blob: 37576b822f06c98c2d3342af1cdecbe0e5a8f25b [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 Rientjes5a3135c2007-10-16 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 Morton6aa30012006-04-18 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 Morton6aa30012006-04-18 22:20:52 -0700384 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000385 }
Andrew Morton6aa30012006-04-18 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 Piggin92be2e32006-01-06 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 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800540void __init __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800541{
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 Piggin92be2e32006-01-06 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 Piggincc102502006-12-06 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 Gorman535131e2007-10-16 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
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800893/*
894 * Drain pages of the indicated processor.
895 *
896 * The processor must either be the current processor and the
897 * thread pinned to the current processor or a processor that
898 * is not online.
899 */
900static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901{
Nick Pigginc54ad302006-01-06 00:10:56 -0800902 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
905 for_each_zone(zone) {
906 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800907 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
Christoph Lameterf2e12bb2007-01-05 16:37:02 -0800909 if (!populated_zone(zone))
910 continue;
911
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700912 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800914 pcp = &pset->pcp;
915 local_irq_save(flags);
916 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
917 pcp->count = 0;
918 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 }
920}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800922/*
923 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
924 */
925void drain_local_pages(void *arg)
926{
927 drain_pages(smp_processor_id());
928}
929
930/*
931 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
932 */
933void drain_all_pages(void)
934{
935 on_each_cpu(drain_local_pages, NULL, 0, 1);
936}
937
Rafael J. Wysocki296699d2007-07-29 23:27:18 +0200938#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
940void mark_free_pages(struct zone *zone)
941{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700942 unsigned long pfn, max_zone_pfn;
943 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700944 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 struct list_head *curr;
946
947 if (!zone->spanned_pages)
948 return;
949
950 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700951
952 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
953 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
954 if (pfn_valid(pfn)) {
955 struct page *page = pfn_to_page(pfn);
956
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700957 if (!swsusp_page_is_forbidden(page))
958 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700959 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700961 for_each_migratetype_order(order, t) {
962 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700963 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700965 pfn = page_to_pfn(list_entry(curr, struct page, lru));
966 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700967 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700968 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700969 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 spin_unlock_irqrestore(&zone->lock, flags);
971}
Mel Gormane2c55dc2007-10-16 01:25:50 -0700972#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
974/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 * Free a 0-order page
976 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800977static void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978{
979 struct zone *zone = page_zone(page);
980 struct per_cpu_pages *pcp;
981 unsigned long flags;
982
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 if (PageAnon(page))
984 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800985 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800986 return;
987
Nick Piggin9858db52006-10-11 01:21:30 -0700988 if (!PageHighMem(page))
989 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Nick Piggindafb1362006-10-11 01:21:30 -0700990 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800991 kernel_map_pages(page, 1, 0);
992
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800993 pcp = &zone_pcp(zone, get_cpu())->pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700995 __count_vm_event(PGFREE);
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800996 if (cold)
997 list_add_tail(&page->lru, &pcp->list);
998 else
999 list_add(&page->lru, &pcp->list);
Mel Gorman535131e2007-10-16 01:25:49 -07001000 set_page_private(page, get_pageblock_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -08001002 if (pcp->count >= pcp->high) {
1003 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1004 pcp->count -= pcp->batch;
1005 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 local_irq_restore(flags);
1007 put_cpu();
1008}
1009
Harvey Harrison920c7a52008-02-04 22:29:26 -08001010void free_hot_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011{
1012 free_hot_cold_page(page, 0);
1013}
1014
Harvey Harrison920c7a52008-02-04 22:29:26 -08001015void free_cold_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016{
1017 free_hot_cold_page(page, 1);
1018}
1019
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001020/*
1021 * split_page takes a non-compound higher-order page, and splits it into
1022 * n (1<<order) sub-pages: page[0..n]
1023 * Each sub-page must be freed individually.
1024 *
1025 * Note: this is probably too low level an operation for use in drivers.
1026 * Please consult with lkml before using this in your driver.
1027 */
1028void split_page(struct page *page, unsigned int order)
1029{
1030 int i;
1031
Nick Piggin725d7042006-09-25 23:30:55 -07001032 VM_BUG_ON(PageCompound(page));
1033 VM_BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -08001034 for (i = 1; i < (1 << order); i++)
1035 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001036}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001037
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038/*
1039 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1040 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1041 * or two.
1042 */
Nick Piggina74609f2006-01-06 00:11:20 -08001043static struct page *buffered_rmqueue(struct zonelist *zonelist,
1044 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045{
1046 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001047 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -08001049 int cpu;
Mel Gorman64c5e132007-10-16 01:25:59 -07001050 int migratetype = allocflags_to_migratetype(gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
Hugh Dickins689bceb2005-11-21 21:32:20 -08001052again:
Nick Piggina74609f2006-01-06 00:11:20 -08001053 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -08001054 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 struct per_cpu_pages *pcp;
1056
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001057 pcp = &zone_pcp(zone, cpu)->pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001059 if (!pcp->count) {
nkalmala941c7102006-11-02 22:07:04 -08001060 pcp->count = rmqueue_bulk(zone, 0,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001061 pcp->batch, &pcp->list, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001062 if (unlikely(!pcp->count))
1063 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 }
Mel Gorman535131e2007-10-16 01:25:49 -07001065
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001066 /* Find a page of the appropriate migrate type */
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001067 if (cold) {
1068 list_for_each_entry_reverse(page, &pcp->list, lru)
1069 if (page_private(page) == migratetype)
1070 break;
1071 } else {
1072 list_for_each_entry(page, &pcp->list, lru)
1073 if (page_private(page) == migratetype)
1074 break;
1075 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001076
1077 /* Allocate more to the pcp list if necessary */
1078 if (unlikely(&page->lru == &pcp->list)) {
Mel Gorman535131e2007-10-16 01:25:49 -07001079 pcp->count += rmqueue_bulk(zone, 0,
1080 pcp->batch, &pcp->list, migratetype);
1081 page = list_entry(pcp->list.next, struct page, lru);
Mel Gorman535131e2007-10-16 01:25:49 -07001082 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001083
1084 list_del(&page->lru);
1085 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001086 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001088 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001089 spin_unlock(&zone->lock);
1090 if (!page)
1091 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 }
1093
Christoph Lameterf8891e52006-06-30 01:55:45 -07001094 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Christoph Lameterca889e62006-06-30 01:55:44 -07001095 zone_statistics(zonelist, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001096 local_irq_restore(flags);
1097 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098
Nick Piggin725d7042006-09-25 23:30:55 -07001099 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001100 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001101 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001103
1104failed:
1105 local_irq_restore(flags);
1106 put_cpu();
1107 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108}
1109
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001110#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -08001111#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1112#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1113#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1114#define ALLOC_HARDER 0x10 /* try to alloc harder */
1115#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1116#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001117
Akinobu Mita933e3122006-12-08 02:39:45 -08001118#ifdef CONFIG_FAIL_PAGE_ALLOC
1119
1120static struct fail_page_alloc_attr {
1121 struct fault_attr attr;
1122
1123 u32 ignore_gfp_highmem;
1124 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001125 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001126
1127#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1128
1129 struct dentry *ignore_gfp_highmem_file;
1130 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001131 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001132
1133#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1134
1135} fail_page_alloc = {
1136 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001137 .ignore_gfp_wait = 1,
1138 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001139 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001140};
1141
1142static int __init setup_fail_page_alloc(char *str)
1143{
1144 return setup_fault_attr(&fail_page_alloc.attr, str);
1145}
1146__setup("fail_page_alloc=", setup_fail_page_alloc);
1147
1148static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1149{
Akinobu Mita54114992007-07-15 23:40:23 -07001150 if (order < fail_page_alloc.min_order)
1151 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001152 if (gfp_mask & __GFP_NOFAIL)
1153 return 0;
1154 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1155 return 0;
1156 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1157 return 0;
1158
1159 return should_fail(&fail_page_alloc.attr, 1 << order);
1160}
1161
1162#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1163
1164static int __init fail_page_alloc_debugfs(void)
1165{
1166 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1167 struct dentry *dir;
1168 int err;
1169
1170 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1171 "fail_page_alloc");
1172 if (err)
1173 return err;
1174 dir = fail_page_alloc.attr.dentries.dir;
1175
1176 fail_page_alloc.ignore_gfp_wait_file =
1177 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1178 &fail_page_alloc.ignore_gfp_wait);
1179
1180 fail_page_alloc.ignore_gfp_highmem_file =
1181 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1182 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001183 fail_page_alloc.min_order_file =
1184 debugfs_create_u32("min-order", mode, dir,
1185 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001186
1187 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001188 !fail_page_alloc.ignore_gfp_highmem_file ||
1189 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001190 err = -ENOMEM;
1191 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1192 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001193 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001194 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1195 }
1196
1197 return err;
1198}
1199
1200late_initcall(fail_page_alloc_debugfs);
1201
1202#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1203
1204#else /* CONFIG_FAIL_PAGE_ALLOC */
1205
1206static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1207{
1208 return 0;
1209}
1210
1211#endif /* CONFIG_FAIL_PAGE_ALLOC */
1212
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213/*
1214 * Return 1 if free pages are above 'mark'. This takes into account the order
1215 * of the allocation.
1216 */
1217int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001218 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219{
1220 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001221 long min = mark;
1222 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 int o;
1224
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001225 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001227 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 min -= min / 4;
1229
1230 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1231 return 0;
1232 for (o = 0; o < order; o++) {
1233 /* At the next order, this order's pages become unavailable */
1234 free_pages -= z->free_area[o].nr_free << o;
1235
1236 /* Require fewer higher order pages to be free */
1237 min >>= 1;
1238
1239 if (free_pages <= min)
1240 return 0;
1241 }
1242 return 1;
1243}
1244
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001245#ifdef CONFIG_NUMA
1246/*
1247 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1248 * skip over zones that are not allowed by the cpuset, or that have
1249 * been recently (in last second) found to be nearly full. See further
1250 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001251 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001252 *
1253 * If the zonelist cache is present in the passed in zonelist, then
1254 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001255 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001256 *
1257 * If the zonelist cache is not available for this zonelist, does
1258 * nothing and returns NULL.
1259 *
1260 * If the fullzones BITMAP in the zonelist cache is stale (more than
1261 * a second since last zap'd) then we zap it out (clear its bits.)
1262 *
1263 * We hold off even calling zlc_setup, until after we've checked the
1264 * first zone in the zonelist, on the theory that most allocations will
1265 * be satisfied from that first zone, so best to examine that zone as
1266 * quickly as we can.
1267 */
1268static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1269{
1270 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1271 nodemask_t *allowednodes; /* zonelist_cache approximation */
1272
1273 zlc = zonelist->zlcache_ptr;
1274 if (!zlc)
1275 return NULL;
1276
1277 if (jiffies - zlc->last_full_zap > 1 * HZ) {
1278 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1279 zlc->last_full_zap = jiffies;
1280 }
1281
1282 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1283 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001284 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001285 return allowednodes;
1286}
1287
1288/*
1289 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1290 * if it is worth looking at further for free memory:
1291 * 1) Check that the zone isn't thought to be full (doesn't have its
1292 * bit set in the zonelist_cache fullzones BITMAP).
1293 * 2) Check that the zones node (obtained from the zonelist_cache
1294 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1295 * Return true (non-zero) if zone is worth looking at further, or
1296 * else return false (zero) if it is not.
1297 *
1298 * This check -ignores- the distinction between various watermarks,
1299 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1300 * found to be full for any variation of these watermarks, it will
1301 * be considered full for up to one second by all requests, unless
1302 * we are so low on memory on all allowed nodes that we are forced
1303 * into the second scan of the zonelist.
1304 *
1305 * In the second scan we ignore this zonelist cache and exactly
1306 * apply the watermarks to all zones, even it is slower to do so.
1307 * We are low on memory in the second scan, and should leave no stone
1308 * unturned looking for a free page.
1309 */
1310static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1311 nodemask_t *allowednodes)
1312{
1313 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1314 int i; /* index of *z in zonelist zones */
1315 int n; /* node that zone *z is on */
1316
1317 zlc = zonelist->zlcache_ptr;
1318 if (!zlc)
1319 return 1;
1320
1321 i = z - zonelist->zones;
1322 n = zlc->z_to_n[i];
1323
1324 /* This zone is worth trying if it is allowed but not full */
1325 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1326}
1327
1328/*
1329 * Given 'z' scanning a zonelist, set the corresponding bit in
1330 * zlc->fullzones, so that subsequent attempts to allocate a page
1331 * from that zone don't waste time re-examining it.
1332 */
1333static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1334{
1335 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1336 int i; /* index of *z in zonelist zones */
1337
1338 zlc = zonelist->zlcache_ptr;
1339 if (!zlc)
1340 return;
1341
1342 i = z - zonelist->zones;
1343
1344 set_bit(i, zlc->fullzones);
1345}
1346
1347#else /* CONFIG_NUMA */
1348
1349static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1350{
1351 return NULL;
1352}
1353
1354static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1355 nodemask_t *allowednodes)
1356{
1357 return 1;
1358}
1359
1360static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1361{
1362}
1363#endif /* CONFIG_NUMA */
1364
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001365/*
Paul Jackson0798e512006-12-06 20:31:38 -08001366 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001367 * a page.
1368 */
1369static struct page *
1370get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1371 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -07001372{
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001373 struct zone **z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001374 struct page *page = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001375 int classzone_idx = zone_idx(zonelist->zones[0]);
Christoph Lameter1192d522006-09-25 23:31:45 -07001376 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001377 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1378 int zlc_active = 0; /* set if using zonelist_cache */
1379 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gormanb377fd32007-08-22 14:02:05 -07001380 enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001381
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001382zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001383 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001384 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001385 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1386 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001387 z = zonelist->zones;
1388
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001389 do {
Mel Gormanb377fd32007-08-22 14:02:05 -07001390 /*
1391 * In NUMA, this could be a policy zonelist which contains
1392 * zones that may not be allowed by the current gfp_mask.
