method must be used. This function removes
3562 * all currently registered regions.
3563 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003564void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07003565{
3566 memset(early_node_map, 0, sizeof(early_node_map));
3567 nr_nodemap_entries = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07003568#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3569 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3570 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3571#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -07003572}
3573
3574/* Compare two active node_active_regions */
3575static int __init cmp_node_active_region(const void *a, const void *b)
3576{
3577 struct node_active_region *arange = (struct node_active_region *)a;
3578 struct node_active_region *brange = (struct node_active_region *)b;
3579
3580 /* Done this way to avoid overflows */
3581 if (arange->start_pfn > brange->start_pfn)
3582 return 1;
3583 if (arange->start_pfn < brange->start_pfn)
3584 return -1;
3585
3586 return 0;
3587}
3588
3589/* sort the node_map by start_pfn */
3590static void __init sort_node_map(void)
3591{
3592 sort(early_node_map, (size_t)nr_nodemap_entries,
3593 sizeof(struct node_active_region),
3594 cmp_node_active_region, NULL);
3595}
3596
Mel Gormana6af2bc2007-02-10 01:42:57 -08003597/* Find the lowest pfn for a node */
Mel Gormanc7132162006-09-27 01:49:43 -07003598unsigned long __init find_min_pfn_for_node(unsigned long nid)
3599{
3600 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08003601 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00003602
Mel Gormanc7132162006-09-27 01:49:43 -07003603 /* Assuming a sorted map, the first range found has the starting pfn */
3604 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08003605 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003606
Mel Gormana6af2bc2007-02-10 01:42:57 -08003607 if (min_pfn == ULONG_MAX) {
3608 printk(KERN_WARNING
3609 "Could not find start_pfn for node %lu\n", nid);
3610 return 0;
3611 }
3612
3613 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003614}
3615
3616/**
3617 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3618 *
3619 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003620 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07003621 */
3622unsigned long __init find_min_pfn_with_active_regions(void)
3623{
3624 return find_min_pfn_for_node(MAX_NUMNODES);
3625}
3626
3627/**
3628 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3629 *
3630 * It returns the maximum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003631 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07003632 */
3633unsigned long __init find_max_pfn_with_active_regions(void)
3634{
3635 int i;
3636 unsigned long max_pfn = 0;
3637
3638 for (i = 0; i < nr_nodemap_entries; i++)
3639 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
3640
3641 return max_pfn;
3642}
3643
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003644/*
3645 * early_calculate_totalpages()
3646 * Sum pages in active regions for movable zone.
3647 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3648 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07003649static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07003650{
3651 int i;
3652 unsigned long totalpages = 0;
3653
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003654 for (i = 0; i < nr_nodemap_entries; i++) {
3655 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07003656 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003657 totalpages += pages;
3658 if (pages)
3659 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3660 }
3661 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07003662}
3663
Mel Gorman2a1e2742007-07-17 04:03:12 -07003664/*
3665 * Find the PFN the Movable zone begins in each node. Kernel memory
3666 * is spread evenly between nodes as long as the nodes have enough
3667 * memory. When they don't, some nodes will have more kernelcore than
3668 * others
3669 */
3670void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3671{
3672 int i, nid;
3673 unsigned long usable_startpfn;
3674 unsigned long kernelcore_node, kernelcore_remaining;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003675 unsigned long totalpages = early_calculate_totalpages();
3676 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07003677
Mel Gorman7e63efe2007-07-17 04:03:15 -07003678 /*
3679 * If movablecore was specified, calculate what size of
3680 * kernelcore that corresponds so that memory usable for
3681 * any allocation type is evenly spread. If both kernelcore
3682 * and movablecore are specified, then the value of kernelcore
3683 * will be used for required_kernelcore if it's greater than
3684 * what movablecore would have allowed.
