blob: af982f7cdb2a8d409315f8fd95b140277e62c3af [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
KOSAKI Motohiro10ed2732008-03-04 14:28:32 -080022#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/bootmem.h>
24#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070025#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/module.h>
27#include <linux/suspend.h>
28#include <linux/pagevec.h>
29#include <linux/blkdev.h>
30#include <linux/slab.h>
David Rientjes5a3135c2007-10-16 23:25:53 -070031#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/notifier.h>
33#include <linux/topology.h>
34#include <linux/sysctl.h>
35#include <linux/cpu.h>
36#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070037#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <linux/nodemask.h>
39#include <linux/vmalloc.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080040#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070041#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070042#include <linux/sort.h>
43#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070044#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080045#include <linux/fault-inject.h>
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -070046#include <linux/page-isolation.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080047#include <linux/memcontrol.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070048#include <linux/debugobjects.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
50#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070051#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include "internal.h"
53
54/*
Christoph Lameter13808912007-10-16 01:25:27 -070055 * Array of node states.
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 */
Christoph Lameter13808912007-10-16 01:25:27 -070057nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
58 [N_POSSIBLE] = NODE_MASK_ALL,
59 [N_ONLINE] = { { [0] = 1UL } },
60#ifndef CONFIG_NUMA
61 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
62#ifdef CONFIG_HIGHMEM
63 [N_HIGH_MEMORY] = { { [0] = 1UL } },
64#endif
65 [N_CPU] = { { [0] = 1UL } },
66#endif /* NUMA */
67};
68EXPORT_SYMBOL(node_states);
69
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070070unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070071unsigned long totalreserve_pages __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070072long nr_swap_pages;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080073int percpu_pagelist_fraction;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
Mel Gormand9c23402007-10-16 01:26:01 -070075#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
76int pageblock_order __read_mostly;
77#endif
78
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080079static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080080
Linus Torvalds1da177e2005-04-16 15:20:36 -070081/*
82 * results with 256, 32 in the lowmem_reserve sysctl:
83 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
84 * 1G machine -> (16M dma, 784M normal, 224M high)
85 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
86 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
87 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010088 *
89 * TBD: should special case ZONE_DMA32 machines here - in those we normally
90 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -070092int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -080093#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -070094 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -080095#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -070096#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -070097 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -070098#endif
Christoph Lametere53ef382006-09-25 23:31:14 -070099#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700100 32,
Christoph Lametere53ef382006-09-25 23:31:14 -0700101#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700102 32,
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700103};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104
105EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
Helge Deller15ad7cd2006-12-06 20:40:36 -0800107static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -0800108#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700109 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -0800110#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700111#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700112 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -0700113#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700114 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -0700115#ifdef CONFIG_HIGHMEM
Mel Gorman2a1e2742007-07-17 04:03:12 -0700116 "HighMem",
Christoph Lametere53ef382006-09-25 23:31:14 -0700117#endif
Mel Gorman2a1e2742007-07-17 04:03:12 -0700118 "Movable",
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700119};
120
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121int min_free_kbytes = 1024;
122
Yasunori Goto86356ab2006-06-23 02:03:09 -0700123unsigned long __meminitdata nr_kernel_pages;
124unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700125static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Mel Gormanc7132162006-09-27 01:49:43 -0700127#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
128 /*
Simon Arlott183ff222007-10-20 01:27:18 +0200129 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
Mel Gormanc7132162006-09-27 01:49:43 -0700130 * ranges of memory (RAM) that may be registered with add_active_range().
131 * Ranges passed to add_active_range() will be merged if possible
132 * so the number of times add_active_range() can be called is
133 * related to the number of nodes and the number of holes
134 */
135 #ifdef CONFIG_MAX_ACTIVE_REGIONS
136 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
137 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
138 #else
139 #if MAX_NUMNODES >= 32
140 /* If there can be many nodes, allow up to 50 holes per node */
141 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
142 #else
143 /* By default, allow up to 256 distinct regions */
144 #define MAX_ACTIVE_REGIONS 256
145 #endif
146 #endif
147
Jan Beulich98011f52007-07-15 23:38:17 -0700148 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
149 static int __meminitdata nr_nodemap_entries;
150 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
151 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Mel Gormanfb014392006-09-27 01:49:59 -0700152#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
Jan Beulich98011f52007-07-15 23:38:17 -0700153 static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
154 static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
Mel Gormanfb014392006-09-27 01:49:59 -0700155#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700156 static unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700157 static unsigned long __initdata required_movablecore;
Adrian Bunkb69a7282008-07-23 21:28:12 -0700158 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
Mel Gorman2a1e2742007-07-17 04:03:12 -0700159
160 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
161 int movable_zone;
162 EXPORT_SYMBOL(movable_zone);
Mel Gormanc7132162006-09-27 01:49:43 -0700163#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
164
Miklos Szeredi418508c2007-05-23 13:57:55 -0700165#if MAX_NUMNODES > 1
166int nr_node_ids __read_mostly = MAX_NUMNODES;
167EXPORT_SYMBOL(nr_node_ids);
168#endif
169
Mel Gorman9ef9acb2007-10-16 01:25:54 -0700170int page_group_by_mobility_disabled __read_mostly;
171
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700172static void set_pageblock_migratetype(struct page *page, int migratetype)
173{
174 set_pageblock_flags_group(page, (unsigned long)migratetype,
175 PB_migrate, PB_migrate_end);
176}
177
Nick Piggin13e74442006-01-06 00:10:58 -0800178#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700179static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700181 int ret = 0;
182 unsigned seq;
183 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700184
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700185 do {
186 seq = zone_span_seqbegin(zone);
187 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
188 ret = 1;
189 else if (pfn < zone->zone_start_pfn)
190 ret = 1;
191 } while (zone_span_seqretry(zone, seq));
192
193 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700194}
195
196static int page_is_consistent(struct zone *zone, struct page *page)
197{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700198 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700199 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700201 return 0;
202
203 return 1;
204}
205/*
206 * Temporary debugging check for pages not lying within a given zone.
207 */
208static int bad_range(struct zone *zone, struct page *page)
209{
210 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700212 if (!page_is_consistent(zone, page))
213 return 1;
214
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 return 0;
216}
Nick Piggin13e74442006-01-06 00:10:58 -0800217#else
218static inline int bad_range(struct zone *zone, struct page *page)
219{
220 return 0;
221}
222#endif
223
Nick Piggin224abf92006-01-06 00:11:11 -0800224static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
Hugh Dickins9442ec9d2008-03-04 14:29:07 -0800226 void *pc = page_get_page_cgroup(page);
227
228 printk(KERN_EMERG "Bad page state in process '%s'\n" KERN_EMERG
229 "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
Nick Piggin224abf92006-01-06 00:11:11 -0800230 current->comm, page, (int)(2*sizeof(unsigned long)),
231 (unsigned long)page->flags, page->mapping,
232 page_mapcount(page), page_count(page));
Hugh Dickins9442ec9d2008-03-04 14:29:07 -0800233 if (pc) {
234 printk(KERN_EMERG "cgroup:%p\n", pc);
235 page_reset_bad_cgroup(page);
236 }
237 printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
238 KERN_EMERG "Backtrace:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 dump_stack();
Russ Andersondfa7e202008-06-09 11:18:45 -0500240 page->flags &= ~PAGE_FLAGS_CLEAR_WHEN_BAD;
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
Andi Kleen01ad1c02008-07-23 21:27:46 -0700267void prep_compound_page(struct page *page, unsigned long order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268{
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))
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700435 break;
Nick Piggin13e74442006-01-06 00:10:58 -0800436
Andy Whitcroft3c82d0c2008-07-23 21:27:11 -0700437 /* Our buddy is free, merge with it and move up one order. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700439 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800441 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 page = page + (combined_idx - page_idx);
443 page_idx = combined_idx;
444 order++;
445 }
446 set_page_order(page, order);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700447 list_add(&page->lru,
448 &zone->free_area[order].free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 zone->free_area[order].nr_free++;
450}
451
Nick Piggin224abf92006-01-06 00:11:11 -0800452static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
Nick Piggin92be2e32006-01-06 00:10:57 -0800454 if (unlikely(page_mapcount(page) |
455 (page->mapping != NULL) |
Hugh Dickins9442ec9d2008-03-04 14:29:07 -0800456 (page_get_page_cgroup(page) != NULL) |
Nick Piggin92be2e32006-01-06 00:10:57 -0800457 (page_count(page) != 0) |
Russ Andersondfa7e202008-06-09 11:18:45 -0500458 (page->flags & PAGE_FLAGS_CHECK_AT_FREE)))
Nick Piggin224abf92006-01-06 00:11:11 -0800459 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700461 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800462 /*
463 * For now, we report if PG_reserved was found set, but do not
464 * clear it, and do not free the page. But we shall soon need
465 * to do more, for when the ZERO_PAGE count wraps negative.
466 */
467 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468}
469
470/*
471 * Frees a list of pages.
472 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700473 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 *
475 * If the zone was previously in an "all pages pinned" state then look to
476 * see if this freeing clears that state.
477 *
478 * And clear the zone's pages_scanned counter, to hold off the "all pages are
479 * pinned" detection logic.
480 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800481static void free_pages_bulk(struct zone *zone, int count,
482 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483{
Nick Pigginc54ad302006-01-06 00:10:56 -0800484 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700485 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800487 while (count--) {
488 struct page *page;
489
Nick Piggin725d7042006-09-25 23:30:55 -0700490 VM_BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800492 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800494 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800496 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497}
498
Nick Piggin48db57f2006-01-08 01:00:42 -0800499static void free_one_page(struct zone *zone, struct page *page, int order)
500{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700501 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700502 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700503 zone->pages_scanned = 0;
Paul Jackson0798e512006-12-06 20:31:38 -0800504 __free_one_page(page, zone, order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700505 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800506}
507
508static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509{
Nick Pigginc54ad302006-01-06 00:10:56 -0800510 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800512 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800515 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800516 if (reserved)
517 return;
518
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700519 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700520 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700521 debug_check_no_obj_freed(page_address(page),
522 PAGE_SIZE << order);
523 }
Nick Piggindafb1362006-10-11 01:21:30 -0700524 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800525 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700526
Nick Pigginc54ad302006-01-06 00:10:56 -0800527 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700528 __count_vm_events(PGFREE, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800529 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800530 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531}
532
David Howellsa226f6c2006-01-06 00:11:08 -0800533/*
534 * permit the bootmem allocator to evade page validation on high-order frees
535 */
Yasunori Gotoaf370fb2008-07-23 21:28:17 -0700536void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800537{
538 if (order == 0) {
539 __ClearPageReserved(page);
540 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800541 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800542 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800543 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800544 int loop;
545
Nick Piggin545b1ea2006-03-22 00:08:07 -0800546 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800547 for (loop = 0; loop < BITS_PER_LONG; loop++) {
548 struct page *p = &page[loop];
549
Nick Piggin545b1ea2006-03-22 00:08:07 -0800550 if (loop + 1 < BITS_PER_LONG)
551 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800552 __ClearPageReserved(p);
553 set_page_count(p, 0);
554 }
555
Nick Piggin7835e982006-03-22 00:08:40 -0800556 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800557 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800558 }
559}
560
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
562/*
563 * The order of subdivision here is critical for the IO subsystem.
564 * Please do not alter this order without good reasons and regression
565 * testing. Specifically, as large blocks of memory are subdivided,
566 * the order in which smaller blocks are delivered depends on the order
567 * they're subdivided in this function. This is the primary factor
568 * influencing the order in which pages are delivered to the IO
569 * subsystem according to empirical testing, and this is also justified
570 * by considering the behavior of a buddy system containing a single
571 * large block of memory acted on by a series of small allocations.
572 * This behavior is a critical factor in sglist merging's success.
573 *
574 * -- wli
575 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800576static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700577 int low, int high, struct free_area *area,
578 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579{
580 unsigned long size = 1 << high;
581
582 while (high > low) {
583 area--;
584 high--;
585 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700586 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700587 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 area->nr_free++;
589 set_page_order(&page[size], high);
590 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591}
592
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593/*
594 * This page is about to be returned from the page allocator
595 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800596static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597{
Nick Piggin92be2e32006-01-06 00:10:57 -0800598 if (unlikely(page_mapcount(page) |
599 (page->mapping != NULL) |
Hugh Dickins9442ec9d2008-03-04 14:29:07 -0800600 (page_get_page_cgroup(page) != NULL) |
Nick Piggin92be2e32006-01-06 00:10:57 -0800601 (page_count(page) != 0) |
Russ Andersondfa7e202008-06-09 11:18:45 -0500602 (page->flags & PAGE_FLAGS_CHECK_AT_PREP)))
Nick Piggin224abf92006-01-06 00:11:11 -0800603 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604
Hugh Dickins689bceb2005-11-21 21:32:20 -0800605 /*
606 * For now, we report if PG_reserved was found set, but do not
607 * clear it, and do not allocate the page: as a safety net.
608 */
609 if (PageReserved(page))
610 return 1;
611
Christoph Lameter0a128b22008-04-28 02:12:52 -0700612 page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_reclaim |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 1 << PG_referenced | 1 << PG_arch_1 |
Nick Piggin5409bae2007-02-28 20:12:27 -0800614 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700615 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800616 set_page_refcounted(page);
Nick Piggincc102502006-12-06 20:32:00 -0800617
618 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800620
621 if (gfp_flags & __GFP_ZERO)
622 prep_zero_page(page, order, gfp_flags);
623
624 if (order && (gfp_flags & __GFP_COMP))
625 prep_compound_page(page, order);
626
Hugh Dickins689bceb2005-11-21 21:32:20 -0800627 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628}
629
Mel Gorman56fd56b2007-10-16 01:25:58 -0700630/*
631 * Go through the free lists for the given migratetype and remove
632 * the smallest available page from the freelists
633 */
634static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
635 int migratetype)
636{
637 unsigned int current_order;
638 struct free_area * area;
639 struct page *page;
640
641 /* Find a page of the appropriate size in the preferred list */
642 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
643 area = &(zone->free_area[current_order]);
644 if (list_empty(&area->free_list[migratetype]))
645 continue;
646
647 page = list_entry(area->free_list[migratetype].next,
648 struct page, lru);
649 list_del(&page->lru);
650 rmv_page_order(page);
651 area->nr_free--;
652 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
653 expand(zone, page, order, current_order, area, migratetype);
654 return page;
655 }
656
657 return NULL;
658}
659
660
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700661/*
662 * This array describes the order lists are fallen back to when
663 * the free lists for the desirable migrate type are depleted
664 */
665static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700666 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
667 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
668 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
669 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700670};
671
Mel Gormanc361be52007-10-16 01:25:51 -0700672/*
673 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700674 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700675 * boundary. If alignment is required, use move_freepages_block()
676 */
Adrian Bunkb69a7282008-07-23 21:28:12 -0700677static int move_freepages(struct zone *zone,
678 struct page *start_page, struct page *end_page,
679 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700680{
681 struct page *page;
682 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700683 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700684
685#ifndef CONFIG_HOLES_IN_ZONE
686 /*
687 * page_zone is not safe to call in this context when
688 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
689 * anyway as we check zone boundaries in move_freepages_block().
690 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700691 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700692 */
693 BUG_ON(page_zone(start_page) != page_zone(end_page));
694#endif
695
696 for (page = start_page; page <= end_page;) {
697 if (!pfn_valid_within(page_to_pfn(page))) {
698 page++;
699 continue;
700 }
701
702 if (!PageBuddy(page)) {
703 page++;
704 continue;
705 }
706
707 order = page_order(page);
708 list_del(&page->lru);
709 list_add(&page->lru,
710 &zone->free_area[order].free_list[migratetype]);
711 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700712 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700713 }
714
Mel Gormand1003132007-10-16 01:26:00 -0700715 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700716}
717
Adrian Bunkb69a7282008-07-23 21:28:12 -0700718static int move_freepages_block(struct zone *zone, struct page *page,
719 int migratetype)
Mel Gormanc361be52007-10-16 01:25:51 -0700720{
721 unsigned long start_pfn, end_pfn;
722 struct page *start_page, *end_page;
723
724 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700725 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700726 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700727 end_page = start_page + pageblock_nr_pages - 1;
728 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700729
730 /* Do not cross zone boundaries */
731 if (start_pfn < zone->zone_start_pfn)
732 start_page = page;
733 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
734 return 0;
735
736 return move_freepages(zone, start_page, end_page, migratetype);
737}
738
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700739/* Remove an element from the buddy allocator from the fallback list */
740static struct page *__rmqueue_fallback(struct zone *zone, int order,
741 int start_migratetype)
742{
743 struct free_area * area;
744 int current_order;
745 struct page *page;
746 int migratetype, i;
747
748 /* Find the largest possible block of pages in the other list */
749 for (current_order = MAX_ORDER-1; current_order >= order;
750 --current_order) {
751 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
752 migratetype = fallbacks[start_migratetype][i];
753
Mel Gorman56fd56b2007-10-16 01:25:58 -0700754 /* MIGRATE_RESERVE handled later if necessary */
755 if (migratetype == MIGRATE_RESERVE)
756 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700757
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700758 area = &(zone->free_area[current_order]);
759 if (list_empty(&area->free_list[migratetype]))
760 continue;
761
762 page = list_entry(area->free_list[migratetype].next,
763 struct page, lru);
764 area->nr_free--;
765
766 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700767 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700768 * pages to the preferred allocation list. If falling
769 * back for a reclaimable kernel allocation, be more
770 * agressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700771 */
Mel Gormand9c23402007-10-16 01:26:01 -0700772 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gorman46dafbc2007-10-16 01:25:55 -0700773 start_migratetype == MIGRATE_RECLAIMABLE) {
774 unsigned long pages;
775 pages = move_freepages_block(zone, page,
776 start_migratetype);
777
778 /* Claim the whole block if over half of it is free */
Mel Gormand9c23402007-10-16 01:26:01 -0700779 if (pages >= (1 << (pageblock_order-1)))
Mel Gorman46dafbc2007-10-16 01:25:55 -0700780 set_pageblock_migratetype(page,
781 start_migratetype);
782
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700783 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700784 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700785
786 /* Remove the page from the freelists */
787 list_del(&page->lru);
788 rmv_page_order(page);
789 __mod_zone_page_state(zone, NR_FREE_PAGES,
790 -(1UL << order));
791
Mel Gormand9c23402007-10-16 01:26:01 -0700792 if (current_order == pageblock_order)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700793 set_pageblock_migratetype(page,
794 start_migratetype);
795
796 expand(zone, page, order, current_order, area, migratetype);
797 return page;
798 }
799 }
800
Mel Gorman56fd56b2007-10-16 01:25:58 -0700801 /* Use MIGRATE_RESERVE rather than fail an allocation */
802 return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700803}
804
Mel Gorman56fd56b2007-10-16 01:25:58 -0700805/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 * Do the hard work of removing an element from the buddy allocator.