1393 * Check the zone is allowed by the current flags
1394 */
1395 if (unlikely(alloc_should_filter_zonelist(zonelist))) {
1396 if (highest_zoneidx == -1)
1397 highest_zoneidx = gfp_zone(gfp_mask);
1398 if (zone_idx(*z) > highest_zoneidx)
1399 continue;
1400 }
1401
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001402 if (NUMA_BUILD && zlc_active &&
1403 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1404 continue;
Christoph Lameter1192d522006-09-25 23:31:45 -07001405 zone = *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001406 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001407 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001408 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001409
1410 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001411 unsigned long mark;
1412 if (alloc_flags & ALLOC_WMARK_MIN)
Christoph Lameter1192d522006-09-25 23:31:45 -07001413 mark = zone->pages_min;
Nick Piggin31488902005-11-28 13:44:03 -08001414 else if (alloc_flags & ALLOC_WMARK_LOW)
Christoph Lameter1192d522006-09-25 23:31:45 -07001415 mark = zone->pages_low;
Nick Piggin31488902005-11-28 13:44:03 -08001416 else
Christoph Lameter1192d522006-09-25 23:31:45 -07001417 mark = zone->pages_high;
Paul Jackson0798e512006-12-06 20:31:38 -08001418 if (!zone_watermark_ok(zone, order, mark,
1419 classzone_idx, alloc_flags)) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001420 if (!zone_reclaim_mode ||
Christoph Lameter1192d522006-09-25 23:31:45 -07001421 !zone_reclaim(zone, gfp_mask, order))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001422 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001423 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001424 }
1425
Christoph Lameter1192d522006-09-25 23:31:45 -07001426 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
Paul Jackson0798e512006-12-06 20:31:38 -08001427 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001428 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001429this_zone_full:
1430 if (NUMA_BUILD)
1431 zlc_mark_zone_full(zonelist, z);
1432try_next_zone:
1433 if (NUMA_BUILD && !did_zlc_setup) {
1434 /* we do zlc_setup after the first zone is tried */
1435 allowednodes = zlc_setup(zonelist, alloc_flags);
1436 zlc_active = 1;
1437 did_zlc_setup = 1;
1438 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001439 } while (*(++z) != NULL);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001440
1441 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1442 /* Disable zlc cache for second zonelist scan */
1443 zlc_active = 0;
1444 goto zonelist_scan;
1445 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001446 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001447}
1448
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449/*
1450 * This is the 'heart' of the zoned buddy allocator.
1451 */
1452struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +01001453__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 struct zonelist *zonelist)
1455{
Al Viro260b2362005-10-21 03:22:44 -04001456 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001457 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 struct page *page;
1459 struct reclaim_state reclaim_state;
1460 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001462 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 int did_some_progress;
1464
1465 might_sleep_if(wait);
1466
Akinobu Mita933e3122006-12-08 02:39:45 -08001467 if (should_fail_alloc_page(gfp_mask, order))
1468 return NULL;
1469
Jens Axboe6b1de912005-11-17 21:35:02 +01001470restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001471 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001473 if (unlikely(*z == NULL)) {
Christoph Lameter523b9452007-10-16 01:25:37 -07001474 /*
1475 * Happens if we have an empty zonelist as a result of
1476 * GFP_THISNODE being used on a memoryless node
1477 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 return NULL;
1479 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001480
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001481 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001482 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001483 if (page)
1484 goto got_pg;
1485
Christoph Lameter952f3b52006-12-06 20:33:26 -08001486 /*
1487 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1488 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1489 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1490 * using a larger set of nodes after it has established that the
1491 * allowed per node queues are empty and that nodes are
1492 * over allocated.
1493 */
1494 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1495 goto nopage;
1496
Paul Jackson0798e512006-12-06 20:31:38 -08001497 for (z = zonelist->zones; *z; z++)
Chris Wright43b0bc02006-06-25 05:47:55 -07001498 wakeup_kswapd(*z, order);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001499
Paul Jackson9bf22292005-09-06 15:18:12 -07001500 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001501 * OK, we're below the kswapd watermark and have kicked background
1502 * reclaim. Now things get more complex, so set up alloc_flags according
1503 * to how we want to proceed.
1504 *
1505 * The caller may dip into page reserves a bit more if the caller
1506 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -08001507 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1508 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -07001509 */
Nick Piggin31488902005-11-28 13:44:03 -08001510 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001511 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1512 alloc_flags |= ALLOC_HARDER;
1513 if (gfp_mask & __GFP_HIGH)
1514 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -07001515 if (wait)
1516 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517
1518 /*
1519 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001520 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 *
1522 * This is the last chance, in general, before the goto nopage.
1523 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -07001524 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001526 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1527 if (page)
1528 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
1530 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -07001531
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001532rebalance:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001533 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1534 && !in_interrupt()) {
1535 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -08001536nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001537 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001538 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -08001539 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001540 if (page)
1541 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -08001542 if (gfp_mask & __GFP_NOFAIL) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001543 congestion_wait(WRITE, HZ/50);
Kirill Korotaev885036d2005-11-13 16:06:41 -08001544 goto nofail_alloc;
1545 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 }
1547 goto nopage;
1548 }
1549
1550 /* Atomic allocations - we can't balance anything */
1551 if (!wait)
1552 goto nopage;
1553
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 cond_resched();
1555
1556 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001557 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 p->flags |= PF_MEMALLOC;
1559 reclaim_state.reclaimed_slab = 0;
1560 p->reclaim_state = &reclaim_state;
1561
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001562 did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563
1564 p->reclaim_state = NULL;
1565 p->flags &= ~PF_MEMALLOC;
1566
1567 cond_resched();
1568
Mel Gormane2c55dc2007-10-16 01:25:50 -07001569 if (order != 0)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001570 drain_all_pages();
Mel Gormane2c55dc2007-10-16 01:25:50 -07001571
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001573 page = get_page_from_freelist(gfp_mask, order,
1574 zonelist, alloc_flags);
1575 if (page)
1576 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
David Rientjesff0ceb92007-10-16 23:25:56 -07001578 if (!try_set_zone_oom(zonelist)) {
1579 schedule_timeout_uninterruptible(1);
1580 goto restart;
1581 }
1582
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 /*
1584 * Go through the zonelist yet one more time, keep
1585 * very high watermark here, this is only to catch
1586 * a parallel oom killing, we must fail if we're still
1587 * under heavy pressure.
1588 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001589 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001590 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
David Rientjesff0ceb92007-10-16 23:25:56 -07001591 if (page) {
1592 clear_zonelist_oom(zonelist);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001593 goto got_pg;
David Rientjesff0ceb92007-10-16 23:25:56 -07001594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595
Mel Gormana8bbf722007-07-31 00:37:30 -07001596 /* The OOM killer will not help higher order allocs so fail */
David Rientjesff0ceb92007-10-16 23:25:56 -07001597 if (order > PAGE_ALLOC_COSTLY_ORDER) {
1598 clear_zonelist_oom(zonelist);
Mel Gormana8bbf722007-07-31 00:37:30 -07001599 goto nopage;
David Rientjesff0ceb92007-10-16 23:25:56 -07001600 }
Mel Gormana8bbf722007-07-31 00:37:30 -07001601
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001602 out_of_memory(zonelist, gfp_mask, order);
David Rientjesff0ceb92007-10-16 23:25:56 -07001603 clear_zonelist_oom(zonelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 goto restart;
1605 }
1606
1607 /*
1608 * Don't let big-order allocations loop unless the caller explicitly
1609 * requests that. Wait for some write requests to complete then retry.
1610 *
1611 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1612 * <= 3, but that may not be true in other implementations.
1613 */
1614 do_retry = 0;
1615 if (!(gfp_mask & __GFP_NORETRY)) {
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001616 if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
1617 (gfp_mask & __GFP_REPEAT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 do_retry = 1;
1619 if (gfp_mask & __GFP_NOFAIL)
1620 do_retry = 1;
1621 }
1622 if (do_retry) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001623 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 goto rebalance;
1625 }
1626
1627nopage:
1628 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1629 printk(KERN_WARNING "%s: page allocation failure."
1630 " order:%d, mode:0x%x\n",
1631 p->comm, order, gfp_mask);
1632 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001633 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 return page;
1637}
1638
1639EXPORT_SYMBOL(__alloc_pages);
1640
1641/*
1642 * Common helper functions.
1643 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001644unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645{
1646 struct page * page;
1647 page = alloc_pages(gfp_mask, order);
1648 if (!page)
1649 return 0;
1650 return (unsigned long) page_address(page);
1651}
1652
1653EXPORT_SYMBOL(__get_free_pages);
1654
Harvey Harrison920c7a52008-02-04 22:29:26 -08001655unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656{
1657 struct page * page;
1658
1659 /*
1660 * get_zeroed_page() returns a 32-bit address, which cannot represent
1661 * a highmem page
1662 */
Nick Piggin725d7042006-09-25 23:30:55 -07001663 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
1665 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1666 if (page)
1667 return (unsigned long) page_address(page);
1668 return 0;
1669}
1670
1671EXPORT_SYMBOL(get_zeroed_page);
1672
1673void __pagevec_free(struct pagevec *pvec)
1674{
1675 int i = pagevec_count(pvec);
1676
1677 while (--i >= 0)
1678 free_hot_cold_page(pvec->pages[i], pvec->cold);
1679}
1680
Harvey Harrison920c7a52008-02-04 22:29:26 -08001681void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682{
Nick Pigginb5810032005-10-29 18:16:12 -07001683 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 if (order == 0)
1685 free_hot_page(page);
1686 else
1687 __free_pages_ok(page, order);
1688 }
1689}
1690
1691EXPORT_SYMBOL(__free_pages);
1692
Harvey Harrison920c7a52008-02-04 22:29:26 -08001693void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694{
1695 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001696 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 __free_pages(virt_to_page((void *)addr), order);
1698 }
1699}
1700
1701EXPORT_SYMBOL(free_pages);
1702
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703static unsigned int nr_free_zone_pages(int offset)
1704{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001705 /* Just pick one node, since fallback list is circular */
1706 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 unsigned int sum = 0;
1708
Martin J. Blighe310fd42005-07-29 22:59:18 -07001709 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1710 struct zone **zonep = zonelist->zones;
1711 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712
Martin J. Blighe310fd42005-07-29 22:59:18 -07001713 for (zone = *zonep++; zone; zone = *zonep++) {
1714 unsigned long size = zone->present_pages;
1715 unsigned long high = zone->pages_high;
1716 if (size > high)
1717 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 }
1719
1720 return sum;
1721}
1722
1723/*
1724 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1725 */
1726unsigned int nr_free_buffer_pages(void)
1727{
Al Viroaf4ca452005-10-21 02:55:38 -04001728 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729}
Meelap Shahc2f1a552007-07-17 04:04:39 -07001730EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731
1732/*
1733 * Amount of free RAM allocatable within all zones
1734 */
1735unsigned int nr_free_pagecache_pages(void)
1736{
Mel Gorman2a1e2742007-07-17 04:03:12 -07001737 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001739
1740static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001742 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08001743 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746void si_meminfo(struct sysinfo *val)
1747{
1748 val->totalram = totalram_pages;
1749 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001750 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 val->totalhigh = totalhigh_pages;
1753 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 val->mem_unit = PAGE_SIZE;
1755}
1756
1757EXPORT_SYMBOL(si_meminfo);
1758
1759#ifdef CONFIG_NUMA
1760void si_meminfo_node(struct sysinfo *val, int nid)
1761{
1762 pg_data_t *pgdat = NODE_DATA(nid);
1763
1764 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001765 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001766#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001768 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1769 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001770#else
1771 val->totalhigh = 0;
1772 val->freehigh = 0;
1773#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 val->mem_unit = PAGE_SIZE;
1775}
1776#endif
1777
1778#define K(x) ((x) << (PAGE_SHIFT-10))
1779
1780/*
1781 * Show free area list (used inside shift_scroll-lock stuff)
1782 * We also calculate the percentage fragmentation. We do this by counting the
1783 * memory on each free list with the exception of the first item on the list.
1784 */
1785void show_free_areas(void)
1786{
Jes Sorensenc7241912006-09-27 01:50:05 -07001787 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 struct zone *zone;
1789
1790 for_each_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07001791 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07001793
1794 show_node(zone);
1795 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796
Dave Jones6b482c62005-11-10 15:45:56 -05001797 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 struct per_cpu_pageset *pageset;
1799
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001800 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001802 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1803 cpu, pageset->pcp.high,
1804 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 }
1806 }
1807
Andrew Mortona25700a2007-02-08 14:20:40 -08001808 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
Christoph Lameterd23ad422007-02-10 01:43:02 -08001809 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
Christoph Lameter65e458d42007-02-10 01:43:05 -08001810 global_page_state(NR_ACTIVE),
1811 global_page_state(NR_INACTIVE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07001812 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07001813 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001814 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08001815 global_page_state(NR_FREE_PAGES),
Christoph Lameter972d1a72006-09-25 23:31:51 -07001816 global_page_state(NR_SLAB_RECLAIMABLE) +
1817 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001818 global_page_state(NR_FILE_MAPPED),
Andrew Mortona25700a2007-02-08 14:20:40 -08001819 global_page_state(NR_PAGETABLE),
1820 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821
1822 for_each_zone(zone) {
1823 int i;
1824
Jes Sorensenc7241912006-09-27 01:50:05 -07001825 if (!populated_zone(zone))
1826 continue;
1827
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 show_node(zone);
1829 printk("%s"
1830 " free:%lukB"
1831 " min:%lukB"
1832 " low:%lukB"
1833 " high:%lukB"
1834 " active:%lukB"
1835 " inactive:%lukB"
1836 " present:%lukB"
1837 " pages_scanned:%lu"
1838 " all_unreclaimable? %s"
1839 "\n",
1840 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08001841 K(zone_page_state(zone, NR_FREE_PAGES)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 K(zone->pages_min),
1843 K(zone->pages_low),
1844 K(zone->pages_high),
Christoph Lameterc8785382007-02-10 01:43:01 -08001845 K(zone_page_state(zone, NR_ACTIVE)),
1846 K(zone_page_state(zone, NR_INACTIVE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 K(zone->present_pages),
1848 zone->pages_scanned,
David Rientjese815af92007-10-16 23:25:54 -07001849 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 );
1851 printk("lowmem_reserve[]:");
1852 for (i = 0; i < MAX_NR_ZONES; i++)
1853 printk(" %lu", zone->lowmem_reserve[i]);
1854 printk("\n");
1855 }
1856
1857 for_each_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001858 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859
Jes Sorensenc7241912006-09-27 01:50:05 -07001860 if (!populated_zone(zone))
1861 continue;
1862
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 show_node(zone);
1864 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865
1866 spin_lock_irqsave(&zone->lock, flags);
1867 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001868 nr[order] = zone->free_area[order].nr_free;
1869 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 }
1871 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001872 for (order = 0; order < MAX_ORDER; order++)
1873 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 printk("= %lukB\n", K(total));
1875 }
1876
Larry Woodmane6f36022008-02-04 22:29:30 -08001877 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
1878
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 show_swap_cache_info();
1880}
1881
1882/*
1883 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001884 *
1885 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001887static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
1888 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889{
Christoph Lameter1a932052006-01-06 00:11:16 -08001890 struct zone *zone;
1891
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001892 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001893 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001894
1895 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001896 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08001897 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001898 if (populated_zone(zone)) {
Christoph Lameter070f8032006-01-06 00:11:19 -08001899 zonelist->zones[nr_zones++] = zone;
1900 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08001902
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001903 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08001904 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905}
1906
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001907
1908/*
1909 * zonelist_order:
1910 * 0 = automatic detection of better ordering.
