blob: 79ac4afc908cff9ed4b7220e4fd9851955d73783 [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 Rientjes5a3135c22007-10-16 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 */
Mel Gorman2a1e2742007-07-17 04:03:12 -0700156 unsigned long __initdata required_kernelcore;
Adrian Bunk484f51f2007-10-16 01:26:03 -0700157 static unsigned long __initdata required_movablecore;
Paul Mundte2289292007-07-20 00:31:44 -0700158 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 Dickins9442ec92008-03-04 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 Dickins9442ec92008-03-04 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267static void prep_compound_page(struct page *page, unsigned long order)
268{
269 int i;
270 int nr_pages = 1 << order;
271
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800272 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700273 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700274 __SetPageHead(page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700275 for (i = 1; i < nr_pages; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 struct page *p = page + i;
277
Christoph Lameterd85f3382007-05-06 14:49:39 -0700278 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700279 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 }
281}
282
283static void destroy_compound_page(struct page *page, unsigned long order)
284{
285 int i;
286 int nr_pages = 1 << order;
287
Christoph Lameterd85f3382007-05-06 14:49:39 -0700288 if (unlikely(compound_order(page) != order))
Nick Piggin224abf92006-01-06 00:11:11 -0800289 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
Christoph Lameter6d777952007-05-06 14:49:40 -0700291 if (unlikely(!PageHead(page)))
Christoph Lameterd85f3382007-05-06 14:49:39 -0700292 bad_page(page);
Christoph Lameter6d777952007-05-06 14:49:40 -0700293 __ClearPageHead(page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700294 for (i = 1; i < nr_pages; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 struct page *p = page + i;
296
Christoph Lameter6d777952007-05-06 14:49:40 -0700297 if (unlikely(!PageTail(p) |
Christoph Lameterd85f3382007-05-06 14:49:39 -0700298 (p->first_page != page)))
Nick Piggin224abf92006-01-06 00:11:11 -0800299 bad_page(page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700300 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 }
302}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
Nick Piggin17cf4402006-03-22 00:08:41 -0800304static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
305{
306 int i;
307
Andrew Morton6626c5d2006-03-22 00:08:42 -0800308 /*
309 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
310 * and __GFP_HIGHMEM from hard or soft interrupt context.
311 */
Nick Piggin725d7042006-09-25 23:30:55 -0700312 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800313 for (i = 0; i < (1 << order); i++)
314 clear_highpage(page + i);
315}
316
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700317static inline void set_page_order(struct page *page, int order)
318{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700319 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000320 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322
323static inline void rmv_page_order(struct page *page)
324{
Nick Piggin676165a2006-04-10 11:21:48 +1000325 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700326 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327}
328
329/*
330 * Locate the struct page for both the matching buddy in our
331 * pair (buddy1) and the combined O(n+1) page they form (page).
332 *
333 * 1) Any buddy B1 will have an order O twin B2 which satisfies
334 * the following equation:
335 * B2 = B1 ^ (1 << O)
336 * For example, if the starting buddy (buddy2) is #8 its order
337 * 1 buddy is #10:
338 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
339 *
340 * 2) Any buddy B will have an order O+1 parent P which
341 * satisfies the following equation:
342 * P = B & ~(1 << O)
343 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200344 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 */
346static inline struct page *
347__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
348{
349 unsigned long buddy_idx = page_idx ^ (1 << order);
350
351 return page + (buddy_idx - page_idx);
352}
353
354static inline unsigned long
355__find_combined_index(unsigned long page_idx, unsigned int order)
356{
357 return (page_idx & ~(1 << order));
358}
359
360/*
361 * This function checks whether a page is free && is the buddy
362 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800363 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000364 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700365 * (c) a page and its buddy have the same order &&
366 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 *
Nick Piggin676165a2006-04-10 11:21:48 +1000368 * For recording whether a page is in the buddy system, we use PG_buddy.
369 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
370 *
371 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700373static inline int page_is_buddy(struct page *page, struct page *buddy,
374 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700376 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800377 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800378
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700379 if (page_zone_id(page) != page_zone_id(buddy))
380 return 0;
381
382 if (PageBuddy(buddy) && page_order(buddy) == order) {
383 BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700384 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000385 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700386 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387}
388
389/*
390 * Freeing function for a buddy system allocator.
391 *
392 * The concept of a buddy system is to maintain direct-mapped table
393 * (containing bit values) for memory blocks of various "orders".
394 * The bottom level table contains the map for the smallest allocatable
395 * units of memory (here, pages), and each level above it describes
396 * pairs of units from the levels below, hence, "buddies".
397 * At a high level, all that happens here is marking the table entry
398 * at the bottom level available, and propagating the changes upward
399 * as necessary, plus some accounting needed to play nicely with other
400 * parts of the VM system.