807 * Call me with the zone->lock already held.
808 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700809static struct page *__rmqueue(struct zone *zone, unsigned int order,
810 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 struct page *page;
813
Mel Gorman56fd56b2007-10-16 01:25:58 -0700814 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
Mel Gorman56fd56b2007-10-16 01:25:58 -0700816 if (unlikely(!page))
817 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700818
819 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820}
821
822/*
823 * Obtain a specified number of elements from the buddy allocator, all under
824 * a single hold of the lock, for efficiency. Add them to the supplied list.
825 * Returns the number of new pages which were placed at *list.
826 */
827static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700828 unsigned long count, struct list_head *list,
829 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832
Nick Pigginc54ad302006-01-06 00:10:56 -0800833 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700835 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800836 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -0800838
839 /*
840 * Split buddy pages returned by expand() are received here
841 * in physical page order. The page is added to the callers and
842 * list and the list head then moves forward. From the callers
843 * perspective, the linked list is ordered by page number in
844 * some conditions. This is useful for IO devices that can
845 * merge IO requests if the physical pages are ordered
846 * properly.
847 */
Mel Gorman535131e2007-10-16 01:25:49 -0700848 list_add(&page->lru, list);
849 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -0800850 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800852 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800853 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854}
855
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700856#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800857/*
Christoph Lameter4037d452007-05-09 02:35:14 -0700858 * Called from the vmstat counter updater to drain pagesets of this
859 * currently executing processor on remote nodes after they have
860 * expired.
861 *
Christoph Lameter879336c2006-03-22 00:09:08 -0800862 * Note that this function must be called with the thread pinned to
863 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800864 */
Christoph Lameter4037d452007-05-09 02:35:14 -0700865void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700866{
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700867 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -0700868 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700869
Christoph Lameter4037d452007-05-09 02:35:14 -0700870 local_irq_save(flags);
871 if (pcp->count >= pcp->batch)
872 to_drain = pcp->batch;
873 else
874 to_drain = pcp->count;
875 free_pages_bulk(zone, to_drain, &pcp->list, 0);
876 pcp->count -= to_drain;
877 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700878}
879#endif
880
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800881/*
882 * Drain pages of the indicated processor.
883 *
884 * The processor must either be the current processor and the
885 * thread pinned to the current processor or a processor that
886 * is not online.
887 */
888static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889{
Nick Pigginc54ad302006-01-06 00:10:56 -0800890 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892
893 for_each_zone(zone) {
894 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800895 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
Christoph Lameterf2e12bb2007-01-05 16:37:02 -0800897 if (!populated_zone(zone))
898 continue;
899
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700900 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800902 pcp = &pset->pcp;
903 local_irq_save(flags);
904 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
905 pcp->count = 0;
906 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 }
908}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800910/*
911 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
912 */
913void drain_local_pages(void *arg)
914{
915 drain_pages(smp_processor_id());
916}
917
918/*
919 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
920 */
921void drain_all_pages(void)
922{
Jens Axboe15c8b6c2008-05-09 09:39:44 +0200923 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800924}
925
Rafael J. Wysocki296699d2007-07-29 23:27:18 +0200926#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
928void mark_free_pages(struct zone *zone)
929{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700930 unsigned long pfn, max_zone_pfn;
931 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700932 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 struct list_head *curr;
934
935 if (!zone->spanned_pages)
936 return;
937
938 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700939
940 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
941 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
942 if (pfn_valid(pfn)) {
943 struct page *page = pfn_to_page(pfn);
944
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700945 if (!swsusp_page_is_forbidden(page))
946 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700947 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700949 for_each_migratetype_order(order, t) {
950 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700951 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700953 pfn = page_to_pfn(list_entry(curr, struct page, lru));
954 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700955 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700956 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700957 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 spin_unlock_irqrestore(&zone->lock, flags);
959}
Mel Gormane2c55dc2007-10-16 01:25:50 -0700960#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
962/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 * Free a 0-order page
964 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800965static void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966{
967 struct zone *zone = page_zone(page);
968 struct per_cpu_pages *pcp;
969 unsigned long flags;
970
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 if (PageAnon(page))
972 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800973 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800974 return;
975
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700976 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700977 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700978 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
979 }
Nick Piggindafb1362006-10-11 01:21:30 -0700980 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800981 kernel_map_pages(page, 1, 0);
982
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800983 pcp = &zone_pcp(zone, get_cpu())->pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700985 __count_vm_event(PGFREE);
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800986 if (cold)
987 list_add_tail(&page->lru, &pcp->list);
988 else
989 list_add(&page->lru, &pcp->list);
Mel Gorman535131e2007-10-16 01:25:49 -0700990 set_page_private(page, get_pageblock_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800992 if (pcp->count >= pcp->high) {
993 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
994 pcp->count -= pcp->batch;
995 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 local_irq_restore(flags);
997 put_cpu();
998}
999
Harvey Harrison920c7a52008-02-04 22:29:26 -08001000void free_hot_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001{
1002 free_hot_cold_page(page, 0);
1003}
1004
Harvey Harrison920c7a52008-02-04 22:29:26 -08001005void free_cold_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006{
1007 free_hot_cold_page(page, 1);
1008}
1009
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001010/*
1011 * split_page takes a non-compound higher-order page, and splits it into
1012 * n (1<<order) sub-pages: page[0..n]
1013 * Each sub-page must be freed individually.
1014 *
1015 * Note: this is probably too low level an operation for use in drivers.
1016 * Please consult with lkml before using this in your driver.
1017 */
1018void split_page(struct page *page, unsigned int order)
1019{
1020 int i;
1021
Nick Piggin725d7042006-09-25 23:30:55 -07001022 VM_BUG_ON(PageCompound(page));
1023 VM_BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -08001024 for (i = 1; i < (1 << order); i++)
1025 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001026}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028/*
1029 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1030 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1031 * or two.
1032 */
Mel Gorman18ea7e72008-04-28 02:12:14 -07001033static struct page *buffered_rmqueue(struct zone *preferred_zone,
Nick Piggina74609f2006-01-06 00:11:20 -08001034 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035{
1036 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001037 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -08001039 int cpu;
Mel Gorman64c5e132007-10-16 01:25:59 -07001040 int migratetype = allocflags_to_migratetype(gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
Hugh Dickins689bceb2005-11-21 21:32:20 -08001042again:
Nick Piggina74609f2006-01-06 00:11:20 -08001043 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -08001044 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 struct per_cpu_pages *pcp;
1046
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001047 pcp = &zone_pcp(zone, cpu)->pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001049 if (!pcp->count) {
nkalmala941c7102006-11-02 22:07:04 -08001050 pcp->count = rmqueue_bulk(zone, 0,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001051 pcp->batch, &pcp->list, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001052 if (unlikely(!pcp->count))
1053 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 }
Mel Gorman535131e2007-10-16 01:25:49 -07001055
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001056 /* Find a page of the appropriate migrate type */
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001057 if (cold) {
1058 list_for_each_entry_reverse(page, &pcp->list, lru)
1059 if (page_private(page) == migratetype)
1060 break;
1061 } else {
1062 list_for_each_entry(page, &pcp->list, lru)
1063 if (page_private(page) == migratetype)
1064 break;
1065 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001066
1067 /* Allocate more to the pcp list if necessary */
1068 if (unlikely(&page->lru == &pcp->list)) {
Mel Gorman535131e2007-10-16 01:25:49 -07001069 pcp->count += rmqueue_bulk(zone, 0,
1070 pcp->batch, &pcp->list, migratetype);
1071 page = list_entry(pcp->list.next, struct page, lru);
Mel Gorman535131e2007-10-16 01:25:49 -07001072 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001073
1074 list_del(&page->lru);
1075 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001076 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001078 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001079 spin_unlock(&zone->lock);
1080 if (!page)
1081 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 }
1083
Christoph Lameterf8891e52006-06-30 01:55:45 -07001084 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Mel Gorman18ea7e72008-04-28 02:12:14 -07001085 zone_statistics(preferred_zone, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001086 local_irq_restore(flags);
1087 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
Nick Piggin725d7042006-09-25 23:30:55 -07001089 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001090 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001091 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001093
1094failed:
1095 local_irq_restore(flags);
1096 put_cpu();
1097 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098}
1099
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001100#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -08001101#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1102#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1103#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1104#define ALLOC_HARDER 0x10 /* try to alloc harder */
1105#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1106#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001107
Akinobu Mita933e3122006-12-08 02:39:45 -08001108#ifdef CONFIG_FAIL_PAGE_ALLOC
1109
1110static struct fail_page_alloc_attr {
1111 struct fault_attr attr;
1112
1113 u32 ignore_gfp_highmem;
1114 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001115 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001116
1117#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1118
1119 struct dentry *ignore_gfp_highmem_file;
1120 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001121 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001122
1123#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1124
1125} fail_page_alloc = {
1126 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001127 .ignore_gfp_wait = 1,
1128 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001129 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001130};
1131
1132static int __init setup_fail_page_alloc(char *str)
1133{
1134 return setup_fault_attr(&fail_page_alloc.attr, str);
1135}
1136__setup("fail_page_alloc=", setup_fail_page_alloc);
1137
1138static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1139{
Akinobu Mita54114992007-07-15 23:40:23 -07001140 if (order < fail_page_alloc.min_order)
1141 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001142 if (gfp_mask & __GFP_NOFAIL)
1143 return 0;
1144 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1145 return 0;
1146 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1147 return 0;
1148
1149 return should_fail(&fail_page_alloc.attr, 1 << order);
1150}
1151
1152#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1153
1154static int __init fail_page_alloc_debugfs(void)
1155{
1156 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1157 struct dentry *dir;
1158 int err;
1159
1160 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1161 "fail_page_alloc");
1162 if (err)
1163 return err;
1164 dir = fail_page_alloc.attr.dentries.dir;
1165
1166 fail_page_alloc.ignore_gfp_wait_file =
1167 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1168 &fail_page_alloc.ignore_gfp_wait);
1169
1170 fail_page_alloc.ignore_gfp_highmem_file =
1171 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1172 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001173 fail_page_alloc.min_order_file =
1174 debugfs_create_u32("min-order", mode, dir,
1175 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001176
1177 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001178 !fail_page_alloc.ignore_gfp_highmem_file ||
1179 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001180 err = -ENOMEM;
1181 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1182 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001183 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001184 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1185 }
1186
1187 return err;
1188}
1189
1190late_initcall(fail_page_alloc_debugfs);
1191
1192#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1193
1194#else /* CONFIG_FAIL_PAGE_ALLOC */
1195
1196static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1197{
1198 return 0;
1199}
1200
1201#endif /* CONFIG_FAIL_PAGE_ALLOC */
1202
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203/*
1204 * Return 1 if free pages are above 'mark'. This takes into account the order
1205 * of the allocation.
1206 */
1207int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001208 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209{
1210 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001211 long min = mark;
1212 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 int o;
1214
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001215 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001217 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 min -= min / 4;
1219
1220 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1221 return 0;
1222 for (o = 0; o < order; o++) {
1223 /* At the next order, this order's pages become unavailable */
1224 free_pages -= z->free_area[o].nr_free << o;
1225
1226 /* Require fewer higher order pages to be free */
1227 min >>= 1;
1228
1229 if (free_pages <= min)
1230 return 0;
1231 }
1232 return 1;
1233}
1234
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001235#ifdef CONFIG_NUMA
1236/*
1237 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1238 * skip over zones that are not allowed by the cpuset, or that have
1239 * been recently (in last second) found to be nearly full. See further
1240 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001241 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001242 *
1243 * If the zonelist cache is present in the passed in zonelist, then
1244 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001245 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001246 *
1247 * If the zonelist cache is not available for this zonelist, does
1248 * nothing and returns NULL.
1249 *
1250 * If the fullzones BITMAP in the zonelist cache is stale (more than
1251 * a second since last zap'd) then we zap it out (clear its bits.)
1252 *
1253 * We hold off even calling zlc_setup, until after we've checked the
1254 * first zone in the zonelist, on the theory that most allocations will
1255 * be satisfied from that first zone, so best to examine that zone as
1256 * quickly as we can.
1257 */
1258static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1259{
1260 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1261 nodemask_t *allowednodes; /* zonelist_cache approximation */
1262
1263 zlc = zonelist->zlcache_ptr;
1264 if (!zlc)
1265 return NULL;
1266
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001267 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001268 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1269 zlc->last_full_zap = jiffies;
1270 }
1271
1272 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1273 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001274 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001275 return allowednodes;
1276}
1277
1278/*
1279 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1280 * if it is worth looking at further for free memory:
1281 * 1) Check that the zone isn't thought to be full (doesn't have its
1282 * bit set in the zonelist_cache fullzones BITMAP).
1283 * 2) Check that the zones node (obtained from the zonelist_cache
1284 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1285 * Return true (non-zero) if zone is worth looking at further, or
1286 * else return false (zero) if it is not.
1287 *
1288 * This check -ignores- the distinction between various watermarks,
1289 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1290 * found to be full for any variation of these watermarks, it will
1291 * be considered full for up to one second by all requests, unless
1292 * we are so low on memory on all allowed nodes that we are forced
1293 * into the second scan of the zonelist.
1294 *
1295 * In the second scan we ignore this zonelist cache and exactly
1296 * apply the watermarks to all zones, even it is slower to do so.
1297 * We are low on memory in the second scan, and should leave no stone
1298 * unturned looking for a free page.
1299 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001300static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001301 nodemask_t *allowednodes)
1302{
1303 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1304 int i; /* index of *z in zonelist zones */
1305 int n; /* node that zone *z is on */
1306
1307 zlc = zonelist->zlcache_ptr;
1308 if (!zlc)
1309 return 1;
1310
Mel Gormandd1a2392008-04-28 02:12:17 -07001311 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001312 n = zlc->z_to_n[i];
1313
1314 /* This zone is worth trying if it is allowed but not full */
1315 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1316}
1317
1318/*
1319 * Given 'z' scanning a zonelist, set the corresponding bit in
1320 * zlc->fullzones, so that subsequent attempts to allocate a page
1321 * from that zone don't waste time re-examining it.
1322 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001323static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001324{
1325 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1326 int i; /* index of *z in zonelist zones */
1327
1328 zlc = zonelist->zlcache_ptr;
1329 if (!zlc)
1330 return;
1331
Mel Gormandd1a2392008-04-28 02:12:17 -07001332 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001333
1334 set_bit(i, zlc->fullzones);
1335}
1336
1337#else /* CONFIG_NUMA */
1338
1339static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1340{
1341 return NULL;
1342}
1343
Mel Gormandd1a2392008-04-28 02:12:17 -07001344static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001345 nodemask_t *allowednodes)
1346{
1347 return 1;
1348}
1349
Mel Gormandd1a2392008-04-28 02:12:17 -07001350static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001351{
1352}
1353#endif /* CONFIG_NUMA */
1354
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001355/*
Paul Jackson0798e512006-12-06 20:31:38 -08001356 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001357 * a page.