1911 * 1 = order by ([node] distance, -zonetype)
1912 * 2 = order by (-zonetype, [node] distance)
1913 *
1914 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1915 * the same zonelist. So only NUMA can configure this param.
1916 */
1917#define ZONELIST_ORDER_DEFAULT 0
1918#define ZONELIST_ORDER_NODE 1
1919#define ZONELIST_ORDER_ZONE 2
1920
1921/* zonelist order in the kernel.
1922 * set_zonelist_order() will set this to NODE or ZONE.
1923 */
1924static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
1925static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
1926
1927
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001929/* The value user specified ....changed by config */
1930static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1931/* string for sysctl */
1932#define NUMA_ZONELIST_ORDER_LEN 16
1933char numa_zonelist_order[16] = "default";
1934
1935/*
1936 * interface for configure zonelist ordering.
1937 * command line option "numa_zonelist_order"
1938 * = "[dD]efault - default, automatic configuration.
1939 * = "[nN]ode - order by node locality, then by zone within node
1940 * = "[zZ]one - order by zone, then by locality within zone
1941 */
1942
1943static int __parse_numa_zonelist_order(char *s)
1944{
1945 if (*s == 'd' || *s == 'D') {
1946 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1947 } else if (*s == 'n' || *s == 'N') {
1948 user_zonelist_order = ZONELIST_ORDER_NODE;
1949 } else if (*s == 'z' || *s == 'Z') {
1950 user_zonelist_order = ZONELIST_ORDER_ZONE;
1951 } else {
1952 printk(KERN_WARNING
1953 "Ignoring invalid numa_zonelist_order value: "
1954 "%s\n", s);
1955 return -EINVAL;
1956 }
1957 return 0;
1958}
1959
1960static __init int setup_numa_zonelist_order(char *s)
1961{
1962 if (s)
1963 return __parse_numa_zonelist_order(s);
1964 return 0;
1965}
1966early_param("numa_zonelist_order", setup_numa_zonelist_order);
1967
1968/*
1969 * sysctl handler for numa_zonelist_order
1970 */
1971int numa_zonelist_order_handler(ctl_table *table, int write,
1972 struct file *file, void __user *buffer, size_t *length,
1973 loff_t *ppos)
1974{
1975 char saved_string[NUMA_ZONELIST_ORDER_LEN];
1976 int ret;
1977
1978 if (write)
1979 strncpy(saved_string, (char*)table->data,
1980 NUMA_ZONELIST_ORDER_LEN);
1981 ret = proc_dostring(table, write, file, buffer, length, ppos);
1982 if (ret)
1983 return ret;
1984 if (write) {
1985 int oldval = user_zonelist_order;
1986 if (__parse_numa_zonelist_order((char*)table->data)) {
1987 /*
1988 * bogus value. restore saved string
1989 */
1990 strncpy((char*)table->data, saved_string,
1991 NUMA_ZONELIST_ORDER_LEN);
1992 user_zonelist_order = oldval;
1993 } else if (oldval != user_zonelist_order)
1994 build_all_zonelists();
1995 }
1996 return 0;
1997}
1998
1999
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000#define MAX_NODE_LOAD (num_online_nodes())
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002001static int node_load[MAX_NUMNODES];
2002
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002004 * 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 -07002005 * @node: node whose fallback list we're appending
2006 * @used_node_mask: nodemask_t of already used nodes
2007 *
2008 * We use a number of factors to determine which is the next node that should
2009 * appear on a given node's fallback list. The node should not have appeared
2010 * already in @node's fallback list, and it should be the next closest node
2011 * according to the distance array (which contains arbitrary distance values
2012 * from each node to each node in the system), and should also prefer nodes
2013 * with no CPUs, since presumably they'll have very little allocation pressure
2014 * on them otherwise.
2015 * It returns -1 if no node is found.
2016 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002017static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002019 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 int min_val = INT_MAX;
2021 int best_node = -1;
2022
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002023 /* Use the local node if we haven't already */
2024 if (!node_isset(node, *used_node_mask)) {
2025 node_set(node, *used_node_mask);
2026 return node;
2027 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002029 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002030 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031
2032 /* Don't want a node to appear more than once */
2033 if (node_isset(n, *used_node_mask))
2034 continue;
2035
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 /* Use the distance array to find the distance */
2037 val = node_distance(node, n);
2038
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002039 /* Penalize nodes under us ("prefer the next node") */
2040 val += (n < node);
2041
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 /* Give preference to headless and unused nodes */
2043 tmp = node_to_cpumask(n);
2044 if (!cpus_empty(tmp))
2045 val += PENALTY_FOR_NODE_WITH_CPUS;
2046
2047 /* Slight preference for less loaded node */
2048 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2049 val += node_load[n];
2050
2051 if (val < min_val) {
2052 min_val = val;
2053 best_node = n;
2054 }
2055 }
2056
2057 if (best_node >= 0)
2058 node_set(best_node, *used_node_mask);
2059
2060 return best_node;
2061}
2062
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002063
2064/*
2065 * Build zonelists ordered by node and zones within node.
2066 * This results in maximum locality--normal zone overflows into local
2067 * DMA zone, if any--but risks exhausting DMA zone.
2068 */
2069static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070{
Christoph Lameter19655d32006-09-25 23:31:19 -07002071 enum zone_type i;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002072 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002074
2075 for (i = 0; i < MAX_NR_ZONES; i++) {
2076 zonelist = pgdat->node_zonelists + i;
2077 for (j = 0; zonelist->zones[j] != NULL; j++)
2078 ;
2079 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2080 zonelist->zones[j] = NULL;
2081 }
2082}
2083
2084/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002085 * Build gfp_thisnode zonelists
2086 */
2087static void build_thisnode_zonelists(pg_data_t *pgdat)
2088{
2089 enum zone_type i;
2090 int j;
2091 struct zonelist *zonelist;
2092
2093 for (i = 0; i < MAX_NR_ZONES; i++) {
2094 zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i;
2095 j = build_zonelists_node(pgdat, zonelist, 0, i);
2096 zonelist->zones[j] = NULL;
2097 }
2098}
2099
2100/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002101 * Build zonelists ordered by zone and nodes within zones.
2102 * This results in conserving DMA zone[s] until all Normal memory is
2103 * exhausted, but results in overflowing to remote node while memory
2104 * may still exist in local DMA zone.
2105 */
2106static int node_order[MAX_NUMNODES];
2107
2108static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2109{
2110 enum zone_type i;
2111 int pos, j, node;
2112 int zone_type; /* needs to be signed */
2113 struct zone *z;
2114 struct zonelist *zonelist;
2115
2116 for (i = 0; i < MAX_NR_ZONES; i++) {
2117 zonelist = pgdat->node_zonelists + i;
2118 pos = 0;
2119 for (zone_type = i; zone_type >= 0; zone_type--) {
2120 for (j = 0; j < nr_nodes; j++) {
2121 node = node_order[j];
2122 z = &NODE_DATA(node)->node_zones[zone_type];
2123 if (populated_zone(z)) {
2124 zonelist->zones[pos++] = z;
2125 check_highest_zone(zone_type);
2126 }
2127 }
2128 }
2129 zonelist->zones[pos] = NULL;
2130 }
2131}
2132
2133static int default_zonelist_order(void)
2134{
2135 int nid, zone_type;
2136 unsigned long low_kmem_size,total_size;
2137 struct zone *z;
2138 int average_size;
2139 /*
2140 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2141 * If they are really small and used heavily, the system can fall
2142 * into OOM very easily.
2143 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2144 */
2145 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2146 low_kmem_size = 0;
2147 total_size = 0;
2148 for_each_online_node(nid) {
2149 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2150 z = &NODE_DATA(nid)->node_zones[zone_type];
2151 if (populated_zone(z)) {
2152 if (zone_type < ZONE_NORMAL)
2153 low_kmem_size += z->present_pages;
2154 total_size += z->present_pages;
2155 }
2156 }
2157 }
2158 if (!low_kmem_size || /* there are no DMA area. */
2159 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2160 return ZONELIST_ORDER_NODE;
2161 /*
2162 * look into each node's config.
2163 * If there is a node whose DMA/DMA32 memory is very big area on
2164 * local memory, NODE_ORDER may be suitable.
2165 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002166 average_size = total_size /
2167 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002168 for_each_online_node(nid) {
2169 low_kmem_size = 0;
2170 total_size = 0;
2171 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2172 z = &NODE_DATA(nid)->node_zones[zone_type];
2173 if (populated_zone(z)) {
2174 if (zone_type < ZONE_NORMAL)
2175 low_kmem_size += z->present_pages;
2176 total_size += z->present_pages;
2177 }
2178 }
2179 if (low_kmem_size &&
2180 total_size > average_size && /* ignore small node */
2181 low_kmem_size > total_size * 70/100)
2182 return ZONELIST_ORDER_NODE;
2183 }
2184 return ZONELIST_ORDER_ZONE;
2185}
2186
2187static void set_zonelist_order(void)
2188{
2189 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2190 current_zonelist_order = default_zonelist_order();
2191 else
2192 current_zonelist_order = user_zonelist_order;
2193}
2194
2195static void build_zonelists(pg_data_t *pgdat)
2196{
2197 int j, node, load;
2198 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002200 int local_node, prev_node;
2201 struct zonelist *zonelist;
2202 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203
2204 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002205 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 zonelist = pgdat->node_zonelists + i;
2207 zonelist->zones[0] = NULL;
2208 }
2209
2210 /* NUMA-aware ordering of nodes */
2211 local_node = pgdat->node_id;
2212 load = num_online_nodes();
2213 prev_node = local_node;
2214 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002215
2216 memset(node_load, 0, sizeof(node_load));
2217 memset(node_order, 0, sizeof(node_order));
2218 j = 0;
2219
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002221 int distance = node_distance(local_node, node);
2222
2223 /*
2224 * If another node is sufficiently far away then it is better
2225 * to reclaim pages in a zone before going off node.
2226 */
2227 if (distance > RECLAIM_DISTANCE)
2228 zone_reclaim_mode = 1;
2229
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 /*
2231 * We don't want to pressure a particular node.
2232 * So adding penalty to the first node in same
2233 * distance group to make it round-robin.
2234 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002235 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002236 node_load[node] = load;
2237
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 prev_node = node;
2239 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002240 if (order == ZONELIST_ORDER_NODE)
2241 build_zonelists_in_node_order(pgdat, node);
2242 else
2243 node_order[j++] = node; /* remember order */
2244 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002246 if (order == ZONELIST_ORDER_ZONE) {
2247 /* calculate node order -- i.e., DMA last! */
2248 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002250
2251 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252}
2253
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002254/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002255static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002256{
2257 int i;
2258
2259 for (i = 0; i < MAX_NR_ZONES; i++) {
2260 struct zonelist *zonelist;
2261 struct zonelist_cache *zlc;
2262 struct zone **z;
2263
2264 zonelist = pgdat->node_zonelists + i;
2265 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2266 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2267 for (z = zonelist->zones; *z; z++)
2268 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
2269 }
2270}
2271
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002272
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273#else /* CONFIG_NUMA */
2274
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002275static void set_zonelist_order(void)
2276{
2277 current_zonelist_order = ZONELIST_ORDER_ZONE;
2278}
2279
2280static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281{
Christoph Lameter19655d32006-09-25 23:31:19 -07002282 int node, local_node;
2283 enum zone_type i,j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284
2285 local_node = pgdat->node_id;
Christoph Lameter19655d32006-09-25 23:31:19 -07002286 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 struct zonelist *zonelist;
2288
2289 zonelist = pgdat->node_zonelists + i;
2290
Christoph Lameter19655d32006-09-25 23:31:19 -07002291 j = build_zonelists_node(pgdat, zonelist, 0, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 /*
2293 * Now we build the zonelist so that it contains the zones
2294 * of all the other nodes.
