blob: b2838c24e582c11b8e0680ca5d5b5f97389145ba [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 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200126 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
Mel Gormanc7132162006-09-27 01:49:43 -0700127 * 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
Andrew Morton6626c5d2006-03-22 00:08:42 -0800308 /*
309 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
310 * and __GFP_HIGHMEM from hard or soft interrupt context.
311 */
Nick Piggin725d7042006-09-25 23:30:55 -0700312 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800313 for (i = 0; i < (1 << order); i++)
314 clear_highpage(page + i);
315}
316
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700317static inline void set_page_order(struct page *page, int order)
318{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700319 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000320 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322
323static inline void rmv_page_order(struct page *page)
324{
Nick Piggin676165a2006-04-10 11:21:48 +1000325 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700326 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327}
328
329/*
330 * Locate the struct page for both the matching buddy in our
331 * pair (buddy1) and the combined O(n+1) page they form (page).
332 *
333 * 1) Any buddy B1 will have an order O twin B2 which satisfies
334 * the following equation:
335 * B2 = B1 ^ (1 << O)
336 * For example, if the starting buddy (buddy2) is #8 its order
337 * 1 buddy is #10:
338 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
339 *
340 * 2) Any buddy B will have an order O+1 parent P which
341 * satisfies the following equation:
342 * P = B & ~(1 << O)
343 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200344 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 */
346static inline struct page *
347__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
348{
349 unsigned long buddy_idx = page_idx ^ (1 << order);
350
351 return page + (buddy_idx - page_idx);
352}
353
354static inline unsigned long
355__find_combined_index(unsigned long page_idx, unsigned int order)
356{
357 return (page_idx & ~(1 << order));
358}
359
360/*
361 * This function checks whether a page is free && is the buddy
362 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800363 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000364 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700365 * (c) a page and its buddy have the same order &&
366 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 *
Nick Piggin676165a2006-04-10 11:21:48 +1000368 * For recording whether a page is in the buddy system, we use PG_buddy.
369 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
370 *
371 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700373static inline int page_is_buddy(struct page *page, struct page *buddy,
374 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700376 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800377 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800378
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700379 if (page_zone_id(page) != page_zone_id(buddy))
380 return 0;
381
382 if (PageBuddy(buddy) && page_order(buddy) == order) {
383 BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700384 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000385 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700386 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387}
388
389/*
390 * Freeing function for a buddy system allocator.
391 *
392 * The concept of a buddy system is to maintain direct-mapped table
393 * (containing bit values) for memory blocks of various "orders".
394 * The bottom level table contains the map for the smallest allocatable
395 * units of memory (here, pages), and each level above it describes
396 * pairs of units from the levels below, hence, "buddies".
397 * At a high level, all that happens here is marking the table entry
398 * at the bottom level available, and propagating the changes upward
399 * as necessary, plus some accounting needed to play nicely with other
400 * parts of the VM system.
401 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000402 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700403 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 * So when we are allocating or freeing one, we can derive the state of the
405 * other. That is, if we allocate a small block, and both were
406 * free, the remainder of the region must be split into blocks.
407 * If a block is freed, and its buddy is also free, then this
408 * triggers coalescing into a block of larger size.
409 *
410 * -- wli
411 */
412
Nick Piggin48db57f2006-01-08 01:00:42 -0800413static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 struct zone *zone, unsigned int order)
415{
416 unsigned long page_idx;
417 int order_size = 1 << order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700418 int migratetype = get_pageblock_migratetype(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
Nick Piggin224abf92006-01-06 00:11:11 -0800420 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 destroy_compound_page(page, order);
422
423 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
424
Nick Piggin725d7042006-09-25 23:30:55 -0700425 VM_BUG_ON(page_idx & (order_size - 1));
426 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
Christoph Lameterd23ad422007-02-10 01:43:02 -0800428 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 while (order < MAX_ORDER-1) {
430 unsigned long combined_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 struct page *buddy;
432
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700434 if (!page_is_buddy(page, buddy, order))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700438 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800440 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 page = page + (combined_idx - page_idx);
442 page_idx = combined_idx;
443 order++;
444 }
445 set_page_order(page, order);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700446 list_add(&page->lru,
447 &zone->free_area[order].free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 zone->free_area[order].nr_free++;
449}
450
Nick Piggin224abf92006-01-06 00:11:11 -0800451static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452{
Nick Piggin92be2e332006-01-06 00:10:57 -0800453 if (unlikely(page_mapcount(page) |
454 (page->mapping != NULL) |
455 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 (page->flags & (
457 1 << PG_lru |
458 1 << PG_private |
459 1 << PG_locked |
460 1 << PG_active |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 1 << PG_slab |
462 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700463 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000464 1 << PG_reserved |
465 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800466 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700468 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800469 /*
470 * For now, we report if PG_reserved was found set, but do not
471 * clear it, and do not free the page. But we shall soon need
472 * to do more, for when the ZERO_PAGE count wraps negative.
473 */
474 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475}
476
477/*
478 * Frees a list of pages.
479 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700480 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 *
482 * If the zone was previously in an "all pages pinned" state then look to
483 * see if this freeing clears that state.
484 *
485 * And clear the zone's pages_scanned counter, to hold off the "all pages are
486 * pinned" detection logic.
487 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800488static void free_pages_bulk(struct zone *zone, int count,
489 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490{
Nick Pigginc54ad302006-01-06 00:10:56 -0800491 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700492 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800494 while (count--) {
495 struct page *page;
496
Nick Piggin725d7042006-09-25 23:30:55 -0700497 VM_BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800499 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800501 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800503 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504}
505
Nick Piggin48db57f2006-01-08 01:00:42 -0800506static void free_one_page(struct zone *zone, struct page *page, int order)
507{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700508 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700509 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700510 zone->pages_scanned = 0;
Paul Jackson0798e512006-12-06 20:31:38 -0800511 __free_one_page(page, zone, order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700512 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800513}
514
515static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516{
Nick Pigginc54ad302006-01-06 00:10:56 -0800517 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800519 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800522 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800523 if (reserved)
524 return;
525
Nick Piggin9858db52006-10-11 01:21:30 -0700526 if (!PageHighMem(page))
527 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Nick Piggindafb1362006-10-11 01:21:30 -0700528 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800529 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700530
Nick Pigginc54ad302006-01-06 00:10:56 -0800531 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700532 __count_vm_events(PGFREE, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800533 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800534 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535}
536
David Howellsa226f6c2006-01-06 00:11:08 -0800537/*
538 * permit the bootmem allocator to evade page validation on high-order frees
539 */
540void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
541{
542 if (order == 0) {
543 __ClearPageReserved(page);
544 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800545 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800546 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800547 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800548 int loop;
549
Nick Piggin545b1ea2006-03-22 00:08:07 -0800550 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800551 for (loop = 0; loop < BITS_PER_LONG; loop++) {
552 struct page *p = &page[loop];
553
Nick Piggin545b1ea2006-03-22 00:08:07 -0800554 if (loop + 1 < BITS_PER_LONG)
555 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800556 __ClearPageReserved(p);
557 set_page_count(p, 0);
558 }
559
Nick Piggin7835e982006-03-22 00:08:40 -0800560 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800561 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800562 }
563}
564
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
566/*
567 * The order of subdivision here is critical for the IO subsystem.
568 * Please do not alter this order without good reasons and regression
569 * testing. Specifically, as large blocks of memory are subdivided,
570 * the order in which smaller blocks are delivered depends on the order
571 * they're subdivided in this function. This is the primary factor
572 * influencing the order in which pages are delivered to the IO
573 * subsystem according to empirical testing, and this is also justified
574 * by considering the behavior of a buddy system containing a single
575 * large block of memory acted on by a series of small allocations.
576 * This behavior is a critical factor in sglist merging's success.
577 *
578 * -- wli
579 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800580static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700581 int low, int high, struct free_area *area,
582 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583{
584 unsigned long size = 1 << high;
585
586 while (high > low) {
587 area--;
588 high--;
589 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700590 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700591 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 area->nr_free++;
593 set_page_order(&page[size], high);
594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595}
596
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597/*
598 * This page is about to be returned from the page allocator
599 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800600static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601{
Nick Piggin92be2e332006-01-06 00:10:57 -0800602 if (unlikely(page_mapcount(page) |
603 (page->mapping != NULL) |
604 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700605 (page->flags & (
606 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 1 << PG_private |
608 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 1 << PG_active |
610 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700611 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700613 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000614 1 << PG_reserved |
615 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800616 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
Hugh Dickins689bceb2005-11-21 21:32:20 -0800618 /*
619 * For now, we report if PG_reserved was found set, but do not
620 * clear it, and do not allocate the page: as a safety net.
621 */
622 if (PageReserved(page))
623 return 1;
624
Fengguang Wud77c2d72007-07-19 01:47:55 -0700625 page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 1 << PG_referenced | 1 << PG_arch_1 |
Nick Piggin5409bae2007-02-28 20:12:27 -0800627 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700628 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800629 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800630
631 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800633
634 if (gfp_flags & __GFP_ZERO)
635 prep_zero_page(page, order, gfp_flags);
636
637 if (order && (gfp_flags & __GFP_COMP))
638 prep_compound_page(page, order);
639
Hugh Dickins689bceb2005-11-21 21:32:20 -0800640 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641}
642
Mel Gorman56fd56b2007-10-16 01:25:58 -0700643/*
644 * Go through the free lists for the given migratetype and remove
645 * the smallest available page from the freelists
646 */
647static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
648 int migratetype)
649{
650 unsigned int current_order;
651 struct free_area * area;
652 struct page *page;
653
654 /* Find a page of the appropriate size in the preferred list */
655 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
656 area = &(zone->free_area[current_order]);
657 if (list_empty(&area->free_list[migratetype]))
658 continue;
659
660 page = list_entry(area->free_list[migratetype].next,
661 struct page, lru);
662 list_del(&page->lru);
663 rmv_page_order(page);
664 area->nr_free--;
665 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
666 expand(zone, page, order, current_order, area, migratetype);
667 return page;
668 }
669
670 return NULL;
671}
672
673
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700674/*
675 * This array describes the order lists are fallen back to when
676 * the free lists for the desirable migrate type are depleted
677 */
678static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700679 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
680 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
681 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
682 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700683};
684
Mel Gormanc361be52007-10-16 01:25:51 -0700685/*
686 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700687 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700688 * boundary. If alignment is required, use move_freepages_block()
689 */
690int move_freepages(struct zone *zone,
691 struct page *start_page, struct page *end_page,
692 int migratetype)
693{
694 struct page *page;
695 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700696 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700697
698#ifndef CONFIG_HOLES_IN_ZONE
699 /*
700 * page_zone is not safe to call in this context when
701 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
702 * anyway as we check zone boundaries in move_freepages_block().
703 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700704 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700705 */
706 BUG_ON(page_zone(start_page) != page_zone(end_page));
707#endif
708
709 for (page = start_page; page <= end_page;) {
710 if (!pfn_valid_within(page_to_pfn(page))) {
711 page++;
712 continue;
713 }
714
715 if (!PageBuddy(page)) {
716 page++;
717 continue;
718 }
719
720 order = page_order(page);
721 list_del(&page->lru);
722 list_add(&page->lru,
723 &zone->free_area[order].free_list[migratetype]);
724 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700725 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700726 }
727
Mel Gormand1003132007-10-16 01:26:00 -0700728 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700729}
730
731int move_freepages_block(struct zone *zone, struct page *page, int migratetype)
732{
733 unsigned long start_pfn, end_pfn;
734 struct page *start_page, *end_page;
735
736 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700737 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700738 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700739 end_page = start_page + pageblock_nr_pages - 1;
740 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700741
742 /* Do not cross zone boundaries */
743 if (start_pfn < zone->zone_start_pfn)
744 start_page = page;
745 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
746 return 0;
747
748 return move_freepages(zone, start_page, end_page, migratetype);
749}
750
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700751/* Remove an element from the buddy allocator from the fallback list */
752static struct page *__rmqueue_fallback(struct zone *zone, int order,
753 int start_migratetype)
754{
755 struct free_area * area;
756 int current_order;
757 struct page *page;
758 int migratetype, i;
759
760 /* Find the largest possible block of pages in the other list */
761 for (current_order = MAX_ORDER-1; current_order >= order;
762 --current_order) {
763 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
764 migratetype = fallbacks[start_migratetype][i];
765
Mel Gorman56fd56b2007-10-16 01:25:58 -0700766 /* MIGRATE_RESERVE handled later if necessary */
767 if (migratetype == MIGRATE_RESERVE)
768 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700769
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700770 area = &(zone->free_area[current_order]);
771 if (list_empty(&area->free_list[migratetype]))
772 continue;
773
774 page = list_entry(area->free_list[migratetype].next,
775 struct page, lru);
776 area->nr_free--;
777
778 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700779 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700780 * pages to the preferred allocation list. If falling
781 * back for a reclaimable kernel allocation, be more
782 * agressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700783 */
Mel Gormand9c23402007-10-16 01:26:01 -0700784 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gorman46dafbc2007-10-16 01:25:55 -0700785 start_migratetype == MIGRATE_RECLAIMABLE) {
786 unsigned long pages;
787 pages = move_freepages_block(zone, page,
788 start_migratetype);
789
790 /* Claim the whole block if over half of it is free */
Mel Gormand9c23402007-10-16 01:26:01 -0700791 if (pages >= (1 << (pageblock_order-1)))
Mel Gorman46dafbc2007-10-16 01:25:55 -0700792 set_pageblock_migratetype(page,
793 start_migratetype);
794
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700795 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700796 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700797
798 /* Remove the page from the freelists */
799 list_del(&page->lru);
800 rmv_page_order(page);
801 __mod_zone_page_state(zone, NR_FREE_PAGES,
802 -(1UL << order));
803
Mel Gormand9c23402007-10-16 01:26:01 -0700804 if (current_order == pageblock_order)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700805 set_pageblock_migratetype(page,
806 start_migratetype);
807
808 expand(zone, page, order, current_order, area, migratetype);
809 return page;
810 }
811 }
812
Mel Gorman56fd56b2007-10-16 01:25:58 -0700813 /* Use MIGRATE_RESERVE rather than fail an allocation */
814 return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700815}
816
Mel Gorman56fd56b2007-10-16 01:25:58 -0700817/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 * Do the hard work of removing an element from the buddy allocator.
819 * Call me with the zone->lock already held.
820 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700821static struct page *__rmqueue(struct zone *zone, unsigned int order,
822 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 struct page *page;
825
Mel Gorman56fd56b2007-10-16 01:25:58 -0700826 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
Mel Gorman56fd56b2007-10-16 01:25:58 -0700828 if (unlikely(!page))
829 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700830
831 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832}
833
834/*
835 * Obtain a specified number of elements from the buddy allocator, all under
836 * a single hold of the lock, for efficiency. Add them to the supplied list.
837 * Returns the number of new pages which were placed at *list.
838 */
839static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700840 unsigned long count, struct list_head *list,
841 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844
Nick Pigginc54ad302006-01-06 00:10:56 -0800845 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700847 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800848 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -0800850
851 /*
852 * Split buddy pages returned by expand() are received here
853 * in physical page order. The page is added to the callers and
854 * list and the list head then moves forward. From the callers
855 * perspective, the linked list is ordered by page number in
856 * some conditions. This is useful for IO devices that can
857 * merge IO requests if the physical pages are ordered
858 * properly.
859 */
Mel Gorman535131e62007-10-16 01:25:49 -0700860 list_add(&page->lru, list);
861 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -0800862 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800864 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800865 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866}
867
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700868#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800869/*
Christoph Lameter4037d452007-05-09 02:35:14 -0700870 * Called from the vmstat counter updater to drain pagesets of this
871 * currently executing processor on remote nodes after they have
872 * expired.