1358 */
1359static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001360get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001361 struct zonelist *zonelist, int high_zoneidx, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -07001362{
Mel Gormandd1a2392008-04-28 02:12:17 -07001363 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001364 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001365 int classzone_idx;
Mel Gorman18ea7e72008-04-28 02:12:14 -07001366 struct zone *zone, *preferred_zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001367 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1368 int zlc_active = 0; /* set if using zonelist_cache */
1369 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001370
Mel Gorman19770b32008-04-28 02:12:18 -07001371 (void)first_zones_zonelist(zonelist, high_zoneidx, nodemask,
1372 &preferred_zone);
Andy Whitcroft7eb54822008-05-23 13:04:50 -07001373 if (!preferred_zone)
1374 return NULL;
1375
Mel Gorman19770b32008-04-28 02:12:18 -07001376 classzone_idx = zone_idx(preferred_zone);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001377
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001378zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001379 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001380 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001381 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1382 */
Mel Gorman19770b32008-04-28 02:12:18 -07001383 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1384 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001385 if (NUMA_BUILD && zlc_active &&
1386 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1387 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001388 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001389 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001390 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001391
1392 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001393 unsigned long mark;
1394 if (alloc_flags & ALLOC_WMARK_MIN)
Christoph Lameter1192d522006-09-25 23:31:45 -07001395 mark = zone->pages_min;
Nick Piggin31488902005-11-28 13:44:03 -08001396 else if (alloc_flags & ALLOC_WMARK_LOW)
Christoph Lameter1192d522006-09-25 23:31:45 -07001397 mark = zone->pages_low;
Nick Piggin31488902005-11-28 13:44:03 -08001398 else
Christoph Lameter1192d522006-09-25 23:31:45 -07001399 mark = zone->pages_high;
Paul Jackson0798e512006-12-06 20:31:38 -08001400 if (!zone_watermark_ok(zone, order, mark,
1401 classzone_idx, alloc_flags)) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001402 if (!zone_reclaim_mode ||
Christoph Lameter1192d522006-09-25 23:31:45 -07001403 !zone_reclaim(zone, gfp_mask, order))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001404 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001405 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001406 }
1407
Mel Gorman18ea7e72008-04-28 02:12:14 -07001408 page = buffered_rmqueue(preferred_zone, zone, order, gfp_mask);
Paul Jackson0798e512006-12-06 20:31:38 -08001409 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001410 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001411this_zone_full:
1412 if (NUMA_BUILD)
1413 zlc_mark_zone_full(zonelist, z);
1414try_next_zone:
1415 if (NUMA_BUILD && !did_zlc_setup) {
1416 /* we do zlc_setup after the first zone is tried */
1417 allowednodes = zlc_setup(zonelist, alloc_flags);
1418 zlc_active = 1;
1419 did_zlc_setup = 1;
1420 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001421 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001422
1423 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1424 /* Disable zlc cache for second zonelist scan */
1425 zlc_active = 0;
1426 goto zonelist_scan;
1427 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001428 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001429}
1430
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431/*
1432 * This is the 'heart' of the zoned buddy allocator.
1433 */
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -07001434struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001435__alloc_pages_internal(gfp_t gfp_mask, unsigned int order,
1436 struct zonelist *zonelist, nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437{
Al Viro260b2362005-10-21 03:22:44 -04001438 const gfp_t wait = gfp_mask & __GFP_WAIT;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001439 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gormandd1a2392008-04-28 02:12:17 -07001440 struct zoneref *z;
1441 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 struct page *page;
1443 struct reclaim_state reclaim_state;
1444 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001446 int alloc_flags;
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001447 unsigned long did_some_progress;
1448 unsigned long pages_reclaimed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
1450 might_sleep_if(wait);
1451
Akinobu Mita933e3122006-12-08 02:39:45 -08001452 if (should_fail_alloc_page(gfp_mask, order))
1453 return NULL;
1454
Jens Axboe6b1de912005-11-17 21:35:02 +01001455restart:
Mel Gormandd1a2392008-04-28 02:12:17 -07001456 z = zonelist->_zonerefs; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
Mel Gormandd1a2392008-04-28 02:12:17 -07001458 if (unlikely(!z->zone)) {
Christoph Lameter523b9452007-10-16 01:25:37 -07001459 /*
1460 * Happens if we have an empty zonelist as a result of
1461 * GFP_THISNODE being used on a memoryless node
1462 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 return NULL;
1464 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001465
Mel Gorman19770b32008-04-28 02:12:18 -07001466 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001467 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001468 if (page)
1469 goto got_pg;
1470
Christoph Lameter952f3b52006-12-06 20:33:26 -08001471 /*
1472 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1473 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1474 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1475 * using a larger set of nodes after it has established that the
1476 * allowed per node queues are empty and that nodes are
1477 * over allocated.
1478 */
1479 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1480 goto nopage;
1481
Mel Gormandd1a2392008-04-28 02:12:17 -07001482 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1483 wakeup_kswapd(zone, order);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001484
Paul Jackson9bf22292005-09-06 15:18:12 -07001485 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001486 * OK, we're below the kswapd watermark and have kicked background
1487 * reclaim. Now things get more complex, so set up alloc_flags according
1488 * to how we want to proceed.
1489 *
1490 * The caller may dip into page reserves a bit more if the caller
1491 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -08001492 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1493 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -07001494 */
Nick Piggin31488902005-11-28 13:44:03 -08001495 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001496 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1497 alloc_flags |= ALLOC_HARDER;
1498 if (gfp_mask & __GFP_HIGH)
1499 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -07001500 if (wait)
1501 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502
1503 /*
1504 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001505 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 *
1507 * This is the last chance, in general, before the goto nopage.
1508 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -07001509 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 */
Mel Gorman19770b32008-04-28 02:12:18 -07001511 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001512 high_zoneidx, alloc_flags);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001513 if (page)
1514 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
1516 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -07001517
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001518rebalance:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001519 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1520 && !in_interrupt()) {
1521 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -08001522nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001523 /* go through the zonelist yet again, ignoring mins */
Mel Gorman19770b32008-04-28 02:12:18 -07001524 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001525 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001526 if (page)
1527 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -08001528 if (gfp_mask & __GFP_NOFAIL) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001529 congestion_wait(WRITE, HZ/50);
Kirill Korotaev885036d2005-11-13 16:06:41 -08001530 goto nofail_alloc;
1531 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 }
1533 goto nopage;
1534 }
1535
1536 /* Atomic allocations - we can't balance anything */
1537 if (!wait)
1538 goto nopage;
1539
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 cond_resched();
1541
1542 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001543 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 p->flags |= PF_MEMALLOC;
1545 reclaim_state.reclaimed_slab = 0;
1546 p->reclaim_state = &reclaim_state;
1547
Mel Gormandac1d272008-04-28 02:12:12 -07001548 did_some_progress = try_to_free_pages(zonelist, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
1550 p->reclaim_state = NULL;
1551 p->flags &= ~PF_MEMALLOC;
1552
1553 cond_resched();
1554
Mel Gormane2c55dc2007-10-16 01:25:50 -07001555 if (order != 0)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001556 drain_all_pages();
Mel Gormane2c55dc2007-10-16 01:25:50 -07001557
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 if (likely(did_some_progress)) {
Mel Gorman19770b32008-04-28 02:12:18 -07001559 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001560 zonelist, high_zoneidx, alloc_flags);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001561 if (page)
1562 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07001564 if (!try_set_zone_oom(zonelist, gfp_mask)) {
David Rientjesff0ceb92007-10-16 23:25:56 -07001565 schedule_timeout_uninterruptible(1);
1566 goto restart;
1567 }
1568
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 /*
1570 * Go through the zonelist yet one more time, keep
1571 * very high watermark here, this is only to catch
1572 * a parallel oom killing, we must fail if we're still
1573 * under heavy pressure.
1574 */
Mel Gorman19770b32008-04-28 02:12:18 -07001575 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1576 order, zonelist, high_zoneidx,
1577 ALLOC_WMARK_HIGH|ALLOC_CPUSET);
David Rientjesff0ceb92007-10-16 23:25:56 -07001578 if (page) {
Mel Gormandd1a2392008-04-28 02:12:17 -07001579 clear_zonelist_oom(zonelist, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001580 goto got_pg;
David Rientjesff0ceb92007-10-16 23:25:56 -07001581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582
Mel Gormana8bbf722007-07-31 00:37:30 -07001583 /* The OOM killer will not help higher order allocs so fail */
David Rientjesff0ceb92007-10-16 23:25:56 -07001584 if (order > PAGE_ALLOC_COSTLY_ORDER) {
Mel Gormandd1a2392008-04-28 02:12:17 -07001585 clear_zonelist_oom(zonelist, gfp_mask);
Mel Gormana8bbf722007-07-31 00:37:30 -07001586 goto nopage;
David Rientjesff0ceb92007-10-16 23:25:56 -07001587 }
Mel Gormana8bbf722007-07-31 00:37:30 -07001588
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001589 out_of_memory(zonelist, gfp_mask, order);
Mel Gormandd1a2392008-04-28 02:12:17 -07001590 clear_zonelist_oom(zonelist, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 goto restart;
1592 }
1593
1594 /*
1595 * Don't let big-order allocations loop unless the caller explicitly
1596 * requests that. Wait for some write requests to complete then retry.
1597 *
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001598 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1599 * means __GFP_NOFAIL, but that may not be true in other
Nishanth Aravamudanab857d02008-04-29 00:58:23 -07001600 * implementations.
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001601 *
1602 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1603 * specified, then we retry until we no longer reclaim any pages
1604 * (above), or we've reclaimed an order of pages at least as
1605 * large as the allocation's order. In both cases, if the
1606 * allocation still fails, we stop retrying.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001608 pages_reclaimed += did_some_progress;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 do_retry = 0;
1610 if (!(gfp_mask & __GFP_NORETRY)) {
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001611 if (order <= PAGE_ALLOC_COSTLY_ORDER) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 do_retry = 1;
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001613 } else {
1614 if (gfp_mask & __GFP_REPEAT &&
1615 pages_reclaimed < (1 << order))
1616 do_retry = 1;
1617 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 if (gfp_mask & __GFP_NOFAIL)
1619 do_retry = 1;
1620 }
1621 if (do_retry) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001622 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 goto rebalance;
1624 }
1625
1626nopage:
1627 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1628 printk(KERN_WARNING "%s: page allocation failure."
1629 " order:%d, mode:0x%x\n",
1630 p->comm, order, gfp_mask);
1631 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001632 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 return page;
1636}
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -07001637EXPORT_SYMBOL(__alloc_pages_internal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638
1639/*
1640 * Common helper functions.
1641 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001642unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643{
1644 struct page * page;
1645 page = alloc_pages(gfp_mask, order);
1646 if (!page)
1647 return 0;
1648 return (unsigned long) page_address(page);
1649}
1650
1651EXPORT_SYMBOL(__get_free_pages);
1652
Harvey Harrison920c7a52008-02-04 22:29:26 -08001653unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654{
1655 struct page * page;
1656
1657 /*
1658 * get_zeroed_page() returns a 32-bit address, which cannot represent
1659 * a highmem page
1660 */
Nick Piggin725d7042006-09-25 23:30:55 -07001661 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662
1663 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1664 if (page)
1665 return (unsigned long) page_address(page);
1666 return 0;
1667}
1668
1669EXPORT_SYMBOL(get_zeroed_page);
1670
1671void __pagevec_free(struct pagevec *pvec)
1672{
1673 int i = pagevec_count(pvec);
1674
1675 while (--i >= 0)
1676 free_hot_cold_page(pvec->pages[i], pvec->cold);
1677}
1678
Harvey Harrison920c7a52008-02-04 22:29:26 -08001679void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680{
Nick Pigginb5810032005-10-29 18:16:12 -07001681 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 if (order == 0)
1683 free_hot_page(page);
1684 else
1685 __free_pages_ok(page, order);
1686 }
1687}
1688
1689EXPORT_SYMBOL(__free_pages);
1690
Harvey Harrison920c7a52008-02-04 22:29:26 -08001691void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692{
1693 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001694 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 __free_pages(virt_to_page((void *)addr), order);
1696 }
1697}
1698
1699EXPORT_SYMBOL(free_pages);
1700
Timur Tabi2be0ffe2008-07-23 21:28:11 -07001701/**
1702 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1703 * @size: the number of bytes to allocate
1704 * @gfp_mask: GFP flags for the allocation
1705 *
1706 * This function is similar to alloc_pages(), except that it allocates the
1707 * minimum number of pages to satisfy the request. alloc_pages() can only
1708 * allocate memory in power-of-two pages.
1709 *
1710 * This function is also limited by MAX_ORDER.
1711 *
1712 * Memory allocated by this function must be released by free_pages_exact().
1713 */
1714void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
1715{
1716 unsigned int order = get_order(size);
1717 unsigned long addr;
1718
1719 addr = __get_free_pages(gfp_mask, order);
1720 if (addr) {
1721 unsigned long alloc_end = addr + (PAGE_SIZE << order);
1722 unsigned long used = addr + PAGE_ALIGN(size);
1723
1724 split_page(virt_to_page(addr), order);
1725 while (used < alloc_end) {
1726 free_page(used);
1727 used += PAGE_SIZE;
1728 }
1729 }
1730
1731 return (void *)addr;
1732}
1733EXPORT_SYMBOL(alloc_pages_exact);
1734
1735/**
1736 * free_pages_exact - release memory allocated via alloc_pages_exact()
1737 * @virt: the value returned by alloc_pages_exact.
1738 * @size: size of allocation, same value as passed to alloc_pages_exact().
1739 *
1740 * Release the memory allocated by a previous call to alloc_pages_exact.
1741 */
1742void free_pages_exact(void *virt, size_t size)
1743{
1744 unsigned long addr = (unsigned long)virt;
1745 unsigned long end = addr + PAGE_ALIGN(size);
1746
1747 while (addr < end) {
1748 free_page(addr);
1749 addr += PAGE_SIZE;
1750 }
1751}
1752EXPORT_SYMBOL(free_pages_exact);
1753
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754static unsigned int nr_free_zone_pages(int offset)
1755{
Mel Gormandd1a2392008-04-28 02:12:17 -07001756 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001757 struct zone *zone;
1758
Martin J. Blighe310fd42005-07-29 22:59:18 -07001759 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 unsigned int sum = 0;
1761
Mel Gorman0e884602008-04-28 02:12:14 -07001762 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763
Mel Gorman54a6eb52008-04-28 02:12:16 -07001764 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07001765 unsigned long size = zone->present_pages;
1766 unsigned long high = zone->pages_high;
1767 if (size > high)
1768 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 }
1770
1771 return sum;
1772}
1773
1774/*
1775 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1776 */
1777unsigned int nr_free_buffer_pages(void)
1778{
Al Viroaf4ca452005-10-21 02:55:38 -04001779 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780}
Meelap Shahc2f1a552007-07-17 04:04:39 -07001781EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782
1783/*
1784 * Amount of free RAM allocatable within all zones
1785 */
1786unsigned int nr_free_pagecache_pages(void)
1787{
Mel Gorman2a1e2742007-07-17 04:03:12 -07001788 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001790
1791static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001793 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08001794 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797void si_meminfo(struct sysinfo *val)
1798{
1799 val->totalram = totalram_pages;
1800 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001801 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 val->totalhigh = totalhigh_pages;
1804 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 val->mem_unit = PAGE_SIZE;
1806}
1807
1808EXPORT_SYMBOL(si_meminfo);
1809
1810#ifdef CONFIG_NUMA
1811void si_meminfo_node(struct sysinfo *val, int nid)
1812{
1813 pg_data_t *pgdat = NODE_DATA(nid);
1814
1815 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001816 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001817#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001819 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1820 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001821#else
1822 val->totalhigh = 0;
1823 val->freehigh = 0;
1824#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 val->mem_unit = PAGE_SIZE;
1826}
1827#endif
1828
1829#define K(x) ((x) << (PAGE_SHIFT-10))
1830
1831/*
1832 * Show free area list (used inside shift_scroll-lock stuff)
1833 * We also calculate the percentage fragmentation. We do this by counting the
1834 * memory on each free list with the exception of the first item on the list.
1835 */
1836void show_free_areas(void)
1837{
Jes Sorensenc7241912006-09-27 01:50:05 -07001838 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 struct zone *zone;
1840
1841 for_each_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07001842 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07001844
1845 show_node(zone);
1846 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
Dave Jones6b482c62005-11-10 15:45:56 -05001848 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 struct per_cpu_pageset *pageset;
1850
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001851 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001853 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1854 cpu, pageset->pcp.high,
1855 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 }
1857 }
1858
Andrew Mortona25700a2007-02-08 14:20:40 -08001859 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
Christoph Lameterd23ad422007-02-10 01:43:02 -08001860 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
Christoph Lameter65e458d42007-02-10 01:43:05 -08001861 global_page_state(NR_ACTIVE),
1862 global_page_state(NR_INACTIVE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07001863 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07001864 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001865 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08001866 global_page_state(NR_FREE_PAGES),
Christoph Lameter972d1a72006-09-25 23:31:51 -07001867 global_page_state(NR_SLAB_RECLAIMABLE) +
1868 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001869 global_page_state(NR_FILE_MAPPED),
Andrew Mortona25700a2007-02-08 14:20:40 -08001870 global_page_state(NR_PAGETABLE),
1871 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872
1873 for_each_zone(zone) {
1874 int i;
1875
Jes Sorensenc7241912006-09-27 01:50:05 -07001876 if (!populated_zone(zone))
1877 continue;
1878
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 show_node(zone);
1880 printk("%s"
1881 " free:%lukB"
1882 " min:%lukB"
1883 " low:%lukB"
1884 " high:%lukB"
1885 " active:%lukB"
1886 " inactive:%lukB"
1887 " present:%lukB"
1888 " pages_scanned:%lu"
1889 " all_unreclaimable? %s"
1890 "\n",
1891 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08001892 K(zone_page_state(zone, NR_FREE_PAGES)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 K(zone->pages_min),
1894 K(zone->pages_low),
1895 K(zone->pages_high),
Christoph Lameterc8785382007-02-10 01:43:01 -08001896 K(zone_page_state(zone, NR_ACTIVE)),
1897 K(zone_page_state(zone, NR_INACTIVE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 K(zone->present_pages),
1899 zone->pages_scanned,
David Rientjese815af92007-10-16 23:25:54 -07001900 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 );
1902 printk("lowmem_reserve[]:");
1903 for (i = 0; i < MAX_NR_ZONES; i++)
1904 printk(" %lu", zone->lowmem_reserve[i]);
1905 printk("\n");
1906 }
1907
1908 for_each_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001909 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910
Jes Sorensenc7241912006-09-27 01:50:05 -07001911 if (!populated_zone(zone))
1912 continue;
1913
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 show_node(zone);
1915 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
1917 spin_lock_irqsave(&zone->lock, flags);
1918 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001919 nr[order] = zone->free_area[order].nr_free;
1920 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 }
1922 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001923 for (order = 0; order < MAX_ORDER; order++)
1924 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 printk("= %lukB\n", K(total));
1926 }
1927
Larry Woodmane6f36022008-02-04 22:29:30 -08001928 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
1929
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 show_swap_cache_info();
1931}
1932
Mel Gorman19770b32008-04-28 02:12:18 -07001933static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
1934{
1935 zoneref->zone = zone;
1936 zoneref->zone_idx = zone_idx(zone);
1937}
1938
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939/*
1940 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001941 *
1942 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001944static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
1945 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946{
Christoph Lameter1a932052006-01-06 00:11:16 -08001947 struct zone *zone;
1948
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001949 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001950 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001951
1952 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001953 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08001954 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001955 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07001956 zoneref_set_zone(zone,
1957 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08001958 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08001960
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001961 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08001962 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963}
1964
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001965
1966/*
1967 * zonelist_order:
1968 * 0 = automatic detection of better ordering.