2295 * We don't want to pressure a particular node, so when
2296 * building the zones for node N, we make sure that the
2297 * zones coming right after the local ones are those from
2298 * node N+1 (modulo N)
2299 */
2300 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2301 if (!node_online(node))
2302 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07002303 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 }
2305 for (node = 0; node < local_node; node++) {
2306 if (!node_online(node))
2307 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07002308 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 }
2310
2311 zonelist->zones[j] = NULL;
2312 }
2313}
2314
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002315/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002316static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002317{
2318 int i;
2319
2320 for (i = 0; i < MAX_NR_ZONES; i++)
2321 pgdat->node_zonelists[i].zlcache_ptr = NULL;
2322}
2323
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324#endif /* CONFIG_NUMA */
2325
Yasunori Goto68113782006-06-23 02:03:11 -07002326/* return values int ....just for stop_machine_run() */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002327static int __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328{
Yasunori Goto68113782006-06-23 02:03:11 -07002329 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002330
2331 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07002332 pg_data_t *pgdat = NODE_DATA(nid);
2333
2334 build_zonelists(pgdat);
2335 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002336 }
Yasunori Goto68113782006-06-23 02:03:11 -07002337 return 0;
2338}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002340void build_all_zonelists(void)
Yasunori Goto68113782006-06-23 02:03:11 -07002341{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002342 set_zonelist_order();
2343
Yasunori Goto68113782006-06-23 02:03:11 -07002344 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07002345 __build_all_zonelists(NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07002346 cpuset_init_current_mems_allowed();
2347 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02002348 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07002349 of zonelist */
2350 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
2351 /* cpuset refresh routine should be here */
2352 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07002353 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002354 /*
2355 * Disable grouping by mobility if the number of pages in the
2356 * system is too low to allow the mechanism to work. It would be
2357 * more accurate, but expensive to check per-zone. This check is
2358 * made on memory-hotadd so a system can start with mobility
2359 * disabled and enable it later
2360 */
Mel Gormand9c23402007-10-16 01:26:01 -07002361 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002362 page_group_by_mobility_disabled = 1;
2363 else
2364 page_group_by_mobility_disabled = 0;
2365
2366 printk("Built %i zonelists in %s order, mobility grouping %s. "
2367 "Total pages: %ld\n",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002368 num_online_nodes(),
2369 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002370 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002371 vm_total_pages);
2372#ifdef CONFIG_NUMA
2373 printk("Policy zone: %s\n", zone_names[policy_zone]);
2374#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375}
2376
2377/*
2378 * Helper functions to size the waitqueue hash table.
2379 * Essentially these want to choose hash table sizes sufficiently
2380 * large so that collisions trying to wait on pages are rare.
2381 * But in fact, the number of active page waitqueues on typical
2382 * systems is ridiculously low, less than 200. So this is even
2383 * conservative, even though it seems large.
2384 *
2385 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2386 * waitqueues, i.e. the size of the waitq table given the number of pages.
2387 */
2388#define PAGES_PER_WAITQUEUE 256
2389
Yasunori Gotocca448f2006-06-23 02:03:10 -07002390#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07002391static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392{
2393 unsigned long size = 1;
2394
2395 pages /= PAGES_PER_WAITQUEUE;
2396
2397 while (size < pages)
2398 size <<= 1;
2399
2400 /*
2401 * Once we have dozens or even hundreds of threads sleeping
2402 * on IO we've got bigger problems than wait queue collision.
2403 * Limit the size of the wait table to a reasonable size.
2404 */
2405 size = min(size, 4096UL);
2406
2407 return max(size, 4UL);
2408}
Yasunori Gotocca448f2006-06-23 02:03:10 -07002409#else
2410/*
2411 * A zone's size might be changed by hot-add, so it is not possible to determine
2412 * a suitable size for its wait_table. So we use the maximum size now.
2413 *
2414 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2415 *
2416 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2417 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2418 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2419 *
2420 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2421 * or more by the traditional way. (See above). It equals:
2422 *
2423 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2424 * ia64(16K page size) : = ( 8G + 4M)byte.
2425 * powerpc (64K page size) : = (32G +16M)byte.
2426 */
2427static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2428{
2429 return 4096UL;
2430}
2431#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432
2433/*
2434 * This is an integer logarithm so that shifts can be used later
2435 * to extract the more random high bits from the multiplicative
2436 * hash function before the remainder is taken.
2437 */
2438static inline unsigned long wait_table_bits(unsigned long size)
2439{
2440 return ffz(~size);
2441}
2442
2443#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2444
Mel Gorman56fd56b2007-10-16 01:25:58 -07002445/*
Mel Gormand9c23402007-10-16 01:26:01 -07002446 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman56fd56b2007-10-16 01:25:58 -07002447 * of blocks reserved is based on zone->pages_min. The memory within the
2448 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2449 * higher will lead to a bigger reserve which will get freed as contiguous
2450 * blocks as reclaim kicks in
2451 */
2452static void setup_zone_migrate_reserve(struct zone *zone)
2453{
2454 unsigned long start_pfn, pfn, end_pfn;
2455 struct page *page;
2456 unsigned long reserve, block_migratetype;
2457
2458 /* Get the start pfn, end pfn and the number of blocks to reserve */
2459 start_pfn = zone->zone_start_pfn;
2460 end_pfn = start_pfn + zone->spanned_pages;
Mel Gormand9c23402007-10-16 01:26:01 -07002461 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2462 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002463
Mel Gormand9c23402007-10-16 01:26:01 -07002464 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07002465 if (!pfn_valid(pfn))
2466 continue;
2467 page = pfn_to_page(pfn);
2468
2469 /* Blocks with reserved pages will never free, skip them. */
2470 if (PageReserved(page))
2471 continue;
2472
2473 block_migratetype = get_pageblock_migratetype(page);
2474
2475 /* If this block is reserved, account for it */
2476 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2477 reserve--;
2478 continue;
2479 }
2480
2481 /* Suitable for reserving if this block is movable */
2482 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2483 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2484 move_freepages_block(zone, page, MIGRATE_RESERVE);
2485 reserve--;
2486 continue;
2487 }
2488
2489 /*
2490 * If the reserve is met and this is a previous reserved block,
2491 * take it back
2492 */
2493 if (block_migratetype == MIGRATE_RESERVE) {
2494 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2495 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2496 }
2497 }
2498}
Mel Gormanac0e5b72007-10-16 01:25:58 -07002499
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500/*
2501 * Initially all pages are reserved - free ones are freed
2502 * up by free_all_bootmem() once the early boot process is
2503 * done. Non-atomic initialization, single-pass.
2504 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002505void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa02007-01-10 23:15:30 -08002506 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07002509 unsigned long end_pfn = start_pfn + size;
2510 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511
Greg Ungerercbe8dd42006-01-12 01:05:24 -08002512 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa02007-01-10 23:15:30 -08002513 /*
2514 * There can be holes in boot-time mem_map[]s
2515 * handed to this function. They do not
2516 * exist on hotplugged memory.
2517 */
2518 if (context == MEMMAP_EARLY) {
2519 if (!early_pfn_valid(pfn))
2520 continue;
2521 if (!early_pfn_in_nid(pfn, nid))
2522 continue;
2523 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002524 page = pfn_to_page(pfn);
2525 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08002526 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 reset_page_mapcount(page);
2528 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002529
2530 /*
2531 * Mark the block movable so that blocks are reserved for
2532 * movable at startup. This will force kernel allocations
2533 * to reserve their blocks rather than leaking throughout
2534 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07002535 * kernel allocations are made. Later some blocks near
2536 * the start are marked MIGRATE_RESERVE by
2537 * setup_zone_migrate_reserve()
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002538 */
Mel Gormand9c23402007-10-16 01:26:01 -07002539 if ((pfn & (pageblock_nr_pages-1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07002540 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002541
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 INIT_LIST_HEAD(&page->lru);
2543#ifdef WANT_PAGE_VIRTUAL
2544 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2545 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07002546 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 }
2549}
2550
Andi Kleen1e548de2008-02-04 22:29:26 -08002551static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002553 int order, t;
2554 for_each_migratetype_order(order, t) {
2555 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 zone->free_area[order].nr_free = 0;
2557 }
2558}
2559
2560#ifndef __HAVE_ARCH_MEMMAP_INIT
2561#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa02007-01-10 23:15:30 -08002562 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563#endif
2564
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08002565static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002566{
2567 int batch;
2568
2569 /*
2570 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07002571 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002572 *
2573 * OK, so we don't know how big the cache is. So guess.
2574 */
2575 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07002576 if (batch * PAGE_SIZE > 512 * 1024)
2577 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002578 batch /= 4; /* We effectively *= 4 below */
2579 if (batch < 1)
2580 batch = 1;
2581
2582 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002583 * Clamp the batch to a 2^n - 1 value. Having a power
2584 * of 2 value was found to be more likely to have
2585 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002586 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002587 * For example if 2 tasks are alternately allocating
2588 * batches of pages, one task can end up with a lot
2589 * of pages of one half of the possible page colors
2590 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002591 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002592 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07002593
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002594 return batch;
2595}
2596
Christoph Lameter2caaad42005-06-21 17:15:00 -07002597inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2598{
2599 struct per_cpu_pages *pcp;
2600
Magnus Damm1c6fe942005-10-26 01:58:59 -07002601 memset(p, 0, sizeof(*p));
2602
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002603 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002604 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002605 pcp->high = 6 * batch;
2606 pcp->batch = max(1UL, 1 * batch);
2607 INIT_LIST_HEAD(&pcp->list);
Christoph Lameter2caaad42005-06-21 17:15:00 -07002608}
2609
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002610/*
2611 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2612 * to the value high for the pageset p.
2613 */
2614
2615static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2616 unsigned long high)
2617{
2618 struct per_cpu_pages *pcp;
2619
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002620 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002621 pcp->high = high;
2622 pcp->batch = max(1UL, high/4);
2623 if ((high/4) > (PAGE_SHIFT * 8))
2624 pcp->batch = PAGE_SHIFT * 8;
2625}
2626
2627
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002628#ifdef CONFIG_NUMA
2629/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07002630 * Boot pageset table. One per cpu which is going to be used for all
2631 * zones and all nodes. The parameters will be set in such a way
2632 * that an item put on a list will immediately be handed over to
2633 * the buddy list. This is safe since pageset manipulation is done
2634 * with interrupts disabled.
2635 *
2636 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07002637 *
2638 * The boot_pagesets must be kept even after bootup is complete for
2639 * unused processors and/or zones. They do play a role for bootstrapping
2640 * hotplugged processors.
2641 *
2642 * zoneinfo_show() and maybe other functions do
2643 * not check if the processor is online before following the pageset pointer.
2644 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07002645 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08002646static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07002647
2648/*
2649 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002650 * per cpu pageset array in struct zone.
2651 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08002652static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002653{
2654 struct zone *zone, *dzone;
Christoph Lameter37c07082007-10-16 01:25:36 -07002655 int node = cpu_to_node(cpu);
2656
2657 node_set_state(node, N_CPU); /* this node has a cpu */
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002658
2659 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002660
Christoph Lameter66a55032006-09-27 01:50:09 -07002661 if (!populated_zone(zone))
2662 continue;
2663
Nick Piggin23316bc2006-01-08 01:00:41 -08002664 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lameter37c07082007-10-16 01:25:36 -07002665 GFP_KERNEL, node);
Nick Piggin23316bc2006-01-08 01:00:41 -08002666 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002667 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002668
Nick Piggin23316bc2006-01-08 01:00:41 -08002669 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002670
2671 if (percpu_pagelist_fraction)
2672 setup_pagelist_highmark(zone_pcp(zone, cpu),
2673 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002674 }
2675
2676 return 0;
2677bad:
2678 for_each_zone(dzone) {
Andrew Morton64191682007-08-30 23:56:17 -07002679 if (!populated_zone(dzone))
2680 continue;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002681 if (dzone == zone)
2682 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08002683 kfree(zone_pcp(dzone, cpu));
2684 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002685 }
2686 return -ENOMEM;
2687}
2688
2689static inline void free_zone_pagesets(int cpu)
2690{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002691 struct zone *zone;
2692
2693 for_each_zone(zone) {
2694 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2695
David Rientjesf3ef9ea2006-09-25 16:24:57 -07002696 /* Free per_cpu_pageset if it is slab allocated */
2697 if (pset != &boot_pageset[cpu])
2698 kfree(pset);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002699 zone_pcp(zone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002700 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002701}
2702
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07002703static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002704 unsigned long action,
2705 void *hcpu)
2706{
2707 int cpu = (long)hcpu;
2708 int ret = NOTIFY_OK;
2709
2710 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08002711 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002712 case CPU_UP_PREPARE_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002713 if (process_zones(cpu))
2714 ret = NOTIFY_BAD;
2715 break;
2716 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002717 case CPU_UP_CANCELED_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002718 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002719 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002720 free_zone_pagesets(cpu);
2721 break;
2722 default:
2723 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002724 }
2725 return ret;
2726}
2727
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07002728static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002729 { &pageset_cpuup_callback, NULL, 0 };
2730
Al Viro78d99552005-12-15 09:18:25 +00002731void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002732{
2733 int err;
2734
2735 /* Initialize per_cpu_pageset for cpu 0.
2736 * A cpuup callback will do this for every cpu
2737 * as it comes online
2738 */
2739 err = process_zones(smp_processor_id());
2740 BUG_ON(err);
2741 register_cpu_notifier(&pageset_notifier);
2742}
2743
2744#endif
2745
Sam Ravnborg577a32f2007-05-17 23:29:25 +02002746static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07002747int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07002748{
2749 int i;
2750 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002751 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07002752
2753 /*
2754 * The per-page waitqueue mechanism uses hashed waitqueues
2755 * per zone.
2756 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07002757 zone->wait_table_hash_nr_entries =
2758 wait_table_hash_nr_entries(zone_size_pages);
2759 zone->wait_table_bits =
2760 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002761 alloc_size = zone->wait_table_hash_nr_entries
2762 * sizeof(wait_queue_head_t);
2763
2764 if (system_state == SYSTEM_BOOTING) {
2765 zone->wait_table = (wait_queue_head_t *)
2766 alloc_bootmem_node(pgdat, alloc_size);
2767 } else {
2768 /*
2769 * This case means that a zone whose size was 0 gets new memory
2770 * via memory hot-add.