3685 */
3686 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07003687 unsigned long corepages;
3688
3689 /*
3690 * Round-up so that ZONE_MOVABLE is at least as large as what
3691 * was requested by the user
3692 */
3693 required_movablecore =
3694 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3695 corepages = totalpages - required_movablecore;
3696
3697 required_kernelcore = max(required_kernelcore, corepages);
3698 }
3699
Mel Gorman2a1e2742007-07-17 04:03:12 -07003700 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3701 if (!required_kernelcore)
3702 return;
3703
3704 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3705 find_usable_zone_for_movable();
3706 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3707
3708restart:
3709 /* Spread kernelcore memory as evenly as possible throughout nodes */
3710 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003711 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07003712 /*
3713 * Recalculate kernelcore_node if the division per node
3714 * now exceeds what is necessary to satisfy the requested
3715 * amount of memory for the kernel
3716 */
3717 if (required_kernelcore < kernelcore_node)
3718 kernelcore_node = required_kernelcore / usable_nodes;
3719
3720 /*
3721 * As the map is walked, we track how much memory is usable
3722 * by the kernel using kernelcore_remaining. When it is
3723 * 0, the rest of the node is usable by ZONE_MOVABLE
3724 */
3725 kernelcore_remaining = kernelcore_node;
3726
3727 /* Go through each range of PFNs within this node */
3728 for_each_active_range_index_in_nid(i, nid) {
3729 unsigned long start_pfn, end_pfn;
3730 unsigned long size_pages;
3731
3732 start_pfn = max(early_node_map[i].start_pfn,
3733 zone_movable_pfn[nid]);
3734 end_pfn = early_node_map[i].end_pfn;
3735 if (start_pfn >= end_pfn)
3736 continue;
3737
3738 /* Account for what is only usable for kernelcore */
3739 if (start_pfn < usable_startpfn) {
3740 unsigned long kernel_pages;
3741 kernel_pages = min(end_pfn, usable_startpfn)
3742 - start_pfn;
3743
3744 kernelcore_remaining -= min(kernel_pages,
3745 kernelcore_remaining);
3746 required_kernelcore -= min(kernel_pages,
3747 required_kernelcore);
3748
3749 /* Continue if range is now fully accounted */
3750 if (end_pfn <= usable_startpfn) {
3751
3752 /*
3753 * Push zone_movable_pfn to the end so
3754 * that if we have to rebalance
3755 * kernelcore across nodes, we will
3756 * not double account here
3757 */
3758 zone_movable_pfn[nid] = end_pfn;
3759 continue;
3760 }
3761 start_pfn = usable_startpfn;
3762 }
3763
3764 /*
3765 * The usable PFN range for ZONE_MOVABLE is from
3766 * start_pfn->end_pfn. Calculate size_pages as the
3767 * number of pages used as kernelcore
3768 */
3769 size_pages = end_pfn - start_pfn;
3770 if (size_pages > kernelcore_remaining)
3771 size_pages = kernelcore_remaining;
3772 zone_movable_pfn[nid] = start_pfn + size_pages;
3773
3774 /*
3775 * Some kernelcore has been met, update counts and
3776 * break if the kernelcore for this node has been
3777 * satisified
3778 */
3779 required_kernelcore -= min(required_kernelcore,
3780 size_pages);
3781 kernelcore_remaining -= size_pages;
3782 if (!kernelcore_remaining)
3783 break;
3784 }
3785 }
3786
3787 /*
3788 * If there is still required_kernelcore, we do another pass with one
3789 * less node in the count. This will push zone_movable_pfn[nid] further
3790 * along on the nodes that still have memory until kernelcore is
3791 * satisified
3792 */
3793 usable_nodes--;
3794 if (usable_nodes && required_kernelcore > usable_nodes)
3795 goto restart;
3796
3797 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3798 for (nid = 0; nid < MAX_NUMNODES; nid++)
3799 zone_movable_pfn[nid] =
3800 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3801}
3802
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003803/* Any regular memory on that node ? */
3804static void check_for_regular_memory(pg_data_t *pgdat)
3805{
3806#ifdef CONFIG_HIGHMEM
3807 enum zone_type zone_type;
3808
3809 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3810 struct zone *zone = &pgdat->node_zones[zone_type];
3811 if (zone->present_pages)
3812 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3813 }
3814#endif
3815}
3816
Mel Gormanc7132162006-09-27 01:49:43 -07003817/**
3818 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003819 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07003820 *
3821 * This will call free_area_init_node() for each active node in the system.