873 *
Christoph Lameter879336c2006-03-22 00:09:08 -0800874 * Note that this function must be called with the thread pinned to
875 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800876 */
Christoph Lameter4037d452007-05-09 02:35:14 -0700877void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700878{
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700879 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -0700880 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700881
Christoph Lameter4037d452007-05-09 02:35:14 -0700882 local_irq_save(flags);
883 if (pcp->count >= pcp->batch)
884 to_drain = pcp->batch;
885 else
886 to_drain = pcp->count;
887 free_pages_bulk(zone, to_drain, &pcp->list, 0);
888 pcp->count -= to_drain;
889 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700890}
891#endif
892
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893static void __drain_pages(unsigned int cpu)
894{
Nick Pigginc54ad302006-01-06 00:10:56 -0800895 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 struct zone *zone;
897 int i;
898
899 for_each_zone(zone) {
900 struct per_cpu_pageset *pset;
901
Christoph Lameterf2e12bb2007-01-05 16:37:02 -0800902 if (!populated_zone(zone))
903 continue;
904
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700905 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
907 struct per_cpu_pages *pcp;
908
909 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800910 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800911 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
912 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800913 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 }
915 }
916}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Rafael J. Wysocki296699d2007-07-29 23:27:18 +0200918#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
920void mark_free_pages(struct zone *zone)
921{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700922 unsigned long pfn, max_zone_pfn;
923 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700924 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 struct list_head *curr;
926
927 if (!zone->spanned_pages)
928 return;
929
930 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700931
932 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
933 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
934 if (pfn_valid(pfn)) {
935 struct page *page = pfn_to_page(pfn);
936
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700937 if (!swsusp_page_is_forbidden(page))
938 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700939 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700941 for_each_migratetype_order(order, t) {
942 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700943 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700945 pfn = page_to_pfn(list_entry(curr, struct page, lru));
946 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700947 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700948 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700949 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 spin_unlock_irqrestore(&zone->lock, flags);
951}
Mel Gormane2c55dc2007-10-16 01:25:50 -0700952#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
954/*
955 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
956 */
957void drain_local_pages(void)
958{
959 unsigned long flags;
960
961 local_irq_save(flags);
962 __drain_pages(smp_processor_id());
963 local_irq_restore(flags);
964}
Mel Gormane2c55dc2007-10-16 01:25:50 -0700965
966void smp_drain_local_pages(void *arg)
967{
968 drain_local_pages();
969}
970
971/*
972 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
973 */
974void drain_all_local_pages(void)
975{
976 unsigned long flags;
977
978 local_irq_save(flags);
979 __drain_pages(smp_processor_id());
980 local_irq_restore(flags);
981
982 smp_call_function(smp_drain_local_pages, NULL, 0, 1);
983}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985/*
986 * Free a 0-order page
987 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988static void fastcall free_hot_cold_page(struct page *page, int cold)
989{
990 struct zone *zone = page_zone(page);
991 struct per_cpu_pages *pcp;
992 unsigned long flags;
993
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 if (PageAnon(page))
995 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800996 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800997 return;
998
Nick Piggin9858db52006-10-11 01:21:30 -0700999 if (!PageHighMem(page))
1000 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Nick Piggindafb1362006-10-11 01:21:30 -07001001 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -08001002 kernel_map_pages(page, 1, 0);
1003
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001004 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -07001006 __count_vm_event(PGFREE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 list_add(&page->lru, &pcp->list);
Mel Gorman535131e62007-10-16 01:25:49 -07001008 set_page_private(page, get_pageblock_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001010 if (pcp->count >= pcp->high) {
1011 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1012 pcp->count -= pcp->batch;
1013 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 local_irq_restore(flags);
1015 put_cpu();
1016}
1017
1018void fastcall free_hot_page(struct page *page)
1019{
1020 free_hot_cold_page(page, 0);
1021}
1022
1023void fastcall free_cold_page(struct page *page)
1024{
1025 free_hot_cold_page(page, 1);
1026}
1027
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001028/*
1029 * split_page takes a non-compound higher-order page, and splits it into
1030 * n (1<<order) sub-pages: page[0..n]
1031 * Each sub-page must be freed individually.
1032 *
1033 * Note: this is probably too low level an operation for use in drivers.
1034 * Please consult with lkml before using this in your driver.
1035 */
1036void split_page(struct page *page, unsigned int order)
1037{
1038 int i;
1039
Nick Piggin725d7042006-09-25 23:30:55 -07001040 VM_BUG_ON(PageCompound(page));
1041 VM_BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -08001042 for (i = 1; i < (1 << order); i++)
1043 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001044}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001045
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046/*
1047 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1048 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1049 * or two.
1050 */
Nick Piggina74609f2006-01-06 00:11:20 -08001051static struct page *buffered_rmqueue(struct zonelist *zonelist,
1052 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053{
1054 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001055 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -08001057 int cpu;
Mel Gorman64c5e132007-10-16 01:25:59 -07001058 int migratetype = allocflags_to_migratetype(gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059
Hugh Dickins689bceb2005-11-21 21:32:20 -08001060again:
Nick Piggina74609f2006-01-06 00:11:20 -08001061 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -08001062 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 struct per_cpu_pages *pcp;
1064
Nick Piggina74609f2006-01-06 00:11:20 -08001065 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001067 if (!pcp->count) {
nkalmala941c7102006-11-02 22:07:04 -08001068 pcp->count = rmqueue_bulk(zone, 0,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001069 pcp->batch, &pcp->list, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001070 if (unlikely(!pcp->count))
1071 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 }
Mel Gorman535131e62007-10-16 01:25:49 -07001073
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001074 /* Find a page of the appropriate migrate type */
1075 list_for_each_entry(page, &pcp->list, lru)
1076 if (page_private(page) == migratetype)
1077 break;
1078
1079 /* Allocate more to the pcp list if necessary */
1080 if (unlikely(&page->lru == &pcp->list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001081 pcp->count += rmqueue_bulk(zone, 0,
1082 pcp->batch, &pcp->list, migratetype);
1083 page = list_entry(pcp->list.next, struct page, lru);
Mel Gorman535131e62007-10-16 01:25:49 -07001084 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001085
1086 list_del(&page->lru);
1087 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001088 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001090 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001091 spin_unlock(&zone->lock);
1092 if (!page)
1093 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 }
1095
Christoph Lameterf8891e52006-06-30 01:55:45 -07001096 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Christoph Lameterca889e62006-06-30 01:55:44 -07001097 zone_statistics(zonelist, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001098 local_irq_restore(flags);
1099 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
Nick Piggin725d7042006-09-25 23:30:55 -07001101 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001102 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001103 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001105
1106failed:
1107 local_irq_restore(flags);
1108 put_cpu();
1109 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110}
1111
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001112#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -08001113#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1114#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1115#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1116#define ALLOC_HARDER 0x10 /* try to alloc harder */
1117#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1118#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001119
Akinobu Mita933e3122006-12-08 02:39:45 -08001120#ifdef CONFIG_FAIL_PAGE_ALLOC
1121
1122static struct fail_page_alloc_attr {
1123 struct fault_attr attr;
1124
1125 u32 ignore_gfp_highmem;
1126 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001127 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001128
1129#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1130
1131 struct dentry *ignore_gfp_highmem_file;
1132 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001133 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001134
1135#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1136
1137} fail_page_alloc = {
1138 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001139 .ignore_gfp_wait = 1,
1140 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001141 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001142};
1143
1144static int __init setup_fail_page_alloc(char *str)
1145{
1146 return setup_fault_attr(&fail_page_alloc.attr, str);
1147}
1148__setup("fail_page_alloc=", setup_fail_page_alloc);
1149
1150static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1151{
Akinobu Mita54114992007-07-15 23:40:23 -07001152 if (order < fail_page_alloc.min_order)
1153 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001154 if (gfp_mask & __GFP_NOFAIL)
1155 return 0;
1156 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1157 return 0;
1158 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1159 return 0;
1160
1161 return should_fail(&fail_page_alloc.attr, 1 << order);
1162}
1163
1164#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1165
1166static int __init fail_page_alloc_debugfs(void)
1167{
1168 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1169 struct dentry *dir;
1170 int err;
1171
1172 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1173 "fail_page_alloc");
1174 if (err)
1175 return err;
1176 dir = fail_page_alloc.attr.dentries.dir;
1177
1178 fail_page_alloc.ignore_gfp_wait_file =
1179 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1180 &fail_page_alloc.ignore_gfp_wait);
1181
1182 fail_page_alloc.ignore_gfp_highmem_file =
1183 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1184 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001185 fail_page_alloc.min_order_file =
1186 debugfs_create_u32("min-order", mode, dir,
1187 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001188
1189 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001190 !fail_page_alloc.ignore_gfp_highmem_file ||
1191 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001192 err = -ENOMEM;
1193 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1194 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001195 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001196 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1197 }
1198
1199 return err;
1200}
1201
1202late_initcall(fail_page_alloc_debugfs);
1203
1204#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1205
1206#else /* CONFIG_FAIL_PAGE_ALLOC */
1207
1208static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1209{
1210 return 0;
1211}
1212
1213#endif /* CONFIG_FAIL_PAGE_ALLOC */
1214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215/*
1216 * Return 1 if free pages are above 'mark'. This takes into account the order
1217 * of the allocation.
1218 */
1219int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001220 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221{
1222 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001223 long min = mark;
1224 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 int o;
1226
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001227 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001229 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 min -= min / 4;
1231
1232 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1233 return 0;
1234 for (o = 0; o < order; o++) {
1235 /* At the next order, this order's pages become unavailable */
1236 free_pages -= z->free_area[o].nr_free << o;
1237
1238 /* Require fewer higher order pages to be free */
1239 min >>= 1;
1240
1241 if (free_pages <= min)
1242 return 0;
1243 }
1244 return 1;
1245}
1246
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001247#ifdef CONFIG_NUMA
1248/*
1249 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1250 * skip over zones that are not allowed by the cpuset, or that have
1251 * been recently (in last second) found to be nearly full. See further
1252 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001253 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001254 *
1255 * If the zonelist cache is present in the passed in zonelist, then
1256 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001257 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001258 *
1259 * If the zonelist cache is not available for this zonelist, does
1260 * nothing and returns NULL.
1261 *
1262 * If the fullzones BITMAP in the zonelist cache is stale (more than
1263 * a second since last zap'd) then we zap it out (clear its bits.)
1264 *
1265 * We hold off even calling zlc_setup, until after we've checked the
1266 * first zone in the zonelist, on the theory that most allocations will
1267 * be satisfied from that first zone, so best to examine that zone as
1268 * quickly as we can.
1269 */
1270static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1271{
1272 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1273 nodemask_t *allowednodes; /* zonelist_cache approximation */
1274
1275 zlc = zonelist->zlcache_ptr;
1276 if (!zlc)
1277 return NULL;
1278
1279 if (jiffies - zlc->last_full_zap > 1 * HZ) {
1280 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1281 zlc->last_full_zap = jiffies;
1282 }
1283
1284 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1285 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001286 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001287 return allowednodes;
1288}
1289
1290/*
1291 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1292 * if it is worth looking at further for free memory:
1293 * 1) Check that the zone isn't thought to be full (doesn't have its
1294 * bit set in the zonelist_cache fullzones BITMAP).
1295 * 2) Check that the zones node (obtained from the zonelist_cache
1296 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1297 * Return true (non-zero) if zone is worth looking at further, or
1298 * else return false (zero) if it is not.
1299 *
1300 * This check -ignores- the distinction between various watermarks,
1301 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1302 * found to be full for any variation of these watermarks, it will
1303 * be considered full for up to one second by all requests, unless
1304 * we are so low on memory on all allowed nodes that we are forced
1305 * into the second scan of the zonelist.
1306 *
1307 * In the second scan we ignore this zonelist cache and exactly
1308 * apply the watermarks to all zones, even it is slower to do so.
1309 * We are low on memory in the second scan, and should leave no stone
1310 * unturned looking for a free page.
1311 */
1312static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1313 nodemask_t *allowednodes)
1314{
1315 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1316 int i; /* index of *z in zonelist zones */
1317 int n; /* node that zone *z is on */
1318
1319 zlc = zonelist->zlcache_ptr;
1320 if (!zlc)
1321 return 1;
1322
1323 i = z - zonelist->zones;
1324 n = zlc->z_to_n[i];
1325
1326 /* This zone is worth trying if it is allowed but not full */
1327 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1328}
1329
1330/*
1331 * Given 'z' scanning a zonelist, set the corresponding bit in
1332 * zlc->fullzones, so that subsequent attempts to allocate a page
1333 * from that zone don't waste time re-examining it.
1334 */
1335static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1336{
1337 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1338 int i; /* index of *z in zonelist zones */
1339
1340 zlc = zonelist->zlcache_ptr;
1341 if (!zlc)
1342 return;
1343
1344 i = z - zonelist->zones;
1345
1346 set_bit(i, zlc->fullzones);
1347}
1348
1349#else /* CONFIG_NUMA */
1350
1351static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1352{
1353 return NULL;
1354}
1355
1356static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1357 nodemask_t *allowednodes)
1358{
1359 return 1;
1360}
1361
1362static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1363{
1364}
1365#endif /* CONFIG_NUMA */
1366
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001367/*
Paul Jackson0798e512006-12-06 20:31:38 -08001368 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001369 * a page.
1370 */
1371static struct page *
1372get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1373 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -07001374{
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001375 struct zone **z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001376 struct page *page = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001377 int classzone_idx = zone_idx(zonelist->zones[0]);
Christoph Lameter1192d522006-09-25 23:31:45 -07001378 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001379 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1380 int zlc_active = 0; /* set if using zonelist_cache */
1381 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gormanb377fd32007-08-22 14:02:05 -07001382 enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001383
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001384zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001385 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001386 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001387 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1388 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001389 z = zonelist->zones;
1390
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001391 do {
Mel Gormanb377fd32007-08-22 14:02:05 -07001392 /*
1393 * In NUMA, this could be a policy zonelist which contains
1394 * zones that may not be allowed by the current gfp_mask.
1395 * Check the zone is allowed by the current flags
1396 */
1397 if (unlikely(alloc_should_filter_zonelist(zonelist))) {
1398 if (highest_zoneidx == -1)
1399 highest_zoneidx = gfp_zone(gfp_mask);
1400 if (zone_idx(*z) > highest_zoneidx)
1401 continue;
1402 }
1403
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001404 if (NUMA_BUILD && zlc_active &&
1405 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1406 continue;
Christoph Lameter1192d522006-09-25 23:31:45 -07001407 zone = *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001408 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001409 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001410 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001411
1412 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001413 unsigned long mark;
1414 if (alloc_flags & ALLOC_WMARK_MIN)
Christoph Lameter1192d522006-09-25 23:31:45 -07001415 mark = zone->pages_min;
Nick Piggin31488902005-11-28 13:44:03 -08001416 else if (alloc_flags & ALLOC_WMARK_LOW)
Christoph Lameter1192d522006-09-25 23:31:45 -07001417 mark = zone->pages_low;
Nick Piggin31488902005-11-28 13:44:03 -08001418 else
Christoph Lameter1192d522006-09-25 23:31:45 -07001419 mark = zone->pages_high;
Paul Jackson0798e512006-12-06 20:31:38 -08001420 if (!zone_watermark_ok(zone, order, mark,
1421 classzone_idx, alloc_flags)) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001422 if (!zone_reclaim_mode ||
Christoph Lameter1192d522006-09-25 23:31:45 -07001423 !zone_reclaim(zone, gfp_mask, order))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001424 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001425 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001426 }
1427
Christoph Lameter1192d522006-09-25 23:31:45 -07001428 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
Paul Jackson0798e512006-12-06 20:31:38 -08001429 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001430 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001431this_zone_full:
1432 if (NUMA_BUILD)
1433 zlc_mark_zone_full(zonelist, z);
1434try_next_zone:
1435 if (NUMA_BUILD && !did_zlc_setup) {
1436 /* we do zlc_setup after the first zone is tried */
1437 allowednodes = zlc_setup(zonelist, alloc_flags);
1438 zlc_active = 1;
1439 did_zlc_setup = 1;
1440 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001441 } while (*(++z) != NULL);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001442
1443 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1444 /* Disable zlc cache for second zonelist scan */
1445 zlc_active = 0;
1446 goto zonelist_scan;
1447 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001448 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001449}
1450
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451/*
1452 * This is the 'heart' of the zoned buddy allocator.