401 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000402 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700403 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 * So when we are allocating or freeing one, we can derive the state of the
405 * other. That is, if we allocate a small block, and both were
406 * free, the remainder of the region must be split into blocks.
407 * If a block is freed, and its buddy is also free, then this
408 * triggers coalescing into a block of larger size.
409 *
410 * -- wli
411 */
412
Nick Piggin48db57f2006-01-08 01:00:42 -0800413static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 struct zone *zone, unsigned int order)
415{
416 unsigned long page_idx;
417 int order_size = 1 << order;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700418 int migratetype = get_pageblock_migratetype(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
Nick Piggin224abf92006-01-06 00:11:11 -0800420 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 destroy_compound_page(page, order);
422
423 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
424
Nick Piggin725d7042006-09-25 23:30:55 -0700425 VM_BUG_ON(page_idx & (order_size - 1));
426 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
Christoph Lameterd23ad422007-02-10 01:43:02 -0800428 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 while (order < MAX_ORDER-1) {
430 unsigned long combined_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 struct page *buddy;
432
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700434 if (!page_is_buddy(page, buddy, order))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 list_del(&buddy->lru);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700438 zone->free_area[order].nr_free--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800440 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 page = page + (combined_idx - page_idx);
442 page_idx = combined_idx;
443 order++;
444 }
445 set_page_order(page, order);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700446 list_add(&page->lru,
447 &zone->free_area[order].free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 zone->free_area[order].nr_free++;
449}
450
Nick Piggin224abf92006-01-06 00:11:11 -0800451static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452{
Nick Piggin92be2e332006-01-06 00:10:57 -0800453 if (unlikely(page_mapcount(page) |
454 (page->mapping != NULL) |
Hugh Dickins9442ec92008-03-04 14:29:07 -0800455 (page_get_page_cgroup(page) != NULL) |
Nick Piggin92be2e332006-01-06 00:10:57 -0800456 (page_count(page) != 0) |
Russ Andersondfa7e202008-06-09 11:18:45 -0500457 (page->flags & PAGE_FLAGS_CHECK_AT_FREE)))
Nick Piggin224abf92006-01-06 00:11:11 -0800458 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700460 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800461 /*
462 * For now, we report if PG_reserved was found set, but do not
463 * clear it, and do not free the page. But we shall soon need
464 * to do more, for when the ZERO_PAGE count wraps negative.
465 */
466 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467}
468
469/*
470 * Frees a list of pages.
471 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700472 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 *
474 * If the zone was previously in an "all pages pinned" state then look to
475 * see if this freeing clears that state.
476 *
477 * And clear the zone's pages_scanned counter, to hold off the "all pages are
478 * pinned" detection logic.
479 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800480static void free_pages_bulk(struct zone *zone, int count,
481 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482{
Nick Pigginc54ad302006-01-06 00:10:56 -0800483 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700484 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800486 while (count--) {
487 struct page *page;
488
Nick Piggin725d7042006-09-25 23:30:55 -0700489 VM_BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800491 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800493 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800495 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496}
497
Nick Piggin48db57f2006-01-08 01:00:42 -0800498static void free_one_page(struct zone *zone, struct page *page, int order)
499{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700500 spin_lock(&zone->lock);
David Rientjese815af92007-10-16 23:25:54 -0700501 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700502 zone->pages_scanned = 0;
Paul Jackson0798e512006-12-06 20:31:38 -0800503 __free_one_page(page, zone, order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700504 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800505}
506
507static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508{
Nick Pigginc54ad302006-01-06 00:10:56 -0800509 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800511 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800514 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800515 if (reserved)
516 return;
517
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700518 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700519 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700520 debug_check_no_obj_freed(page_address(page),
521 PAGE_SIZE << order);
522 }
Nick Piggindafb1362006-10-11 01:21:30 -0700523 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800524 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700525
Nick Pigginc54ad302006-01-06 00:10:56 -0800526 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700527 __count_vm_events(PGFREE, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800528 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800529 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530}
531
David Howellsa226f6c2006-01-06 00:11:08 -0800532/*
533 * permit the bootmem allocator to evade page validation on high-order frees
534 */
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700535void __free_pages_bootmem(struct page *page, unsigned int order)
David Howellsa226f6c2006-01-06 00:11:08 -0800536{
537 if (order == 0) {
538 __ClearPageReserved(page);
539 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800540 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800541 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800542 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800543 int loop;
544
Nick Piggin545b1ea2006-03-22 00:08:07 -0800545 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800546 for (loop = 0; loop < BITS_PER_LONG; loop++) {
547 struct page *p = &page[loop];
548
Nick Piggin545b1ea2006-03-22 00:08:07 -0800549 if (loop + 1 < BITS_PER_LONG)
550 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800551 __ClearPageReserved(p);
552 set_page_count(p, 0);
553 }
554
Nick Piggin7835e982006-03-22 00:08:40 -0800555 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800556 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800557 }
558}
559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
561/*
562 * The order of subdivision here is critical for the IO subsystem.