2771 * But it may be the case that a new node was hot-added. In
2772 * this case vmalloc() will not be able to use this new node's
2773 * memory - this wait_table must be initialized to use this new
2774 * node itself as well.
2775 * To use this new node's memory, further consideration will be
2776 * necessary.
2777 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07002778 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002779 }
2780 if (!zone->wait_table)
2781 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07002782
Yasunori Goto02b694d2006-06-23 02:03:08 -07002783 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07002784 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002785
2786 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002787}
2788
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002789static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07002790{
2791 int cpu;
2792 unsigned long batch = zone_batchsize(zone);
2793
2794 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2795#ifdef CONFIG_NUMA
2796 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002797 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002798 setup_pageset(&boot_pageset[cpu],0);
2799#else
2800 setup_pageset(zone_pcp(zone,cpu), batch);
2801#endif
2802 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08002803 if (zone->present_pages)
2804 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2805 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07002806}
2807
Yasunori Goto718127c2006-06-23 02:03:10 -07002808__meminit int init_currently_empty_zone(struct zone *zone,
2809 unsigned long zone_start_pfn,
Dave Hansena2f3aa02007-01-10 23:15:30 -08002810 unsigned long size,
2811 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07002812{
2813 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002814 int ret;
2815 ret = zone_wait_table_init(zone, size);
2816 if (ret)
2817 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07002818 pgdat->nr_zones = zone_idx(zone) + 1;
2819
Dave Hansened8ece22005-10-29 18:16:50 -07002820 zone->zone_start_pfn = zone_start_pfn;
2821
2822 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2823
Andi Kleen1e548de2008-02-04 22:29:26 -08002824 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07002825
2826 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002827}
2828
Mel Gormanc7132162006-09-27 01:49:43 -07002829#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2830/*
2831 * Basic iterator support. Return the first range of PFNs for a node
2832 * Note: nid == MAX_NUMNODES returns first region regardless of node
2833 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002834static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07002835{
2836 int i;
2837
2838 for (i = 0; i < nr_nodemap_entries; i++)
2839 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2840 return i;
2841
2842 return -1;
2843}
2844
2845/*
2846 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02002847 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07002848 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002849static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07002850{
2851 for (index = index + 1; index < nr_nodemap_entries; index++)
2852 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2853 return index;
2854
2855 return -1;
2856}
2857
2858#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2859/*
2860 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2861 * Architectures may implement their own version but if add_active_range()
2862 * was used and there are no special requirements, this is a convenient
2863 * alternative
2864 */
Stephen Rothwell6f076f52007-05-10 03:15:27 -07002865int __meminit early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07002866{
2867 int i;
2868
2869 for (i = 0; i < nr_nodemap_entries; i++) {
2870 unsigned long start_pfn = early_node_map[i].start_pfn;
2871 unsigned long end_pfn = early_node_map[i].end_pfn;
2872
2873 if (start_pfn <= pfn && pfn < end_pfn)
2874 return early_node_map[i].nid;
2875 }
2876
2877 return 0;
2878}
2879#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2880
2881/* Basic iterator support to walk early_node_map[] */
2882#define for_each_active_range_index_in_nid(i, nid) \
2883 for (i = first_active_region_index_in_nid(nid); i != -1; \
2884 i = next_active_region_index_in_nid(i, nid))
2885
2886/**
2887 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002888 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2889 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07002890 *
2891 * If an architecture guarantees that all ranges registered with
2892 * add_active_ranges() contain no holes and may be freed, this
2893 * this function may be used instead of calling free_bootmem() manually.
2894 */
2895void __init free_bootmem_with_active_regions(int nid,
2896 unsigned long max_low_pfn)
2897{
2898 int i;
2899
2900 for_each_active_range_index_in_nid(i, nid) {
2901 unsigned long size_pages = 0;
2902 unsigned long end_pfn = early_node_map[i].end_pfn;
2903
2904 if (early_node_map[i].start_pfn >= max_low_pfn)
2905 continue;
2906
2907 if (end_pfn > max_low_pfn)
2908 end_pfn = max_low_pfn;
2909
2910 size_pages = end_pfn - early_node_map[i].start_pfn;
2911 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2912 PFN_PHYS(early_node_map[i].start_pfn),
2913 size_pages << PAGE_SHIFT);
2914 }
2915}
2916
2917/**
2918 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002919 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07002920 *
2921 * If an architecture guarantees that all ranges registered with
2922 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002923 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07002924 */
2925void __init sparse_memory_present_with_active_regions(int nid)
2926{
2927 int i;
2928
2929 for_each_active_range_index_in_nid(i, nid)
2930 memory_present(early_node_map[i].nid,
2931 early_node_map[i].start_pfn,
2932 early_node_map[i].end_pfn);
2933}
2934
2935/**
Mel Gormanfb014392006-09-27 01:49:59 -07002936 * push_node_boundaries - Push node boundaries to at least the requested boundary
2937 * @nid: The nid of the node to push the boundary for
2938 * @start_pfn: The start pfn of the node
2939 * @end_pfn: The end pfn of the node
2940 *
2941 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2942 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2943 * be hotplugged even though no physical memory exists. This function allows
2944 * an arch to push out the node boundaries so mem_map is allocated that can
2945 * be used later.
2946 */
2947#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2948void __init push_node_boundaries(unsigned int nid,
2949 unsigned long start_pfn, unsigned long end_pfn)
2950{
2951 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2952 nid, start_pfn, end_pfn);
2953
2954 /* Initialise the boundary for this node if necessary */
2955 if (node_boundary_end_pfn[nid] == 0)
2956 node_boundary_start_pfn[nid] = -1UL;
2957
2958 /* Update the boundaries */
2959 if (node_boundary_start_pfn[nid] > start_pfn)
2960 node_boundary_start_pfn[nid] = start_pfn;
2961 if (node_boundary_end_pfn[nid] < end_pfn)
2962 node_boundary_end_pfn[nid] = end_pfn;
2963}
2964
2965/* If necessary, push the node boundary out for reserve hotadd */
Jan Beulich98011f52007-07-15 23:38:17 -07002966static void __meminit account_node_boundary(unsigned int nid,
Mel Gormanfb014392006-09-27 01:49:59 -07002967 unsigned long *start_pfn, unsigned long *end_pfn)
2968{
2969 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2970 nid, *start_pfn, *end_pfn);
2971
2972 /* Return if boundary information has not been provided */
2973 if (node_boundary_end_pfn[nid] == 0)
2974 return;
2975
2976 /* Check the boundaries and update if necessary */
2977 if (node_boundary_start_pfn[nid] < *start_pfn)
2978 *start_pfn = node_boundary_start_pfn[nid];
2979 if (node_boundary_end_pfn[nid] > *end_pfn)
2980 *end_pfn = node_boundary_end_pfn[nid];
2981}
2982#else
2983void __init push_node_boundaries(unsigned int nid,
2984 unsigned long start_pfn, unsigned long end_pfn) {}
2985
Jan Beulich98011f52007-07-15 23:38:17 -07002986static void __meminit account_node_boundary(unsigned int nid,
Mel Gormanfb014392006-09-27 01:49:59 -07002987 unsigned long *start_pfn, unsigned long *end_pfn) {}
2988#endif
2989
2990
2991/**
Mel Gormanc7132162006-09-27 01:49:43 -07002992 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002993 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2994 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2995 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07002996 *
2997 * It returns the start and end page frame of a node based on information
2998 * provided by an arch calling add_active_range(). If called for a node
2999 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003000 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003001 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003002void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003003 unsigned long *start_pfn, unsigned long *end_pfn)
3004{
3005 int i;
3006 *start_pfn = -1UL;
3007 *end_pfn = 0;
3008
3009 for_each_active_range_index_in_nid(i, nid) {
3010 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3011 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3012 }
3013
Christoph Lameter633c0662007-10-16 01:25:37 -07003014 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003015 *start_pfn = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07003016
3017 /* Push the node boundaries out if requested */
3018 account_node_boundary(nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003019}
3020
3021/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003022 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3023 * assumption is made that zones within a node are ordered in monotonic
3024 * increasing memory addresses so that the "highest" populated zone is used
3025 */
3026void __init find_usable_zone_for_movable(void)
3027{
3028 int zone_index;
3029 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3030 if (zone_index == ZONE_MOVABLE)
3031 continue;
3032
3033 if (arch_zone_highest_possible_pfn[zone_index] >
3034 arch_zone_lowest_possible_pfn[zone_index])
3035 break;
3036 }
3037
3038 VM_BUG_ON(zone_index == -1);
3039 movable_zone = zone_index;
3040}
3041
3042/*
3043 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3044 * because it is sized independant of architecture. Unlike the other zones,
3045 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3046 * in each node depending on the size of each node and how evenly kernelcore
3047 * is distributed. This helper function adjusts the zone ranges
3048 * provided by the architecture for a given node by using the end of the
3049 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3050 * zones within a node are in order of monotonic increases memory addresses
3051 */
3052void __meminit adjust_zone_range_for_zone_movable(int nid,
3053 unsigned long zone_type,
3054 unsigned long node_start_pfn,
3055 unsigned long node_end_pfn,
3056 unsigned long *zone_start_pfn,
3057 unsigned long *zone_end_pfn)
3058{
3059 /* Only adjust if ZONE_MOVABLE is on this node */
3060 if (zone_movable_pfn[nid]) {
3061 /* Size ZONE_MOVABLE */
3062 if (zone_type == ZONE_MOVABLE) {
3063 *zone_start_pfn = zone_movable_pfn[nid];
3064 *zone_end_pfn = min(node_end_pfn,
3065 arch_zone_highest_possible_pfn[movable_zone]);
3066
3067 /* Adjust for ZONE_MOVABLE starting within this range */
3068 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3069 *zone_end_pfn > zone_movable_pfn[nid]) {
3070 *zone_end_pfn = zone_movable_pfn[nid];
3071
3072 /* Check if this whole range is within ZONE_MOVABLE */
3073 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3074 *zone_start_pfn = *zone_end_pfn;
3075 }
3076}
3077
3078/*
Mel Gormanc7132162006-09-27 01:49:43 -07003079 * Return the number of pages a zone spans in a node, including holes
3080 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3081 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003082static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003083 unsigned long zone_type,
3084 unsigned long *ignored)
3085{
3086 unsigned long node_start_pfn, node_end_pfn;
3087 unsigned long zone_start_pfn, zone_end_pfn;
3088
3089 /* Get the start and end of the node and zone */
3090 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3091 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3092 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003093 adjust_zone_range_for_zone_movable(nid, zone_type,
3094 node_start_pfn, node_end_pfn,
3095 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003096
3097 /* Check that this node has pages within the zone's required range */
3098 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3099 return 0;
3100
3101 /* Move the zone boundaries inside the node if necessary */
3102 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3103 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3104
3105 /* Return the spanned pages */
3106 return zone_end_pfn - zone_start_pfn;
3107}
3108
3109/*
3110 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003111 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003112 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003113unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003114 unsigned long range_start_pfn,
3115 unsigned long range_end_pfn)
3116{
3117 int i = 0;
3118 unsigned long prev_end_pfn = 0, hole_pages = 0;
3119 unsigned long start_pfn;
3120
3121 /* Find the end_pfn of the first active range of pfns in the node */
3122 i = first_active_region_index_in_nid(nid);
3123 if (i == -1)
3124 return 0;
3125
Mel Gormanb5445f92007-07-26 10:41:18 -07003126 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3127
Mel Gorman9c7cd682006-09-27 01:49:58 -07003128 /* Account for ranges before physical memory on this node */
3129 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003130 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003131
3132 /* Find all holes for the zone within the node */
3133 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3134
3135 /* No need to continue if prev_end_pfn is outside the zone */
3136 if (prev_end_pfn >= range_end_pfn)
3137 break;
3138
3139 /* Make sure the end of the zone is not within the hole */
3140 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3141 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3142
3143 /* Update the hole size cound and move on */
3144 if (start_pfn > range_start_pfn) {
3145 BUG_ON(prev_end_pfn > start_pfn);
3146 hole_pages += start_pfn - prev_end_pfn;
3147 }
3148 prev_end_pfn = early_node_map[i].end_pfn;
3149 }
3150
Mel Gorman9c7cd682006-09-27 01:49:58 -07003151 /* Account for ranges past physical memory on this node */
3152 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003153 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003154 max(range_start_pfn, prev_end_pfn);
3155
Mel Gormanc7132162006-09-27 01:49:43 -07003156 return hole_pages;
3157}
3158
3159/**
3160 * absent_pages_in_range - Return number of page frames in holes within a range
3161 * @start_pfn: The start PFN to start searching for holes
3162 * @end_pfn: The end PFN to stop searching for holes
3163 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003164 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003165 */
3166unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3167 unsigned long end_pfn)
3168{
3169 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3170}
3171
3172/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003173static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003174 unsigned long zone_type,
3175 unsigned long *ignored)
3176{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003177 unsigned long node_start_pfn, node_end_pfn;
3178 unsigned long zone_start_pfn, zone_end_pfn;
3179
3180 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3181 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3182 node_start_pfn);
3183 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3184 node_end_pfn);
3185
Mel Gorman2a1e2742007-07-17 04:03:12 -07003186 adjust_zone_range_for_zone_movable(nid, zone_type,
3187 node_start_pfn, node_end_pfn,
3188 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07003189 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003190}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003191
Mel Gormanc7132162006-09-27 01:49:43 -07003192#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07003193static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003194 unsigned long zone_type,
3195 unsigned long *zones_size)
3196{
3197 return zones_size[zone_type];
3198}
3199
Paul Mundt6ea6e682007-07-15 23:38:20 -07003200static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003201 unsigned long zone_type,
3202 unsigned long *zholes_size)
3203{
3204 if (!zholes_size)
3205 return 0;
3206
3207 return zholes_size[zone_type];
3208}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003209
Mel Gormanc7132162006-09-27 01:49:43 -07003210#endif
3211
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003212static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07003213 unsigned long *zones_size, unsigned long *zholes_size)
3214{
3215 unsigned long realtotalpages, totalpages = 0;
3216 enum zone_type i;
3217
3218 for (i = 0; i < MAX_NR_ZONES; i++)
3219 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3220 zones_size);
3221 pgdat->node_spanned_pages = totalpages;
3222
3223 realtotalpages = totalpages;
3224 for (i = 0; i < MAX_NR_ZONES; i++)
3225 realtotalpages -=
3226 zone_absent_pages_in_node(pgdat->node_id, i,
3227 zholes_size);
3228 pgdat->node_present_pages = realtotalpages;
3229 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3230 realtotalpages);
3231}
3232
Mel Gorman835c1342007-10-16 01:25:47 -07003233#ifndef CONFIG_SPARSEMEM
3234/*
3235 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07003236 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3237 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07003238 * round what is now in bits to nearest long in bits, then return it in
3239 * bytes.