3822 * Using the page ranges provided by add_active_range(), the size of each
3823 * zone in each node and their holes is calculated. If the maximum PFN
3824 * between two adjacent zones match, it is assumed that the zone is empty.
3825 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3826 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3827 * starts where the previous one ended. For example, ZONE_DMA32 starts
3828 * at arch_max_dma_pfn.
3829 */
3830void __init free_area_init_nodes(unsigned long *max_zone_pfn)
3831{
3832 unsigned long nid;
3833 enum zone_type i;
3834
Mel Gormana6af2bc2007-02-10 01:42:57 -08003835 /* Sort early_node_map as initialisation assumes it is sorted */
3836 sort_node_map();
3837
Mel Gormanc7132162006-09-27 01:49:43 -07003838 /* Record where the zone boundaries are */
3839 memset(arch_zone_lowest_possible_pfn, 0,
3840 sizeof(arch_zone_lowest_possible_pfn));
3841 memset(arch_zone_highest_possible_pfn, 0,
3842 sizeof(arch_zone_highest_possible_pfn));
3843 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
3844 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
3845 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07003846 if (i == ZONE_MOVABLE)
3847 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07003848 arch_zone_lowest_possible_pfn[i] =
3849 arch_zone_highest_possible_pfn[i-1];
3850 arch_zone_highest_possible_pfn[i] =
3851 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
3852 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07003853 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
3854 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
3855
3856 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3857 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
3858 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003859
Mel Gormanc7132162006-09-27 01:49:43 -07003860 /* Print out the zone ranges */
3861 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07003862 for (i = 0; i < MAX_NR_ZONES; i++) {
3863 if (i == ZONE_MOVABLE)
3864 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07003865 printk(" %-8s %8lu -> %8lu\n",
3866 zone_names[i],
3867 arch_zone_lowest_possible_pfn[i],
3868 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07003869 }
3870
3871 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3872 printk("Movable zone start PFN for each node\n");
3873 for (i = 0; i < MAX_NUMNODES; i++) {
3874 if (zone_movable_pfn[i])
3875 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
3876 }
Mel Gormanc7132162006-09-27 01:49:43 -07003877
3878 /* Print out the early_node_map[] */
3879 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
3880 for (i = 0; i < nr_nodemap_entries; i++)
3881 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
3882 early_node_map[i].start_pfn,
3883 early_node_map[i].end_pfn);
3884
3885 /* Initialise every node */
Christoph Lameter8ef82862007-02-20 13:57:52 -08003886 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07003887 for_each_online_node(nid) {
3888 pg_data_t *pgdat = NODE_DATA(nid);
3889 free_area_init_node(nid, pgdat, NULL,
3890 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003891
3892 /* Any memory on that node */
3893 if (pgdat->node_present_pages)
3894 node_set_state(nid, N_HIGH_MEMORY);
3895 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07003896 }
3897}
Mel Gorman2a1e2742007-07-17 04:03:12 -07003898
Mel Gorman7e63efe2007-07-17 04:03:15 -07003899static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003900{
3901 unsigned long long coremem;
3902 if (!p)
3903 return -EINVAL;
3904
3905 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07003906 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07003907
Mel Gorman7e63efe2007-07-17 04:03:15 -07003908 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07003909 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
3910
3911 return 0;
3912}
Mel Gormaned7ed362007-07-17 04:03:14 -07003913
Mel Gorman7e63efe2007-07-17 04:03:15 -07003914/*
3915 * kernelcore=size sets the amount of memory for use for allocations that
3916 * cannot be reclaimed or migrated.