1969 * 1 = order by ([node] distance, -zonetype)
1970 * 2 = order by (-zonetype, [node] distance)
1971 *
1972 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1973 * the same zonelist. So only NUMA can configure this param.
1974 */
1975#define ZONELIST_ORDER_DEFAULT 0
1976#define ZONELIST_ORDER_NODE 1
1977#define ZONELIST_ORDER_ZONE 2
1978
1979/* zonelist order in the kernel.
1980 * set_zonelist_order() will set this to NODE or ZONE.
1981 */
1982static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
1983static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
1984
1985
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001987/* The value user specified ....changed by config */
1988static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1989/* string for sysctl */
1990#define NUMA_ZONELIST_ORDER_LEN 16
1991char numa_zonelist_order[16] = "default";
1992
1993/*
1994 * interface for configure zonelist ordering.
1995 * command line option "numa_zonelist_order"
1996 * = "[dD]efault - default, automatic configuration.
1997 * = "[nN]ode - order by node locality, then by zone within node
1998 * = "[zZ]one - order by zone, then by locality within zone
1999 */
2000
2001static int __parse_numa_zonelist_order(char *s)
2002{
2003 if (*s == 'd' || *s == 'D') {
2004 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2005 } else if (*s == 'n' || *s == 'N') {
2006 user_zonelist_order = ZONELIST_ORDER_NODE;
2007 } else if (*s == 'z' || *s == 'Z') {
2008 user_zonelist_order = ZONELIST_ORDER_ZONE;
2009 } else {
2010 printk(KERN_WARNING
2011 "Ignoring invalid numa_zonelist_order value: "
2012 "%s\n", s);
2013 return -EINVAL;
2014 }
2015 return 0;
2016}
2017
2018static __init int setup_numa_zonelist_order(char *s)
2019{
2020 if (s)
2021 return __parse_numa_zonelist_order(s);
2022 return 0;
2023}
2024early_param("numa_zonelist_order", setup_numa_zonelist_order);
2025
2026/*
2027 * sysctl handler for numa_zonelist_order
2028 */
2029int numa_zonelist_order_handler(ctl_table *table, int write,
2030 struct file *file, void __user *buffer, size_t *length,
2031 loff_t *ppos)
2032{
2033 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2034 int ret;
2035
2036 if (write)
2037 strncpy(saved_string, (char*)table->data,
2038 NUMA_ZONELIST_ORDER_LEN);
2039 ret = proc_dostring(table, write, file, buffer, length, ppos);
2040 if (ret)
2041 return ret;
2042 if (write) {
2043 int oldval = user_zonelist_order;
2044 if (__parse_numa_zonelist_order((char*)table->data)) {
2045 /*
2046 * bogus value. restore saved string
2047 */
2048 strncpy((char*)table->data, saved_string,
2049 NUMA_ZONELIST_ORDER_LEN);
2050 user_zonelist_order = oldval;
2051 } else if (oldval != user_zonelist_order)
2052 build_all_zonelists();
2053 }
2054 return 0;
2055}
2056
2057
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058#define MAX_NODE_LOAD (num_online_nodes())
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002059static int node_load[MAX_NUMNODES];
2060
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002062 * 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 -07002063 * @node: node whose fallback list we're appending
2064 * @used_node_mask: nodemask_t of already used nodes
2065 *
2066 * We use a number of factors to determine which is the next node that should
2067 * appear on a given node's fallback list. The node should not have appeared
2068 * already in @node's fallback list, and it should be the next closest node
2069 * according to the distance array (which contains arbitrary distance values
2070 * from each node to each node in the system), and should also prefer nodes
2071 * with no CPUs, since presumably they'll have very little allocation pressure
2072 * on them otherwise.
2073 * It returns -1 if no node is found.
2074 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002075static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002077 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 int min_val = INT_MAX;
2079 int best_node = -1;
Mike Travisc5f59f02008-04-04 18:11:10 -07002080 node_to_cpumask_ptr(tmp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002082 /* Use the local node if we haven't already */
2083 if (!node_isset(node, *used_node_mask)) {
2084 node_set(node, *used_node_mask);
2085 return node;
2086 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002088 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089
2090 /* Don't want a node to appear more than once */
2091 if (node_isset(n, *used_node_mask))
2092 continue;
2093
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 /* Use the distance array to find the distance */
2095 val = node_distance(node, n);
2096
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002097 /* Penalize nodes under us ("prefer the next node") */
2098 val += (n < node);
2099
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 /* Give preference to headless and unused nodes */
Mike Travisc5f59f02008-04-04 18:11:10 -07002101 node_to_cpumask_ptr_next(tmp, n);
2102 if (!cpus_empty(*tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 val += PENALTY_FOR_NODE_WITH_CPUS;
2104
2105 /* Slight preference for less loaded node */
2106 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2107 val += node_load[n];
2108
2109 if (val < min_val) {
2110 min_val = val;
2111 best_node = n;
2112 }
2113 }
2114
2115 if (best_node >= 0)
2116 node_set(best_node, *used_node_mask);
2117
2118 return best_node;
2119}
2120
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002121
2122/*
2123 * Build zonelists ordered by node and zones within node.
2124 * This results in maximum locality--normal zone overflows into local
2125 * DMA zone, if any--but risks exhausting DMA zone.
2126 */
2127static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002129 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002131
Mel Gorman54a6eb52008-04-28 02:12:16 -07002132 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002133 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002134 ;
2135 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2136 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002137 zonelist->_zonerefs[j].zone = NULL;
2138 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002139}
2140
2141/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002142 * Build gfp_thisnode zonelists
2143 */
2144static void build_thisnode_zonelists(pg_data_t *pgdat)
2145{
Christoph Lameter523b9452007-10-16 01:25:37 -07002146 int j;
2147 struct zonelist *zonelist;
2148
Mel Gorman54a6eb52008-04-28 02:12:16 -07002149 zonelist = &pgdat->node_zonelists[1];
2150 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002151 zonelist->_zonerefs[j].zone = NULL;
2152 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002153}
2154
2155/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002156 * Build zonelists ordered by zone and nodes within zones.
2157 * This results in conserving DMA zone[s] until all Normal memory is
2158 * exhausted, but results in overflowing to remote node while memory
2159 * may still exist in local DMA zone.
2160 */
2161static int node_order[MAX_NUMNODES];
2162
2163static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2164{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002165 int pos, j, node;
2166 int zone_type; /* needs to be signed */
2167 struct zone *z;
2168 struct zonelist *zonelist;
2169
Mel Gorman54a6eb52008-04-28 02:12:16 -07002170 zonelist = &pgdat->node_zonelists[0];
2171 pos = 0;
2172 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2173 for (j = 0; j < nr_nodes; j++) {
2174 node = node_order[j];
2175 z = &NODE_DATA(node)->node_zones[zone_type];
2176 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002177 zoneref_set_zone(z,
2178 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002179 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002180 }
2181 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002182 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002183 zonelist->_zonerefs[pos].zone = NULL;
2184 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002185}
2186
2187static int default_zonelist_order(void)
2188{
2189 int nid, zone_type;
2190 unsigned long low_kmem_size,total_size;
2191 struct zone *z;
2192 int average_size;
2193 /*
2194 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2195 * If they are really small and used heavily, the system can fall
2196 * into OOM very easily.
2197 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2198 */
2199 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2200 low_kmem_size = 0;
2201 total_size = 0;
2202 for_each_online_node(nid) {
2203 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2204 z = &NODE_DATA(nid)->node_zones[zone_type];
2205 if (populated_zone(z)) {
2206 if (zone_type < ZONE_NORMAL)
2207 low_kmem_size += z->present_pages;
2208 total_size += z->present_pages;
2209 }
2210 }
2211 }
2212 if (!low_kmem_size || /* there are no DMA area. */
2213 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2214 return ZONELIST_ORDER_NODE;
2215 /*
2216 * look into each node's config.
2217 * If there is a node whose DMA/DMA32 memory is very big area on
2218 * local memory, NODE_ORDER may be suitable.
2219 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002220 average_size = total_size /
2221 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002222 for_each_online_node(nid) {
2223 low_kmem_size = 0;
2224 total_size = 0;
2225 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2226 z = &NODE_DATA(nid)->node_zones[zone_type];
2227 if (populated_zone(z)) {
2228 if (zone_type < ZONE_NORMAL)
2229 low_kmem_size += z->present_pages;
2230 total_size += z->present_pages;
2231 }
2232 }
2233 if (low_kmem_size &&
2234 total_size > average_size && /* ignore small node */
2235 low_kmem_size > total_size * 70/100)
2236 return ZONELIST_ORDER_NODE;
2237 }
2238 return ZONELIST_ORDER_ZONE;
2239}
2240
2241static void set_zonelist_order(void)
2242{
2243 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2244 current_zonelist_order = default_zonelist_order();
2245 else
2246 current_zonelist_order = user_zonelist_order;
2247}
2248
2249static void build_zonelists(pg_data_t *pgdat)
2250{
2251 int j, node, load;
2252 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002254 int local_node, prev_node;
2255 struct zonelist *zonelist;
2256 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257
2258 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002259 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07002261 zonelist->_zonerefs[0].zone = NULL;
2262 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 }
2264
2265 /* NUMA-aware ordering of nodes */
2266 local_node = pgdat->node_id;
2267 load = num_online_nodes();
2268 prev_node = local_node;
2269 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002270
2271 memset(node_load, 0, sizeof(node_load));
2272 memset(node_order, 0, sizeof(node_order));
2273 j = 0;
2274
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002276 int distance = node_distance(local_node, node);
2277
2278 /*
2279 * If another node is sufficiently far away then it is better
2280 * to reclaim pages in a zone before going off node.
2281 */
2282 if (distance > RECLAIM_DISTANCE)
2283 zone_reclaim_mode = 1;
2284
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 /*
2286 * We don't want to pressure a particular node.
2287 * So adding penalty to the first node in same
2288 * distance group to make it round-robin.
2289 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002290 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002291 node_load[node] = load;
2292
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 prev_node = node;
2294 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002295 if (order == ZONELIST_ORDER_NODE)
2296 build_zonelists_in_node_order(pgdat, node);
2297 else
2298 node_order[j++] = node; /* remember order */
2299 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002301 if (order == ZONELIST_ORDER_ZONE) {
2302 /* calculate node order -- i.e., DMA last! */
2303 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002305
2306 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307}
2308
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002309/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002310static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002311{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002312 struct zonelist *zonelist;
2313 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07002314 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002315
Mel Gorman54a6eb52008-04-28 02:12:16 -07002316 zonelist = &pgdat->node_zonelists[0];
2317 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2318 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07002319 for (z = zonelist->_zonerefs; z->zone; z++)
2320 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002321}
2322
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002323
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324#else /* CONFIG_NUMA */
2325
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002326static void set_zonelist_order(void)
2327{
2328 current_zonelist_order = ZONELIST_ORDER_ZONE;
2329}
2330
2331static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332{
Christoph Lameter19655d32006-09-25 23:31:19 -07002333 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002334 enum zone_type j;
2335 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336
2337 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338
Mel Gorman54a6eb52008-04-28 02:12:16 -07002339 zonelist = &pgdat->node_zonelists[0];
2340 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
Mel Gorman54a6eb52008-04-28 02:12:16 -07002342 /*
2343 * Now we build the zonelist so that it contains the zones
2344 * of all the other nodes.
2345 * We don't want to pressure a particular node, so when
2346 * building the zones for node N, we make sure that the
2347 * zones coming right after the local ones are those from
2348 * node N+1 (modulo N)
2349 */
2350 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2351 if (!node_online(node))
2352 continue;
2353 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2354 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002356 for (node = 0; node < local_node; node++) {
2357 if (!node_online(node))
2358 continue;
2359 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2360 MAX_NR_ZONES - 1);
2361 }
2362
Mel Gormandd1a2392008-04-28 02:12:17 -07002363 zonelist->_zonerefs[j].zone = NULL;
2364 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365}
2366
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002367/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002368static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002369{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002370 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002371}
2372
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373#endif /* CONFIG_NUMA */
2374
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002375/* return values int ....just for stop_machine() */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002376static int __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377{
Yasunori Goto68113782006-06-23 02:03:11 -07002378 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002379
2380 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07002381 pg_data_t *pgdat = NODE_DATA(nid);
2382
2383 build_zonelists(pgdat);
2384 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002385 }
Yasunori Goto68113782006-06-23 02:03:11 -07002386 return 0;
2387}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002389void build_all_zonelists(void)
Yasunori Goto68113782006-06-23 02:03:11 -07002390{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002391 set_zonelist_order();
2392
Yasunori Goto68113782006-06-23 02:03:11 -07002393 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07002394 __build_all_zonelists(NULL);
Mel Gorman68ad8df2008-07-23 21:26:52 -07002395 mminit_verify_zonelist();
Yasunori Goto68113782006-06-23 02:03:11 -07002396 cpuset_init_current_mems_allowed();
2397 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02002398 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07002399 of zonelist */
Rusty Russell9b1a4d32008-07-28 12:16:30 -05002400 stop_machine(__build_all_zonelists, NULL, NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07002401 /* cpuset refresh routine should be here */
2402 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07002403 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002404 /*
2405 * Disable grouping by mobility if the number of pages in the
2406 * system is too low to allow the mechanism to work. It would be
2407 * more accurate, but expensive to check per-zone. This check is
2408 * made on memory-hotadd so a system can start with mobility
2409 * disabled and enable it later
2410 */
Mel Gormand9c23402007-10-16 01:26:01 -07002411 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002412 page_group_by_mobility_disabled = 1;
2413 else
2414 page_group_by_mobility_disabled = 0;
2415
2416 printk("Built %i zonelists in %s order, mobility grouping %s. "
2417 "Total pages: %ld\n",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002418 num_online_nodes(),
2419 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002420 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002421 vm_total_pages);
2422#ifdef CONFIG_NUMA
2423 printk("Policy zone: %s\n", zone_names[policy_zone]);
2424#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425}
2426
2427/*
2428 * Helper functions to size the waitqueue hash table.
2429 * Essentially these want to choose hash table sizes sufficiently
2430 * large so that collisions trying to wait on pages are rare.
2431 * But in fact, the number of active page waitqueues on typical
2432 * systems is ridiculously low, less than 200. So this is even
2433 * conservative, even though it seems large.
2434 *
2435 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2436 * waitqueues, i.e. the size of the waitq table given the number of pages.
2437 */
2438#define PAGES_PER_WAITQUEUE 256
2439
Yasunori Gotocca448f2006-06-23 02:03:10 -07002440#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07002441static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442{
2443 unsigned long size = 1;
2444
2445 pages /= PAGES_PER_WAITQUEUE;
2446
2447 while (size < pages)
2448 size <<= 1;
2449
2450 /*
2451 * Once we have dozens or even hundreds of threads sleeping
2452 * on IO we've got bigger problems than wait queue collision.
2453 * Limit the size of the wait table to a reasonable size.
2454 */
2455 size = min(size, 4096UL);
2456
2457 return max(size, 4UL);
2458}
Yasunori Gotocca448f2006-06-23 02:03:10 -07002459#else
2460/*
2461 * A zone's size might be changed by hot-add, so it is not possible to determine
2462 * a suitable size for its wait_table. So we use the maximum size now.
2463 *
2464 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2465 *
2466 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2467 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2468 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2469 *
2470 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2471 * or more by the traditional way. (See above). It equals:
2472 *
2473 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2474 * ia64(16K page size) : = ( 8G + 4M)byte.
2475 * powerpc (64K page size) : = (32G +16M)byte.
2476 */
2477static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2478{
2479 return 4096UL;
2480}
2481#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482
2483/*
2484 * This is an integer logarithm so that shifts can be used later
2485 * to extract the more random high bits from the multiplicative
2486 * hash function before the remainder is taken.