3240 */
3241static unsigned long __init usemap_size(unsigned long zonesize)
3242{
3243 unsigned long usemapsize;
3244
Mel Gormand9c23402007-10-16 01:26:01 -07003245 usemapsize = roundup(zonesize, pageblock_nr_pages);
3246 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07003247 usemapsize *= NR_PAGEBLOCK_BITS;
3248 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3249
3250 return usemapsize / 8;
3251}
3252
3253static void __init setup_usemap(struct pglist_data *pgdat,
3254 struct zone *zone, unsigned long zonesize)
3255{
3256 unsigned long usemapsize = usemap_size(zonesize);
3257 zone->pageblock_flags = NULL;
3258 if (usemapsize) {
3259 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3260 memset(zone->pageblock_flags, 0, usemapsize);
3261 }
3262}
3263#else
3264static void inline setup_usemap(struct pglist_data *pgdat,
3265 struct zone *zone, unsigned long zonesize) {}
3266#endif /* CONFIG_SPARSEMEM */
3267
Mel Gormand9c23402007-10-16 01:26:01 -07003268#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08003269
3270/* Return a sensible default order for the pageblock size. */
3271static inline int pageblock_default_order(void)
3272{
3273 if (HPAGE_SHIFT > PAGE_SHIFT)
3274 return HUGETLB_PAGE_ORDER;
3275
3276 return MAX_ORDER-1;
3277}
3278
Mel Gormand9c23402007-10-16 01:26:01 -07003279/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3280static inline void __init set_pageblock_order(unsigned int order)
3281{
3282 /* Check that pageblock_nr_pages has not already been setup */
3283 if (pageblock_order)
3284 return;
3285
3286 /*
3287 * Assume the largest contiguous order of interest is a huge page.
3288 * This value may be variable depending on boot parameters on IA64
3289 */
3290 pageblock_order = order;
3291}
3292#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3293
Mel Gormanba72cb82007-11-28 16:21:13 -08003294/*
3295 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3296 * and pageblock_default_order() are unused as pageblock_order is set
3297 * at compile-time. See include/linux/pageblock-flags.h for the values of
3298 * pageblock_order based on the kernel config
3299 */
3300static inline int pageblock_default_order(unsigned int order)
3301{
3302 return MAX_ORDER-1;
3303}
Mel Gormand9c23402007-10-16 01:26:01 -07003304#define set_pageblock_order(x) do {} while (0)
3305
3306#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3307
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308/*
3309 * Set up the zone data structures:
3310 * - mark all pages reserved
3311 * - mark all memory queues empty
3312 * - clear the memory bitmaps
3313 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07003314static void __meminit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 unsigned long *zones_size, unsigned long *zholes_size)
3316{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07003317 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07003318 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07003320 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321
Dave Hansen208d54e2005-10-29 18:16:52 -07003322 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323 pgdat->nr_zones = 0;
3324 init_waitqueue_head(&pgdat->kswapd_wait);
3325 pgdat->kswapd_max_order = 0;
3326
3327 for (j = 0; j < MAX_NR_ZONES; j++) {
3328 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003329 unsigned long size, realsize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330
Mel Gormanc7132162006-09-27 01:49:43 -07003331 size = zone_spanned_pages_in_node(nid, j, zones_size);
3332 realsize = size - zone_absent_pages_in_node(nid, j,
3333 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334
Mel Gorman0e0b8642006-09-27 01:49:56 -07003335 /*
3336 * Adjust realsize so that it accounts for how much memory
3337 * is used by this zone for memmap. This affects the watermark
3338 * and per-cpu initialisations
3339 */
3340 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
3341 if (realsize >= memmap_pages) {
3342 realsize -= memmap_pages;
3343 printk(KERN_DEBUG
3344 " %s zone: %lu pages used for memmap\n",
3345 zone_names[j], memmap_pages);
3346 } else
3347 printk(KERN_WARNING
3348 " %s zone: %lu pages exceeds realsize %lu\n",
3349 zone_names[j], memmap_pages, realsize);
3350
Christoph Lameter62672762007-02-10 01:43:07 -08003351 /* Account for reserved pages */
3352 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07003353 realsize -= dma_reserve;
Christoph Lameter62672762007-02-10 01:43:07 -08003354 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3355 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003356 }
3357
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003358 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359 nr_kernel_pages += realsize;
3360 nr_all_pages += realsize;
3361
3362 zone->spanned_pages = size;
3363 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07003364#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07003365 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07003366 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07003367 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07003368 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07003369#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370 zone->name = zone_names[j];
3371 spin_lock_init(&zone->lock);
3372 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07003373 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375
Martin Bligh3bb1a852006-10-28 10:38:24 -07003376 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377
Dave Hansened8ece22005-10-29 18:16:50 -07003378 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379 INIT_LIST_HEAD(&zone->active_list);
3380 INIT_LIST_HEAD(&zone->inactive_list);
3381 zone->nr_scan_active = 0;
3382 zone->nr_scan_inactive = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07003383 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07003384 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385 if (!size)
3386 continue;
3387
Mel Gormanba72cb82007-11-28 16:21:13 -08003388 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07003389 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa02007-01-10 23:15:30 -08003390 ret = init_currently_empty_zone(zone, zone_start_pfn,
3391 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07003392 BUG_ON(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394 }
3395}
3396
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003397static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399 /* Skip empty nodes */
3400 if (!pgdat->node_spanned_pages)
3401 return;
3402
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003403#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07003404 /* ia64 gets its own node_mem_map, before this, without bootmem */
3405 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07003406 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003407 struct page *map;
3408
Bob Piccoe984bb42006-05-20 15:00:31 -07003409 /*
3410 * The zone's endpoints aren't required to be MAX_ORDER
3411 * aligned but the node_mem_map endpoints must be in order
3412 * for the buddy allocator to function correctly.
3413 */
3414 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3415 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3416 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3417 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07003418 map = alloc_remap(pgdat->node_id, size);
3419 if (!map)
3420 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07003421 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422 }
Roman Zippel12d810c2007-05-31 00:40:54 -07003423#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424 /*
3425 * With no DISCONTIG, the global mem_map is just set as node 0's
3426 */
Mel Gormanc7132162006-09-27 01:49:43 -07003427 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07003429#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3430 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08003431 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07003432#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3433 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003435#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436}
3437
Yasunori Goto86356ab2006-06-23 02:03:09 -07003438void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003439 unsigned long *zones_size, unsigned long node_start_pfn,
3440 unsigned long *zholes_size)
3441{
3442 pgdat->node_id = nid;
3443 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003444 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445
3446 alloc_node_mem_map(pgdat);
3447
3448 free_area_init_core(pgdat, zones_size, zholes_size);
3449}
3450
Mel Gormanc7132162006-09-27 01:49:43 -07003451#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07003452
3453#if MAX_NUMNODES > 1
3454/*
3455 * Figure out the number of possible node ids.
3456 */
3457static void __init setup_nr_node_ids(void)
3458{
3459 unsigned int node;
3460 unsigned int highest = 0;
3461
3462 for_each_node_mask(node, node_possible_map)
3463 highest = node;
3464 nr_node_ids = highest + 1;
3465}
3466#else
3467static inline void setup_nr_node_ids(void)
3468{
3469}
3470#endif
3471
Mel Gormanc7132162006-09-27 01:49:43 -07003472/**
3473 * add_active_range - Register a range of PFNs backed by physical memory
3474 * @nid: The node ID the range resides on
3475 * @start_pfn: The start PFN of the available physical memory
3476 * @end_pfn: The end PFN of the available physical memory
3477 *
3478 * These ranges are stored in an early_node_map[] and later used by
3479 * free_area_init_nodes() to calculate zone sizes and holes. If the
3480 * range spans a memory hole, it is up to the architecture to ensure
3481 * the memory is not freed by the bootmem allocator. If possible
3482 * the range being registered will be merged with existing ranges.
3483 */
3484void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3485 unsigned long end_pfn)
3486{
3487 int i;
3488
3489 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
3490 "%d entries of %d used\n",
3491 nid, start_pfn, end_pfn,
3492 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3493
3494 /* Merge with existing active regions if possible */
3495 for (i = 0; i < nr_nodemap_entries; i++) {
3496 if (early_node_map[i].nid != nid)
3497 continue;
3498
3499 /* Skip if an existing region covers this new one */
3500 if (start_pfn >= early_node_map[i].start_pfn &&
3501 end_pfn <= early_node_map[i].end_pfn)
3502 return;
3503
3504 /* Merge forward if suitable */
3505 if (start_pfn <= early_node_map[i].end_pfn &&
3506 end_pfn > early_node_map[i].end_pfn) {
3507 early_node_map[i].end_pfn = end_pfn;
3508 return;
3509 }
3510
3511 /* Merge backward if suitable */
3512 if (start_pfn < early_node_map[i].end_pfn &&
3513 end_pfn >= early_node_map[i].start_pfn) {
3514 early_node_map[i].start_pfn = start_pfn;
3515 return;
3516 }
3517 }
3518
3519 /* Check that early_node_map is large enough */
3520 if (i >= MAX_ACTIVE_REGIONS) {
3521 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3522 MAX_ACTIVE_REGIONS);
3523 return;
3524 }
3525
3526 early_node_map[i].nid = nid;
3527 early_node_map[i].start_pfn = start_pfn;
3528 early_node_map[i].end_pfn = end_pfn;
3529 nr_nodemap_entries = i + 1;
3530}
3531
3532/**
3533 * shrink_active_range - Shrink an existing registered range of PFNs
3534 * @nid: The node id the range is on that should be shrunk
3535 * @old_end_pfn: The old end PFN of the range
3536 * @new_end_pfn: The new PFN of the range
3537 *
3538 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3539 * The map is kept at the end physical page range that has already been
3540 * registered with add_active_range(). This function allows an arch to shrink
3541 * an existing registered range.
3542 */
3543void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
3544 unsigned long new_end_pfn)
3545{
3546 int i;
3547
3548 /* Find the old active region end and shrink */
3549 for_each_active_range_index_in_nid(i, nid)
3550 if (early_node_map[i].end_pfn == old_end_pfn) {
3551 early_node_map[i].end_pfn = new_end_pfn;
3552 break;
3553 }
3554}
3555
3556/**
3557 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003558 *
Mel Gormanc7132162006-09-27 01:49:43 -07003559 * During discovery, it may be found that a table like SRAT is invalid
3560 * and an alternative discovery method must be used. This function removes
3561 * all currently registered regions.
3562 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003563void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07003564{
3565 memset(early_node_map, 0, sizeof(early_node_map));
3566 nr_nodemap_entries = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07003567#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3568 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3569 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3570#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -07003571}
3572
3573/* Compare two active node_active_regions */
3574static int __init cmp_node_active_region(const void *a, const void *b)
3575{
3576 struct node_active_region *arange = (struct node_active_region *)a;
3577 struct node_active_region *brange = (struct node_active_region *)b;
3578
3579 /* Done this way to avoid overflows */
3580 if (arange->start_pfn > brange->start_pfn)
3581 return 1;
3582 if (arange->start_pfn < brange->start_pfn)
3583 return -1;
3584
3585 return 0;
3586}
3587
3588/* sort the node_map by start_pfn */
3589static void __init sort_node_map(void)
3590{
3591 sort(early_node_map, (size_t)nr_nodemap_entries,
3592 sizeof(struct node_active_region),
3593 cmp_node_active_region, NULL);
3594}
3595
Mel Gormana6af2bc2007-02-10 01:42:57 -08003596/* Find the lowest pfn for a node */
Mel Gormanc7132162006-09-27 01:49:43 -07003597unsigned long __init find_min_pfn_for_node(unsigned long nid)
3598{
3599 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08003600 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00003601
Mel Gormanc7132162006-09-27 01:49:43 -07003602 /* Assuming a sorted map, the first range found has the starting pfn */
3603 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08003604 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003605
Mel Gormana6af2bc2007-02-10 01:42:57 -08003606 if (min_pfn == ULONG_MAX) {
3607 printk(KERN_WARNING
3608 "Could not find start_pfn for node %lu\n", nid);
3609 return 0;
3610 }
3611
3612 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003613}
3614
3615/**
3616 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3617 *
3618 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003619 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07003620 */
3621unsigned long __init find_min_pfn_with_active_regions(void)
3622{
3623 return find_min_pfn_for_node(MAX_NUMNODES);
3624}
3625
3626/**
3627 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3628 *
3629 * It returns the maximum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003630 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07003631 */
3632unsigned long __init find_max_pfn_with_active_regions(void)
3633{
3634 int i;
3635 unsigned long max_pfn = 0;
3636
3637 for (i = 0; i < nr_nodemap_entries; i++)
3638 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
3639
3640 return max_pfn;
3641}
3642
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003643/*
3644 * early_calculate_totalpages()
3645 * Sum pages in active regions for movable zone.