1453 */
1454struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +01001455__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 struct zonelist *zonelist)
1457{
Al Viro260b2362005-10-21 03:22:44 -04001458 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001459 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 struct page *page;
1461 struct reclaim_state reclaim_state;
1462 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001464 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 int did_some_progress;
1466
1467 might_sleep_if(wait);
1468
Akinobu Mita933e3122006-12-08 02:39:45 -08001469 if (should_fail_alloc_page(gfp_mask, order))
1470 return NULL;
1471
Jens Axboe6b1de912005-11-17 21:35:02 +01001472restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001473 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001475 if (unlikely(*z == NULL)) {
Christoph Lameter523b9452007-10-16 01:25:37 -07001476 /*
1477 * Happens if we have an empty zonelist as a result of
1478 * GFP_THISNODE being used on a memoryless node
1479 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 return NULL;
1481 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001482
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001483 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001484 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001485 if (page)
1486 goto got_pg;
1487
Christoph Lameter952f3b52006-12-06 20:33:26 -08001488 /*
1489 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1490 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1491 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1492 * using a larger set of nodes after it has established that the
1493 * allowed per node queues are empty and that nodes are
1494 * over allocated.
1495 */
1496 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1497 goto nopage;
1498
Paul Jackson0798e512006-12-06 20:31:38 -08001499 for (z = zonelist->zones; *z; z++)
Chris Wright43b0bc02006-06-25 05:47:55 -07001500 wakeup_kswapd(*z, order);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001501
Paul Jackson9bf22292005-09-06 15:18:12 -07001502 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001503 * OK, we're below the kswapd watermark and have kicked background
1504 * reclaim. Now things get more complex, so set up alloc_flags according
1505 * to how we want to proceed.
1506 *
1507 * The caller may dip into page reserves a bit more if the caller
1508 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -08001509 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1510 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -07001511 */
Nick Piggin31488902005-11-28 13:44:03 -08001512 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001513 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1514 alloc_flags |= ALLOC_HARDER;
1515 if (gfp_mask & __GFP_HIGH)
1516 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -07001517 if (wait)
1518 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519
1520 /*
1521 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001522 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 *
1524 * This is the last chance, in general, before the goto nopage.
1525 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -07001526 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001528 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1529 if (page)
1530 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531
1532 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -07001533
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001534rebalance:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001535 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1536 && !in_interrupt()) {
1537 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -08001538nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001539 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001540 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -08001541 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001542 if (page)
1543 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -08001544 if (gfp_mask & __GFP_NOFAIL) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001545 congestion_wait(WRITE, HZ/50);
Kirill Korotaev885036d2005-11-13 16:06:41 -08001546 goto nofail_alloc;
1547 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 }
1549 goto nopage;
1550 }
1551
1552 /* Atomic allocations - we can't balance anything */
1553 if (!wait)
1554 goto nopage;
1555
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 cond_resched();
1557
1558 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001559 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 p->flags |= PF_MEMALLOC;
1561 reclaim_state.reclaimed_slab = 0;
1562 p->reclaim_state = &reclaim_state;
1563
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001564 did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565
1566 p->reclaim_state = NULL;
1567 p->flags &= ~PF_MEMALLOC;
1568
1569 cond_resched();
1570
Mel Gormane2c55dc2007-10-16 01:25:50 -07001571 if (order != 0)
1572 drain_all_local_pages();
1573
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001575 page = get_page_from_freelist(gfp_mask, order,
1576 zonelist, alloc_flags);
1577 if (page)
1578 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
David Rientjesff0ceb92007-10-16 23:25:56 -07001580 if (!try_set_zone_oom(zonelist)) {
1581 schedule_timeout_uninterruptible(1);
1582 goto restart;
1583 }
1584
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 /*
1586 * Go through the zonelist yet one more time, keep
1587 * very high watermark here, this is only to catch
1588 * a parallel oom killing, we must fail if we're still
1589 * under heavy pressure.
1590 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001591 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001592 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
David Rientjesff0ceb92007-10-16 23:25:56 -07001593 if (page) {
1594 clear_zonelist_oom(zonelist);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001595 goto got_pg;
David Rientjesff0ceb92007-10-16 23:25:56 -07001596 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597
Mel Gormana8bbf722007-07-31 00:37:30 -07001598 /* The OOM killer will not help higher order allocs so fail */
David Rientjesff0ceb92007-10-16 23:25:56 -07001599 if (order > PAGE_ALLOC_COSTLY_ORDER) {
1600 clear_zonelist_oom(zonelist);
Mel Gormana8bbf722007-07-31 00:37:30 -07001601 goto nopage;
David Rientjesff0ceb92007-10-16 23:25:56 -07001602 }
Mel Gormana8bbf722007-07-31 00:37:30 -07001603
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001604 out_of_memory(zonelist, gfp_mask, order);
David Rientjesff0ceb92007-10-16 23:25:56 -07001605 clear_zonelist_oom(zonelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 goto restart;
1607 }
1608
1609 /*
1610 * Don't let big-order allocations loop unless the caller explicitly
1611 * requests that. Wait for some write requests to complete then retry.
1612 *
1613 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1614 * <= 3, but that may not be true in other implementations.
1615 */
1616 do_retry = 0;
1617 if (!(gfp_mask & __GFP_NORETRY)) {
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001618 if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
1619 (gfp_mask & __GFP_REPEAT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 do_retry = 1;
1621 if (gfp_mask & __GFP_NOFAIL)
1622 do_retry = 1;
1623 }
1624 if (do_retry) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001625 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 goto rebalance;
1627 }
1628
1629nopage:
1630 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1631 printk(KERN_WARNING "%s: page allocation failure."
1632 " order:%d, mode:0x%x\n",
1633 p->comm, order, gfp_mask);
1634 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001635 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 return page;
1639}
1640
1641EXPORT_SYMBOL(__alloc_pages);
1642
1643/*
1644 * Common helper functions.
1645 */
Al Virodd0fc662005-10-07 07:46:04 +01001646fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647{
1648 struct page * page;
1649 page = alloc_pages(gfp_mask, order);
1650 if (!page)
1651 return 0;
1652 return (unsigned long) page_address(page);
1653}
1654
1655EXPORT_SYMBOL(__get_free_pages);
1656
Al Virodd0fc662005-10-07 07:46:04 +01001657fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658{
1659 struct page * page;
1660
1661 /*
1662 * get_zeroed_page() returns a 32-bit address, which cannot represent
1663 * a highmem page
1664 */
Nick Piggin725d7042006-09-25 23:30:55 -07001665 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666
1667 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1668 if (page)
1669 return (unsigned long) page_address(page);
1670 return 0;
1671}
1672
1673EXPORT_SYMBOL(get_zeroed_page);
1674
1675void __pagevec_free(struct pagevec *pvec)
1676{
1677 int i = pagevec_count(pvec);
1678
1679 while (--i >= 0)
1680 free_hot_cold_page(pvec->pages[i], pvec->cold);
1681}
1682
1683fastcall void __free_pages(struct page *page, unsigned int order)
1684{
Nick Pigginb5810032005-10-29 18:16:12 -07001685 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 if (order == 0)
1687 free_hot_page(page);
1688 else
1689 __free_pages_ok(page, order);
1690 }
1691}
1692
1693EXPORT_SYMBOL(__free_pages);
1694
1695fastcall void free_pages(unsigned long addr, unsigned int order)
1696{
1697 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001698 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 __free_pages(virt_to_page((void *)addr), order);
1700 }
1701}
1702
1703EXPORT_SYMBOL(free_pages);
1704
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705static unsigned int nr_free_zone_pages(int offset)
1706{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001707 /* Just pick one node, since fallback list is circular */
1708 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 unsigned int sum = 0;
1710
Martin J. Blighe310fd42005-07-29 22:59:18 -07001711 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1712 struct zone **zonep = zonelist->zones;
1713 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
Martin J. Blighe310fd42005-07-29 22:59:18 -07001715 for (zone = *zonep++; zone; zone = *zonep++) {
1716 unsigned long size = zone->present_pages;
1717 unsigned long high = zone->pages_high;
1718 if (size > high)
1719 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 }
1721
1722 return sum;
1723}
1724
1725/*
1726 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1727 */
1728unsigned int nr_free_buffer_pages(void)
1729{
Al Viroaf4ca452005-10-21 02:55:38 -04001730 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731}
Meelap Shahc2f1a552007-07-17 04:04:39 -07001732EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
1734/*
1735 * Amount of free RAM allocatable within all zones
1736 */
1737unsigned int nr_free_pagecache_pages(void)
1738{
Mel Gorman2a1e2742007-07-17 04:03:12 -07001739 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001741
1742static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001744 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08001745 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748void si_meminfo(struct sysinfo *val)
1749{
1750 val->totalram = totalram_pages;
1751 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001752 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 val->totalhigh = totalhigh_pages;
1755 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 val->mem_unit = PAGE_SIZE;
1757}
1758
1759EXPORT_SYMBOL(si_meminfo);
1760
1761#ifdef CONFIG_NUMA
1762void si_meminfo_node(struct sysinfo *val, int nid)
1763{
1764 pg_data_t *pgdat = NODE_DATA(nid);
1765
1766 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001767 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001768#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001770 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1771 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001772#else
1773 val->totalhigh = 0;
1774 val->freehigh = 0;
1775#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 val->mem_unit = PAGE_SIZE;
1777}
1778#endif
1779
1780#define K(x) ((x) << (PAGE_SHIFT-10))
1781
1782/*
1783 * Show free area list (used inside shift_scroll-lock stuff)
1784 * We also calculate the percentage fragmentation. We do this by counting the
1785 * memory on each free list with the exception of the first item on the list.
1786 */
1787void show_free_areas(void)
1788{
Jes Sorensenc7241912006-09-27 01:50:05 -07001789 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 struct zone *zone;
1791
1792 for_each_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07001793 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07001795
1796 show_node(zone);
1797 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798
Dave Jones6b482c62005-11-10 15:45:56 -05001799 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 struct per_cpu_pageset *pageset;
1801
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001802 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
Jes Sorensenc7241912006-09-27 01:50:05 -07001804 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1805 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1806 cpu, pageset->pcp[0].high,
1807 pageset->pcp[0].batch, pageset->pcp[0].count,
1808 pageset->pcp[1].high, pageset->pcp[1].batch,
1809 pageset->pcp[1].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 }
1811 }
1812
Andrew Mortona25700a2007-02-08 14:20:40 -08001813 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
Christoph Lameterd23ad422007-02-10 01:43:02 -08001814 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
Christoph Lameter65e458d42007-02-10 01:43:05 -08001815 global_page_state(NR_ACTIVE),
1816 global_page_state(NR_INACTIVE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07001817 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07001818 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001819 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08001820 global_page_state(NR_FREE_PAGES),
Christoph Lameter972d1a72006-09-25 23:31:51 -07001821 global_page_state(NR_SLAB_RECLAIMABLE) +
1822 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001823 global_page_state(NR_FILE_MAPPED),
Andrew Mortona25700a2007-02-08 14:20:40 -08001824 global_page_state(NR_PAGETABLE),
1825 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
1827 for_each_zone(zone) {
1828 int i;
1829
Jes Sorensenc7241912006-09-27 01:50:05 -07001830 if (!populated_zone(zone))
1831 continue;
1832
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 show_node(zone);
1834 printk("%s"
1835 " free:%lukB"
1836 " min:%lukB"
1837 " low:%lukB"
1838 " high:%lukB"
1839 " active:%lukB"
1840 " inactive:%lukB"
1841 " present:%lukB"
1842 " pages_scanned:%lu"
1843 " all_unreclaimable? %s"
1844 "\n",
1845 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08001846 K(zone_page_state(zone, NR_FREE_PAGES)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 K(zone->pages_min),
1848 K(zone->pages_low),
1849 K(zone->pages_high),
Christoph Lameterc8785382007-02-10 01:43:01 -08001850 K(zone_page_state(zone, NR_ACTIVE)),
1851 K(zone_page_state(zone, NR_INACTIVE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 K(zone->present_pages),
1853 zone->pages_scanned,
David Rientjese815af92007-10-16 23:25:54 -07001854 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 );
1856 printk("lowmem_reserve[]:");
1857 for (i = 0; i < MAX_NR_ZONES; i++)
1858 printk(" %lu", zone->lowmem_reserve[i]);
1859 printk("\n");
1860 }
1861
1862 for_each_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001863 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
Jes Sorensenc7241912006-09-27 01:50:05 -07001865 if (!populated_zone(zone))
1866 continue;
1867
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 show_node(zone);
1869 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
1871 spin_lock_irqsave(&zone->lock, flags);
1872 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001873 nr[order] = zone->free_area[order].nr_free;
1874 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 }
1876 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001877 for (order = 0; order < MAX_ORDER; order++)
1878 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 printk("= %lukB\n", K(total));
1880 }
1881
1882 show_swap_cache_info();
1883}
1884
1885/*
1886 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001887 *
1888 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001890static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
1891 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892{
Christoph Lameter1a932052006-01-06 00:11:16 -08001893 struct zone *zone;
1894
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001895 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001896 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001897
1898 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001899 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08001900 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001901 if (populated_zone(zone)) {
Christoph Lameter070f8032006-01-06 00:11:19 -08001902 zonelist->zones[nr_zones++] = zone;
1903 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08001905
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001906 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08001907 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908}
1909
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001910
1911/*
1912 * zonelist_order:
1913 * 0 = automatic detection of better ordering.
1914 * 1 = order by ([node] distance, -zonetype)
1915 * 2 = order by (-zonetype, [node] distance)
1916 *
1917 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1918 * the same zonelist. So only NUMA can configure this param.
1919 */
1920#define ZONELIST_ORDER_DEFAULT 0
1921#define ZONELIST_ORDER_NODE 1
1922#define ZONELIST_ORDER_ZONE 2
1923
1924/* zonelist order in the kernel.
1925 * set_zonelist_order() will set this to NODE or ZONE.
1926 */
1927static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
1928static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
1929
1930
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001932/* The value user specified ....changed by config */
1933static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1934/* string for sysctl */
1935#define NUMA_ZONELIST_ORDER_LEN 16
1936char numa_zonelist_order[16] = "default";
1937
1938/*
1939 * interface for configure zonelist ordering.
1940 * command line option "numa_zonelist_order"
1941 * = "[dD]efault - default, automatic configuration.