2487 */
2488static inline unsigned long wait_table_bits(unsigned long size)
2489{
2490 return ffz(~size);
2491}
2492
2493#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2494
Mel Gorman56fd56b2007-10-16 01:25:58 -07002495/*
Mel Gormand9c23402007-10-16 01:26:01 -07002496 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman56fd56b2007-10-16 01:25:58 -07002497 * of blocks reserved is based on zone->pages_min. The memory within the
2498 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2499 * higher will lead to a bigger reserve which will get freed as contiguous
2500 * blocks as reclaim kicks in
2501 */
2502static void setup_zone_migrate_reserve(struct zone *zone)
2503{
2504 unsigned long start_pfn, pfn, end_pfn;
2505 struct page *page;
2506 unsigned long reserve, block_migratetype;
2507
2508 /* Get the start pfn, end pfn and the number of blocks to reserve */
2509 start_pfn = zone->zone_start_pfn;
2510 end_pfn = start_pfn + zone->spanned_pages;
Mel Gormand9c23402007-10-16 01:26:01 -07002511 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2512 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002513
Mel Gormand9c23402007-10-16 01:26:01 -07002514 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07002515 if (!pfn_valid(pfn))
2516 continue;
2517 page = pfn_to_page(pfn);
2518
2519 /* Blocks with reserved pages will never free, skip them. */
2520 if (PageReserved(page))
2521 continue;
2522
2523 block_migratetype = get_pageblock_migratetype(page);
2524
2525 /* If this block is reserved, account for it */
2526 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2527 reserve--;
2528 continue;
2529 }
2530
2531 /* Suitable for reserving if this block is movable */
2532 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2533 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2534 move_freepages_block(zone, page, MIGRATE_RESERVE);
2535 reserve--;
2536 continue;
2537 }
2538
2539 /*
2540 * If the reserve is met and this is a previous reserved block,
2541 * take it back
2542 */
2543 if (block_migratetype == MIGRATE_RESERVE) {
2544 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2545 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2546 }
2547 }
2548}
Mel Gormanac0e5b72007-10-16 01:25:58 -07002549
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550/*
2551 * Initially all pages are reserved - free ones are freed
2552 * up by free_all_bootmem() once the early boot process is
2553 * done. Non-atomic initialization, single-pass.
2554 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002555void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa02007-01-10 23:15:30 -08002556 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07002559 unsigned long end_pfn = start_pfn + size;
2560 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002561 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002563 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08002564 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa02007-01-10 23:15:30 -08002565 /*
2566 * There can be holes in boot-time mem_map[]s
2567 * handed to this function. They do not
2568 * exist on hotplugged memory.
2569 */
2570 if (context == MEMMAP_EARLY) {
2571 if (!early_pfn_valid(pfn))
2572 continue;
2573 if (!early_pfn_in_nid(pfn, nid))
2574 continue;
2575 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002576 page = pfn_to_page(pfn);
2577 set_page_links(page, zone, nid, pfn);
Mel Gorman708614e2008-07-23 21:26:51 -07002578 mminit_verify_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08002579 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 reset_page_mapcount(page);
2581 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002582 /*
2583 * Mark the block movable so that blocks are reserved for
2584 * movable at startup. This will force kernel allocations
2585 * to reserve their blocks rather than leaking throughout
2586 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07002587 * kernel allocations are made. Later some blocks near
2588 * the start are marked MIGRATE_RESERVE by
2589 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002590 *
2591 * bitmap is created for zone's valid pfn range. but memmap
2592 * can be created for invalid pages (for alignment)
2593 * check here not to call set_pageblock_migratetype() against
2594 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002595 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002596 if ((z->zone_start_pfn <= pfn)
2597 && (pfn < z->zone_start_pfn + z->spanned_pages)
2598 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07002599 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002600
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 INIT_LIST_HEAD(&page->lru);
2602#ifdef WANT_PAGE_VIRTUAL
2603 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2604 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07002605 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 }
2608}
2609
Andi Kleen1e548de2008-02-04 22:29:26 -08002610static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002612 int order, t;
2613 for_each_migratetype_order(order, t) {
2614 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 zone->free_area[order].nr_free = 0;
2616 }
2617}
2618
2619#ifndef __HAVE_ARCH_MEMMAP_INIT
2620#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa02007-01-10 23:15:30 -08002621 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622#endif
2623
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08002624static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002625{
2626 int batch;
2627
2628 /*
2629 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07002630 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002631 *
2632 * OK, so we don't know how big the cache is. So guess.
2633 */
2634 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07002635 if (batch * PAGE_SIZE > 512 * 1024)
2636 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002637 batch /= 4; /* We effectively *= 4 below */
2638 if (batch < 1)
2639 batch = 1;
2640
2641 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002642 * Clamp the batch to a 2^n - 1 value. Having a power
2643 * of 2 value was found to be more likely to have
2644 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002645 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002646 * For example if 2 tasks are alternately allocating
2647 * batches of pages, one task can end up with a lot
2648 * of pages of one half of the possible page colors
2649 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002650 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002651 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07002652
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002653 return batch;
2654}
2655
Adrian Bunkb69a7282008-07-23 21:28:12 -07002656static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
Christoph Lameter2caaad42005-06-21 17:15:00 -07002657{
2658 struct per_cpu_pages *pcp;
2659
Magnus Damm1c6fe942005-10-26 01:58:59 -07002660 memset(p, 0, sizeof(*p));
2661
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002662 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002663 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002664 pcp->high = 6 * batch;
2665 pcp->batch = max(1UL, 1 * batch);
2666 INIT_LIST_HEAD(&pcp->list);
Christoph Lameter2caaad42005-06-21 17:15:00 -07002667}
2668
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002669/*
2670 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2671 * to the value high for the pageset p.
2672 */
2673
2674static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2675 unsigned long high)
2676{
2677 struct per_cpu_pages *pcp;
2678
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002679 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002680 pcp->high = high;
2681 pcp->batch = max(1UL, high/4);
2682 if ((high/4) > (PAGE_SHIFT * 8))
2683 pcp->batch = PAGE_SHIFT * 8;
2684}
2685
2686
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002687#ifdef CONFIG_NUMA
2688/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07002689 * Boot pageset table. One per cpu which is going to be used for all
2690 * zones and all nodes. The parameters will be set in such a way
2691 * that an item put on a list will immediately be handed over to
2692 * the buddy list. This is safe since pageset manipulation is done
2693 * with interrupts disabled.
2694 *
2695 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07002696 *
2697 * The boot_pagesets must be kept even after bootup is complete for
2698 * unused processors and/or zones. They do play a role for bootstrapping
2699 * hotplugged processors.
2700 *
2701 * zoneinfo_show() and maybe other functions do
2702 * not check if the processor is online before following the pageset pointer.
2703 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07002704 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08002705static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07002706
2707/*
2708 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002709 * per cpu pageset array in struct zone.
2710 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08002711static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002712{
2713 struct zone *zone, *dzone;
Christoph Lameter37c07082007-10-16 01:25:36 -07002714 int node = cpu_to_node(cpu);
2715
2716 node_set_state(node, N_CPU); /* this node has a cpu */
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002717
2718 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002719
Christoph Lameter66a55032006-09-27 01:50:09 -07002720 if (!populated_zone(zone))
2721 continue;
2722
Nick Piggin23316bc2006-01-08 01:00:41 -08002723 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lameter37c07082007-10-16 01:25:36 -07002724 GFP_KERNEL, node);
Nick Piggin23316bc2006-01-08 01:00:41 -08002725 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002726 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002727
Nick Piggin23316bc2006-01-08 01:00:41 -08002728 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002729
2730 if (percpu_pagelist_fraction)
2731 setup_pagelist_highmark(zone_pcp(zone, cpu),
2732 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002733 }
2734
2735 return 0;
2736bad:
2737 for_each_zone(dzone) {
Andrew Morton64191682007-08-30 23:56:17 -07002738 if (!populated_zone(dzone))
2739 continue;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002740 if (dzone == zone)
2741 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08002742 kfree(zone_pcp(dzone, cpu));
2743 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002744 }
2745 return -ENOMEM;
2746}
2747
2748static inline void free_zone_pagesets(int cpu)
2749{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002750 struct zone *zone;
2751
2752 for_each_zone(zone) {
2753 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2754
David Rientjesf3ef9ea2006-09-25 16:24:57 -07002755 /* Free per_cpu_pageset if it is slab allocated */
2756 if (pset != &boot_pageset[cpu])
2757 kfree(pset);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002758 zone_pcp(zone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002759 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002760}
2761
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07002762static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002763 unsigned long action,
2764 void *hcpu)
2765{
2766 int cpu = (long)hcpu;
2767 int ret = NOTIFY_OK;
2768
2769 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08002770 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002771 case CPU_UP_PREPARE_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002772 if (process_zones(cpu))
2773 ret = NOTIFY_BAD;
2774 break;
2775 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002776 case CPU_UP_CANCELED_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002777 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002778 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002779 free_zone_pagesets(cpu);
2780 break;
2781 default:
2782 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002783 }
2784 return ret;
2785}
2786
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07002787static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002788 { &pageset_cpuup_callback, NULL, 0 };
2789
Al Viro78d99552005-12-15 09:18:25 +00002790void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002791{
2792 int err;
2793
2794 /* Initialize per_cpu_pageset for cpu 0.
2795 * A cpuup callback will do this for every cpu
2796 * as it comes online
2797 */
2798 err = process_zones(smp_processor_id());
2799 BUG_ON(err);
2800 register_cpu_notifier(&pageset_notifier);
2801}
2802
2803#endif
2804
Sam Ravnborg577a32f2007-05-17 23:29:25 +02002805static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07002806int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07002807{
2808 int i;
2809 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002810 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07002811
2812 /*
2813 * The per-page waitqueue mechanism uses hashed waitqueues
2814 * per zone.
2815 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07002816 zone->wait_table_hash_nr_entries =
2817 wait_table_hash_nr_entries(zone_size_pages);
2818 zone->wait_table_bits =
2819 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002820 alloc_size = zone->wait_table_hash_nr_entries
2821 * sizeof(wait_queue_head_t);
2822
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07002823 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07002824 zone->wait_table = (wait_queue_head_t *)
2825 alloc_bootmem_node(pgdat, alloc_size);
2826 } else {
2827 /*
2828 * This case means that a zone whose size was 0 gets new memory
2829 * via memory hot-add.
2830 * But it may be the case that a new node was hot-added. In
2831 * this case vmalloc() will not be able to use this new node's
2832 * memory - this wait_table must be initialized to use this new
2833 * node itself as well.
2834 * To use this new node's memory, further consideration will be
2835 * necessary.
2836 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07002837 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002838 }
2839 if (!zone->wait_table)
2840 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07002841
Yasunori Goto02b694d2006-06-23 02:03:08 -07002842 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07002843 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002844
2845 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002846}
2847
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002848static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07002849{
2850 int cpu;
2851 unsigned long batch = zone_batchsize(zone);
2852
2853 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2854#ifdef CONFIG_NUMA
2855 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002856 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002857 setup_pageset(&boot_pageset[cpu],0);
2858#else
2859 setup_pageset(zone_pcp(zone,cpu), batch);
2860#endif
2861 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08002862 if (zone->present_pages)
2863 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2864 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07002865}
2866
Yasunori Goto718127c2006-06-23 02:03:10 -07002867__meminit int init_currently_empty_zone(struct zone *zone,
2868 unsigned long zone_start_pfn,
Dave Hansena2f3aa02007-01-10 23:15:30 -08002869 unsigned long size,
2870 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07002871{
2872 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002873 int ret;
2874 ret = zone_wait_table_init(zone, size);
2875 if (ret)
2876 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07002877 pgdat->nr_zones = zone_idx(zone) + 1;
2878
Dave Hansened8ece22005-10-29 18:16:50 -07002879 zone->zone_start_pfn = zone_start_pfn;
2880
Mel Gorman708614e2008-07-23 21:26:51 -07002881 mminit_dprintk(MMINIT_TRACE, "memmap_init",
2882 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2883 pgdat->node_id,
2884 (unsigned long)zone_idx(zone),
2885 zone_start_pfn, (zone_start_pfn + size));
2886
Andi Kleen1e548de2008-02-04 22:29:26 -08002887 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07002888
2889 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002890}
2891
Mel Gormanc7132162006-09-27 01:49:43 -07002892#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2893/*
2894 * Basic iterator support. Return the first range of PFNs for a node
2895 * Note: nid == MAX_NUMNODES returns first region regardless of node
2896 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002897static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07002898{
2899 int i;
2900
2901 for (i = 0; i < nr_nodemap_entries; i++)
2902 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2903 return i;
2904
2905 return -1;
2906}
2907
2908/*
2909 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02002910 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07002911 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002912static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07002913{
2914 for (index = index + 1; index < nr_nodemap_entries; index++)
2915 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2916 return index;
2917
2918 return -1;
2919}
2920
2921#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2922/*
2923 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2924 * Architectures may implement their own version but if add_active_range()
2925 * was used and there are no special requirements, this is a convenient
2926 * alternative
2927 */
Stephen Rothwell6f076f52007-05-10 03:15:27 -07002928int __meminit early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07002929{
2930 int i;
2931
2932 for (i = 0; i < nr_nodemap_entries; i++) {
2933 unsigned long start_pfn = early_node_map[i].start_pfn;
2934 unsigned long end_pfn = early_node_map[i].end_pfn;
2935
2936 if (start_pfn <= pfn && pfn < end_pfn)
2937 return early_node_map[i].nid;
2938 }
2939
2940 return 0;
2941}
2942#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2943
2944/* Basic iterator support to walk early_node_map[] */
2945#define for_each_active_range_index_in_nid(i, nid) \
2946 for (i = first_active_region_index_in_nid(nid); i != -1; \
2947 i = next_active_region_index_in_nid(i, nid))
2948
2949/**
2950 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002951 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2952 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07002953 *
2954 * If an architecture guarantees that all ranges registered with
2955 * add_active_ranges() contain no holes and may be freed, this
2956 * this function may be used instead of calling free_bootmem() manually.
2957 */
2958void __init free_bootmem_with_active_regions(int nid,
2959 unsigned long max_low_pfn)
2960{
2961 int i;
2962
2963 for_each_active_range_index_in_nid(i, nid) {
2964 unsigned long size_pages = 0;
2965 unsigned long end_pfn = early_node_map[i].end_pfn;
2966
2967 if (early_node_map[i].start_pfn >= max_low_pfn)
2968 continue;
2969
2970 if (end_pfn > max_low_pfn)
2971 end_pfn = max_low_pfn;
2972
2973 size_pages = end_pfn - early_node_map[i].start_pfn;
2974 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2975 PFN_PHYS(early_node_map[i].start_pfn),
2976 size_pages << PAGE_SHIFT);
2977 }
2978}
2979
Yinghai Lub5bc6c02008-06-14 18:32:52 -07002980void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
2981{
2982 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07002983 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07002984
Yinghai Lud52d53b2008-06-16 20:10:55 -07002985 for_each_active_range_index_in_nid(i, nid) {
2986 ret = work_fn(early_node_map[i].start_pfn,
2987 early_node_map[i].end_pfn, data);
2988 if (ret)
2989 break;
2990 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07002991}
Mel Gormanc7132162006-09-27 01:49:43 -07002992/**
2993 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002994 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07002995 *
2996 * If an architecture guarantees that all ranges registered with
2997 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002998 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07002999 */
3000void __init sparse_memory_present_with_active_regions(int nid)
3001{
3002 int i;
3003
3004 for_each_active_range_index_in_nid(i, nid)
3005 memory_present(early_node_map[i].nid,
3006 early_node_map[i].start_pfn,
3007 early_node_map[i].end_pfn);
3008}
3009
3010/**
Mel Gormanfb014392006-09-27 01:49:59 -07003011 * push_node_boundaries - Push node boundaries to at least the requested boundary
3012 * @nid: The nid of the node to push the boundary for
3013 * @start_pfn: The start pfn of the node
3014 * @end_pfn: The end pfn of the node
3015 *
3016 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
3017 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
3018 * be hotplugged even though no physical memory exists. This function allows
3019 * an arch to push out the node boundaries so mem_map is allocated that can
3020 * be used later.