3646 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3647 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07003648static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07003649{
3650 int i;
3651 unsigned long totalpages = 0;
3652
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003653 for (i = 0; i < nr_nodemap_entries; i++) {
3654 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07003655 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003656 totalpages += pages;
3657 if (pages)
3658 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3659 }
3660 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07003661}
3662
Mel Gorman2a1e2742007-07-17 04:03:12 -07003663/*
3664 * Find the PFN the Movable zone begins in each node. Kernel memory
3665 * is spread evenly between nodes as long as the nodes have enough
3666 * memory. When they don't, some nodes will have more kernelcore than
3667 * others
3668 */
3669void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3670{
3671 int i, nid;
3672 unsigned long usable_startpfn;
3673 unsigned long kernelcore_node, kernelcore_remaining;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003674 unsigned long totalpages = early_calculate_totalpages();
3675 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07003676
Mel Gorman7e63efe2007-07-17 04:03:15 -07003677 /*
3678 * If movablecore was specified, calculate what size of
3679 * kernelcore that corresponds so that memory usable for
3680 * any allocation type is evenly spread. If both kernelcore
3681 * and movablecore are specified, then the value of kernelcore
3682 * will be used for required_kernelcore if it's greater than
3683 * what movablecore would have allowed.
3684 */
3685 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07003686 unsigned long corepages;
3687
3688 /*
3689 * Round-up so that ZONE_MOVABLE is at least as large as what
3690 * was requested by the user
3691 */
3692 required_movablecore =
3693 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3694 corepages = totalpages - required_movablecore;
3695
3696 required_kernelcore = max(required_kernelcore, corepages);
3697 }
3698
Mel Gorman2a1e2742007-07-17 04:03:12 -07003699 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3700 if (!required_kernelcore)
3701 return;
3702
3703 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3704 find_usable_zone_for_movable();
3705 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3706
3707restart:
3708 /* Spread kernelcore memory as evenly as possible throughout nodes */
3709 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003710 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07003711 /*
3712 * Recalculate kernelcore_node if the division per node
3713 * now exceeds what is necessary to satisfy the requested
3714 * amount of memory for the kernel
3715 */
3716 if (required_kernelcore < kernelcore_node)
3717 kernelcore_node = required_kernelcore / usable_nodes;
3718
3719 /*
3720 * As the map is walked, we track how much memory is usable
3721 * by the kernel using kernelcore_remaining. When it is
3722 * 0, the rest of the node is usable by ZONE_MOVABLE
3723 */
3724 kernelcore_remaining = kernelcore_node;
3725
3726 /* Go through each range of PFNs within this node */
3727 for_each_active_range_index_in_nid(i, nid) {
3728 unsigned long start_pfn, end_pfn;
3729 unsigned long size_pages;
3730
3731 start_pfn = max(early_node_map[i].start_pfn,
3732 zone_movable_pfn[nid]);
3733 end_pfn = early_node_map[i].end_pfn;
3734 if (start_pfn >= end_pfn)
3735 continue;
3736
3737 /* Account for what is only usable for kernelcore */
3738 if (start_pfn < usable_startpfn) {
3739 unsigned long kernel_pages;
3740 kernel_pages = min(end_pfn, usable_startpfn)
3741 - start_pfn;
3742
3743 kernelcore_remaining -= min(kernel_pages,
3744 kernelcore_remaining);
3745 required_kernelcore -= min(kernel_pages,
3746 required_kernelcore);
3747
3748 /* Continue if range is now fully accounted */
3749 if (end_pfn <= usable_startpfn) {
3750
3751 /*
3752 * Push zone_movable_pfn to the end so
3753 * that if we have to rebalance
3754 * kernelcore across nodes, we will
3755 * not double account here
3756 */
3757 zone_movable_pfn[nid] = end_pfn;
3758 continue;
3759 }
3760 start_pfn = usable_startpfn;
3761 }
3762
3763 /*
3764 * The usable PFN range for ZONE_MOVABLE is from
3765 * start_pfn->end_pfn. Calculate size_pages as the
3766 * number of pages used as kernelcore
3767 */
3768 size_pages = end_pfn - start_pfn;
3769 if (size_pages > kernelcore_remaining)
3770 size_pages = kernelcore_remaining;
3771 zone_movable_pfn[nid] = start_pfn + size_pages;
3772
3773 /*
3774 * Some kernelcore has been met, update counts and
3775 * break if the kernelcore for this node has been
3776 * satisified
3777 */
3778 required_kernelcore -= min(required_kernelcore,
3779 size_pages);
3780 kernelcore_remaining -= size_pages;
3781 if (!kernelcore_remaining)
3782 break;
3783 }
3784 }
3785
3786 /*
3787 * If there is still required_kernelcore, we do another pass with one
3788 * less node in the count. This will push zone_movable_pfn[nid] further
3789 * along on the nodes that still have memory until kernelcore is
3790 * satisified
3791 */
3792 usable_nodes--;
3793 if (usable_nodes && required_kernelcore > usable_nodes)
3794 goto restart;
3795
3796 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3797 for (nid = 0; nid < MAX_NUMNODES; nid++)
3798 zone_movable_pfn[nid] =
3799 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3800}
3801
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003802/* Any regular memory on that node ? */
3803static void check_for_regular_memory(pg_data_t *pgdat)
3804{
3805#ifdef CONFIG_HIGHMEM
3806 enum zone_type zone_type;
3807
3808 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3809 struct zone *zone = &pgdat->node_zones[zone_type];
3810 if (zone->present_pages)
3811 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3812 }
3813#endif
3814}
3815
Mel Gormanc7132162006-09-27 01:49:43 -07003816/**
3817 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003818 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07003819 *
3820 * This will call free_area_init_node() for each active node in the system.
3821 * Using the page ranges provided by add_active_range(), the size of each
3822 * zone in each node and their holes is calculated. If the maximum PFN
3823 * between two adjacent zones match, it is assumed that the zone is empty.
3824 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3825 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3826 * starts where the previous one ended. For example, ZONE_DMA32 starts
3827 * at arch_max_dma_pfn.
3828 */
3829void __init free_area_init_nodes(unsigned long *max_zone_pfn)
3830{
3831 unsigned long nid;
3832 enum zone_type i;
3833
Mel Gormana6af2bc2007-02-10 01:42:57 -08003834 /* Sort early_node_map as initialisation assumes it is sorted */
3835 sort_node_map();
3836
Mel Gormanc7132162006-09-27 01:49:43 -07003837 /* Record where the zone boundaries are */
3838 memset(arch_zone_lowest_possible_pfn, 0,
3839 sizeof(arch_zone_lowest_possible_pfn));
3840 memset(arch_zone_highest_possible_pfn, 0,
3841 sizeof(arch_zone_highest_possible_pfn));
3842 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
3843 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
3844 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07003845 if (i == ZONE_MOVABLE)
3846 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07003847 arch_zone_lowest_possible_pfn[i] =
3848 arch_zone_highest_possible_pfn[i-1];
3849 arch_zone_highest_possible_pfn[i] =
3850 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
3851 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07003852 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
3853 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
3854
3855 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3856 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
3857 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003858
Mel Gormanc7132162006-09-27 01:49:43 -07003859 /* Print out the zone ranges */
3860 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07003861 for (i = 0; i < MAX_NR_ZONES; i++) {
3862 if (i == ZONE_MOVABLE)
3863 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07003864 printk(" %-8s %8lu -> %8lu\n",
3865 zone_names[i],
3866 arch_zone_lowest_possible_pfn[i],
3867 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07003868 }
3869
3870 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3871 printk("Movable zone start PFN for each node\n");
3872 for (i = 0; i < MAX_NUMNODES; i++) {
3873 if (zone_movable_pfn[i])
3874 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
3875 }
Mel Gormanc7132162006-09-27 01:49:43 -07003876
3877 /* Print out the early_node_map[] */
3878 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
3879 for (i = 0; i < nr_nodemap_entries; i++)
3880 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
3881 early_node_map[i].start_pfn,
3882 early_node_map[i].end_pfn);
3883
3884 /* Initialise every node */
Christoph Lameter8ef82862007-02-20 13:57:52 -08003885 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07003886 for_each_online_node(nid) {
3887 pg_data_t *pgdat = NODE_DATA(nid);
3888 free_area_init_node(nid, pgdat, NULL,
3889 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003890
3891 /* Any memory on that node */
3892 if (pgdat->node_present_pages)
3893 node_set_state(nid, N_HIGH_MEMORY);
3894 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07003895 }
3896}
Mel Gorman2a1e2742007-07-17 04:03:12 -07003897
Mel Gorman7e63efe2007-07-17 04:03:15 -07003898static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003899{
3900 unsigned long long coremem;
3901 if (!p)
3902 return -EINVAL;
3903
3904 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07003905 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07003906
Mel Gorman7e63efe2007-07-17 04:03:15 -07003907 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07003908 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
3909
3910 return 0;
3911}
Mel Gormaned7ed362007-07-17 04:03:14 -07003912
Mel Gorman7e63efe2007-07-17 04:03:15 -07003913/*
3914 * kernelcore=size sets the amount of memory for use for allocations that
3915 * cannot be reclaimed or migrated.
3916 */
3917static int __init cmdline_parse_kernelcore(char *p)
3918{
3919 return cmdline_parse_core(p, &required_kernelcore);
3920}
3921
3922/*
3923 * movablecore=size sets the amount of memory for use for allocations that
3924 * can be reclaimed or migrated.
3925 */
3926static int __init cmdline_parse_movablecore(char *p)
3927{
3928 return cmdline_parse_core(p, &required_movablecore);
3929}
3930
Mel Gormaned7ed362007-07-17 04:03:14 -07003931early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07003932early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07003933
Mel Gormanc7132162006-09-27 01:49:43 -07003934#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3935
Mel Gorman0e0b8642006-09-27 01:49:56 -07003936/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003937 * set_dma_reserve - set the specified number of pages reserved in the first zone
3938 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07003939 *
3940 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3941 * In the DMA zone, a significant percentage may be consumed by kernel image
3942 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003943 * function may optionally be used to account for unfreeable pages in the
3944 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3945 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07003946 */
3947void __init set_dma_reserve(unsigned long new_dma_reserve)
3948{
3949 dma_reserve = new_dma_reserve;
3950}
3951
Dave Hansen93b75042005-06-23 00:07:47 -07003952#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003953static bootmem_data_t contig_bootmem_data;
3954struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
3955
3956EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07003957#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003958
3959void __init free_area_init(unsigned long *zones_size)
3960{
Dave Hansen93b75042005-06-23 00:07:47 -07003961 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003962 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
3963}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003964
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965static int page_alloc_cpu_notify(struct notifier_block *self,
3966 unsigned long action, void *hcpu)
3967{
3968 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003969
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003970 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08003971 drain_pages(cpu);
3972
3973 /*
3974 * Spill the event counters of the dead processor
3975 * into the current processors event counters.
3976 * This artificially elevates the count of the current
3977 * processor.
3978 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07003979 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08003980
3981 /*
3982 * Zero the differential counters of the dead processor
3983 * so that the vm statistics are consistent.
3984 *
3985 * This is only okay since the processor is dead and cannot
3986 * race with what we are doing.
3987 */
Christoph Lameter2244b952006-06-30 01:55:33 -07003988 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989 }
3990 return NOTIFY_OK;
3991}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003992
3993void __init page_alloc_init(void)
3994{
3995 hotcpu_notifier(page_alloc_cpu_notify, 0);
3996}
3997
3998/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003999 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4000 * or min_free_kbytes changes.
4001 */
4002static void calculate_totalreserve_pages(void)
4003{
4004 struct pglist_data *pgdat;
4005 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004006 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004007
4008 for_each_online_pgdat(pgdat) {
4009 for (i = 0; i < MAX_NR_ZONES; i++) {
4010 struct zone *zone = pgdat->node_zones + i;
4011 unsigned long max = 0;
4012
4013 /* Find valid and maximum lowmem_reserve in the zone */
4014 for (j = i; j < MAX_NR_ZONES; j++) {
4015 if (zone->lowmem_reserve[j] > max)
4016 max = zone->lowmem_reserve[j];
4017 }
4018
4019 /* we treat pages_high as reserved pages. */
4020 max += zone->pages_high;
4021
4022 if (max > zone->present_pages)
4023 max = zone->present_pages;
4024 reserve_pages += max;
4025 }
4026 }
4027 totalreserve_pages = reserve_pages;
4028}
4029
4030/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004031 * setup_per_zone_lowmem_reserve - called whenever
4032 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4033 * has a correct pages reserved value, so an adequate number of
4034 * pages are left in the zone after a successful __alloc_pages().
4035 */
4036static void setup_per_zone_lowmem_reserve(void)
4037{
4038 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004039 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004040
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004041 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042 for (j = 0; j < MAX_NR_ZONES; j++) {
4043 struct zone *zone = pgdat->node_zones + j;
4044 unsigned long present_pages = zone->present_pages;
4045
4046 zone->lowmem_reserve[j] = 0;
4047
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004048 idx = j;
4049 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004050 struct zone *lower_zone;
4051
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004052 idx--;
4053
Linus Torvalds1da177e2005-04-16 15:20:36 -07004054 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4055 sysctl_lowmem_reserve_ratio[idx] = 1;
4056
4057 lower_zone = pgdat->node_zones + idx;
4058 lower_zone->lowmem_reserve[j] = present_pages /
4059 sysctl_lowmem_reserve_ratio[idx];
4060 present_pages += lower_zone->present_pages;
4061 }
4062 }
4063 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004064
4065 /* update totalreserve_pages */
4066 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067}
4068
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004069/**
4070 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4071 *
4072 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4073 * with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004074 */
Dave Hansen3947be12005-10-29 18:16:54 -07004075void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004076{
4077 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4078 unsigned long lowmem_pages = 0;
4079 struct zone *zone;
4080 unsigned long flags;
4081
4082 /* Calculate total number of !ZONE_HIGHMEM pages */
4083 for_each_zone(zone) {
4084 if (!is_highmem(zone))
4085 lowmem_pages += zone->present_pages;
4086 }
4087
4088 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004089 u64 tmp;
4090
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004092 tmp = (u64)pages_min * zone->present_pages;
4093 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094 if (is_highmem(zone)) {
4095 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004096 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4097 * need highmem pages, so cap pages_min to a small
4098 * value here.