1942 * = "[nN]ode - order by node locality, then by zone within node
1943 * = "[zZ]one - order by zone, then by locality within zone
1944 */
1945
1946static int __parse_numa_zonelist_order(char *s)
1947{
1948 if (*s == 'd' || *s == 'D') {
1949 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1950 } else if (*s == 'n' || *s == 'N') {
1951 user_zonelist_order = ZONELIST_ORDER_NODE;
1952 } else if (*s == 'z' || *s == 'Z') {
1953 user_zonelist_order = ZONELIST_ORDER_ZONE;
1954 } else {
1955 printk(KERN_WARNING
1956 "Ignoring invalid numa_zonelist_order value: "
1957 "%s\n", s);
1958 return -EINVAL;
1959 }
1960 return 0;
1961}
1962
1963static __init int setup_numa_zonelist_order(char *s)
1964{
1965 if (s)
1966 return __parse_numa_zonelist_order(s);
1967 return 0;
1968}
1969early_param("numa_zonelist_order", setup_numa_zonelist_order);
1970
1971/*
1972 * sysctl handler for numa_zonelist_order
1973 */
1974int numa_zonelist_order_handler(ctl_table *table, int write,
1975 struct file *file, void __user *buffer, size_t *length,
1976 loff_t *ppos)
1977{
1978 char saved_string[NUMA_ZONELIST_ORDER_LEN];
1979 int ret;
1980
1981 if (write)
1982 strncpy(saved_string, (char*)table->data,
1983 NUMA_ZONELIST_ORDER_LEN);
1984 ret = proc_dostring(table, write, file, buffer, length, ppos);
1985 if (ret)
1986 return ret;
1987 if (write) {
1988 int oldval = user_zonelist_order;
1989 if (__parse_numa_zonelist_order((char*)table->data)) {
1990 /*
1991 * bogus value. restore saved string
1992 */
1993 strncpy((char*)table->data, saved_string,
1994 NUMA_ZONELIST_ORDER_LEN);
1995 user_zonelist_order = oldval;
1996 } else if (oldval != user_zonelist_order)
1997 build_all_zonelists();
1998 }
1999 return 0;
2000}
2001
2002
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003#define MAX_NODE_LOAD (num_online_nodes())
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002004static int node_load[MAX_NUMNODES];
2005
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002007 * 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 -07002008 * @node: node whose fallback list we're appending
2009 * @used_node_mask: nodemask_t of already used nodes
2010 *
2011 * We use a number of factors to determine which is the next node that should
2012 * appear on a given node's fallback list. The node should not have appeared
2013 * already in @node's fallback list, and it should be the next closest node
2014 * according to the distance array (which contains arbitrary distance values
2015 * from each node to each node in the system), and should also prefer nodes
2016 * with no CPUs, since presumably they'll have very little allocation pressure
2017 * on them otherwise.
2018 * It returns -1 if no node is found.
2019 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002020static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002022 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 int min_val = INT_MAX;
2024 int best_node = -1;
2025
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002026 /* Use the local node if we haven't already */
2027 if (!node_isset(node, *used_node_mask)) {
2028 node_set(node, *used_node_mask);
2029 return node;
2030 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002032 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002033 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034
2035 /* Don't want a node to appear more than once */
2036 if (node_isset(n, *used_node_mask))
2037 continue;
2038
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 /* Use the distance array to find the distance */
2040 val = node_distance(node, n);
2041
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002042 /* Penalize nodes under us ("prefer the next node") */
2043 val += (n < node);
2044
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 /* Give preference to headless and unused nodes */
2046 tmp = node_to_cpumask(n);
2047 if (!cpus_empty(tmp))
2048 val += PENALTY_FOR_NODE_WITH_CPUS;
2049
2050 /* Slight preference for less loaded node */
2051 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2052 val += node_load[n];
2053
2054 if (val < min_val) {
2055 min_val = val;
2056 best_node = n;
2057 }
2058 }
2059
2060 if (best_node >= 0)
2061 node_set(best_node, *used_node_mask);
2062
2063 return best_node;
2064}
2065
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002066
2067/*
2068 * Build zonelists ordered by node and zones within node.
2069 * This results in maximum locality--normal zone overflows into local
2070 * DMA zone, if any--but risks exhausting DMA zone.
2071 */
2072static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073{
Christoph Lameter19655d32006-09-25 23:31:19 -07002074 enum zone_type i;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002075 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002077
2078 for (i = 0; i < MAX_NR_ZONES; i++) {
2079 zonelist = pgdat->node_zonelists + i;
2080 for (j = 0; zonelist->zones[j] != NULL; j++)
2081 ;
2082 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2083 zonelist->zones[j] = NULL;
2084 }
2085}
2086
2087/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002088 * Build gfp_thisnode zonelists
2089 */
2090static void build_thisnode_zonelists(pg_data_t *pgdat)
2091{
2092 enum zone_type i;
2093 int j;
2094 struct zonelist *zonelist;
2095
2096 for (i = 0; i < MAX_NR_ZONES; i++) {
2097 zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i;
2098 j = build_zonelists_node(pgdat, zonelist, 0, i);
2099 zonelist->zones[j] = NULL;
2100 }
2101}
2102
2103/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002104 * Build zonelists ordered by zone and nodes within zones.
2105 * This results in conserving DMA zone[s] until all Normal memory is
2106 * exhausted, but results in overflowing to remote node while memory
2107 * may still exist in local DMA zone.
2108 */
2109static int node_order[MAX_NUMNODES];
2110
2111static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2112{
2113 enum zone_type i;
2114 int pos, j, node;
2115 int zone_type; /* needs to be signed */
2116 struct zone *z;
2117 struct zonelist *zonelist;
2118
2119 for (i = 0; i < MAX_NR_ZONES; i++) {
2120 zonelist = pgdat->node_zonelists + i;
2121 pos = 0;
2122 for (zone_type = i; zone_type >= 0; zone_type--) {
2123 for (j = 0; j < nr_nodes; j++) {
2124 node = node_order[j];
2125 z = &NODE_DATA(node)->node_zones[zone_type];
2126 if (populated_zone(z)) {
2127 zonelist->zones[pos++] = z;
2128 check_highest_zone(zone_type);
2129 }
2130 }
2131 }
2132 zonelist->zones[pos] = NULL;
2133 }
2134}
2135
2136static int default_zonelist_order(void)
2137{
2138 int nid, zone_type;
2139 unsigned long low_kmem_size,total_size;
2140 struct zone *z;
2141 int average_size;
2142 /*
2143 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2144 * If they are really small and used heavily, the system can fall
2145 * into OOM very easily.
2146 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2147 */
2148 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2149 low_kmem_size = 0;
2150 total_size = 0;
2151 for_each_online_node(nid) {
2152 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2153 z = &NODE_DATA(nid)->node_zones[zone_type];
2154 if (populated_zone(z)) {
2155 if (zone_type < ZONE_NORMAL)
2156 low_kmem_size += z->present_pages;
2157 total_size += z->present_pages;
2158 }
2159 }
2160 }
2161 if (!low_kmem_size || /* there are no DMA area. */
2162 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2163 return ZONELIST_ORDER_NODE;
2164 /*
2165 * look into each node's config.
2166 * If there is a node whose DMA/DMA32 memory is very big area on
2167 * local memory, NODE_ORDER may be suitable.
2168 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002169 average_size = total_size /
2170 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002171 for_each_online_node(nid) {
2172 low_kmem_size = 0;
2173 total_size = 0;
2174 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2175 z = &NODE_DATA(nid)->node_zones[zone_type];
2176 if (populated_zone(z)) {
2177 if (zone_type < ZONE_NORMAL)
2178 low_kmem_size += z->present_pages;
2179 total_size += z->present_pages;
2180 }
2181 }
2182 if (low_kmem_size &&
2183 total_size > average_size && /* ignore small node */
2184 low_kmem_size > total_size * 70/100)
2185 return ZONELIST_ORDER_NODE;
2186 }
2187 return ZONELIST_ORDER_ZONE;
2188}
2189
2190static void set_zonelist_order(void)
2191{
2192 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2193 current_zonelist_order = default_zonelist_order();
2194 else
2195 current_zonelist_order = user_zonelist_order;
2196}
2197
2198static void build_zonelists(pg_data_t *pgdat)
2199{
2200 int j, node, load;
2201 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002203 int local_node, prev_node;
2204 struct zonelist *zonelist;
2205 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206
2207 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002208 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 zonelist = pgdat->node_zonelists + i;
2210 zonelist->zones[0] = NULL;
2211 }
2212
2213 /* NUMA-aware ordering of nodes */
2214 local_node = pgdat->node_id;
2215 load = num_online_nodes();
2216 prev_node = local_node;
2217 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002218
2219 memset(node_load, 0, sizeof(node_load));
2220 memset(node_order, 0, sizeof(node_order));
2221 j = 0;
2222
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002224 int distance = node_distance(local_node, node);
2225
2226 /*
2227 * If another node is sufficiently far away then it is better
2228 * to reclaim pages in a zone before going off node.
2229 */
2230 if (distance > RECLAIM_DISTANCE)
2231 zone_reclaim_mode = 1;
2232
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 /*
2234 * We don't want to pressure a particular node.
2235 * So adding penalty to the first node in same
2236 * distance group to make it round-robin.
2237 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002238 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002239 node_load[node] = load;
2240
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 prev_node = node;
2242 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002243 if (order == ZONELIST_ORDER_NODE)
2244 build_zonelists_in_node_order(pgdat, node);
2245 else
2246 node_order[j++] = node; /* remember order */
2247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002249 if (order == ZONELIST_ORDER_ZONE) {
2250 /* calculate node order -- i.e., DMA last! */
2251 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002253
2254 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255}
2256
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002257/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002258static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002259{
2260 int i;
2261
2262 for (i = 0; i < MAX_NR_ZONES; i++) {
2263 struct zonelist *zonelist;
2264 struct zonelist_cache *zlc;
2265 struct zone **z;
2266
2267 zonelist = pgdat->node_zonelists + i;
2268 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2269 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2270 for (z = zonelist->zones; *z; z++)
2271 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
2272 }
2273}
2274
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002275
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276#else /* CONFIG_NUMA */
2277
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002278static void set_zonelist_order(void)
2279{
2280 current_zonelist_order = ZONELIST_ORDER_ZONE;
2281}
2282
2283static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284{
Christoph Lameter19655d32006-09-25 23:31:19 -07002285 int node, local_node;
2286 enum zone_type i,j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287
2288 local_node = pgdat->node_id;
Christoph Lameter19655d32006-09-25 23:31:19 -07002289 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 struct zonelist *zonelist;
2291
2292 zonelist = pgdat->node_zonelists + i;
2293
Christoph Lameter19655d32006-09-25 23:31:19 -07002294 j = build_zonelists_node(pgdat, zonelist, 0, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 /*
2296 * Now we build the zonelist so that it contains the zones
2297 * of all the other nodes.
2298 * We don't want to pressure a particular node, so when
2299 * building the zones for node N, we make sure that the
2300 * zones coming right after the local ones are those from
2301 * node N+1 (modulo N)
2302 */
2303 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2304 if (!node_online(node))
2305 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07002306 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 }
2308 for (node = 0; node < local_node; node++) {
2309 if (!node_online(node))
2310 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07002311 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 }
2313
2314 zonelist->zones[j] = NULL;
2315 }
2316}
2317
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002318/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002319static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002320{
2321 int i;
2322
2323 for (i = 0; i < MAX_NR_ZONES; i++)
2324 pgdat->node_zonelists[i].zlcache_ptr = NULL;
2325}
2326
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327#endif /* CONFIG_NUMA */
2328
Yasunori Goto68113782006-06-23 02:03:11 -07002329/* return values int ....just for stop_machine_run() */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002330static int __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331{
Yasunori Goto68113782006-06-23 02:03:11 -07002332 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002333
2334 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07002335 pg_data_t *pgdat = NODE_DATA(nid);
2336
2337 build_zonelists(pgdat);
2338 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002339 }
Yasunori Goto68113782006-06-23 02:03:11 -07002340 return 0;
2341}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002343void build_all_zonelists(void)
Yasunori Goto68113782006-06-23 02:03:11 -07002344{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002345 set_zonelist_order();
2346
Yasunori Goto68113782006-06-23 02:03:11 -07002347 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07002348 __build_all_zonelists(NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07002349 cpuset_init_current_mems_allowed();
2350 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02002351 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07002352 of zonelist */
2353 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
2354 /* cpuset refresh routine should be here */
2355 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07002356 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002357 /*
2358 * Disable grouping by mobility if the number of pages in the
2359 * system is too low to allow the mechanism to work. It would be
2360 * more accurate, but expensive to check per-zone. This check is
2361 * made on memory-hotadd so a system can start with mobility
2362 * disabled and enable it later
2363 */
Mel Gormand9c23402007-10-16 01:26:01 -07002364 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002365 page_group_by_mobility_disabled = 1;
2366 else
2367 page_group_by_mobility_disabled = 0;
2368
2369 printk("Built %i zonelists in %s order, mobility grouping %s. "
2370 "Total pages: %ld\n",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002371 num_online_nodes(),
2372 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002373 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002374 vm_total_pages);
2375#ifdef CONFIG_NUMA
2376 printk("Policy zone: %s\n", zone_names[policy_zone]);
2377#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378}
2379
2380/*
2381 * Helper functions to size the waitqueue hash table.
2382 * Essentially these want to choose hash table sizes sufficiently
2383 * large so that collisions trying to wait on pages are rare.
2384 * But in fact, the number of active page waitqueues on typical
2385 * systems is ridiculously low, less than 200. So this is even
2386 * conservative, even though it seems large.
2387 *
2388 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2389 * waitqueues, i.e. the size of the waitq table given the number of pages.
2390 */
2391#define PAGES_PER_WAITQUEUE 256
2392
Yasunori Gotocca448f2006-06-23 02:03:10 -07002393#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07002394static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395{
2396 unsigned long size = 1;
2397
2398 pages /= PAGES_PER_WAITQUEUE;
2399
2400 while (size < pages)
2401 size <<= 1;
2402
2403 /*
2404 * Once we have dozens or even hundreds of threads sleeping
2405 * on IO we've got bigger problems than wait queue collision.
2406 * Limit the size of the wait table to a reasonable size.
2407 */
2408 size = min(size, 4096UL);
2409
2410 return max(size, 4UL);
2411}
Yasunori Gotocca448f2006-06-23 02:03:10 -07002412#else
2413/*
2414 * A zone's size might be changed by hot-add, so it is not possible to determine
2415 * a suitable size for its wait_table. So we use the maximum size now.
2416 *
2417 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2418 *
2419 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2420 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2421 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2422 *
2423 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2424 * or more by the traditional way. (See above). It equals:
2425 *
2426 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2427 * ia64(16K page size) : = ( 8G + 4M)byte.
2428 * powerpc (64K page size) : = (32G +16M)byte.
2429 */
2430static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2431{
2432 return 4096UL;
2433}
2434#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435
2436/*
2437 * This is an integer logarithm so that shifts can be used later
2438 * to extract the more random high bits from the multiplicative
2439 * hash function before the remainder is taken.
2440 */
2441static inline unsigned long wait_table_bits(unsigned long size)
2442{
2443 return ffz(~size);
2444}
2445
2446#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2447
Mel Gorman56fd56b2007-10-16 01:25:58 -07002448/*
Mel Gormand9c23402007-10-16 01:26:01 -07002449 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman56fd56b2007-10-16 01:25:58 -07002450 * of blocks reserved is based on zone->pages_min. The memory within the
2451 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2452 * higher will lead to a bigger reserve which will get freed as contiguous
2453 * blocks as reclaim kicks in
2454 */
2455static void setup_zone_migrate_reserve(struct zone *zone)
2456{
2457 unsigned long start_pfn, pfn, end_pfn;
2458 struct page *page;
2459 unsigned long reserve, block_migratetype;
2460
2461 /* Get the start pfn, end pfn and the number of blocks to reserve */
2462 start_pfn = zone->zone_start_pfn;
2463 end_pfn = start_pfn + zone->spanned_pages;
Mel Gormand9c23402007-10-16 01:26:01 -07002464 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2465 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002466
Mel Gormand9c23402007-10-16 01:26:01 -07002467 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07002468 if (!pfn_valid(pfn))
2469 continue;
2470 page = pfn_to_page(pfn);
2471
2472 /* Blocks with reserved pages will never free, skip them. */
2473 if (PageReserved(page))
2474 continue;
2475
2476 block_migratetype = get_pageblock_migratetype(page);
2477
2478 /* If this block is reserved, account for it */
2479 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2480 reserve--;
2481 continue;
2482 }
2483
2484 /* Suitable for reserving if this block is movable */
2485 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2486 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2487 move_freepages_block(zone, page, MIGRATE_RESERVE);
2488 reserve--;
2489 continue;
2490 }
2491
2492 /*
2493 * If the reserve is met and this is a previous reserved block,
2494 * take it back
2495 */
2496 if (block_migratetype == MIGRATE_RESERVE) {
2497 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2498 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2499 }
2500 }
2501}
Mel Gormanac0e5b72007-10-16 01:25:58 -07002502
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503/*
2504 * Initially all pages are reserved - free ones are freed
2505 * up by free_all_bootmem() once the early boot process is
2506 * done. Non-atomic initialization, single-pass.