3021 */
3022#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3023void __init push_node_boundaries(unsigned int nid,
3024 unsigned long start_pfn, unsigned long end_pfn)
3025{
Mel Gorman6b74ab92008-07-23 21:26:49 -07003026 mminit_dprintk(MMINIT_TRACE, "zoneboundary",
3027 "Entering push_node_boundaries(%u, %lu, %lu)\n",
Mel Gormanfb014392006-09-27 01:49:59 -07003028 nid, start_pfn, end_pfn);
3029
3030 /* Initialise the boundary for this node if necessary */
3031 if (node_boundary_end_pfn[nid] == 0)
3032 node_boundary_start_pfn[nid] = -1UL;
3033
3034 /* Update the boundaries */
3035 if (node_boundary_start_pfn[nid] > start_pfn)
3036 node_boundary_start_pfn[nid] = start_pfn;
3037 if (node_boundary_end_pfn[nid] < end_pfn)
3038 node_boundary_end_pfn[nid] = end_pfn;
3039}
3040
3041/* If necessary, push the node boundary out for reserve hotadd */
Jan Beulich98011f52007-07-15 23:38:17 -07003042static void __meminit account_node_boundary(unsigned int nid,
Mel Gormanfb014392006-09-27 01:49:59 -07003043 unsigned long *start_pfn, unsigned long *end_pfn)
3044{
Mel Gorman6b74ab92008-07-23 21:26:49 -07003045 mminit_dprintk(MMINIT_TRACE, "zoneboundary",
3046 "Entering account_node_boundary(%u, %lu, %lu)\n",
Mel Gormanfb014392006-09-27 01:49:59 -07003047 nid, *start_pfn, *end_pfn);
3048
3049 /* Return if boundary information has not been provided */
3050 if (node_boundary_end_pfn[nid] == 0)
3051 return;
3052
3053 /* Check the boundaries and update if necessary */
3054 if (node_boundary_start_pfn[nid] < *start_pfn)
3055 *start_pfn = node_boundary_start_pfn[nid];
3056 if (node_boundary_end_pfn[nid] > *end_pfn)
3057 *end_pfn = node_boundary_end_pfn[nid];
3058}
3059#else
3060void __init push_node_boundaries(unsigned int nid,
3061 unsigned long start_pfn, unsigned long end_pfn) {}
3062
Jan Beulich98011f52007-07-15 23:38:17 -07003063static void __meminit account_node_boundary(unsigned int nid,
Mel Gormanfb014392006-09-27 01:49:59 -07003064 unsigned long *start_pfn, unsigned long *end_pfn) {}
3065#endif
3066
3067
3068/**
Mel Gormanc7132162006-09-27 01:49:43 -07003069 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003070 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3071 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3072 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003073 *
3074 * It returns the start and end page frame of a node based on information
3075 * provided by an arch calling add_active_range(). If called for a node
3076 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003077 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003078 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003079void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003080 unsigned long *start_pfn, unsigned long *end_pfn)
3081{
3082 int i;
3083 *start_pfn = -1UL;
3084 *end_pfn = 0;
3085
3086 for_each_active_range_index_in_nid(i, nid) {
3087 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3088 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3089 }
3090
Christoph Lameter633c0662007-10-16 01:25:37 -07003091 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003092 *start_pfn = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07003093
3094 /* Push the node boundaries out if requested */
3095 account_node_boundary(nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003096}
3097
3098/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003099 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3100 * assumption is made that zones within a node are ordered in monotonic
3101 * increasing memory addresses so that the "highest" populated zone is used
3102 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003103static void __init find_usable_zone_for_movable(void)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003104{
3105 int zone_index;
3106 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3107 if (zone_index == ZONE_MOVABLE)
3108 continue;
3109
3110 if (arch_zone_highest_possible_pfn[zone_index] >
3111 arch_zone_lowest_possible_pfn[zone_index])
3112 break;
3113 }
3114
3115 VM_BUG_ON(zone_index == -1);
3116 movable_zone = zone_index;
3117}
3118
3119/*
3120 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3121 * because it is sized independant of architecture. Unlike the other zones,
3122 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3123 * in each node depending on the size of each node and how evenly kernelcore
3124 * is distributed. This helper function adjusts the zone ranges
3125 * provided by the architecture for a given node by using the end of the
3126 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3127 * zones within a node are in order of monotonic increases memory addresses
3128 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003129static void __meminit adjust_zone_range_for_zone_movable(int nid,
Mel Gorman2a1e2742007-07-17 04:03:12 -07003130 unsigned long zone_type,
3131 unsigned long node_start_pfn,
3132 unsigned long node_end_pfn,
3133 unsigned long *zone_start_pfn,
3134 unsigned long *zone_end_pfn)
3135{
3136 /* Only adjust if ZONE_MOVABLE is on this node */
3137 if (zone_movable_pfn[nid]) {
3138 /* Size ZONE_MOVABLE */
3139 if (zone_type == ZONE_MOVABLE) {
3140 *zone_start_pfn = zone_movable_pfn[nid];
3141 *zone_end_pfn = min(node_end_pfn,
3142 arch_zone_highest_possible_pfn[movable_zone]);
3143
3144 /* Adjust for ZONE_MOVABLE starting within this range */
3145 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3146 *zone_end_pfn > zone_movable_pfn[nid]) {
3147 *zone_end_pfn = zone_movable_pfn[nid];
3148
3149 /* Check if this whole range is within ZONE_MOVABLE */
3150 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3151 *zone_start_pfn = *zone_end_pfn;
3152 }
3153}
3154
3155/*
Mel Gormanc7132162006-09-27 01:49:43 -07003156 * Return the number of pages a zone spans in a node, including holes
3157 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3158 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003159static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003160 unsigned long zone_type,
3161 unsigned long *ignored)
3162{
3163 unsigned long node_start_pfn, node_end_pfn;
3164 unsigned long zone_start_pfn, zone_end_pfn;
3165
3166 /* Get the start and end of the node and zone */
3167 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3168 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3169 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003170 adjust_zone_range_for_zone_movable(nid, zone_type,
3171 node_start_pfn, node_end_pfn,
3172 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003173
3174 /* Check that this node has pages within the zone's required range */
3175 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3176 return 0;
3177
3178 /* Move the zone boundaries inside the node if necessary */
3179 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3180 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3181
3182 /* Return the spanned pages */
3183 return zone_end_pfn - zone_start_pfn;
3184}
3185
3186/*
3187 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003188 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003189 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003190static unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003191 unsigned long range_start_pfn,
3192 unsigned long range_end_pfn)
3193{
3194 int i = 0;
3195 unsigned long prev_end_pfn = 0, hole_pages = 0;
3196 unsigned long start_pfn;
3197
3198 /* Find the end_pfn of the first active range of pfns in the node */
3199 i = first_active_region_index_in_nid(nid);
3200 if (i == -1)
3201 return 0;
3202
Mel Gormanb5445f92007-07-26 10:41:18 -07003203 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3204
Mel Gorman9c7cd682006-09-27 01:49:58 -07003205 /* Account for ranges before physical memory on this node */
3206 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003207 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003208
3209 /* Find all holes for the zone within the node */
3210 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3211
3212 /* No need to continue if prev_end_pfn is outside the zone */
3213 if (prev_end_pfn >= range_end_pfn)
3214 break;
3215
3216 /* Make sure the end of the zone is not within the hole */
3217 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3218 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3219
3220 /* Update the hole size cound and move on */
3221 if (start_pfn > range_start_pfn) {
3222 BUG_ON(prev_end_pfn > start_pfn);
3223 hole_pages += start_pfn - prev_end_pfn;
3224 }
3225 prev_end_pfn = early_node_map[i].end_pfn;
3226 }
3227
Mel Gorman9c7cd682006-09-27 01:49:58 -07003228 /* Account for ranges past physical memory on this node */
3229 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003230 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003231 max(range_start_pfn, prev_end_pfn);
3232
Mel Gormanc7132162006-09-27 01:49:43 -07003233 return hole_pages;
3234}
3235
3236/**
3237 * absent_pages_in_range - Return number of page frames in holes within a range
3238 * @start_pfn: The start PFN to start searching for holes
3239 * @end_pfn: The end PFN to stop searching for holes
3240 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003241 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003242 */
3243unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3244 unsigned long end_pfn)
3245{
3246 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3247}
3248
3249/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003250static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003251 unsigned long zone_type,
3252 unsigned long *ignored)
3253{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003254 unsigned long node_start_pfn, node_end_pfn;
3255 unsigned long zone_start_pfn, zone_end_pfn;
3256
3257 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3258 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3259 node_start_pfn);
3260 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3261 node_end_pfn);
3262
Mel Gorman2a1e2742007-07-17 04:03:12 -07003263 adjust_zone_range_for_zone_movable(nid, zone_type,
3264 node_start_pfn, node_end_pfn,
3265 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07003266 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003267}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003268
Mel Gormanc7132162006-09-27 01:49:43 -07003269#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07003270static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003271 unsigned long zone_type,
3272 unsigned long *zones_size)
3273{
3274 return zones_size[zone_type];
3275}
3276
Paul Mundt6ea6e682007-07-15 23:38:20 -07003277static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003278 unsigned long zone_type,
3279 unsigned long *zholes_size)
3280{
3281 if (!zholes_size)
3282 return 0;
3283
3284 return zholes_size[zone_type];
3285}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003286
Mel Gormanc7132162006-09-27 01:49:43 -07003287#endif
3288
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003289static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07003290 unsigned long *zones_size, unsigned long *zholes_size)
3291{
3292 unsigned long realtotalpages, totalpages = 0;
3293 enum zone_type i;
3294
3295 for (i = 0; i < MAX_NR_ZONES; i++)
3296 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3297 zones_size);
3298 pgdat->node_spanned_pages = totalpages;
3299
3300 realtotalpages = totalpages;
3301 for (i = 0; i < MAX_NR_ZONES; i++)
3302 realtotalpages -=
3303 zone_absent_pages_in_node(pgdat->node_id, i,
3304 zholes_size);
3305 pgdat->node_present_pages = realtotalpages;
3306 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3307 realtotalpages);
3308}
3309
Mel Gorman835c1342007-10-16 01:25:47 -07003310#ifndef CONFIG_SPARSEMEM
3311/*
3312 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07003313 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3314 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07003315 * round what is now in bits to nearest long in bits, then return it in
3316 * bytes.
3317 */
3318static unsigned long __init usemap_size(unsigned long zonesize)
3319{
3320 unsigned long usemapsize;
3321
Mel Gormand9c23402007-10-16 01:26:01 -07003322 usemapsize = roundup(zonesize, pageblock_nr_pages);
3323 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07003324 usemapsize *= NR_PAGEBLOCK_BITS;
3325 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3326
3327 return usemapsize / 8;
3328}
3329
3330static void __init setup_usemap(struct pglist_data *pgdat,
3331 struct zone *zone, unsigned long zonesize)
3332{
3333 unsigned long usemapsize = usemap_size(zonesize);
3334 zone->pageblock_flags = NULL;
3335 if (usemapsize) {
3336 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3337 memset(zone->pageblock_flags, 0, usemapsize);
3338 }
3339}
3340#else
3341static void inline setup_usemap(struct pglist_data *pgdat,
3342 struct zone *zone, unsigned long zonesize) {}
3343#endif /* CONFIG_SPARSEMEM */
3344
Mel Gormand9c23402007-10-16 01:26:01 -07003345#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08003346
3347/* Return a sensible default order for the pageblock size. */
3348static inline int pageblock_default_order(void)
3349{
3350 if (HPAGE_SHIFT > PAGE_SHIFT)
3351 return HUGETLB_PAGE_ORDER;
3352
3353 return MAX_ORDER-1;
3354}
3355
Mel Gormand9c23402007-10-16 01:26:01 -07003356/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3357static inline void __init set_pageblock_order(unsigned int order)
3358{
3359 /* Check that pageblock_nr_pages has not already been setup */
3360 if (pageblock_order)
3361 return;
3362
3363 /*
3364 * Assume the largest contiguous order of interest is a huge page.
3365 * This value may be variable depending on boot parameters on IA64
3366 */
3367 pageblock_order = order;
3368}
3369#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3370
Mel Gormanba72cb82007-11-28 16:21:13 -08003371/*
3372 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3373 * and pageblock_default_order() are unused as pageblock_order is set
3374 * at compile-time. See include/linux/pageblock-flags.h for the values of
3375 * pageblock_order based on the kernel config
3376 */
3377static inline int pageblock_default_order(unsigned int order)
3378{
3379 return MAX_ORDER-1;
3380}
Mel Gormand9c23402007-10-16 01:26:01 -07003381#define set_pageblock_order(x) do {} while (0)
3382
3383#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3384
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385/*
3386 * Set up the zone data structures:
3387 * - mark all pages reserved
3388 * - mark all memory queues empty
3389 * - clear the memory bitmaps
3390 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08003391static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392 unsigned long *zones_size, unsigned long *zholes_size)
3393{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07003394 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07003395 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07003397 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398
Dave Hansen208d54e2005-10-29 18:16:52 -07003399 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 pgdat->nr_zones = 0;
3401 init_waitqueue_head(&pgdat->kswapd_wait);
3402 pgdat->kswapd_max_order = 0;
3403
3404 for (j = 0; j < MAX_NR_ZONES; j++) {
3405 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003406 unsigned long size, realsize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407
Mel Gormanc7132162006-09-27 01:49:43 -07003408 size = zone_spanned_pages_in_node(nid, j, zones_size);
3409 realsize = size - zone_absent_pages_in_node(nid, j,
3410 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411
Mel Gorman0e0b8642006-09-27 01:49:56 -07003412 /*
3413 * Adjust realsize so that it accounts for how much memory
3414 * is used by this zone for memmap. This affects the watermark
3415 * and per-cpu initialisations
3416 */
Johannes Weinerf7232152008-05-23 13:04:21 -07003417 memmap_pages =
3418 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003419 if (realsize >= memmap_pages) {
3420 realsize -= memmap_pages;
Mel Gorman6b74ab92008-07-23 21:26:49 -07003421 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3422 "%s zone: %lu pages used for memmap\n",
Mel Gorman0e0b8642006-09-27 01:49:56 -07003423 zone_names[j], memmap_pages);
3424 } else
3425 printk(KERN_WARNING
3426 " %s zone: %lu pages exceeds realsize %lu\n",
3427 zone_names[j], memmap_pages, realsize);
3428
Christoph Lameter62672762007-02-10 01:43:07 -08003429 /* Account for reserved pages */
3430 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07003431 realsize -= dma_reserve;
Mel Gorman6b74ab92008-07-23 21:26:49 -07003432 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3433 "%s zone: %lu pages reserved\n",
Christoph Lameter62672762007-02-10 01:43:07 -08003434 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003435 }
3436
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003437 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438 nr_kernel_pages += realsize;
3439 nr_all_pages += realsize;
3440
3441 zone->spanned_pages = size;
3442 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07003443#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07003444 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07003445 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07003446 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07003447 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07003448#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449 zone->name = zone_names[j];
3450 spin_lock_init(&zone->lock);
3451 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07003452 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454
Martin Bligh3bb1a852006-10-28 10:38:24 -07003455 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003456
Dave Hansened8ece22005-10-29 18:16:50 -07003457 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 INIT_LIST_HEAD(&zone->active_list);
3459 INIT_LIST_HEAD(&zone->inactive_list);
3460 zone->nr_scan_active = 0;
3461 zone->nr_scan_inactive = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07003462 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07003463 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003464 if (!size)
3465 continue;
3466
Mel Gormanba72cb82007-11-28 16:21:13 -08003467 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07003468 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa02007-01-10 23:15:30 -08003469 ret = init_currently_empty_zone(zone, zone_start_pfn,
3470 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07003471 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07003472 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003473 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003474 }
3475}
3476
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003477static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003478{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003479 /* Skip empty nodes */
3480 if (!pgdat->node_spanned_pages)
3481 return;
3482
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003483#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484 /* ia64 gets its own node_mem_map, before this, without bootmem */
3485 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07003486 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003487 struct page *map;
3488
Bob Piccoe984bb42006-05-20 15:00:31 -07003489 /*
3490 * The zone's endpoints aren't required to be MAX_ORDER
3491 * aligned but the node_mem_map endpoints must be in order
3492 * for the buddy allocator to function correctly.
3493 */
3494 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3495 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3496 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3497 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07003498 map = alloc_remap(pgdat->node_id, size);
3499 if (!map)
3500 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07003501 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502 }
Roman Zippel12d810c2007-05-31 00:40:54 -07003503#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003504 /*
3505 * With no DISCONTIG, the global mem_map is just set as node 0's
3506 */
Mel Gormanc7132162006-09-27 01:49:43 -07003507 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07003509#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3510 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08003511 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07003512#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3513 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003514#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003515#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516}
3517
Johannes Weiner9109fb72008-07-23 21:27:20 -07003518void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3519 unsigned long node_start_pfn, unsigned long *zholes_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003520{
Johannes Weiner9109fb72008-07-23 21:27:20 -07003521 pg_data_t *pgdat = NODE_DATA(nid);
3522
Linus Torvalds1da177e2005-04-16 15:20:36 -07003523 pgdat->node_id = nid;
3524 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003525 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003526
3527 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07003528#ifdef CONFIG_FLAT_NODE_MEM_MAP
3529 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3530 nid, (unsigned long)pgdat,
3531 (unsigned long)pgdat->node_mem_map);
3532#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003533
3534 free_area_init_core(pgdat, zones_size, zholes_size);
3535}
3536
Mel Gormanc7132162006-09-27 01:49:43 -07003537#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07003538
3539#if MAX_NUMNODES > 1
3540/*
3541 * Figure out the number of possible node ids.
3542 */
3543static void __init setup_nr_node_ids(void)
3544{
3545 unsigned int node;
3546 unsigned int highest = 0;
3547
3548 for_each_node_mask(node, node_possible_map)
3549 highest = node;
3550 nr_node_ids = highest + 1;
3551}
3552#else
3553static inline void setup_nr_node_ids(void)
3554{
3555}
3556#endif
3557
Mel Gormanc7132162006-09-27 01:49:43 -07003558/**
3559 * add_active_range - Register a range of PFNs backed by physical memory
3560 * @nid: The node ID the range resides on
3561 * @start_pfn: The start PFN of the available physical memory
3562 * @end_pfn: The end PFN of the available physical memory
3563 *
3564 * These ranges are stored in an early_node_map[] and later used by
3565 * free_area_init_nodes() to calculate zone sizes and holes. If the
3566 * range spans a memory hole, it is up to the architecture to ensure
3567 * the memory is not freed by the bootmem allocator. If possible
3568 * the range being registered will be merged with existing ranges.