4099 *
4100 * The (pages_high-pages_low) and (pages_low-pages_min)
4101 * deltas controls asynch page reclaim, and so should
4102 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004103 */
4104 int min_pages;
4105
4106 min_pages = zone->present_pages / 1024;
4107 if (min_pages < SWAP_CLUSTER_MAX)
4108 min_pages = SWAP_CLUSTER_MAX;
4109 if (min_pages > 128)
4110 min_pages = 128;
4111 zone->pages_min = min_pages;
4112 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004113 /*
4114 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004115 * proportionate to the zone's size.
4116 */
Nick Piggin669ed172005-11-13 16:06:45 -08004117 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004118 }
4119
Andrew Mortonac924c62006-05-15 09:43:59 -07004120 zone->pages_low = zone->pages_min + (tmp >> 2);
4121 zone->pages_high = zone->pages_min + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004122 setup_zone_migrate_reserve(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004123 spin_unlock_irqrestore(&zone->lru_lock, flags);
4124 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004125
4126 /* update totalreserve_pages */
4127 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004128}
4129
4130/*
4131 * Initialise min_free_kbytes.
4132 *
4133 * For small machines we want it small (128k min). For large machines
4134 * we want it large (64MB max). But it is not linear, because network
4135 * bandwidth does not increase linearly with machine size. We use
4136 *
4137 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4138 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4139 *
4140 * which yields
4141 *
4142 * 16MB: 512k
4143 * 32MB: 724k
4144 * 64MB: 1024k
4145 * 128MB: 1448k
4146 * 256MB: 2048k
4147 * 512MB: 2896k
4148 * 1024MB: 4096k
4149 * 2048MB: 5792k
4150 * 4096MB: 8192k
4151 * 8192MB: 11584k
4152 * 16384MB: 16384k
4153 */
4154static int __init init_per_zone_pages_min(void)
4155{
4156 unsigned long lowmem_kbytes;
4157
4158 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4159
4160 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4161 if (min_free_kbytes < 128)
4162 min_free_kbytes = 128;
4163 if (min_free_kbytes > 65536)
4164 min_free_kbytes = 65536;
4165 setup_per_zone_pages_min();
4166 setup_per_zone_lowmem_reserve();
4167 return 0;
4168}
4169module_init(init_per_zone_pages_min)
4170
4171/*
4172 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4173 * that we can call two helper functions whenever min_free_kbytes
4174 * changes.
4175 */
4176int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4177 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4178{
4179 proc_dointvec(table, write, file, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07004180 if (write)
4181 setup_per_zone_pages_min();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004182 return 0;
4183}
4184
Christoph Lameter96146342006-07-03 00:24:13 -07004185#ifdef CONFIG_NUMA
4186int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4187 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4188{
4189 struct zone *zone;
4190 int rc;
4191
4192 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4193 if (rc)
4194 return rc;
4195
4196 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07004197 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07004198 sysctl_min_unmapped_ratio) / 100;
4199 return 0;
4200}
Christoph Lameter0ff38492006-09-25 23:31:52 -07004201
4202int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4203 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4204{
4205 struct zone *zone;
4206 int rc;
4207
4208 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4209 if (rc)
4210 return rc;
4211
4212 for_each_zone(zone)
4213 zone->min_slab_pages = (zone->present_pages *
4214 sysctl_min_slab_ratio) / 100;
4215 return 0;
4216}
Christoph Lameter96146342006-07-03 00:24:13 -07004217#endif
4218
Linus Torvalds1da177e2005-04-16 15:20:36 -07004219/*
4220 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4221 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4222 * whenever sysctl_lowmem_reserve_ratio changes.
4223 *
4224 * The reserve ratio obviously has absolutely no relation with the
4225 * pages_min watermarks. The lowmem reserve ratio can only make sense
4226 * if in function of the boot time zone sizes.
4227 */
4228int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4229 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4230{
4231 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4232 setup_per_zone_lowmem_reserve();
4233 return 0;
4234}
4235
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004236/*
4237 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4238 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4239 * can have before it gets flushed back to buddy allocator.
4240 */
4241
4242int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4243 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4244{
4245 struct zone *zone;
4246 unsigned int cpu;
4247 int ret;
4248
4249 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4250 if (!write || (ret == -EINVAL))
4251 return ret;
4252 for_each_zone(zone) {
4253 for_each_online_cpu(cpu) {
4254 unsigned long high;
4255 high = zone->present_pages / percpu_pagelist_fraction;
4256 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4257 }
4258 }
4259 return 0;
4260}
4261
David S. Millerf034b5d2006-08-24 03:08:07 -07004262int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004263
4264#ifdef CONFIG_NUMA
4265static int __init set_hashdist(char *str)
4266{
4267 if (!str)
4268 return 0;
4269 hashdist = simple_strtoul(str, &str, 0);
4270 return 1;
4271}
4272__setup("hashdist=", set_hashdist);
4273#endif
4274
4275/*
4276 * allocate a large system hash table from bootmem
4277 * - it is assumed that the hash table must contain an exact power-of-2
4278 * quantity of entries
4279 * - limit is the number of hash buckets, not the total allocation size
4280 */
4281void *__init alloc_large_system_hash(const char *tablename,
4282 unsigned long bucketsize,
4283 unsigned long numentries,
4284 int scale,
4285 int flags,
4286 unsigned int *_hash_shift,
4287 unsigned int *_hash_mask,
4288 unsigned long limit)
4289{
4290 unsigned long long max = limit;
4291 unsigned long log2qty, size;
4292 void *table = NULL;
4293
4294 /* allow the kernel cmdline to have a say */
4295 if (!numentries) {
4296 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08004297 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004298 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4299 numentries >>= 20 - PAGE_SHIFT;
4300 numentries <<= 20 - PAGE_SHIFT;
4301
4302 /* limit to 1 bucket per 2^scale bytes of low memory */
4303 if (scale > PAGE_SHIFT)
4304 numentries >>= (scale - PAGE_SHIFT);
4305 else
4306 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08004307
4308 /* Make sure we've got at least a 0-order allocation.. */
4309 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4310 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004311 }
John Hawkes6e692ed2006-03-25 03:08:02 -08004312 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004313
4314 /* limit allocation size to 1/16 total memory by default */
4315 if (max == 0) {
4316 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4317 do_div(max, bucketsize);
4318 }
4319
4320 if (numentries > max)
4321 numentries = max;
4322
David Howellsf0d1b0b2006-12-08 02:37:49 -08004323 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004324
4325 do {
4326 size = bucketsize << log2qty;
4327 if (flags & HASH_EARLY)
4328 table = alloc_bootmem(size);
4329 else if (hashdist)
4330 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4331 else {
4332 unsigned long order;
4333 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
4334 ;
4335 table = (void*) __get_free_pages(GFP_ATOMIC, order);
Eric Dumazet1037b832007-07-15 23:38:05 -07004336 /*
4337 * If bucketsize is not a power-of-two, we may free
4338 * some pages at the end of hash table.
4339 */
4340 if (table) {
4341 unsigned long alloc_end = (unsigned long)table +
4342 (PAGE_SIZE << order);
4343 unsigned long used = (unsigned long)table +
4344 PAGE_ALIGN(size);
4345 split_page(virt_to_page(table), order);
4346 while (used < alloc_end) {
4347 free_page(used);
4348 used += PAGE_SIZE;
4349 }
4350 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004351 }
4352 } while (!table && size > PAGE_SIZE && --log2qty);
4353
4354 if (!table)
4355 panic("Failed to allocate %s hash table\n", tablename);
4356
Dan Alonib49ad482007-07-15 23:38:23 -07004357 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004358 tablename,
4359 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08004360 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004361 size);
4362
4363 if (_hash_shift)
4364 *_hash_shift = log2qty;
4365 if (_hash_mask)
4366 *_hash_mask = (1 << log2qty) - 1;
4367
4368 return table;
4369}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004370
4371#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004372struct page *pfn_to_page(unsigned long pfn)
4373{
Andy Whitcroft67de6482006-06-23 02:03:12 -07004374 return __pfn_to_page(pfn);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004375}
4376unsigned long page_to_pfn(struct page *page)
4377{
Andy Whitcroft67de6482006-06-23 02:03:12 -07004378 return __page_to_pfn(page);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004379}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004380EXPORT_SYMBOL(pfn_to_page);
4381EXPORT_SYMBOL(page_to_pfn);
4382#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
Andrew Morton6220ec72006-10-19 23:29:05 -07004383
Mel Gorman835c1342007-10-16 01:25:47 -07004384/* Return a pointer to the bitmap storing bits affecting a block of pages */
4385static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4386 unsigned long pfn)
4387{
4388#ifdef CONFIG_SPARSEMEM
4389 return __pfn_to_section(pfn)->pageblock_flags;
4390#else
4391 return zone->pageblock_flags;
4392#endif /* CONFIG_SPARSEMEM */
4393}
Andrew Morton6220ec72006-10-19 23:29:05 -07004394
Mel Gorman835c1342007-10-16 01:25:47 -07004395static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4396{
4397#ifdef CONFIG_SPARSEMEM
4398 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004399 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004400#else
4401 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07004402 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004403#endif /* CONFIG_SPARSEMEM */
4404}
4405
4406/**
Mel Gormand9c23402007-10-16 01:26:01 -07004407 * 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 -07004408 * @page: The page within the block of interest
4409 * @start_bitidx: The first bit of interest to retrieve
4410 * @end_bitidx: The last bit of interest
4411 * returns pageblock_bits flags
4412 */
4413unsigned long get_pageblock_flags_group(struct page *page,
4414 int start_bitidx, int end_bitidx)
4415{
4416 struct zone *zone;
4417 unsigned long *bitmap;
4418 unsigned long pfn, bitidx;
4419 unsigned long flags = 0;
4420 unsigned long value = 1;
4421
4422 zone = page_zone(page);
4423 pfn = page_to_pfn(page);
4424 bitmap = get_pageblock_bitmap(zone, pfn);
4425 bitidx = pfn_to_bitidx(zone, pfn);
4426
4427 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4428 if (test_bit(bitidx + start_bitidx, bitmap))
4429 flags |= value;
4430
4431 return flags;
4432}
4433
4434/**
Mel Gormand9c23402007-10-16 01:26:01 -07004435 * 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 -07004436 * @page: The page within the block of interest
4437 * @start_bitidx: The first bit of interest
4438 * @end_bitidx: The last bit of interest
4439 * @flags: The flags to set
4440 */
4441void set_pageblock_flags_group(struct page *page, unsigned long flags,
4442 int start_bitidx, int end_bitidx)
4443{
4444 struct zone *zone;
4445 unsigned long *bitmap;
4446 unsigned long pfn, bitidx;
4447 unsigned long value = 1;
4448
4449 zone = page_zone(page);
4450 pfn = page_to_pfn(page);
4451 bitmap = get_pageblock_bitmap(zone, pfn);
4452 bitidx = pfn_to_bitidx(zone, pfn);
4453
4454 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4455 if (flags & value)
4456 __set_bit(bitidx + start_bitidx, bitmap);
4457 else
4458 __clear_bit(bitidx + start_bitidx, bitmap);
4459}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004460
4461/*
4462 * This is designed as sub function...plz see page_isolation.c also.
4463 * set/clear page block's type to be ISOLATE.
4464 * page allocater never alloc memory from ISOLATE block.
4465 */
4466
4467int set_migratetype_isolate(struct page *page)
4468{
4469 struct zone *zone;
4470 unsigned long flags;
4471 int ret = -EBUSY;
4472
4473 zone = page_zone(page);
4474 spin_lock_irqsave(&zone->lock, flags);
4475 /*
4476 * In future, more migrate types will be able to be isolation target.
4477 */
4478 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4479 goto out;
4480 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4481 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4482 ret = 0;
4483out:
4484 spin_unlock_irqrestore(&zone->lock, flags);
4485 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004486 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004487 return ret;
4488}
4489
4490void unset_migratetype_isolate(struct page *page)
4491{
4492 struct zone *zone;
4493 unsigned long flags;
4494 zone = page_zone(page);
4495 spin_lock_irqsave(&zone->lock, flags);
4496 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4497 goto out;
4498 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4499 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4500out:
4501 spin_unlock_irqrestore(&zone->lock, flags);
4502}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07004503
4504#ifdef CONFIG_MEMORY_HOTREMOVE
4505/*
4506 * All pages in the range must be isolated before calling this.
4507 */
4508void
4509__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4510{
4511 struct page *page;
4512 struct zone *zone;
4513 int order, i;
4514 unsigned long pfn;
4515 unsigned long flags;
4516 /* find the first valid pfn */
4517 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4518 if (pfn_valid(pfn))
4519 break;
4520 if (pfn == end_pfn)
4521 return;
4522 zone = page_zone(pfn_to_page(pfn));
4523 spin_lock_irqsave(&zone->lock, flags);
4524 pfn = start_pfn;
4525 while (pfn < end_pfn) {
4526 if (!pfn_valid(pfn)) {
4527 pfn++;
4528 continue;
4529 }
4530 page = pfn_to_page(pfn);
4531 BUG_ON(page_count(page));
4532 BUG_ON(!PageBuddy(page));
4533 order = page_order(page);
4534#ifdef CONFIG_DEBUG_VM
4535 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4536 pfn, 1 << order, end_pfn);
4537#endif
4538 list_del(&page->lru);
4539 rmv_page_order(page);
4540 zone->free_area[order].nr_free--;
4541 __mod_zone_page_state(zone, NR_FREE_PAGES,
4542 - (1UL << order));
4543 for (i = 0; i < (1 << order); i++)
4544 SetPageReserved((page+i));
4545 pfn += (1 << order);
4546 }
4547 spin_unlock_irqrestore(&zone->lock, flags);
4548}
4549#endif