2507 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002508void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08002509 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07002512 unsigned long end_pfn = start_pfn + size;
2513 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514
Greg Ungerercbe8dd42006-01-12 01:05:24 -08002515 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08002516 /*
2517 * There can be holes in boot-time mem_map[]s
2518 * handed to this function. They do not
2519 * exist on hotplugged memory.
2520 */
2521 if (context == MEMMAP_EARLY) {
2522 if (!early_pfn_valid(pfn))
2523 continue;
2524 if (!early_pfn_in_nid(pfn, nid))
2525 continue;
2526 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002527 page = pfn_to_page(pfn);
2528 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08002529 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 reset_page_mapcount(page);
2531 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002532
2533 /*
2534 * Mark the block movable so that blocks are reserved for
2535 * movable at startup. This will force kernel allocations
2536 * to reserve their blocks rather than leaking throughout
2537 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07002538 * kernel allocations are made. Later some blocks near
2539 * the start are marked MIGRATE_RESERVE by
2540 * setup_zone_migrate_reserve()
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002541 */
Mel Gormand9c23402007-10-16 01:26:01 -07002542 if ((pfn & (pageblock_nr_pages-1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07002543 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002544
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 INIT_LIST_HEAD(&page->lru);
2546#ifdef WANT_PAGE_VIRTUAL
2547 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2548 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07002549 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 }
2552}
2553
Paul Mundt6ea6e682007-07-15 23:38:20 -07002554static void __meminit zone_init_free_lists(struct pglist_data *pgdat,
2555 struct zone *zone, unsigned long size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002557 int order, t;
2558 for_each_migratetype_order(order, t) {
2559 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 zone->free_area[order].nr_free = 0;
2561 }
2562}
2563
2564#ifndef __HAVE_ARCH_MEMMAP_INIT
2565#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08002566 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567#endif
2568
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08002569static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002570{
2571 int batch;
2572
2573 /*
2574 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07002575 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002576 *
2577 * OK, so we don't know how big the cache is. So guess.
2578 */
2579 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07002580 if (batch * PAGE_SIZE > 512 * 1024)
2581 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002582 batch /= 4; /* We effectively *= 4 below */
2583 if (batch < 1)
2584 batch = 1;
2585
2586 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002587 * Clamp the batch to a 2^n - 1 value. Having a power
2588 * of 2 value was found to be more likely to have
2589 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002590 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002591 * For example if 2 tasks are alternately allocating
2592 * batches of pages, one task can end up with a lot
2593 * of pages of one half of the possible page colors
2594 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002595 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002596 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07002597
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002598 return batch;
2599}
2600
Christoph Lameter2caaad42005-06-21 17:15:00 -07002601inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2602{
2603 struct per_cpu_pages *pcp;
2604
Magnus Damm1c6fe942005-10-26 01:58:59 -07002605 memset(p, 0, sizeof(*p));
2606
Christoph Lameter2caaad42005-06-21 17:15:00 -07002607 pcp = &p->pcp[0]; /* hot */
2608 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002609 pcp->high = 6 * batch;
2610 pcp->batch = max(1UL, 1 * batch);
2611 INIT_LIST_HEAD(&pcp->list);
2612
2613 pcp = &p->pcp[1]; /* cold*/
2614 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002615 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07002616 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07002617 INIT_LIST_HEAD(&pcp->list);
2618}
2619
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002620/*
2621 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2622 * to the value high for the pageset p.
2623 */
2624
2625static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2626 unsigned long high)
2627{
2628 struct per_cpu_pages *pcp;
2629
2630 pcp = &p->pcp[0]; /* hot list */
2631 pcp->high = high;
2632 pcp->batch = max(1UL, high/4);
2633 if ((high/4) > (PAGE_SHIFT * 8))
2634 pcp->batch = PAGE_SHIFT * 8;
2635}
2636
2637
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002638#ifdef CONFIG_NUMA
2639/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07002640 * Boot pageset table. One per cpu which is going to be used for all
2641 * zones and all nodes. The parameters will be set in such a way
2642 * that an item put on a list will immediately be handed over to
2643 * the buddy list. This is safe since pageset manipulation is done
2644 * with interrupts disabled.
2645 *
2646 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07002647 *
2648 * The boot_pagesets must be kept even after bootup is complete for
2649 * unused processors and/or zones. They do play a role for bootstrapping
2650 * hotplugged processors.
2651 *
2652 * zoneinfo_show() and maybe other functions do
2653 * not check if the processor is online before following the pageset pointer.
2654 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07002655 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08002656static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07002657
2658/*
2659 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002660 * per cpu pageset array in struct zone.
2661 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08002662static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002663{
2664 struct zone *zone, *dzone;
Christoph Lameter37c07082007-10-16 01:25:36 -07002665 int node = cpu_to_node(cpu);
2666
2667 node_set_state(node, N_CPU); /* this node has a cpu */
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002668
2669 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002670
Christoph Lameter66a55032006-09-27 01:50:09 -07002671 if (!populated_zone(zone))
2672 continue;
2673
Nick Piggin23316bc2006-01-08 01:00:41 -08002674 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lameter37c07082007-10-16 01:25:36 -07002675 GFP_KERNEL, node);
Nick Piggin23316bc2006-01-08 01:00:41 -08002676 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002677 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002678
Nick Piggin23316bc2006-01-08 01:00:41 -08002679 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002680
2681 if (percpu_pagelist_fraction)
2682 setup_pagelist_highmark(zone_pcp(zone, cpu),
2683 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002684 }
2685
2686 return 0;
2687bad:
2688 for_each_zone(dzone) {
Andrew Morton64191682007-08-30 23:56:17 -07002689 if (!populated_zone(dzone))
2690 continue;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002691 if (dzone == zone)
2692 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08002693 kfree(zone_pcp(dzone, cpu));
2694 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002695 }
2696 return -ENOMEM;
2697}
2698
2699static inline void free_zone_pagesets(int cpu)
2700{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002701 struct zone *zone;
2702
2703 for_each_zone(zone) {
2704 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2705
David Rientjesf3ef9ea2006-09-25 16:24:57 -07002706 /* Free per_cpu_pageset if it is slab allocated */
2707 if (pset != &boot_pageset[cpu])
2708 kfree(pset);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002709 zone_pcp(zone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002710 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002711}
2712
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07002713static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002714 unsigned long action,
2715 void *hcpu)
2716{
2717 int cpu = (long)hcpu;
2718 int ret = NOTIFY_OK;
2719
2720 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08002721 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002722 case CPU_UP_PREPARE_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002723 if (process_zones(cpu))
2724 ret = NOTIFY_BAD;
2725 break;
2726 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002727 case CPU_UP_CANCELED_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002728 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002729 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002730 free_zone_pagesets(cpu);
2731 break;
2732 default:
2733 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002734 }
2735 return ret;
2736}
2737
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07002738static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002739 { &pageset_cpuup_callback, NULL, 0 };
2740
Al Viro78d99552005-12-15 09:18:25 +00002741void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002742{
2743 int err;
2744
2745 /* Initialize per_cpu_pageset for cpu 0.
2746 * A cpuup callback will do this for every cpu
2747 * as it comes online
2748 */
2749 err = process_zones(smp_processor_id());
2750 BUG_ON(err);
2751 register_cpu_notifier(&pageset_notifier);
2752}
2753
2754#endif
2755
Sam Ravnborg577a32f2007-05-17 23:29:25 +02002756static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07002757int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07002758{
2759 int i;
2760 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002761 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07002762
2763 /*
2764 * The per-page waitqueue mechanism uses hashed waitqueues
2765 * per zone.
2766 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07002767 zone->wait_table_hash_nr_entries =
2768 wait_table_hash_nr_entries(zone_size_pages);
2769 zone->wait_table_bits =
2770 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002771 alloc_size = zone->wait_table_hash_nr_entries
2772 * sizeof(wait_queue_head_t);
2773
2774 if (system_state == SYSTEM_BOOTING) {
2775 zone->wait_table = (wait_queue_head_t *)
2776 alloc_bootmem_node(pgdat, alloc_size);
2777 } else {
2778 /*
2779 * This case means that a zone whose size was 0 gets new memory
2780 * via memory hot-add.
2781 * But it may be the case that a new node was hot-added. In
2782 * this case vmalloc() will not be able to use this new node's
2783 * memory - this wait_table must be initialized to use this new
2784 * node itself as well.
2785 * To use this new node's memory, further consideration will be
2786 * necessary.
2787 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07002788 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002789 }
2790 if (!zone->wait_table)
2791 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07002792
Yasunori Goto02b694d2006-06-23 02:03:08 -07002793 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07002794 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002795
2796 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002797}
2798
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002799static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07002800{
2801 int cpu;
2802 unsigned long batch = zone_batchsize(zone);
2803
2804 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2805#ifdef CONFIG_NUMA
2806 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002807 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002808 setup_pageset(&boot_pageset[cpu],0);
2809#else
2810 setup_pageset(zone_pcp(zone,cpu), batch);
2811#endif
2812 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08002813 if (zone->present_pages)
2814 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2815 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07002816}
2817
Yasunori Goto718127c2006-06-23 02:03:10 -07002818__meminit int init_currently_empty_zone(struct zone *zone,
2819 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08002820 unsigned long size,
2821 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07002822{
2823 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002824 int ret;
2825 ret = zone_wait_table_init(zone, size);
2826 if (ret)
2827 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07002828 pgdat->nr_zones = zone_idx(zone) + 1;
2829
Dave Hansened8ece22005-10-29 18:16:50 -07002830 zone->zone_start_pfn = zone_start_pfn;
2831
2832 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2833
2834 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
Yasunori Goto718127c2006-06-23 02:03:10 -07002835
2836 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002837}
2838
Mel Gormanc7132162006-09-27 01:49:43 -07002839#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2840/*
2841 * Basic iterator support. Return the first range of PFNs for a node
2842 * Note: nid == MAX_NUMNODES returns first region regardless of node
2843 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002844static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07002845{
2846 int i;
2847
2848 for (i = 0; i < nr_nodemap_entries; i++)
2849 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2850 return i;
2851
2852 return -1;
2853}
2854
2855/*
2856 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02002857 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07002858 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002859static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07002860{
2861 for (index = index + 1; index < nr_nodemap_entries; index++)
2862 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2863 return index;
2864
2865 return -1;
2866}
2867
2868#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2869/*
2870 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2871 * Architectures may implement their own version but if add_active_range()
2872 * was used and there are no special requirements, this is a convenient
2873 * alternative
2874 */
Stephen Rothwell6f076f52007-05-10 03:15:27 -07002875int __meminit early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07002876{
2877 int i;
2878
2879 for (i = 0; i < nr_nodemap_entries; i++) {
2880 unsigned long start_pfn = early_node_map[i].start_pfn;
2881 unsigned long end_pfn = early_node_map[i].end_pfn;
2882
2883 if (start_pfn <= pfn && pfn < end_pfn)
2884 return early_node_map[i].nid;
2885 }
2886
2887 return 0;
2888}
2889#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2890
2891/* Basic iterator support to walk early_node_map[] */
2892#define for_each_active_range_index_in_nid(i, nid) \
2893 for (i = first_active_region_index_in_nid(nid); i != -1; \
2894 i = next_active_region_index_in_nid(i, nid))
2895
2896/**
2897 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002898 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2899 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07002900 *
2901 * If an architecture guarantees that all ranges registered with
2902 * add_active_ranges() contain no holes and may be freed, this
2903 * this function may be used instead of calling free_bootmem() manually.
2904 */
2905void __init free_bootmem_with_active_regions(int nid,
2906 unsigned long max_low_pfn)
2907{
2908 int i;
2909
2910 for_each_active_range_index_in_nid(i, nid) {
2911 unsigned long size_pages = 0;
2912 unsigned long end_pfn = early_node_map[i].end_pfn;
2913
2914 if (early_node_map[i].start_pfn >= max_low_pfn)
2915 continue;
2916
2917 if (end_pfn > max_low_pfn)
2918 end_pfn = max_low_pfn;
2919
2920 size_pages = end_pfn - early_node_map[i].start_pfn;
2921 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2922 PFN_PHYS(early_node_map[i].start_pfn),
2923 size_pages << PAGE_SHIFT);
2924 }
2925}
2926
2927/**
2928 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002929 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07002930 *
2931 * If an architecture guarantees that all ranges registered with
2932 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002933 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07002934 */
2935void __init sparse_memory_present_with_active_regions(int nid)
2936{
2937 int i;
2938
2939 for_each_active_range_index_in_nid(i, nid)
2940 memory_present(early_node_map[i].nid,
2941 early_node_map[i].start_pfn,
2942 early_node_map[i].end_pfn);
2943}
2944
2945/**
Mel Gormanfb014392006-09-27 01:49:59 -07002946 * push_node_boundaries - Push node boundaries to at least the requested boundary
2947 * @nid: The nid of the node to push the boundary for
2948 * @start_pfn: The start pfn of the node
2949 * @end_pfn: The end pfn of the node
2950 *
2951 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2952 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2953 * be hotplugged even though no physical memory exists. This function allows
2954 * an arch to push out the node boundaries so mem_map is allocated that can
2955 * be used later.