3569 */
3570void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3571 unsigned long end_pfn)
3572{
3573 int i;
3574
Mel Gorman6b74ab92008-07-23 21:26:49 -07003575 mminit_dprintk(MMINIT_TRACE, "memory_register",
3576 "Entering add_active_range(%d, %#lx, %#lx) "
3577 "%d entries of %d used\n",
3578 nid, start_pfn, end_pfn,
3579 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
Mel Gormanc7132162006-09-27 01:49:43 -07003580
Mel Gorman2dbb51c2008-07-23 21:26:52 -07003581 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3582
Mel Gormanc7132162006-09-27 01:49:43 -07003583 /* Merge with existing active regions if possible */
3584 for (i = 0; i < nr_nodemap_entries; i++) {
3585 if (early_node_map[i].nid != nid)
3586 continue;
3587
3588 /* Skip if an existing region covers this new one */
3589 if (start_pfn >= early_node_map[i].start_pfn &&
3590 end_pfn <= early_node_map[i].end_pfn)
3591 return;
3592
3593 /* Merge forward if suitable */
3594 if (start_pfn <= early_node_map[i].end_pfn &&
3595 end_pfn > early_node_map[i].end_pfn) {
3596 early_node_map[i].end_pfn = end_pfn;
3597 return;
3598 }
3599
3600 /* Merge backward if suitable */
3601 if (start_pfn < early_node_map[i].end_pfn &&
3602 end_pfn >= early_node_map[i].start_pfn) {
3603 early_node_map[i].start_pfn = start_pfn;
3604 return;
3605 }
3606 }
3607
3608 /* Check that early_node_map is large enough */
3609 if (i >= MAX_ACTIVE_REGIONS) {
3610 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3611 MAX_ACTIVE_REGIONS);
3612 return;
3613 }
3614
3615 early_node_map[i].nid = nid;
3616 early_node_map[i].start_pfn = start_pfn;
3617 early_node_map[i].end_pfn = end_pfn;
3618 nr_nodemap_entries = i + 1;
3619}
3620
3621/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07003622 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07003623 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07003624 * @start_pfn: The new PFN of the range
3625 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07003626 *
3627 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003628 * The map is kept near the end physical page range that has already been
3629 * registered. This function allows an arch to shrink an existing registered
3630 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07003631 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07003632void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3633 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003634{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003635 int i, j;
3636 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003637
Yinghai Lucc1050b2008-06-13 19:08:52 -07003638 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3639 nid, start_pfn, end_pfn);
3640
Mel Gormanc7132162006-09-27 01:49:43 -07003641 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003642 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07003643 if (early_node_map[i].start_pfn >= start_pfn &&
3644 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003645 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07003646 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003647 early_node_map[i].end_pfn = 0;
3648 removed = 1;
3649 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07003650 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07003651 if (early_node_map[i].start_pfn < start_pfn &&
3652 early_node_map[i].end_pfn > start_pfn) {
3653 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3654 early_node_map[i].end_pfn = start_pfn;
3655 if (temp_end_pfn > end_pfn)
3656 add_active_range(nid, end_pfn, temp_end_pfn);
3657 continue;
3658 }
3659 if (early_node_map[i].start_pfn >= start_pfn &&
3660 early_node_map[i].end_pfn > end_pfn &&
3661 early_node_map[i].start_pfn < end_pfn) {
3662 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003663 continue;
3664 }
3665 }
3666
3667 if (!removed)
3668 return;
3669
3670 /* remove the blank ones */
3671 for (i = nr_nodemap_entries - 1; i > 0; i--) {
3672 if (early_node_map[i].nid != nid)
3673 continue;
3674 if (early_node_map[i].end_pfn)
3675 continue;
3676 /* we found it, get rid of it */
3677 for (j = i; j < nr_nodemap_entries - 1; j++)
3678 memcpy(&early_node_map[j], &early_node_map[j+1],
3679 sizeof(early_node_map[j]));
3680 j = nr_nodemap_entries - 1;
3681 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3682 nr_nodemap_entries--;
3683 }
Mel Gormanc7132162006-09-27 01:49:43 -07003684}
3685
3686/**
3687 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003688 *
Mel Gormanc7132162006-09-27 01:49:43 -07003689 * During discovery, it may be found that a table like SRAT is invalid
3690 * and an alternative discovery method must be used. This function removes
3691 * all currently registered regions.
3692 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003693void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07003694{
3695 memset(early_node_map, 0, sizeof(early_node_map));
3696 nr_nodemap_entries = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07003697#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3698 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3699 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3700#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -07003701}
3702
3703/* Compare two active node_active_regions */
3704static int __init cmp_node_active_region(const void *a, const void *b)
3705{
3706 struct node_active_region *arange = (struct node_active_region *)a;
3707 struct node_active_region *brange = (struct node_active_region *)b;
3708
3709 /* Done this way to avoid overflows */
3710 if (arange->start_pfn > brange->start_pfn)
3711 return 1;
3712 if (arange->start_pfn < brange->start_pfn)
3713 return -1;
3714
3715 return 0;
3716}
3717
3718/* sort the node_map by start_pfn */
3719static void __init sort_node_map(void)
3720{
3721 sort(early_node_map, (size_t)nr_nodemap_entries,
3722 sizeof(struct node_active_region),
3723 cmp_node_active_region, NULL);
3724}
3725
Mel Gormana6af2bc2007-02-10 01:42:57 -08003726/* Find the lowest pfn for a node */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003727static unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003728{
3729 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08003730 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00003731
Mel Gormanc7132162006-09-27 01:49:43 -07003732 /* Assuming a sorted map, the first range found has the starting pfn */
3733 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08003734 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003735
Mel Gormana6af2bc2007-02-10 01:42:57 -08003736 if (min_pfn == ULONG_MAX) {
3737 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07003738 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08003739 return 0;
3740 }
3741
3742 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003743}
3744
3745/**
3746 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3747 *
3748 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003749 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07003750 */
3751unsigned long __init find_min_pfn_with_active_regions(void)
3752{
3753 return find_min_pfn_for_node(MAX_NUMNODES);
3754}
3755
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003756/*
3757 * early_calculate_totalpages()
3758 * Sum pages in active regions for movable zone.
3759 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3760 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07003761static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07003762{
3763 int i;
3764 unsigned long totalpages = 0;
3765
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003766 for (i = 0; i < nr_nodemap_entries; i++) {
3767 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07003768 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003769 totalpages += pages;
3770 if (pages)
3771 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3772 }
3773 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07003774}
3775
Mel Gorman2a1e2742007-07-17 04:03:12 -07003776/*
3777 * Find the PFN the Movable zone begins in each node. Kernel memory
3778 * is spread evenly between nodes as long as the nodes have enough
3779 * memory. When they don't, some nodes will have more kernelcore than
3780 * others
3781 */
Adrian Bunkb69a7282008-07-23 21:28:12 -07003782static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003783{
3784 int i, nid;
3785 unsigned long usable_startpfn;
3786 unsigned long kernelcore_node, kernelcore_remaining;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003787 unsigned long totalpages = early_calculate_totalpages();
3788 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07003789
Mel Gorman7e63efe2007-07-17 04:03:15 -07003790 /*
3791 * If movablecore was specified, calculate what size of
3792 * kernelcore that corresponds so that memory usable for
3793 * any allocation type is evenly spread. If both kernelcore
3794 * and movablecore are specified, then the value of kernelcore
3795 * will be used for required_kernelcore if it's greater than
3796 * what movablecore would have allowed.
3797 */
3798 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07003799 unsigned long corepages;
3800
3801 /*
3802 * Round-up so that ZONE_MOVABLE is at least as large as what
3803 * was requested by the user
3804 */
3805 required_movablecore =
3806 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3807 corepages = totalpages - required_movablecore;
3808
3809 required_kernelcore = max(required_kernelcore, corepages);
3810 }
3811
Mel Gorman2a1e2742007-07-17 04:03:12 -07003812 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3813 if (!required_kernelcore)
3814 return;
3815
3816 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3817 find_usable_zone_for_movable();
3818 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3819
3820restart:
3821 /* Spread kernelcore memory as evenly as possible throughout nodes */
3822 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003823 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07003824 /*
3825 * Recalculate kernelcore_node if the division per node
3826 * now exceeds what is necessary to satisfy the requested
3827 * amount of memory for the kernel
3828 */
3829 if (required_kernelcore < kernelcore_node)
3830 kernelcore_node = required_kernelcore / usable_nodes;
3831
3832 /*
3833 * As the map is walked, we track how much memory is usable
3834 * by the kernel using kernelcore_remaining. When it is
3835 * 0, the rest of the node is usable by ZONE_MOVABLE
3836 */
3837 kernelcore_remaining = kernelcore_node;
3838
3839 /* Go through each range of PFNs within this node */
3840 for_each_active_range_index_in_nid(i, nid) {
3841 unsigned long start_pfn, end_pfn;
3842 unsigned long size_pages;
3843
3844 start_pfn = max(early_node_map[i].start_pfn,
3845 zone_movable_pfn[nid]);
3846 end_pfn = early_node_map[i].end_pfn;
3847 if (start_pfn >= end_pfn)
3848 continue;
3849
3850 /* Account for what is only usable for kernelcore */
3851 if (start_pfn < usable_startpfn) {
3852 unsigned long kernel_pages;
3853 kernel_pages = min(end_pfn, usable_startpfn)
3854 - start_pfn;
3855
3856 kernelcore_remaining -= min(kernel_pages,
3857 kernelcore_remaining);
3858 required_kernelcore -= min(kernel_pages,
3859 required_kernelcore);
3860
3861 /* Continue if range is now fully accounted */
3862 if (end_pfn <= usable_startpfn) {
3863
3864 /*
3865 * Push zone_movable_pfn to the end so
3866 * that if we have to rebalance
3867 * kernelcore across nodes, we will
3868 * not double account here
3869 */
3870 zone_movable_pfn[nid] = end_pfn;
3871 continue;
3872 }
3873 start_pfn = usable_startpfn;
3874 }
3875
3876 /*
3877 * The usable PFN range for ZONE_MOVABLE is from
3878 * start_pfn->end_pfn. Calculate size_pages as the
3879 * number of pages used as kernelcore
3880 */
3881 size_pages = end_pfn - start_pfn;
3882 if (size_pages > kernelcore_remaining)
3883 size_pages = kernelcore_remaining;
3884 zone_movable_pfn[nid] = start_pfn + size_pages;
3885
3886 /*
3887 * Some kernelcore has been met, update counts and
3888 * break if the kernelcore for this node has been
3889 * satisified
3890 */
3891 required_kernelcore -= min(required_kernelcore,
3892 size_pages);
3893 kernelcore_remaining -= size_pages;
3894 if (!kernelcore_remaining)
3895 break;
3896 }
3897 }
3898
3899 /*
3900 * If there is still required_kernelcore, we do another pass with one
3901 * less node in the count. This will push zone_movable_pfn[nid] further
3902 * along on the nodes that still have memory until kernelcore is
3903 * satisified
3904 */
3905 usable_nodes--;
3906 if (usable_nodes && required_kernelcore > usable_nodes)
3907 goto restart;
3908
3909 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3910 for (nid = 0; nid < MAX_NUMNODES; nid++)
3911 zone_movable_pfn[nid] =
3912 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3913}
3914
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003915/* Any regular memory on that node ? */
3916static void check_for_regular_memory(pg_data_t *pgdat)
3917{
3918#ifdef CONFIG_HIGHMEM
3919 enum zone_type zone_type;
3920
3921 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3922 struct zone *zone = &pgdat->node_zones[zone_type];
3923 if (zone->present_pages)
3924 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3925 }
3926#endif
3927}
3928
Mel Gormanc7132162006-09-27 01:49:43 -07003929/**
3930 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003931 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07003932 *
3933 * This will call free_area_init_node() for each active node in the system.
3934 * Using the page ranges provided by add_active_range(), the size of each
3935 * zone in each node and their holes is calculated. If the maximum PFN
3936 * between two adjacent zones match, it is assumed that the zone is empty.
3937 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3938 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3939 * starts where the previous one ended. For example, ZONE_DMA32 starts
3940 * at arch_max_dma_pfn.
3941 */
3942void __init free_area_init_nodes(unsigned long *max_zone_pfn)
3943{
3944 unsigned long nid;
3945 enum zone_type i;
3946
Mel Gormana6af2bc2007-02-10 01:42:57 -08003947 /* Sort early_node_map as initialisation assumes it is sorted */
3948 sort_node_map();
3949
Mel Gormanc7132162006-09-27 01:49:43 -07003950 /* Record where the zone boundaries are */
3951 memset(arch_zone_lowest_possible_pfn, 0,
3952 sizeof(arch_zone_lowest_possible_pfn));
3953 memset(arch_zone_highest_possible_pfn, 0,
3954 sizeof(arch_zone_highest_possible_pfn));
3955 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
3956 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
3957 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07003958 if (i == ZONE_MOVABLE)
3959 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07003960 arch_zone_lowest_possible_pfn[i] =
3961 arch_zone_highest_possible_pfn[i-1];
3962 arch_zone_highest_possible_pfn[i] =
3963 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
3964 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07003965 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
3966 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
3967
3968 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3969 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
3970 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003971
Mel Gormanc7132162006-09-27 01:49:43 -07003972 /* Print out the zone ranges */
3973 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07003974 for (i = 0; i < MAX_NR_ZONES; i++) {
3975 if (i == ZONE_MOVABLE)
3976 continue;
Paul Jackson5dab8ec2008-06-25 05:44:40 -07003977 printk(" %-8s %0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07003978 zone_names[i],
3979 arch_zone_lowest_possible_pfn[i],
3980 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07003981 }
3982
3983 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3984 printk("Movable zone start PFN for each node\n");
3985 for (i = 0; i < MAX_NUMNODES; i++) {
3986 if (zone_movable_pfn[i])
3987 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
3988 }
Mel Gormanc7132162006-09-27 01:49:43 -07003989
3990 /* Print out the early_node_map[] */
3991 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
3992 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07003993 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003994 early_node_map[i].start_pfn,
3995 early_node_map[i].end_pfn);
3996
3997 /* Initialise every node */
Mel Gorman708614e2008-07-23 21:26:51 -07003998 mminit_verify_pageflags_layout();
Christoph Lameter8ef82862007-02-20 13:57:52 -08003999 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07004000 for_each_online_node(nid) {
4001 pg_data_t *pgdat = NODE_DATA(nid);
Johannes Weiner9109fb72008-07-23 21:27:20 -07004002 free_area_init_node(nid, NULL,
Mel Gormanc7132162006-09-27 01:49:43 -07004003 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07004004
4005 /* Any memory on that node */
4006 if (pgdat->node_present_pages)
4007 node_set_state(nid, N_HIGH_MEMORY);
4008 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07004009 }
4010}
Mel Gorman2a1e2742007-07-17 04:03:12 -07004011
Mel Gorman7e63efe2007-07-17 04:03:15 -07004012static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07004013{
4014 unsigned long long coremem;
4015 if (!p)
4016 return -EINVAL;
4017
4018 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004019 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07004020
Mel Gorman7e63efe2007-07-17 04:03:15 -07004021 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07004022 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4023
4024 return 0;
4025}
Mel Gormaned7ed362007-07-17 04:03:14 -07004026
Mel Gorman7e63efe2007-07-17 04:03:15 -07004027/*
4028 * kernelcore=size sets the amount of memory for use for allocations that
4029 * cannot be reclaimed or migrated.
4030 */
4031static int __init cmdline_parse_kernelcore(char *p)
4032{
4033 return cmdline_parse_core(p, &required_kernelcore);
4034}
4035
4036/*
4037 * movablecore=size sets the amount of memory for use for allocations that
4038 * can be reclaimed or migrated.
4039 */
4040static int __init cmdline_parse_movablecore(char *p)
4041{
4042 return cmdline_parse_core(p, &required_movablecore);
4043}
4044
Mel Gormaned7ed362007-07-17 04:03:14 -07004045early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004046early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004047
Mel Gormanc7132162006-09-27 01:49:43 -07004048#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4049
Mel Gorman0e0b8642006-09-27 01:49:56 -07004050/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004051 * set_dma_reserve - set the specified number of pages reserved in the first zone
4052 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004053 *
4054 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4055 * In the DMA zone, a significant percentage may be consumed by kernel image
4056 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004057 * function may optionally be used to account for unfreeable pages in the
4058 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4059 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004060 */
4061void __init set_dma_reserve(unsigned long new_dma_reserve)
4062{
4063 dma_reserve = new_dma_reserve;
4064}
4065
Dave Hansen93b75042005-06-23 00:07:47 -07004066#ifndef CONFIG_NEED_MULTIPLE_NODES
Johannes Weinerb61bfa32008-07-23 21:26:55 -07004067struct pglist_data contig_page_data = { .bdata = &bootmem_node_data[0] };
Linus Torvalds1da177e2005-04-16 15:20:36 -07004068EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07004069#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004070
4071void __init free_area_init(unsigned long *zones_size)
4072{
Johannes Weiner9109fb72008-07-23 21:27:20 -07004073 free_area_init_node(0, zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004074 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4075}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004076
Linus Torvalds1da177e2005-04-16 15:20:36 -07004077static int page_alloc_cpu_notify(struct notifier_block *self,
4078 unsigned long action, void *hcpu)
4079{
4080 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004081
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004082 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004083 drain_pages(cpu);
4084
4085 /*
4086 * Spill the event counters of the dead processor
4087 * into the current processors event counters.