3917 */
3918static int __init cmdline_parse_kernelcore(char *p)
3919{
3920 return cmdline_parse_core(p, &required_kernelcore);
3921}
3922
3923/*
3924 * movablecore=size sets the amount of memory for use for allocations that
3925 * can be reclaimed or migrated.
3926 */
3927static int __init cmdline_parse_movablecore(char *p)
3928{
3929 return cmdline_parse_core(p, &required_movablecore);
3930}
3931
Mel Gormaned7ed362007-07-17 04:03:14 -07003932early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07003933early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07003934
Mel Gormanc7132162006-09-27 01:49:43 -07003935#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3936
Mel Gorman0e0b8642006-09-27 01:49:56 -07003937/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003938 * set_dma_reserve - set the specified number of pages reserved in the first zone
3939 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07003940 *
3941 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3942 * In the DMA zone, a significant percentage may be consumed by kernel image
3943 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003944 * function may optionally be used to account for unfreeable pages in the
3945 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3946 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07003947 */
3948void __init set_dma_reserve(unsigned long new_dma_reserve)
3949{
3950 dma_reserve = new_dma_reserve;
3951}
3952
Dave Hansen93b75042005-06-23 00:07:47 -07003953#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003954static bootmem_data_t contig_bootmem_data;
3955struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
3956
3957EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07003958#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959
3960void __init free_area_init(unsigned long *zones_size)
3961{
Dave Hansen93b75042005-06-23 00:07:47 -07003962 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003963 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
3964}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965
Linus Torvalds1da177e2005-04-16 15:20:36 -07003966static int page_alloc_cpu_notify(struct notifier_block *self,
3967 unsigned long action, void *hcpu)
3968{
3969 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003970
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003971 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972 local_irq_disable();
3973 __drain_pages(cpu);
Christoph Lameterf8891e52006-06-30 01:55:45 -07003974 vm_events_fold_cpu(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003975 local_irq_enable();
Christoph Lameter2244b952006-06-30 01:55:33 -07003976 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003977 }
3978 return NOTIFY_OK;
3979}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003980
3981void __init page_alloc_init(void)
3982{
3983 hotcpu_notifier(page_alloc_cpu_notify, 0);
3984}
3985
3986/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003987 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3988 * or min_free_kbytes changes.
3989 */
3990static void calculate_totalreserve_pages(void)
3991{
3992 struct pglist_data *pgdat;
3993 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003994 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003995
3996 for_each_online_pgdat(pgdat) {
3997 for (i = 0; i < MAX_NR_ZONES; i++) {
3998 struct zone *zone = pgdat->node_zones + i;
3999 unsigned long max = 0;
4000
4001 /* Find valid and maximum lowmem_reserve in the zone */
4002 for (j = i; j < MAX_NR_ZONES; j++) {
4003 if (zone->lowmem_reserve[j] > max)
4004 max = zone->lowmem_reserve[j];
4005 }
4006
4007 /* we treat pages_high as reserved pages. */
4008 max += zone->pages_high;
4009
4010 if (max > zone->present_pages)
4011 max = zone->present_pages;
4012 reserve_pages += max;
4013 }
4014 }
4015 totalreserve_pages = reserve_pages;
4016}
4017
4018/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019 * setup_per_zone_lowmem_reserve - called whenever
4020 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4021 * has a correct pages reserved value, so an adequate number of
4022 * pages are left in the zone after a successful __alloc_pages().