2956 */
2957#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2958void __init push_node_boundaries(unsigned int nid,
2959 unsigned long start_pfn, unsigned long end_pfn)
2960{
2961 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2962 nid, start_pfn, end_pfn);
2963
2964 /* Initialise the boundary for this node if necessary */
2965 if (node_boundary_end_pfn[nid] == 0)
2966 node_boundary_start_pfn[nid] = -1UL;
2967
2968 /* Update the boundaries */
2969 if (node_boundary_start_pfn[nid] > start_pfn)
2970 node_boundary_start_pfn[nid] = start_pfn;
2971 if (node_boundary_end_pfn[nid] < end_pfn)
2972 node_boundary_end_pfn[nid] = end_pfn;
2973}
2974
2975/* If necessary, push the node boundary out for reserve hotadd */
Jan Beulich98011f52007-07-15 23:38:17 -07002976static void __meminit account_node_boundary(unsigned int nid,
Mel Gormanfb014392006-09-27 01:49:59 -07002977 unsigned long *start_pfn, unsigned long *end_pfn)
2978{
2979 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2980 nid, *start_pfn, *end_pfn);
2981
2982 /* Return if boundary information has not been provided */
2983 if (node_boundary_end_pfn[nid] == 0)
2984 return;
2985
2986 /* Check the boundaries and update if necessary */
2987 if (node_boundary_start_pfn[nid] < *start_pfn)
2988 *start_pfn = node_boundary_start_pfn[nid];
2989 if (node_boundary_end_pfn[nid] > *end_pfn)
2990 *end_pfn = node_boundary_end_pfn[nid];
2991}
2992#else
2993void __init push_node_boundaries(unsigned int nid,
2994 unsigned long start_pfn, unsigned long end_pfn) {}
2995
Jan Beulich98011f52007-07-15 23:38:17 -07002996static void __meminit account_node_boundary(unsigned int nid,
Mel Gormanfb014392006-09-27 01:49:59 -07002997 unsigned long *start_pfn, unsigned long *end_pfn) {}
2998#endif
2999
3000
3001/**
Mel Gormanc7132162006-09-27 01:49:43 -07003002 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003003 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3004 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3005 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003006 *
3007 * It returns the start and end page frame of a node based on information
3008 * provided by an arch calling add_active_range(). If called for a node
3009 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003010 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003011 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003012void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003013 unsigned long *start_pfn, unsigned long *end_pfn)
3014{
3015 int i;
3016 *start_pfn = -1UL;
3017 *end_pfn = 0;
3018
3019 for_each_active_range_index_in_nid(i, nid) {
3020 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3021 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3022 }
3023
Christoph Lameter633c0662007-10-16 01:25:37 -07003024 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003025 *start_pfn = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07003026
3027 /* Push the node boundaries out if requested */
3028 account_node_boundary(nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003029}
3030
3031/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003032 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3033 * assumption is made that zones within a node are ordered in monotonic
3034 * increasing memory addresses so that the "highest" populated zone is used
3035 */
3036void __init find_usable_zone_for_movable(void)
3037{
3038 int zone_index;
3039 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3040 if (zone_index == ZONE_MOVABLE)
3041 continue;
3042
3043 if (arch_zone_highest_possible_pfn[zone_index] >
3044 arch_zone_lowest_possible_pfn[zone_index])
3045 break;
3046 }
3047
3048 VM_BUG_ON(zone_index == -1);
3049 movable_zone = zone_index;
3050}
3051
3052/*
3053 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3054 * because it is sized independant of architecture. Unlike the other zones,
3055 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3056 * in each node depending on the size of each node and how evenly kernelcore
3057 * is distributed. This helper function adjusts the zone ranges
3058 * provided by the architecture for a given node by using the end of the
3059 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3060 * zones within a node are in order of monotonic increases memory addresses
3061 */
3062void __meminit adjust_zone_range_for_zone_movable(int nid,
3063 unsigned long zone_type,
3064 unsigned long node_start_pfn,
3065 unsigned long node_end_pfn,
3066 unsigned long *zone_start_pfn,
3067 unsigned long *zone_end_pfn)
3068{
3069 /* Only adjust if ZONE_MOVABLE is on this node */
3070 if (zone_movable_pfn[nid]) {
3071 /* Size ZONE_MOVABLE */
3072 if (zone_type == ZONE_MOVABLE) {
3073 *zone_start_pfn = zone_movable_pfn[nid];
3074 *zone_end_pfn = min(node_end_pfn,
3075 arch_zone_highest_possible_pfn[movable_zone]);
3076
3077 /* Adjust for ZONE_MOVABLE starting within this range */
3078 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3079 *zone_end_pfn > zone_movable_pfn[nid]) {
3080 *zone_end_pfn = zone_movable_pfn[nid];
3081
3082 /* Check if this whole range is within ZONE_MOVABLE */
3083 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3084 *zone_start_pfn = *zone_end_pfn;
3085 }
3086}
3087
3088/*
Mel Gormanc7132162006-09-27 01:49:43 -07003089 * Return the number of pages a zone spans in a node, including holes
3090 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3091 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003092static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003093 unsigned long zone_type,
3094 unsigned long *ignored)
3095{
3096 unsigned long node_start_pfn, node_end_pfn;
3097 unsigned long zone_start_pfn, zone_end_pfn;
3098
3099 /* Get the start and end of the node and zone */
3100 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3101 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3102 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003103 adjust_zone_range_for_zone_movable(nid, zone_type,
3104 node_start_pfn, node_end_pfn,
3105 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003106
3107 /* Check that this node has pages within the zone's required range */
3108 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3109 return 0;
3110
3111 /* Move the zone boundaries inside the node if necessary */
3112 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3113 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3114
3115 /* Return the spanned pages */
3116 return zone_end_pfn - zone_start_pfn;
3117}
3118
3119/*
3120 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003121 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003122 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003123unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003124 unsigned long range_start_pfn,
3125 unsigned long range_end_pfn)
3126{
3127 int i = 0;
3128 unsigned long prev_end_pfn = 0, hole_pages = 0;
3129 unsigned long start_pfn;
3130
3131 /* Find the end_pfn of the first active range of pfns in the node */
3132 i = first_active_region_index_in_nid(nid);
3133 if (i == -1)
3134 return 0;
3135
Mel Gormanb5445f92007-07-26 10:41:18 -07003136 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3137
Mel Gorman9c7cd682006-09-27 01:49:58 -07003138 /* Account for ranges before physical memory on this node */
3139 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003140 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003141
3142 /* Find all holes for the zone within the node */
3143 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3144
3145 /* No need to continue if prev_end_pfn is outside the zone */
3146 if (prev_end_pfn >= range_end_pfn)
3147 break;
3148
3149 /* Make sure the end of the zone is not within the hole */
3150 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3151 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3152
3153 /* Update the hole size cound and move on */
3154 if (start_pfn > range_start_pfn) {
3155 BUG_ON(prev_end_pfn > start_pfn);
3156 hole_pages += start_pfn - prev_end_pfn;
3157 }
3158 prev_end_pfn = early_node_map[i].end_pfn;
3159 }
3160
Mel Gorman9c7cd682006-09-27 01:49:58 -07003161 /* Account for ranges past physical memory on this node */
3162 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003163 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003164 max(range_start_pfn, prev_end_pfn);
3165
Mel Gormanc7132162006-09-27 01:49:43 -07003166 return hole_pages;
3167}
3168
3169/**
3170 * absent_pages_in_range - Return number of page frames in holes within a range
3171 * @start_pfn: The start PFN to start searching for holes
3172 * @end_pfn: The end PFN to stop searching for holes
3173 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003174 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003175 */
3176unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3177 unsigned long end_pfn)
3178{
3179 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3180}
3181
3182/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003183static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003184 unsigned long zone_type,
3185 unsigned long *ignored)
3186{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003187 unsigned long node_start_pfn, node_end_pfn;
3188 unsigned long zone_start_pfn, zone_end_pfn;
3189
3190 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3191 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3192 node_start_pfn);
3193 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3194 node_end_pfn);
3195
Mel Gorman2a1e2742007-07-17 04:03:12 -07003196 adjust_zone_range_for_zone_movable(nid, zone_type,
3197 node_start_pfn, node_end_pfn,
3198 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07003199 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003200}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003201
Mel Gormanc7132162006-09-27 01:49:43 -07003202#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07003203static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003204 unsigned long zone_type,
3205 unsigned long *zones_size)
3206{
3207 return zones_size[zone_type];
3208}
3209
Paul Mundt6ea6e682007-07-15 23:38:20 -07003210static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003211 unsigned long zone_type,
3212 unsigned long *zholes_size)
3213{
3214 if (!zholes_size)
3215 return 0;
3216
3217 return zholes_size[zone_type];
3218}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003219
Mel Gormanc7132162006-09-27 01:49:43 -07003220#endif
3221
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003222static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07003223 unsigned long *zones_size, unsigned long *zholes_size)
3224{
3225 unsigned long realtotalpages, totalpages = 0;
3226 enum zone_type i;
3227
3228 for (i = 0; i < MAX_NR_ZONES; i++)
3229 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3230 zones_size);
3231 pgdat->node_spanned_pages = totalpages;
3232
3233 realtotalpages = totalpages;
3234 for (i = 0; i < MAX_NR_ZONES; i++)
3235 realtotalpages -=
3236 zone_absent_pages_in_node(pgdat->node_id, i,
3237 zholes_size);
3238 pgdat->node_present_pages = realtotalpages;
3239 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3240 realtotalpages);
3241}
3242
Mel Gorman835c1342007-10-16 01:25:47 -07003243#ifndef CONFIG_SPARSEMEM
3244/*
3245 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07003246 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3247 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07003248 * round what is now in bits to nearest long in bits, then return it in
3249 * bytes.
3250 */
3251static unsigned long __init usemap_size(unsigned long zonesize)
3252{
3253 unsigned long usemapsize;
3254
Mel Gormand9c23402007-10-16 01:26:01 -07003255 usemapsize = roundup(zonesize, pageblock_nr_pages);
3256 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07003257 usemapsize *= NR_PAGEBLOCK_BITS;
3258 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3259
3260 return usemapsize / 8;
3261}
3262
3263static void __init setup_usemap(struct pglist_data *pgdat,
3264 struct zone *zone, unsigned long zonesize)
3265{
3266 unsigned long usemapsize = usemap_size(zonesize);
3267 zone->pageblock_flags = NULL;
3268 if (usemapsize) {
3269 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3270 memset(zone->pageblock_flags, 0, usemapsize);
3271 }
3272}
3273#else
3274static void inline setup_usemap(struct pglist_data *pgdat,
3275 struct zone *zone, unsigned long zonesize) {}
3276#endif /* CONFIG_SPARSEMEM */
3277
Mel Gormand9c23402007-10-16 01:26:01 -07003278#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08003279
3280/* Return a sensible default order for the pageblock size. */
3281static inline int pageblock_default_order(void)
3282{
3283 if (HPAGE_SHIFT > PAGE_SHIFT)
3284 return HUGETLB_PAGE_ORDER;
3285
3286 return MAX_ORDER-1;
3287}
3288
Mel Gormand9c23402007-10-16 01:26:01 -07003289/* 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
Mel Gormanba72cb82007-11-28 16:21:13 -08003304/*
3305 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3306 * and pageblock_default_order() are unused as pageblock_order is set
3307 * at compile-time. See include/linux/pageblock-flags.h for the values of
3308 * pageblock_order based on the kernel config
3309 */
3310static inline int pageblock_default_order(unsigned int order)
3311{
3312 return MAX_ORDER-1;
3313}
Mel Gormand9c23402007-10-16 01:26:01 -07003314#define set_pageblock_order(x) do {} while (0)
3315
3316#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3317
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318/*
3319 * Set up the zone data structures:
3320 * - mark all pages reserved
3321 * - mark all memory queues empty
3322 * - clear the memory bitmaps
3323 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07003324static void __meminit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325 unsigned long *zones_size, unsigned long *zholes_size)
3326{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07003327 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07003328 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07003330 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003331
Dave Hansen208d54e2005-10-29 18:16:52 -07003332 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333 pgdat->nr_zones = 0;
3334 init_waitqueue_head(&pgdat->kswapd_wait);
3335 pgdat->kswapd_max_order = 0;
3336
3337 for (j = 0; j < MAX_NR_ZONES; j++) {
3338 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003339 unsigned long size, realsize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003340
Mel Gormanc7132162006-09-27 01:49:43 -07003341 size = zone_spanned_pages_in_node(nid, j, zones_size);
3342 realsize = size - zone_absent_pages_in_node(nid, j,
3343 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003344
Mel Gorman0e0b8642006-09-27 01:49:56 -07003345 /*
3346 * Adjust realsize so that it accounts for how much memory
3347 * is used by this zone for memmap. This affects the watermark
3348 * and per-cpu initialisations
3349 */
3350 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
3351 if (realsize >= memmap_pages) {
3352 realsize -= memmap_pages;
3353 printk(KERN_DEBUG
3354 " %s zone: %lu pages used for memmap\n",
3355 zone_names[j], memmap_pages);
3356 } else
3357 printk(KERN_WARNING
3358 " %s zone: %lu pages exceeds realsize %lu\n",
3359 zone_names[j], memmap_pages, realsize);
3360
Christoph Lameter62672762007-02-10 01:43:07 -08003361 /* Account for reserved pages */
3362 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07003363 realsize -= dma_reserve;
Christoph Lameter62672762007-02-10 01:43:07 -08003364 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3365 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003366 }
3367
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003368 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369 nr_kernel_pages += realsize;
3370 nr_all_pages += realsize;
3371
3372 zone->spanned_pages = size;
3373 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07003374#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07003375 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07003376 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07003377 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07003378 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07003379#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380 zone->name = zone_names[j];
3381 spin_lock_init(&zone->lock);
3382 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07003383 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003384 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385
Martin Bligh3bb1a8522006-10-28 10:38:24 -07003386 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387
Dave Hansened8ece22005-10-29 18:16:50 -07003388 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003389 INIT_LIST_HEAD(&zone->active_list);
3390 INIT_LIST_HEAD(&zone->inactive_list);
3391 zone->nr_scan_active = 0;
3392 zone->nr_scan_inactive = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07003393 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07003394 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395 if (!size)
3396 continue;
3397
Mel Gormanba72cb82007-11-28 16:21:13 -08003398 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07003399 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08003400 ret = init_currently_empty_zone(zone, zone_start_pfn,
3401 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07003402 BUG_ON(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003403 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003404 }
3405}
3406
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003407static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409 /* Skip empty nodes */
3410 if (!pgdat->node_spanned_pages)
3411 return;
3412
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003413#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414 /* ia64 gets its own node_mem_map, before this, without bootmem */
3415 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07003416 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003417 struct page *map;
3418
Bob Piccoe984bb42006-05-20 15:00:31 -07003419 /*
3420 * The zone's endpoints aren't required to be MAX_ORDER
3421 * aligned but the node_mem_map endpoints must be in order
3422 * for the buddy allocator to function correctly.
3423 */
3424 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3425 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3426 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3427 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07003428 map = alloc_remap(pgdat->node_id, size);
3429 if (!map)
3430 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07003431 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432 }
Roman Zippel12d810c2007-05-31 00:40:54 -07003433#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434 /*
3435 * With no DISCONTIG, the global mem_map is just set as node 0's
3436 */
Mel Gormanc7132162006-09-27 01:49:43 -07003437 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07003439#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3440 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08003441 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07003442#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3443 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003445#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446}
3447
Yasunori Goto86356ab2006-06-23 02:03:09 -07003448void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449 unsigned long *zones_size, unsigned long node_start_pfn,
3450 unsigned long *zholes_size)
3451{
3452 pgdat->node_id = nid;
3453 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003454 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003455
3456 alloc_node_mem_map(pgdat);
3457
3458 free_area_init_core(pgdat, zones_size, zholes_size);
3459}
3460
Mel Gormanc7132162006-09-27 01:49:43 -07003461#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07003462
3463#if MAX_NUMNODES > 1
3464/*
3465 * Figure out the number of possible node ids.
3466 */
3467static void __init setup_nr_node_ids(void)
3468{
3469 unsigned int node;
3470 unsigned int highest = 0;
3471
3472 for_each_node_mask(node, node_possible_map)
3473 highest = node;
3474 nr_node_ids = highest + 1;
3475}
3476#else
3477static inline void setup_nr_node_ids(void)
3478{
3479}
3480#endif
3481
Mel Gormanc7132162006-09-27 01:49:43 -07003482/**
3483 * add_active_range - Register a range of PFNs backed by physical memory
3484 * @nid: The node ID the range resides on
3485 * @start_pfn: The start PFN of the available physical memory
3486 * @end_pfn: The end PFN of the available physical memory
3487 *
3488 * These ranges are stored in an early_node_map[] and later used by
3489 * free_area_init_nodes() to calculate zone sizes and holes. If the
3490 * range spans a memory hole, it is up to the architecture to ensure
3491 * the memory is not freed by the bootmem allocator. If possible
3492 * the range being registered will be merged with existing ranges.
3493 */
3494void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3495 unsigned long end_pfn)
3496{
3497 int i;
3498
3499 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
3500 "%d entries of %d used\n",
3501 nid, start_pfn, end_pfn,
3502 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3503
3504 /* Merge with existing active regions if possible */
3505 for (i = 0; i < nr_nodemap_entries; i++) {
3506 if (early_node_map[i].nid != nid)
3507 continue;
3508
3509 /* Skip if an existing region covers this new one */
3510 if (start_pfn >= early_node_map[i].start_pfn &&
3511 end_pfn <= early_node_map[i].end_pfn)
3512 return;
3513
3514 /* Merge forward if suitable */
3515 if (start_pfn <= early_node_map[i].end_pfn &&
3516 end_pfn > early_node_map[i].end_pfn) {
3517 early_node_map[i].end_pfn = end_pfn;
3518 return;
3519 }
3520
3521 /* Merge backward if suitable */
3522 if (start_pfn < early_node_map[i].end_pfn &&
3523 end_pfn >= early_node_map[i].start_pfn) {
3524 early_node_map[i].start_pfn = start_pfn;
3525 return;
3526 }
3527 }
3528
3529 /* Check that early_node_map is large enough */
3530 if (i >= MAX_ACTIVE_REGIONS) {
3531 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3532 MAX_ACTIVE_REGIONS);
3533 return;
3534 }
3535
3536 early_node_map[i].nid = nid;
3537 early_node_map[i].start_pfn = start_pfn;
3538 early_node_map[i].end_pfn = end_pfn;
3539 nr_nodemap_entries = i + 1;
3540}
3541
3542/**
3543 * shrink_active_range - Shrink an existing registered range of PFNs
3544 * @nid: The node id the range is on that should be shrunk
3545 * @old_end_pfn: The old end PFN of the range
3546 * @new_end_pfn: The new PFN of the range
3547 *
3548 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3549 * The map is kept at the end physical page range that has already been
3550 * registered with add_active_range(). This function allows an arch to shrink
3551 * an existing registered range.