563 * Please do not alter this order without good reasons and regression
564 * testing. Specifically, as large blocks of memory are subdivided,
565 * the order in which smaller blocks are delivered depends on the order
566 * they're subdivided in this function. This is the primary factor
567 * influencing the order in which pages are delivered to the IO
568 * subsystem according to empirical testing, and this is also justified
569 * by considering the behavior of a buddy system containing a single
570 * large block of memory acted on by a series of small allocations.
571 * This behavior is a critical factor in sglist merging's success.
572 *
573 * -- wli
574 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800575static inline void expand(struct zone *zone, struct page *page,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700576 int low, int high, struct free_area *area,
577 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578{
579 unsigned long size = 1 << high;
580
581 while (high > low) {
582 area--;
583 high--;
584 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700585 VM_BUG_ON(bad_range(zone, &page[size]));
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700586 list_add(&page[size].lru, &area->free_list[migratetype]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 area->nr_free++;
588 set_page_order(&page[size], high);
589 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590}
591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592/*
593 * This page is about to be returned from the page allocator
594 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800595static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596{
Nick Piggin92be2e332006-01-06 00:10:57 -0800597 if (unlikely(page_mapcount(page) |
598 (page->mapping != NULL) |
Hugh Dickins9442ec92008-03-04 14:29:07 -0800599 (page_get_page_cgroup(page) != NULL) |
Nick Piggin92be2e332006-01-06 00:10:57 -0800600 (page_count(page) != 0) |
Russ Andersondfa7e202008-06-09 11:18:45 -0500601 (page->flags & PAGE_FLAGS_CHECK_AT_PREP)))
Nick Piggin224abf92006-01-06 00:11:11 -0800602 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
Hugh Dickins689bceb2005-11-21 21:32:20 -0800604 /*
605 * For now, we report if PG_reserved was found set, but do not
606 * clear it, and do not allocate the page: as a safety net.
607 */
608 if (PageReserved(page))
609 return 1;
610
Christoph Lameter0a128b22008-04-28 02:12:52 -0700611 page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_reclaim |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 1 << PG_referenced | 1 << PG_arch_1 |
Nick Piggin5409bae2007-02-28 20:12:27 -0800613 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700614 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800615 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800616
617 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800619
620 if (gfp_flags & __GFP_ZERO)
621 prep_zero_page(page, order, gfp_flags);
622
623 if (order && (gfp_flags & __GFP_COMP))
624 prep_compound_page(page, order);
625
Hugh Dickins689bceb2005-11-21 21:32:20 -0800626 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627}
628
Mel Gorman56fd56b2007-10-16 01:25:58 -0700629/*
630 * Go through the free lists for the given migratetype and remove
631 * the smallest available page from the freelists
632 */
633static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
634 int migratetype)
635{
636 unsigned int current_order;
637 struct free_area * area;
638 struct page *page;
639
640 /* Find a page of the appropriate size in the preferred list */
641 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
642 area = &(zone->free_area[current_order]);
643 if (list_empty(&area->free_list[migratetype]))
644 continue;
645
646 page = list_entry(area->free_list[migratetype].next,
647 struct page, lru);
648 list_del(&page->lru);
649 rmv_page_order(page);
650 area->nr_free--;
651 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
652 expand(zone, page, order, current_order, area, migratetype);
653 return page;
654 }
655
656 return NULL;
657}
658
659
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700660/*
661 * This array describes the order lists are fallen back to when
662 * the free lists for the desirable migrate type are depleted
663 */
664static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
Mel Gorman64c5e132007-10-16 01:25:59 -0700665 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
666 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
667 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
668 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700669};
670
Mel Gormanc361be52007-10-16 01:25:51 -0700671/*
672 * Move the free pages in a range to the free lists of the requested type.
Mel Gormand9c23402007-10-16 01:26:01 -0700673 * Note that start_page and end_pages are not aligned on a pageblock
Mel Gormanc361be52007-10-16 01:25:51 -0700674 * boundary. If alignment is required, use move_freepages_block()
675 */
676int move_freepages(struct zone *zone,
677 struct page *start_page, struct page *end_page,
678 int migratetype)
679{
680 struct page *page;
681 unsigned long order;
Mel Gormand1003132007-10-16 01:26:00 -0700682 int pages_moved = 0;
Mel Gormanc361be52007-10-16 01:25:51 -0700683
684#ifndef CONFIG_HOLES_IN_ZONE
685 /*
686 * page_zone is not safe to call in this context when
687 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
688 * anyway as we check zone boundaries in move_freepages_block().