4023 */
4024static void setup_per_zone_lowmem_reserve(void)
4025{
4026 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004027 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004029 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004030 for (j = 0; j < MAX_NR_ZONES; j++) {
4031 struct zone *zone = pgdat->node_zones + j;
4032 unsigned long present_pages = zone->present_pages;
4033
4034 zone->lowmem_reserve[j] = 0;
4035
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004036 idx = j;
4037 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038 struct zone *lower_zone;
4039
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004040 idx--;
4041
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4043 sysctl_lowmem_reserve_ratio[idx] = 1;
4044
4045 lower_zone = pgdat->node_zones + idx;
4046 lower_zone->lowmem_reserve[j] = present_pages /
4047 sysctl_lowmem_reserve_ratio[idx];
4048 present_pages += lower_zone->present_pages;
4049 }
4050 }
4051 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004052
4053 /* update totalreserve_pages */
4054 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004055}
4056
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004057/**
4058 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4059 *
4060 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4061 * with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004062 */
Dave Hansen3947be12005-10-29 18:16:54 -07004063void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004064{
4065 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4066 unsigned long lowmem_pages = 0;
4067 struct zone *zone;
4068 unsigned long flags;
4069
4070 /* Calculate total number of !ZONE_HIGHMEM pages */
4071 for_each_zone(zone) {
4072 if (!is_highmem(zone))
4073 lowmem_pages += zone->present_pages;
4074 }
4075
4076 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004077 u64 tmp;
4078
Linus Torvalds1da177e2005-04-16 15:20:36 -07004079 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004080 tmp = (u64)pages_min * zone->present_pages;
4081 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004082 if (is_highmem(zone)) {
4083 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004084 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4085 * need highmem pages, so cap pages_min to a small
4086 * value here.
4087 *
4088 * The (pages_high-pages_low) and (pages_low-pages_min)
4089 * deltas controls asynch page reclaim, and so should
4090 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091 */
4092 int min_pages;
4093
4094 min_pages = zone->present_pages / 1024;
4095 if (min_pages < SWAP_CLUSTER_MAX)
4096 min_pages = SWAP_CLUSTER_MAX;
4097 if (min_pages > 128)
4098 min_pages = 128;
4099 zone->pages_min = min_pages;
4100 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004101 /*
4102 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004103 * proportionate to the zone's size.
4104 */
Nick Piggin669ed172005-11-13 16:06:45 -08004105 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004106 }
4107
Andrew Mortonac924c62006-05-15 09:43:59 -07004108 zone->pages_low = zone->pages_min + (tmp >> 2);
4109 zone->pages_high = zone->pages_min + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004110 setup_zone_migrate_reserve(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004111 spin_unlock_irqrestore(&zone->lru_lock, flags);
4112 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004113
4114 /* update totalreserve_pages */
4115 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004116}
4117
4118/*
4119 * Initialise min_free_kbytes.
4120 *
4121 * For small machines we want it small (128k min). For large machines
4122 * we want it large (64MB max). But it is not linear, because network
4123 * bandwidth does not increase linearly with machine size. We use
4124 *
4125 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4126 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4127 *
4128 * which yields
4129 *
4130 * 16MB: 512k
4131 * 32MB: 724k
4132 * 64MB: 1024k
4133 * 128MB: 1448k
4134 * 256MB: 2048k
4135 * 512MB: 2896k
4136 * 1024MB: 4096k
4137 * 2048MB: 5792k
4138 * 4096MB: 8192k
4139 * 8192MB: 11584k
4140 * 16384MB: 16384k
4141 */
4142static int __init init_per_zone_pages_min(void)
4143{
4144 unsigned long lowmem_kbytes;
4145
4146 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4147
4148 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4149 if (min_free_kbytes < 128)
4150 min_free_kbytes = 128;
4151 if (min_free_kbytes > 65536)
4152 min_free_kbytes = 65536;
4153 setup_per_zone_pages_min();
4154 setup_per_zone_lowmem_reserve();
4155 return 0;
4156}
4157module_init(init_per_zone_pages_min)
4158
4159/*
4160 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4161 * that we can call two helper functions whenever min_free_kbytes
4162 * changes.