3552 */
3553void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
3554 unsigned long new_end_pfn)
3555{
3556 int i;
3557
3558 /* Find the old active region end and shrink */
3559 for_each_active_range_index_in_nid(i, nid)
3560 if (early_node_map[i].end_pfn == old_end_pfn) {
3561 early_node_map[i].end_pfn = new_end_pfn;
3562 break;
3563 }
3564}
3565
3566/**
3567 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003568 *
Mel Gormanc7132162006-09-27 01:49:43 -07003569 * During discovery, it may be found that a table like SRAT is invalid
3570 * and an alternative discovery method must be used. This function removes
3571 * all currently registered regions.
3572 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003573void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07003574{
3575 memset(early_node_map, 0, sizeof(early_node_map));
3576 nr_nodemap_entries = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07003577#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3578 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3579 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3580#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -07003581}
3582
3583/* Compare two active node_active_regions */
3584static int __init cmp_node_active_region(const void *a, const void *b)
3585{
3586 struct node_active_region *arange = (struct node_active_region *)a;
3587 struct node_active_region *brange = (struct node_active_region *)b;
3588
3589 /* Done this way to avoid overflows */
3590 if (arange->start_pfn > brange->start_pfn)
3591 return 1;
3592 if (arange->start_pfn < brange->start_pfn)
3593 return -1;
3594
3595 return 0;
3596}
3597
3598/* sort the node_map by start_pfn */
3599static void __init sort_node_map(void)
3600{
3601 sort(early_node_map, (size_t)nr_nodemap_entries,
3602 sizeof(struct node_active_region),
3603 cmp_node_active_region, NULL);
3604}
3605
Mel Gormana6af2bc2007-02-10 01:42:57 -08003606/* Find the lowest pfn for a node */
Mel Gormanc7132162006-09-27 01:49:43 -07003607unsigned long __init find_min_pfn_for_node(unsigned long nid)
3608{
3609 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08003610 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00003611
Mel Gormanc7132162006-09-27 01:49:43 -07003612 /* Assuming a sorted map, the first range found has the starting pfn */
3613 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08003614 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003615
Mel Gormana6af2bc2007-02-10 01:42:57 -08003616 if (min_pfn == ULONG_MAX) {
3617 printk(KERN_WARNING
3618 "Could not find start_pfn for node %lu\n", nid);
3619 return 0;
3620 }
3621
3622 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003623}
3624
3625/**
3626 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3627 *
3628 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003629 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07003630 */
3631unsigned long __init find_min_pfn_with_active_regions(void)
3632{
3633 return find_min_pfn_for_node(MAX_NUMNODES);
3634}
3635
3636/**
3637 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3638 *
3639 * It returns the maximum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003640 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07003641 */
3642unsigned long __init find_max_pfn_with_active_regions(void)
3643{
3644 int i;
3645 unsigned long max_pfn = 0;
3646
3647 for (i = 0; i < nr_nodemap_entries; i++)
3648 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
3649
3650 return max_pfn;
3651}
3652
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003653/*
3654 * early_calculate_totalpages()
3655 * Sum pages in active regions for movable zone.
3656 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3657 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07003658static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07003659{
3660 int i;
3661 unsigned long totalpages = 0;
3662
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003663 for (i = 0; i < nr_nodemap_entries; i++) {
3664 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07003665 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003666 totalpages += pages;
3667 if (pages)
3668 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3669 }
3670 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07003671}
3672
Mel Gorman2a1e2742007-07-17 04:03:12 -07003673/*
3674 * Find the PFN the Movable zone begins in each node. Kernel memory
3675 * is spread evenly between nodes as long as the nodes have enough
3676 * memory. When they don't, some nodes will have more kernelcore than
3677 * others
3678 */
3679void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3680{
3681 int i, nid;
3682 unsigned long usable_startpfn;
3683 unsigned long kernelcore_node, kernelcore_remaining;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003684 unsigned long totalpages = early_calculate_totalpages();
3685 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07003686
Mel Gorman7e63efe2007-07-17 04:03:15 -07003687 /*
3688 * If movablecore was specified, calculate what size of
3689 * kernelcore that corresponds so that memory usable for
3690 * any allocation type is evenly spread. If both kernelcore
3691 * and movablecore are specified, then the value of kernelcore
3692 * will be used for required_kernelcore if it's greater than
3693 * what movablecore would have allowed.
3694 */
3695 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07003696 unsigned long corepages;
3697
3698 /*
3699 * Round-up so that ZONE_MOVABLE is at least as large as what
3700 * was requested by the user
3701 */
3702 required_movablecore =
3703 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3704 corepages = totalpages - required_movablecore;
3705
3706 required_kernelcore = max(required_kernelcore, corepages);
3707 }
3708
Mel Gorman2a1e2742007-07-17 04:03:12 -07003709 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3710 if (!required_kernelcore)
3711 return;
3712
3713 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3714 find_usable_zone_for_movable();
3715 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3716
3717restart:
3718 /* Spread kernelcore memory as evenly as possible throughout nodes */
3719 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003720 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07003721 /*
3722 * Recalculate kernelcore_node if the division per node
3723 * now exceeds what is necessary to satisfy the requested
3724 * amount of memory for the kernel
3725 */
3726 if (required_kernelcore < kernelcore_node)
3727 kernelcore_node = required_kernelcore / usable_nodes;
3728
3729 /*
3730 * As the map is walked, we track how much memory is usable
3731 * by the kernel using kernelcore_remaining. When it is
3732 * 0, the rest of the node is usable by ZONE_MOVABLE
3733 */
3734 kernelcore_remaining = kernelcore_node;
3735
3736 /* Go through each range of PFNs within this node */
3737 for_each_active_range_index_in_nid(i, nid) {
3738 unsigned long start_pfn, end_pfn;
3739 unsigned long size_pages;
3740
3741 start_pfn = max(early_node_map[i].start_pfn,
3742 zone_movable_pfn[nid]);
3743 end_pfn = early_node_map[i].end_pfn;
3744 if (start_pfn >= end_pfn)
3745 continue;
3746
3747 /* Account for what is only usable for kernelcore */
3748 if (start_pfn < usable_startpfn) {
3749 unsigned long kernel_pages;
3750 kernel_pages = min(end_pfn, usable_startpfn)
3751 - start_pfn;
3752
3753 kernelcore_remaining -= min(kernel_pages,
3754 kernelcore_remaining);
3755 required_kernelcore -= min(kernel_pages,
3756 required_kernelcore);
3757
3758 /* Continue if range is now fully accounted */
3759 if (end_pfn <= usable_startpfn) {
3760
3761 /*
3762 * Push zone_movable_pfn to the end so
3763 * that if we have to rebalance
3764 * kernelcore across nodes, we will
3765 * not double account here
3766 */
3767 zone_movable_pfn[nid] = end_pfn;
3768 continue;
3769 }
3770 start_pfn = usable_startpfn;
3771 }
3772
3773 /*
3774 * The usable PFN range for ZONE_MOVABLE is from
3775 * start_pfn->end_pfn. Calculate size_pages as the
3776 * number of pages used as kernelcore
3777 */
3778 size_pages = end_pfn - start_pfn;
3779 if (size_pages > kernelcore_remaining)
3780 size_pages = kernelcore_remaining;
3781 zone_movable_pfn[nid] = start_pfn + size_pages;
3782
3783 /*
3784 * Some kernelcore has been met, update counts and
3785 * break if the kernelcore for this node has been
3786 * satisified
3787 */
3788 required_kernelcore -= min(required_kernelcore,
3789 size_pages);
3790 kernelcore_remaining -= size_pages;
3791 if (!kernelcore_remaining)
3792 break;
3793 }
3794 }
3795
3796 /*
3797 * If there is still required_kernelcore, we do another pass with one
3798 * less node in the count. This will push zone_movable_pfn[nid] further
3799 * along on the nodes that still have memory until kernelcore is
3800 * satisified
3801 */
3802 usable_nodes--;
3803 if (usable_nodes && required_kernelcore > usable_nodes)
3804 goto restart;
3805
3806 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3807 for (nid = 0; nid < MAX_NUMNODES; nid++)
3808 zone_movable_pfn[nid] =
3809 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3810}
3811
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003812/* Any regular memory on that node ? */
3813static void check_for_regular_memory(pg_data_t *pgdat)
3814{
3815#ifdef CONFIG_HIGHMEM
3816 enum zone_type zone_type;
3817
3818 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3819 struct zone *zone = &pgdat->node_zones[zone_type];
3820 if (zone->present_pages)
3821 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3822 }
3823#endif
3824}
3825
Mel Gormanc7132162006-09-27 01:49:43 -07003826/**
3827 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003828 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07003829 *
3830 * This will call free_area_init_node() for each active node in the system.
3831 * Using the page ranges provided by add_active_range(), the size of each
3832 * zone in each node and their holes is calculated. If the maximum PFN
3833 * between two adjacent zones match, it is assumed that the zone is empty.
3834 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3835 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3836 * starts where the previous one ended. For example, ZONE_DMA32 starts
3837 * at arch_max_dma_pfn.
3838 */
3839void __init free_area_init_nodes(unsigned long *max_zone_pfn)
3840{
3841 unsigned long nid;
3842 enum zone_type i;
3843
Mel Gormana6af2bc2007-02-10 01:42:57 -08003844 /* Sort early_node_map as initialisation assumes it is sorted */
3845 sort_node_map();
3846
Mel Gormanc7132162006-09-27 01:49:43 -07003847 /* Record where the zone boundaries are */
3848 memset(arch_zone_lowest_possible_pfn, 0,
3849 sizeof(arch_zone_lowest_possible_pfn));
3850 memset(arch_zone_highest_possible_pfn, 0,
3851 sizeof(arch_zone_highest_possible_pfn));
3852 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
3853 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
3854 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07003855 if (i == ZONE_MOVABLE)
3856 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07003857 arch_zone_lowest_possible_pfn[i] =
3858 arch_zone_highest_possible_pfn[i-1];
3859 arch_zone_highest_possible_pfn[i] =
3860 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
3861 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07003862 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
3863 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
3864
3865 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3866 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
3867 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003868
Mel Gormanc7132162006-09-27 01:49:43 -07003869 /* Print out the zone ranges */
3870 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07003871 for (i = 0; i < MAX_NR_ZONES; i++) {
3872 if (i == ZONE_MOVABLE)
3873 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07003874 printk(" %-8s %8lu -> %8lu\n",
3875 zone_names[i],
3876 arch_zone_lowest_possible_pfn[i],
3877 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07003878 }
3879
3880 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3881 printk("Movable zone start PFN for each node\n");
3882 for (i = 0; i < MAX_NUMNODES; i++) {
3883 if (zone_movable_pfn[i])
3884 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
3885 }
Mel Gormanc7132162006-09-27 01:49:43 -07003886
3887 /* Print out the early_node_map[] */
3888 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
3889 for (i = 0; i < nr_nodemap_entries; i++)
3890 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
3891 early_node_map[i].start_pfn,
3892 early_node_map[i].end_pfn);
3893
3894 /* Initialise every node */
Christoph Lameter8ef82862007-02-20 13:57:52 -08003895 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07003896 for_each_online_node(nid) {
3897 pg_data_t *pgdat = NODE_DATA(nid);
3898 free_area_init_node(nid, pgdat, NULL,
3899 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003900
3901 /* Any memory on that node */
3902 if (pgdat->node_present_pages)
3903 node_set_state(nid, N_HIGH_MEMORY);
3904 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07003905 }
3906}
Mel Gorman2a1e2742007-07-17 04:03:12 -07003907
Mel Gorman7e63efe2007-07-17 04:03:15 -07003908static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003909{
3910 unsigned long long coremem;
3911 if (!p)
3912 return -EINVAL;
3913
3914 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07003915 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07003916
Mel Gorman7e63efe2007-07-17 04:03:15 -07003917 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07003918 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
3919
3920 return 0;
3921}
Mel Gormaned7ed362007-07-17 04:03:14 -07003922
Mel Gorman7e63efe2007-07-17 04:03:15 -07003923/*
3924 * kernelcore=size sets the amount of memory for use for allocations that
3925 * cannot be reclaimed or migrated.
3926 */
3927static int __init cmdline_parse_kernelcore(char *p)
3928{
3929 return cmdline_parse_core(p, &required_kernelcore);
3930}
3931
3932/*
3933 * movablecore=size sets the amount of memory for use for allocations that
3934 * can be reclaimed or migrated.
3935 */
3936static int __init cmdline_parse_movablecore(char *p)
3937{
3938 return cmdline_parse_core(p, &required_movablecore);
3939}
3940
Mel Gormaned7ed362007-07-17 04:03:14 -07003941early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07003942early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07003943
Mel Gormanc7132162006-09-27 01:49:43 -07003944#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3945
Mel Gorman0e0b8642006-09-27 01:49:56 -07003946/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003947 * set_dma_reserve - set the specified number of pages reserved in the first zone
3948 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07003949 *
3950 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3951 * In the DMA zone, a significant percentage may be consumed by kernel image
3952 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003953 * function may optionally be used to account for unfreeable pages in the
3954 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3955 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07003956 */
3957void __init set_dma_reserve(unsigned long new_dma_reserve)
3958{
3959 dma_reserve = new_dma_reserve;
3960}
3961
Dave Hansen93b75042005-06-23 00:07:47 -07003962#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003963static bootmem_data_t contig_bootmem_data;
3964struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
3965
3966EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07003967#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968
3969void __init free_area_init(unsigned long *zones_size)
3970{
Dave Hansen93b75042005-06-23 00:07:47 -07003971 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
3973}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974
Linus Torvalds1da177e2005-04-16 15:20:36 -07003975static int page_alloc_cpu_notify(struct notifier_block *self,
3976 unsigned long action, void *hcpu)
3977{
3978 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003980 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003981 local_irq_disable();
3982 __drain_pages(cpu);
Christoph Lameterf8891e52006-06-30 01:55:45 -07003983 vm_events_fold_cpu(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984 local_irq_enable();
Christoph Lameter2244b952006-06-30 01:55:33 -07003985 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003986 }
3987 return NOTIFY_OK;
3988}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989
3990void __init page_alloc_init(void)
3991{
3992 hotcpu_notifier(page_alloc_cpu_notify, 0);
3993}
3994
3995/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003996 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3997 * or min_free_kbytes changes.
3998 */
3999static void calculate_totalreserve_pages(void)
4000{
4001 struct pglist_data *pgdat;
4002 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004003 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004004
4005 for_each_online_pgdat(pgdat) {
4006 for (i = 0; i < MAX_NR_ZONES; i++) {
4007 struct zone *zone = pgdat->node_zones + i;
4008 unsigned long max = 0;
4009
4010 /* Find valid and maximum lowmem_reserve in the zone */
4011 for (j = i; j < MAX_NR_ZONES; j++) {
4012 if (zone->lowmem_reserve[j] > max)
4013 max = zone->lowmem_reserve[j];
4014 }
4015
4016 /* we treat pages_high as reserved pages. */
4017 max += zone->pages_high;
4018
4019 if (max > zone->present_pages)
4020 max = zone->present_pages;
4021 reserve_pages += max;
4022 }
4023 }
4024 totalreserve_pages = reserve_pages;
4025}
4026
4027/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028 * setup_per_zone_lowmem_reserve - called whenever
4029 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4030 * has a correct pages reserved value, so an adequate number of
4031 * pages are left in the zone after a successful __alloc_pages().