4088 * This artificially elevates the count of the current
4089 * processor.
4090 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004091 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004092
4093 /*
4094 * Zero the differential counters of the dead processor
4095 * so that the vm statistics are consistent.
4096 *
4097 * This is only okay since the processor is dead and cannot
4098 * race with what we are doing.
4099 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004100 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004101 }
4102 return NOTIFY_OK;
4103}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004104
4105void __init page_alloc_init(void)
4106{
4107 hotcpu_notifier(page_alloc_cpu_notify, 0);
4108}
4109
4110/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004111 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4112 * or min_free_kbytes changes.
4113 */
4114static void calculate_totalreserve_pages(void)
4115{
4116 struct pglist_data *pgdat;
4117 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004118 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004119
4120 for_each_online_pgdat(pgdat) {
4121 for (i = 0; i < MAX_NR_ZONES; i++) {
4122 struct zone *zone = pgdat->node_zones + i;
4123 unsigned long max = 0;
4124
4125 /* Find valid and maximum lowmem_reserve in the zone */
4126 for (j = i; j < MAX_NR_ZONES; j++) {
4127 if (zone->lowmem_reserve[j] > max)
4128 max = zone->lowmem_reserve[j];
4129 }
4130
4131 /* we treat pages_high as reserved pages. */
4132 max += zone->pages_high;
4133
4134 if (max > zone->present_pages)
4135 max = zone->present_pages;
4136 reserve_pages += max;
4137 }
4138 }
4139 totalreserve_pages = reserve_pages;
4140}
4141
4142/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004143 * setup_per_zone_lowmem_reserve - called whenever
4144 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4145 * has a correct pages reserved value, so an adequate number of
4146 * pages are left in the zone after a successful __alloc_pages().
4147 */
4148static void setup_per_zone_lowmem_reserve(void)
4149{
4150 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004151 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004152
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004153 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004154 for (j = 0; j < MAX_NR_ZONES; j++) {
4155 struct zone *zone = pgdat->node_zones + j;
4156 unsigned long present_pages = zone->present_pages;
4157
4158 zone->lowmem_reserve[j] = 0;
4159
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004160 idx = j;
4161 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004162 struct zone *lower_zone;
4163
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004164 idx--;
4165
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4167 sysctl_lowmem_reserve_ratio[idx] = 1;
4168
4169 lower_zone = pgdat->node_zones + idx;
4170 lower_zone->lowmem_reserve[j] = present_pages /
4171 sysctl_lowmem_reserve_ratio[idx];
4172 present_pages += lower_zone->present_pages;
4173 }
4174 }
4175 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004176
4177 /* update totalreserve_pages */
4178 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004179}
4180
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004181/**
4182 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4183 *
4184 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4185 * with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004186 */
Dave Hansen3947be12005-10-29 18:16:54 -07004187void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188{
4189 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4190 unsigned long lowmem_pages = 0;
4191 struct zone *zone;
4192 unsigned long flags;
4193
4194 /* Calculate total number of !ZONE_HIGHMEM pages */
4195 for_each_zone(zone) {
4196 if (!is_highmem(zone))
4197 lowmem_pages += zone->present_pages;
4198 }
4199
4200 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004201 u64 tmp;
4202
Linus Torvalds1da177e2005-04-16 15:20:36 -07004203 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004204 tmp = (u64)pages_min * zone->present_pages;
4205 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004206 if (is_highmem(zone)) {
4207 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004208 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4209 * need highmem pages, so cap pages_min to a small
4210 * value here.
4211 *
4212 * The (pages_high-pages_low) and (pages_low-pages_min)
4213 * deltas controls asynch page reclaim, and so should
4214 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004215 */
4216 int min_pages;
4217
4218 min_pages = zone->present_pages / 1024;
4219 if (min_pages < SWAP_CLUSTER_MAX)
4220 min_pages = SWAP_CLUSTER_MAX;
4221 if (min_pages > 128)
4222 min_pages = 128;
4223 zone->pages_min = min_pages;
4224 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004225 /*
4226 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004227 * proportionate to the zone's size.
4228 */
Nick Piggin669ed172005-11-13 16:06:45 -08004229 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004230 }
4231
Andrew Mortonac924c62006-05-15 09:43:59 -07004232 zone->pages_low = zone->pages_min + (tmp >> 2);
4233 zone->pages_high = zone->pages_min + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004234 setup_zone_migrate_reserve(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004235 spin_unlock_irqrestore(&zone->lru_lock, flags);
4236 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004237
4238 /* update totalreserve_pages */
4239 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004240}
4241
4242/*
4243 * Initialise min_free_kbytes.
4244 *
4245 * For small machines we want it small (128k min). For large machines
4246 * we want it large (64MB max). But it is not linear, because network
4247 * bandwidth does not increase linearly with machine size. We use
4248 *
4249 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4250 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4251 *
4252 * which yields
4253 *
4254 * 16MB: 512k
4255 * 32MB: 724k
4256 * 64MB: 1024k
4257 * 128MB: 1448k
4258 * 256MB: 2048k
4259 * 512MB: 2896k
4260 * 1024MB: 4096k
4261 * 2048MB: 5792k
4262 * 4096MB: 8192k
4263 * 8192MB: 11584k
4264 * 16384MB: 16384k
4265 */
4266static int __init init_per_zone_pages_min(void)
4267{
4268 unsigned long lowmem_kbytes;
4269
4270 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4271
4272 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4273 if (min_free_kbytes < 128)
4274 min_free_kbytes = 128;
4275 if (min_free_kbytes > 65536)
4276 min_free_kbytes = 65536;
4277 setup_per_zone_pages_min();
4278 setup_per_zone_lowmem_reserve();
4279 return 0;
4280}
4281module_init(init_per_zone_pages_min)
4282
4283/*
4284 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4285 * that we can call two helper functions whenever min_free_kbytes
4286 * changes.
4287 */
4288int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4289 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4290{
4291 proc_dointvec(table, write, file, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07004292 if (write)
4293 setup_per_zone_pages_min();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004294 return 0;
4295}
4296
Christoph Lameter96146342006-07-03 00:24:13 -07004297#ifdef CONFIG_NUMA
4298int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4299 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4300{
4301 struct zone *zone;
4302 int rc;
4303
4304 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4305 if (rc)
4306 return rc;
4307
4308 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07004309 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07004310 sysctl_min_unmapped_ratio) / 100;
4311 return 0;
4312}
Christoph Lameter0ff38492006-09-25 23:31:52 -07004313
4314int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4315 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4316{
4317 struct zone *zone;
4318 int rc;
4319
4320 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4321 if (rc)
4322 return rc;
4323
4324 for_each_zone(zone)
4325 zone->min_slab_pages = (zone->present_pages *
4326 sysctl_min_slab_ratio) / 100;
4327 return 0;
4328}
Christoph Lameter96146342006-07-03 00:24:13 -07004329#endif
4330
Linus Torvalds1da177e2005-04-16 15:20:36 -07004331/*
4332 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4333 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4334 * whenever sysctl_lowmem_reserve_ratio changes.
4335 *
4336 * The reserve ratio obviously has absolutely no relation with the
4337 * pages_min watermarks. The lowmem reserve ratio can only make sense
4338 * if in function of the boot time zone sizes.
4339 */
4340int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4341 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4342{
4343 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4344 setup_per_zone_lowmem_reserve();
4345 return 0;
4346}
4347
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004348/*
4349 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4350 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4351 * can have before it gets flushed back to buddy allocator.
4352 */
4353
4354int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4355 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4356{
4357 struct zone *zone;
4358 unsigned int cpu;
4359 int ret;
4360
4361 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4362 if (!write || (ret == -EINVAL))
4363 return ret;
4364 for_each_zone(zone) {
4365 for_each_online_cpu(cpu) {
4366 unsigned long high;
4367 high = zone->present_pages / percpu_pagelist_fraction;
4368 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4369 }
4370 }
4371 return 0;
4372}
4373
David S. Millerf034b5d2006-08-24 03:08:07 -07004374int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004375
4376#ifdef CONFIG_NUMA
4377static int __init set_hashdist(char *str)
4378{
4379 if (!str)
4380 return 0;
4381 hashdist = simple_strtoul(str, &str, 0);
4382 return 1;
4383}
4384__setup("hashdist=", set_hashdist);
4385#endif
4386
4387/*
4388 * allocate a large system hash table from bootmem
4389 * - it is assumed that the hash table must contain an exact power-of-2
4390 * quantity of entries
4391 * - limit is the number of hash buckets, not the total allocation size
4392 */
4393void *__init alloc_large_system_hash(const char *tablename,
4394 unsigned long bucketsize,
4395 unsigned long numentries,
4396 int scale,
4397 int flags,
4398 unsigned int *_hash_shift,
4399 unsigned int *_hash_mask,
4400 unsigned long limit)
4401{
4402 unsigned long long max = limit;
4403 unsigned long log2qty, size;
4404 void *table = NULL;
4405
4406 /* allow the kernel cmdline to have a say */
4407 if (!numentries) {
4408 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08004409 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004410 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4411 numentries >>= 20 - PAGE_SHIFT;
4412 numentries <<= 20 - PAGE_SHIFT;
4413
4414 /* limit to 1 bucket per 2^scale bytes of low memory */
4415 if (scale > PAGE_SHIFT)
4416 numentries >>= (scale - PAGE_SHIFT);
4417 else
4418 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08004419
4420 /* Make sure we've got at least a 0-order allocation.. */
4421 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4422 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004423 }
John Hawkes6e692ed2006-03-25 03:08:02 -08004424 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004425
4426 /* limit allocation size to 1/16 total memory by default */
4427 if (max == 0) {
4428 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4429 do_div(max, bucketsize);
4430 }
4431
4432 if (numentries > max)
4433 numentries = max;
4434
David Howellsf0d1b0b2006-12-08 02:37:49 -08004435 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004436
4437 do {
4438 size = bucketsize << log2qty;
4439 if (flags & HASH_EARLY)
Jan Beulich74768ed2008-08-12 15:08:39 -07004440 table = alloc_bootmem_nopanic(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004441 else if (hashdist)
4442 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4443 else {
Pavel Machek2309f9e2008-04-28 02:13:35 -07004444 unsigned long order = get_order(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004445 table = (void*) __get_free_pages(GFP_ATOMIC, order);
Eric Dumazet1037b832007-07-15 23:38:05 -07004446 /*
4447 * If bucketsize is not a power-of-two, we may free
4448 * some pages at the end of hash table.
4449 */
4450 if (table) {
4451 unsigned long alloc_end = (unsigned long)table +
4452 (PAGE_SIZE << order);
4453 unsigned long used = (unsigned long)table +
4454 PAGE_ALIGN(size);
4455 split_page(virt_to_page(table), order);
4456 while (used < alloc_end) {
4457 free_page(used);
4458 used += PAGE_SIZE;
4459 }
4460 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004461 }
4462 } while (!table && size > PAGE_SIZE && --log2qty);
4463
4464 if (!table)
4465 panic("Failed to allocate %s hash table\n", tablename);
4466
Dan Alonib49ad482007-07-15 23:38:23 -07004467 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004468 tablename,
4469 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08004470 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004471 size);
4472
4473 if (_hash_shift)
4474 *_hash_shift = log2qty;
4475 if (_hash_mask)
4476 *_hash_mask = (1 << log2qty) - 1;
4477
4478 return table;
4479}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004480
4481#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004482struct page *pfn_to_page(unsigned long pfn)
4483{
Andy Whitcroft67de6482006-06-23 02:03:12 -07004484 return __pfn_to_page(pfn);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004485}
4486unsigned long page_to_pfn(struct page *page)
4487{
Andy Whitcroft67de6482006-06-23 02:03:12 -07004488 return __page_to_pfn(page);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004489}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004490EXPORT_SYMBOL(pfn_to_page);
4491EXPORT_SYMBOL(page_to_pfn);
4492#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
Andrew Morton6220ec72006-10-19 23:29:05 -07004493
Mel Gorman835c1342007-10-16 01:25:47 -07004494/* Return a pointer to the bitmap storing bits affecting a block of pages */
4495static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4496 unsigned long pfn)
4497{
4498#ifdef CONFIG_SPARSEMEM
4499 return __pfn_to_section(pfn)->pageblock_flags;
4500#else
4501 return zone->pageblock_flags;
4502#endif /* CONFIG_SPARSEMEM */
4503}
Andrew Morton6220ec72006-10-19 23:29:05 -07004504
Mel Gorman835c1342007-10-16 01:25:47 -07004505static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4506{
4507#ifdef CONFIG_SPARSEMEM
4508 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004509 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004510#else
4511 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07004512 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004513#endif /* CONFIG_SPARSEMEM */
4514}
4515
4516/**
Mel Gormand9c23402007-10-16 01:26:01 -07004517 * 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 -07004518 * @page: The page within the block of interest
4519 * @start_bitidx: The first bit of interest to retrieve
4520 * @end_bitidx: The last bit of interest
4521 * returns pageblock_bits flags
4522 */
4523unsigned long get_pageblock_flags_group(struct page *page,
4524 int start_bitidx, int end_bitidx)
4525{
4526 struct zone *zone;
4527 unsigned long *bitmap;
4528 unsigned long pfn, bitidx;
4529 unsigned long flags = 0;
4530 unsigned long value = 1;
4531
4532 zone = page_zone(page);
4533 pfn = page_to_pfn(page);
4534 bitmap = get_pageblock_bitmap(zone, pfn);
4535 bitidx = pfn_to_bitidx(zone, pfn);
4536
4537 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4538 if (test_bit(bitidx + start_bitidx, bitmap))
4539 flags |= value;
4540
4541 return flags;
4542}
4543
4544/**
Mel Gormand9c23402007-10-16 01:26:01 -07004545 * 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 -07004546 * @page: The page within the block of interest
4547 * @start_bitidx: The first bit of interest
4548 * @end_bitidx: The last bit of interest
4549 * @flags: The flags to set
4550 */
4551void set_pageblock_flags_group(struct page *page, unsigned long flags,
4552 int start_bitidx, int end_bitidx)
4553{
4554 struct zone *zone;
4555 unsigned long *bitmap;
4556 unsigned long pfn, bitidx;
4557 unsigned long value = 1;
4558
4559 zone = page_zone(page);
4560 pfn = page_to_pfn(page);
4561 bitmap = get_pageblock_bitmap(zone, pfn);
4562 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004563 VM_BUG_ON(pfn < zone->zone_start_pfn);
4564 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07004565
4566 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4567 if (flags & value)
4568 __set_bit(bitidx + start_bitidx, bitmap);
4569 else
4570 __clear_bit(bitidx + start_bitidx, bitmap);
4571}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004572
4573/*
4574 * This is designed as sub function...plz see page_isolation.c also.
4575 * set/clear page block's type to be ISOLATE.
4576 * page allocater never alloc memory from ISOLATE block.
4577 */
4578
4579int set_migratetype_isolate(struct page *page)
4580{
4581 struct zone *zone;
4582 unsigned long flags;
4583 int ret = -EBUSY;
4584
4585 zone = page_zone(page);
4586 spin_lock_irqsave(&zone->lock, flags);
4587 /*
4588 * In future, more migrate types will be able to be isolation target.
4589 */
4590 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4591 goto out;
4592 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4593 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4594 ret = 0;
4595out:
4596 spin_unlock_irqrestore(&zone->lock, flags);
4597 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004598 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004599 return ret;
4600}
4601
4602void unset_migratetype_isolate(struct page *page)
4603{
4604 struct zone *zone;
4605 unsigned long flags;
4606 zone = page_zone(page);
4607 spin_lock_irqsave(&zone->lock, flags);
4608 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4609 goto out;
4610 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4611 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4612out:
4613 spin_unlock_irqrestore(&zone->lock, flags);
4614}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07004615
4616#ifdef CONFIG_MEMORY_HOTREMOVE
4617/*
4618 * All pages in the range must be isolated before calling this.
4619 */
4620void
4621__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4622{
4623 struct page *page;
4624 struct zone *zone;
4625 int order, i;
4626 unsigned long pfn;
4627 unsigned long flags;
4628 /* find the first valid pfn */
4629 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4630 if (pfn_valid(pfn))
4631 break;
4632 if (pfn == end_pfn)
4633 return;
4634 zone = page_zone(pfn_to_page(pfn));
4635 spin_lock_irqsave(&zone->lock, flags);
4636 pfn = start_pfn;
4637 while (pfn < end_pfn) {
4638 if (!pfn_valid(pfn)) {
4639 pfn++;
4640 continue;
4641 }
4642 page = pfn_to_page(pfn);
4643 BUG_ON(page_count(page));
4644 BUG_ON(!PageBuddy(page));
4645 order = page_order(page);
4646#ifdef CONFIG_DEBUG_VM
4647 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4648 pfn, 1 << order, end_pfn);
4649#endif
4650 list_del(&page->lru);
4651 rmv_page_order(page);
4652 zone->free_area[order].nr_free--;
4653 __mod_zone_page_state(zone, NR_FREE_PAGES,
4654 - (1UL << order));
4655 for (i = 0; i < (1 << order); i++)
4656 SetPageReserved((page+i));
4657 pfn += (1 << order);
4658 }
4659 spin_unlock_irqrestore(&zone->lock, flags);
4660}
4661#endif