4163 */
4164int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4165 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4166{
4167 proc_dointvec(table, write, file, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07004168 if (write)
4169 setup_per_zone_pages_min();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004170 return 0;
4171}
4172
Christoph Lameter96146342006-07-03 00:24:13 -07004173#ifdef CONFIG_NUMA
4174int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4175 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4176{
4177 struct zone *zone;
4178 int rc;
4179
4180 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4181 if (rc)
4182 return rc;
4183
4184 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07004185 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07004186 sysctl_min_unmapped_ratio) / 100;
4187 return 0;
4188}
Christoph Lameter0ff38492006-09-25 23:31:52 -07004189
4190int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4191 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4192{
4193 struct zone *zone;
4194 int rc;
4195
4196 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4197 if (rc)
4198 return rc;
4199
4200 for_each_zone(zone)
4201 zone->min_slab_pages = (zone->present_pages *
4202 sysctl_min_slab_ratio) / 100;
4203 return 0;
4204}
Christoph Lameter96146342006-07-03 00:24:13 -07004205#endif
4206
Linus Torvalds1da177e2005-04-16 15:20:36 -07004207/*
4208 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4209 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4210 * whenever sysctl_lowmem_reserve_ratio changes.
4211 *
4212 * The reserve ratio obviously has absolutely no relation with the
4213 * pages_min watermarks. The lowmem reserve ratio can only make sense
4214 * if in function of the boot time zone sizes.
4215 */
4216int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4217 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4218{
4219 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4220 setup_per_zone_lowmem_reserve();
4221 return 0;
4222}
4223
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004224/*
4225 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4226 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4227 * can have before it gets flushed back to buddy allocator.
4228 */
4229
4230int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4231 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4232{
4233 struct zone *zone;
4234 unsigned int cpu;
4235 int ret;
4236
4237 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4238 if (!write || (ret == -EINVAL))
4239 return ret;
4240 for_each_zone(zone) {
4241 for_each_online_cpu(cpu) {
4242 unsigned long high;
4243 high = zone->present_pages / percpu_pagelist_fraction;
4244 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4245 }
4246 }
4247 return 0;
4248}
4249
David S. Millerf034b5d2006-08-24 03:08:07 -07004250int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004251
4252#ifdef CONFIG_NUMA
4253static int __init set_hashdist(char *str)
4254{
4255 if (!str)
4256 return 0;
4257 hashdist = simple_strtoul(str, &str, 0);
4258 return 1;
4259}
4260__setup("hashdist=", set_hashdist);
4261#endif
4262
4263/*
4264 * allocate a large system hash table from bootmem
4265 * - it is assumed that the hash table must contain an exact power-of-2
4266 * quantity of entries
4267 * - limit is the number of hash buckets, not the total allocation size
4268 */
4269void *__init alloc_large_system_hash(const char *tablename,
4270 unsigned long bucketsize,
4271 unsigned long numentries,
4272 int scale,
4273 int flags,
4274 unsigned int *_hash_shift,
4275 unsigned int *_hash_mask,
4276 unsigned long limit)
4277{
4278 unsigned long long max = limit;
4279 unsigned long log2qty, size;
4280 void *table = NULL;
4281
4282 /* allow the kernel cmdline to have a say */
4283 if (!numentries) {
4284 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08004285 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004286 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4287 numentries >>= 20 - PAGE_SHIFT;
4288 numentries <<= 20 - PAGE_SHIFT;
4289
4290 /* limit to 1 bucket per 2^scale bytes of low memory */
4291 if (scale > PAGE_SHIFT)
4292 numentries >>= (scale - PAGE_SHIFT);
4293 else
4294 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08004295
4296 /* Make sure we've got at least a 0-order allocation.. */
4297 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4298 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004299 }
John Hawkes6e692ed2006-03-25 03:08:02 -08004300 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004301
4302 /* limit allocation size to 1/16 total memory by default */
4303 if (max == 0) {
4304 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4305 do_div(max, bucketsize);
4306 }
4307
4308 if (numentries > max)
4309 numentries = max;
4310
David Howellsf0d1b0b2006-12-08 02:37:49 -08004311 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004312
4313 do {
4314 size = bucketsize << log2qty;
4315 if (flags & HASH_EARLY)
4316 table = alloc_bootmem(size);
4317 else if (hashdist)
4318 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4319 else {
4320 unsigned long order;
4321 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
4322 ;
4323 table = (void*) __get_free_pages(GFP_ATOMIC, order);
Eric Dumazet1037b832007-07-15 23:38:05 -07004324 /*
4325 * If bucketsize is not a power-of-two, we may free
4326 * some pages at the end of hash table.