4032 */
4033static void setup_per_zone_lowmem_reserve(void)
4034{
4035 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004036 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004037
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004038 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004039 for (j = 0; j < MAX_NR_ZONES; j++) {
4040 struct zone *zone = pgdat->node_zones + j;
4041 unsigned long present_pages = zone->present_pages;
4042
4043 zone->lowmem_reserve[j] = 0;
4044
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004045 idx = j;
4046 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004047 struct zone *lower_zone;
4048
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004049 idx--;
4050
Linus Torvalds1da177e2005-04-16 15:20:36 -07004051 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4052 sysctl_lowmem_reserve_ratio[idx] = 1;
4053
4054 lower_zone = pgdat->node_zones + idx;
4055 lower_zone->lowmem_reserve[j] = present_pages /
4056 sysctl_lowmem_reserve_ratio[idx];
4057 present_pages += lower_zone->present_pages;
4058 }
4059 }
4060 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004061
4062 /* update totalreserve_pages */
4063 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004064}
4065
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004066/**
4067 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4068 *
4069 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4070 * with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071 */
Dave Hansen3947be12005-10-29 18:16:54 -07004072void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004073{
4074 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4075 unsigned long lowmem_pages = 0;
4076 struct zone *zone;
4077 unsigned long flags;
4078
4079 /* Calculate total number of !ZONE_HIGHMEM pages */
4080 for_each_zone(zone) {
4081 if (!is_highmem(zone))
4082 lowmem_pages += zone->present_pages;
4083 }
4084
4085 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004086 u64 tmp;
4087
Linus Torvalds1da177e2005-04-16 15:20:36 -07004088 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004089 tmp = (u64)pages_min * zone->present_pages;
4090 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091 if (is_highmem(zone)) {
4092 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004093 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4094 * need highmem pages, so cap pages_min to a small
4095 * value here.
4096 *
4097 * The (pages_high-pages_low) and (pages_low-pages_min)
4098 * deltas controls asynch page reclaim, and so should
4099 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004100 */
4101 int min_pages;
4102
4103 min_pages = zone->present_pages / 1024;
4104 if (min_pages < SWAP_CLUSTER_MAX)
4105 min_pages = SWAP_CLUSTER_MAX;
4106 if (min_pages > 128)
4107 min_pages = 128;
4108 zone->pages_min = min_pages;
4109 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004110 /*
4111 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004112 * proportionate to the zone's size.
4113 */
Nick Piggin669ed172005-11-13 16:06:45 -08004114 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004115 }
4116
Andrew Mortonac924c62006-05-15 09:43:59 -07004117 zone->pages_low = zone->pages_min + (tmp >> 2);
4118 zone->pages_high = zone->pages_min + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004119 setup_zone_migrate_reserve(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004120 spin_unlock_irqrestore(&zone->lru_lock, flags);
4121 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004122
4123 /* update totalreserve_pages */
4124 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004125}
4126
4127/*
4128 * Initialise min_free_kbytes.
4129 *
4130 * For small machines we want it small (128k min). For large machines
4131 * we want it large (64MB max). But it is not linear, because network
4132 * bandwidth does not increase linearly with machine size. We use
4133 *
4134 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4135 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4136 *
4137 * which yields
4138 *
4139 * 16MB: 512k
4140 * 32MB: 724k
4141 * 64MB: 1024k
4142 * 128MB: 1448k
4143 * 256MB: 2048k
4144 * 512MB: 2896k
4145 * 1024MB: 4096k
4146 * 2048MB: 5792k
4147 * 4096MB: 8192k
4148 * 8192MB: 11584k
4149 * 16384MB: 16384k
4150 */
4151static int __init init_per_zone_pages_min(void)
4152{
4153 unsigned long lowmem_kbytes;
4154
4155 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4156
4157 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4158 if (min_free_kbytes < 128)
4159 min_free_kbytes = 128;
4160 if (min_free_kbytes > 65536)
4161 min_free_kbytes = 65536;
4162 setup_per_zone_pages_min();
4163 setup_per_zone_lowmem_reserve();
4164 return 0;
4165}
4166module_init(init_per_zone_pages_min)
4167
4168/*
4169 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4170 * that we can call two helper functions whenever min_free_kbytes
4171 * changes.
4172 */
4173int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4174 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4175{
4176 proc_dointvec(table, write, file, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07004177 if (write)
4178 setup_per_zone_pages_min();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004179 return 0;
4180}
4181
Christoph Lameter96146342006-07-03 00:24:13 -07004182#ifdef CONFIG_NUMA
4183int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4184 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4185{
4186 struct zone *zone;
4187 int rc;
4188
4189 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4190 if (rc)
4191 return rc;
4192
4193 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07004194 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07004195 sysctl_min_unmapped_ratio) / 100;
4196 return 0;
4197}
Christoph Lameter0ff38492006-09-25 23:31:52 -07004198
4199int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4200 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4201{
4202 struct zone *zone;
4203 int rc;
4204
4205 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4206 if (rc)
4207 return rc;
4208
4209 for_each_zone(zone)
4210 zone->min_slab_pages = (zone->present_pages *
4211 sysctl_min_slab_ratio) / 100;
4212 return 0;
4213}
Christoph Lameter96146342006-07-03 00:24:13 -07004214#endif
4215
Linus Torvalds1da177e2005-04-16 15:20:36 -07004216/*
4217 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4218 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4219 * whenever sysctl_lowmem_reserve_ratio changes.
4220 *
4221 * The reserve ratio obviously has absolutely no relation with the
4222 * pages_min watermarks. The lowmem reserve ratio can only make sense
4223 * if in function of the boot time zone sizes.
4224 */
4225int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4226 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4227{
4228 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4229 setup_per_zone_lowmem_reserve();
4230 return 0;
4231}
4232
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004233/*
4234 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4235 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4236 * can have before it gets flushed back to buddy allocator.
4237 */
4238
4239int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4240 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4241{
4242 struct zone *zone;
4243 unsigned int cpu;
4244 int ret;
4245
4246 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4247 if (!write || (ret == -EINVAL))
4248 return ret;
4249 for_each_zone(zone) {
4250 for_each_online_cpu(cpu) {
4251 unsigned long high;
4252 high = zone->present_pages / percpu_pagelist_fraction;
4253 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4254 }
4255 }
4256 return 0;
4257}
4258
David S. Millerf034b5d2006-08-24 03:08:07 -07004259int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004260
4261#ifdef CONFIG_NUMA
4262static int __init set_hashdist(char *str)
4263{
4264 if (!str)
4265 return 0;
4266 hashdist = simple_strtoul(str, &str, 0);
4267 return 1;
4268}
4269__setup("hashdist=", set_hashdist);
4270#endif
4271
4272/*
4273 * allocate a large system hash table from bootmem
4274 * - it is assumed that the hash table must contain an exact power-of-2
4275 * quantity of entries
4276 * - limit is the number of hash buckets, not the total allocation size
4277 */
4278void *__init alloc_large_system_hash(const char *tablename,
4279 unsigned long bucketsize,
4280 unsigned long numentries,
4281 int scale,
4282 int flags,
4283 unsigned int *_hash_shift,
4284 unsigned int *_hash_mask,
4285 unsigned long limit)
4286{
4287 unsigned long long max = limit;
4288 unsigned long log2qty, size;
4289 void *table = NULL;
4290
4291 /* allow the kernel cmdline to have a say */
4292 if (!numentries) {
4293 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08004294 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004295 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4296 numentries >>= 20 - PAGE_SHIFT;
4297 numentries <<= 20 - PAGE_SHIFT;
4298
4299 /* limit to 1 bucket per 2^scale bytes of low memory */
4300 if (scale > PAGE_SHIFT)
4301 numentries >>= (scale - PAGE_SHIFT);
4302 else
4303 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08004304
4305 /* Make sure we've got at least a 0-order allocation.. */
4306 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4307 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004308 }
John Hawkes6e692ed2006-03-25 03:08:02 -08004309 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004310
4311 /* limit allocation size to 1/16 total memory by default */
4312 if (max == 0) {
4313 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4314 do_div(max, bucketsize);
4315 }
4316
4317 if (numentries > max)
4318 numentries = max;
4319
David Howellsf0d1b0b2006-12-08 02:37:49 -08004320 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004321
4322 do {
4323 size = bucketsize << log2qty;
4324 if (flags & HASH_EARLY)
4325 table = alloc_bootmem(size);
4326 else if (hashdist)
4327 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4328 else {
4329 unsigned long order;
4330 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
4331 ;
4332 table = (void*) __get_free_pages(GFP_ATOMIC, order);
Eric Dumazet1037b832007-07-15 23:38:05 -07004333 /*
4334 * If bucketsize is not a power-of-two, we may free
4335 * some pages at the end of hash table.
4336 */
4337 if (table) {
4338 unsigned long alloc_end = (unsigned long)table +
4339 (PAGE_SIZE << order);
4340 unsigned long used = (unsigned long)table +
4341 PAGE_ALIGN(size);
4342 split_page(virt_to_page(table), order);
4343 while (used < alloc_end) {
4344 free_page(used);
4345 used += PAGE_SIZE;
4346 }
4347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004348 }
4349 } while (!table && size > PAGE_SIZE && --log2qty);
4350
4351 if (!table)
4352 panic("Failed to allocate %s hash table\n", tablename);
4353
Dan Alonib49ad482007-07-15 23:38:23 -07004354 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004355 tablename,
4356 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08004357 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004358 size);
4359
4360 if (_hash_shift)
4361 *_hash_shift = log2qty;
4362 if (_hash_mask)
4363 *_hash_mask = (1 << log2qty) - 1;
4364
4365 return table;
4366}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004367
4368#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004369struct page *pfn_to_page(unsigned long pfn)
4370{
Andy Whitcroft67de6482006-06-23 02:03:12 -07004371 return __pfn_to_page(pfn);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004372}
4373unsigned long page_to_pfn(struct page *page)
4374{
Andy Whitcroft67de6482006-06-23 02:03:12 -07004375 return __page_to_pfn(page);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004376}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004377EXPORT_SYMBOL(pfn_to_page);
4378EXPORT_SYMBOL(page_to_pfn);
4379#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
Andrew Morton6220ec72006-10-19 23:29:05 -07004380
Mel Gorman835c1342007-10-16 01:25:47 -07004381/* Return a pointer to the bitmap storing bits affecting a block of pages */
4382static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4383 unsigned long pfn)
4384{
4385#ifdef CONFIG_SPARSEMEM
4386 return __pfn_to_section(pfn)->pageblock_flags;
4387#else
4388 return zone->pageblock_flags;
4389#endif /* CONFIG_SPARSEMEM */
4390}
Andrew Morton6220ec72006-10-19 23:29:05 -07004391
Mel Gorman835c1342007-10-16 01:25:47 -07004392static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4393{
4394#ifdef CONFIG_SPARSEMEM
4395 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004396 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004397#else
4398 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07004399 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004400#endif /* CONFIG_SPARSEMEM */
4401}
4402
4403/**
Mel Gormand9c23402007-10-16 01:26:01 -07004404 * 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 -07004405 * @page: The page within the block of interest
4406 * @start_bitidx: The first bit of interest to retrieve
4407 * @end_bitidx: The last bit of interest
4408 * returns pageblock_bits flags
4409 */
4410unsigned long get_pageblock_flags_group(struct page *page,
4411 int start_bitidx, int end_bitidx)
4412{
4413 struct zone *zone;
4414 unsigned long *bitmap;
4415 unsigned long pfn, bitidx;
4416 unsigned long flags = 0;
4417 unsigned long value = 1;
4418
4419 zone = page_zone(page);
4420 pfn = page_to_pfn(page);
4421 bitmap = get_pageblock_bitmap(zone, pfn);
4422 bitidx = pfn_to_bitidx(zone, pfn);
4423
4424 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4425 if (test_bit(bitidx + start_bitidx, bitmap))
4426 flags |= value;
4427
4428 return flags;
4429}
4430
4431/**
Mel Gormand9c23402007-10-16 01:26:01 -07004432 * 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 -07004433 * @page: The page within the block of interest
4434 * @start_bitidx: The first bit of interest
4435 * @end_bitidx: The last bit of interest
4436 * @flags: The flags to set
4437 */
4438void set_pageblock_flags_group(struct page *page, unsigned long flags,
4439 int start_bitidx, int end_bitidx)
4440{
4441 struct zone *zone;
4442 unsigned long *bitmap;
4443 unsigned long pfn, bitidx;
4444 unsigned long value = 1;
4445
4446 zone = page_zone(page);
4447 pfn = page_to_pfn(page);
4448 bitmap = get_pageblock_bitmap(zone, pfn);
4449 bitidx = pfn_to_bitidx(zone, pfn);
4450
4451 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4452 if (flags & value)
4453 __set_bit(bitidx + start_bitidx, bitmap);
4454 else
4455 __clear_bit(bitidx + start_bitidx, bitmap);
4456}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004457
4458/*
4459 * This is designed as sub function...plz see page_isolation.c also.
4460 * set/clear page block's type to be ISOLATE.
4461 * page allocater never alloc memory from ISOLATE block.
4462 */
4463
4464int set_migratetype_isolate(struct page *page)
4465{
4466 struct zone *zone;
4467 unsigned long flags;
4468 int ret = -EBUSY;
4469
4470 zone = page_zone(page);
4471 spin_lock_irqsave(&zone->lock, flags);
4472 /*
4473 * In future, more migrate types will be able to be isolation target.
4474 */
4475 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4476 goto out;
4477 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4478 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4479 ret = 0;
4480out:
4481 spin_unlock_irqrestore(&zone->lock, flags);
4482 if (!ret)
4483 drain_all_local_pages();
4484 return ret;
4485}
4486
4487void unset_migratetype_isolate(struct page *page)
4488{
4489 struct zone *zone;
4490 unsigned long flags;
4491 zone = page_zone(page);
4492 spin_lock_irqsave(&zone->lock, flags);
4493 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4494 goto out;
4495 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4496 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4497out:
4498 spin_unlock_irqrestore(&zone->lock, flags);
4499}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07004500
4501#ifdef CONFIG_MEMORY_HOTREMOVE
4502/*
4503 * All pages in the range must be isolated before calling this.
4504 */
4505void
4506__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4507{
4508 struct page *page;
4509 struct zone *zone;
4510 int order, i;
4511 unsigned long pfn;
4512 unsigned long flags;
4513 /* find the first valid pfn */
4514 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4515 if (pfn_valid(pfn))
4516 break;
4517 if (pfn == end_pfn)
4518 return;
4519 zone = page_zone(pfn_to_page(pfn));
4520 spin_lock_irqsave(&zone->lock, flags);
4521 pfn = start_pfn;
4522 while (pfn < end_pfn) {
4523 if (!pfn_valid(pfn)) {
4524 pfn++;
4525 continue;
4526 }
4527 page = pfn_to_page(pfn);
4528 BUG_ON(page_count(page));
4529 BUG_ON(!PageBuddy(page));
4530 order = page_order(page);
4531#ifdef CONFIG_DEBUG_VM
4532 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4533 pfn, 1 << order, end_pfn);
4534#endif
4535 list_del(&page->lru);
4536 rmv_page_order(page);
4537 zone->free_area[order].nr_free--;
4538 __mod_zone_page_state(zone, NR_FREE_PAGES,
4539 - (1UL << order));
4540 for (i = 0; i < (1 << order); i++)
4541 SetPageReserved((page+i));
4542 pfn += (1 << order);
4543 }
4544 spin_unlock_irqrestore(&zone->lock, flags);
4545}
4546#endif