4327 */
4328 if (table) {
4329 unsigned long alloc_end = (unsigned long)table +
4330 (PAGE_SIZE << order);
4331 unsigned long used = (unsigned long)table +
4332 PAGE_ALIGN(size);
4333 split_page(virt_to_page(table), order);
4334 while (used < alloc_end) {
4335 free_page(used);
4336 used += PAGE_SIZE;
4337 }
4338 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004339 }
4340 } while (!table && size > PAGE_SIZE && --log2qty);
4341
4342 if (!table)
4343 panic("Failed to allocate %s hash table\n", tablename);
4344
Dan Alonib49ad482007-07-15 23:38:23 -07004345 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004346 tablename,
4347 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08004348 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004349 size);
4350
4351 if (_hash_shift)
4352 *_hash_shift = log2qty;
4353 if (_hash_mask)
4354 *_hash_mask = (1 << log2qty) - 1;
4355
4356 return table;
4357}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004358
4359#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004360struct page *pfn_to_page(unsigned long pfn)
4361{
Andy Whitcroft67de6482006-06-23 02:03:12 -07004362 return __pfn_to_page(pfn);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004363}
4364unsigned long page_to_pfn(struct page *page)
4365{
Andy Whitcroft67de6482006-06-23 02:03:12 -07004366 return __page_to_pfn(page);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004367}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004368EXPORT_SYMBOL(pfn_to_page);
4369EXPORT_SYMBOL(page_to_pfn);
4370#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
Andrew Morton6220ec72006-10-19 23:29:05 -07004371
Mel Gorman835c1342007-10-16 01:25:47 -07004372/* Return a pointer to the bitmap storing bits affecting a block of pages */
4373static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4374 unsigned long pfn)
4375{
4376#ifdef CONFIG_SPARSEMEM
4377 return __pfn_to_section(pfn)->pageblock_flags;
4378#else
4379 return zone->pageblock_flags;
4380#endif /* CONFIG_SPARSEMEM */
4381}
Andrew Morton6220ec72006-10-19 23:29:05 -07004382
Mel Gorman835c1342007-10-16 01:25:47 -07004383static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4384{
4385#ifdef CONFIG_SPARSEMEM
4386 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004387 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004388#else
4389 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07004390 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004391#endif /* CONFIG_SPARSEMEM */
4392}
4393
4394/**
Mel Gormand9c23402007-10-16 01:26:01 -07004395 * 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 -07004396 * @page: The page within the block of interest
4397 * @start_bitidx: The first bit of interest to retrieve
4398 * @end_bitidx: The last bit of interest
4399 * returns pageblock_bits flags
4400 */
4401unsigned long get_pageblock_flags_group(struct page *page,
4402 int start_bitidx, int end_bitidx)
4403{
4404 struct zone *zone;
4405 unsigned long *bitmap;
4406 unsigned long pfn, bitidx;
4407 unsigned long flags = 0;
4408 unsigned long value = 1;
4409
4410 zone = page_zone(page);
4411 pfn = page_to_pfn(page);
4412 bitmap = get_pageblock_bitmap(zone, pfn);
4413 bitidx = pfn_to_bitidx(zone, pfn);
4414
4415 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4416 if (test_bit(bitidx + start_bitidx, bitmap))
4417 flags |= value;
4418
4419 return flags;
4420}
4421
4422/**
Mel Gormand9c23402007-10-16 01:26:01 -07004423 * 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 -07004424 * @page: The page within the block of interest
4425 * @start_bitidx: The first bit of interest
4426 * @end_bitidx: The last bit of interest
4427 * @flags: The flags to set
4428 */
4429void set_pageblock_flags_group(struct page *page, unsigned long flags,
4430 int start_bitidx, int end_bitidx)
4431{
4432 struct zone *zone;
4433 unsigned long *bitmap;
4434 unsigned long pfn, bitidx;
4435 unsigned long value = 1;
4436
4437 zone = page_zone(page);
4438 pfn = page_to_pfn(page);
4439 bitmap = get_pageblock_bitmap(zone, pfn);
4440 bitidx = pfn_to_bitidx(zone, pfn);
4441
4442 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4443 if (flags & value)
4444 __set_bit(bitidx + start_bitidx, bitmap);
4445 else
4446 __clear_bit(bitidx + start_bitidx, bitmap);
4447}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004448
4449/*
4450 * This is designed as sub function...plz see page_isolation.c also.
4451 * set/clear page block's type to be ISOLATE.
4452 * page allocater never alloc memory from ISOLATE block.
4453 */
4454
4455int set_migratetype_isolate(struct page *page)
4456{
4457 struct zone *zone;
4458 unsigned long flags;
4459 int ret = -EBUSY;
4460
4461 zone = page_zone(page);
4462 spin_lock_irqsave(&zone->lock, flags);
4463 /*
4464 * In future, more migrate types will be able to be isolation target.
4465 */
4466 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4467 goto out;
4468 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4469 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4470 ret = 0;
4471out:
4472 spin_unlock_irqrestore(&zone->lock, flags);
4473 if (!ret)
4474 drain_all_local_pages();
4475 return ret;
4476}
4477
4478void unset_migratetype_isolate(struct page *page)
4479{
4480 struct zone *zone;
4481 unsigned long flags;
4482 zone = page_zone(page);
4483 spin_lock_irqsave(&zone->lock, flags);
4484 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4485 goto out;
4486 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4487 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4488out:
4489 spin_unlock_irqrestore(&zone->lock, flags);
4490}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07004491
4492#ifdef CONFIG_MEMORY_HOTREMOVE
4493/*
4494 * All pages in the range must be isolated before calling this.
4495 */
4496void
4497__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4498{
4499 struct page *page;
4500 struct zone *zone;
4501 int order, i;
4502 unsigned long pfn;
4503 unsigned long flags;
4504 /* find the first valid pfn */
4505 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4506 if (pfn_valid(pfn))
4507 break;
4508 if (pfn == end_pfn)
4509 return;
4510 zone = page_zone(pfn_to_page(pfn));
4511 spin_lock_irqsave(&zone->lock, flags);
4512 pfn = start_pfn;
4513 while (pfn < end_pfn) {
4514 if (!pfn_valid(pfn)) {
4515 pfn++;
4516 continue;
4517 }
4518 page = pfn_to_page(pfn);
4519 BUG_ON(page_count(page));
4520 BUG_ON(!PageBuddy(page));
4521 order = page_order(page);
4522#ifdef CONFIG_DEBUG_VM
4523 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4524 pfn, 1 << order, end_pfn);
4525#endif
4526 list_del(&page->lru);
4527 rmv_page_order(page);
4528 zone->free_area[order].nr_free--;
4529 __mod_zone_page_state(zone, NR_FREE_PAGES,
4530 - (1UL << order));
4531 for (i = 0; i < (1 << order); i++)
4532 SetPageReserved((page+i));
4533 pfn += (1 << order);
4534 }
4535 spin_unlock_irqrestore(&zone->lock, flags);
4536}
4537#endif