689 * Remove at a later date when no bug reports exist related to
Mel Gormanac0e5b72007-10-16 01:25:58 -0700690 * grouping pages by mobility
Mel Gormanc361be52007-10-16 01:25:51 -0700691 */
692 BUG_ON(page_zone(start_page) != page_zone(end_page));
693#endif
694
695 for (page = start_page; page <= end_page;) {
696 if (!pfn_valid_within(page_to_pfn(page))) {
697 page++;
698 continue;
699 }
700
701 if (!PageBuddy(page)) {
702 page++;
703 continue;
704 }
705
706 order = page_order(page);
707 list_del(&page->lru);
708 list_add(&page->lru,
709 &zone->free_area[order].free_list[migratetype]);
710 page += 1 << order;
Mel Gormand1003132007-10-16 01:26:00 -0700711 pages_moved += 1 << order;
Mel Gormanc361be52007-10-16 01:25:51 -0700712 }
713
Mel Gormand1003132007-10-16 01:26:00 -0700714 return pages_moved;
Mel Gormanc361be52007-10-16 01:25:51 -0700715}
716
717int move_freepages_block(struct zone *zone, struct page *page, int migratetype)
718{
719 unsigned long start_pfn, end_pfn;
720 struct page *start_page, *end_page;
721
722 start_pfn = page_to_pfn(page);
Mel Gormand9c23402007-10-16 01:26:01 -0700723 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
Mel Gormanc361be52007-10-16 01:25:51 -0700724 start_page = pfn_to_page(start_pfn);
Mel Gormand9c23402007-10-16 01:26:01 -0700725 end_page = start_page + pageblock_nr_pages - 1;
726 end_pfn = start_pfn + pageblock_nr_pages - 1;
Mel Gormanc361be52007-10-16 01:25:51 -0700727
728 /* Do not cross zone boundaries */
729 if (start_pfn < zone->zone_start_pfn)
730 start_page = page;
731 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
732 return 0;
733
734 return move_freepages(zone, start_page, end_page, migratetype);
735}
736
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700737/* Remove an element from the buddy allocator from the fallback list */
738static struct page *__rmqueue_fallback(struct zone *zone, int order,
739 int start_migratetype)
740{
741 struct free_area * area;
742 int current_order;
743 struct page *page;
744 int migratetype, i;
745
746 /* Find the largest possible block of pages in the other list */
747 for (current_order = MAX_ORDER-1; current_order >= order;
748 --current_order) {
749 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
750 migratetype = fallbacks[start_migratetype][i];
751
Mel Gorman56fd56b2007-10-16 01:25:58 -0700752 /* MIGRATE_RESERVE handled later if necessary */
753 if (migratetype == MIGRATE_RESERVE)
754 continue;
Mel Gormane0104872007-10-16 01:25:53 -0700755
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700756 area = &(zone->free_area[current_order]);
757 if (list_empty(&area->free_list[migratetype]))
758 continue;
759
760 page = list_entry(area->free_list[migratetype].next,
761 struct page, lru);
762 area->nr_free--;
763
764 /*
Mel Gormanc361be52007-10-16 01:25:51 -0700765 * If breaking a large block of pages, move all free
Mel Gorman46dafbc2007-10-16 01:25:55 -0700766 * pages to the preferred allocation list. If falling
767 * back for a reclaimable kernel allocation, be more
768 * agressive about taking ownership of free pages
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700769 */
Mel Gormand9c23402007-10-16 01:26:01 -0700770 if (unlikely(current_order >= (pageblock_order >> 1)) ||
Mel Gorman46dafbc2007-10-16 01:25:55 -0700771 start_migratetype == MIGRATE_RECLAIMABLE) {
772 unsigned long pages;
773 pages = move_freepages_block(zone, page,
774 start_migratetype);
775
776 /* Claim the whole block if over half of it is free */
Mel Gormand9c23402007-10-16 01:26:01 -0700777 if (pages >= (1 << (pageblock_order-1)))
Mel Gorman46dafbc2007-10-16 01:25:55 -0700778 set_pageblock_migratetype(page,
779 start_migratetype);
780
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700781 migratetype = start_migratetype;
Mel Gormanc361be52007-10-16 01:25:51 -0700782 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700783
784 /* Remove the page from the freelists */
785 list_del(&page->lru);
786 rmv_page_order(page);
787 __mod_zone_page_state(zone, NR_FREE_PAGES,
788 -(1UL << order));
789
Mel Gormand9c23402007-10-16 01:26:01 -0700790 if (current_order == pageblock_order)
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700791 set_pageblock_migratetype(page,
792 start_migratetype);
793
794 expand(zone, page, order, current_order, area, migratetype);
795 return page;
796 }
797 }
798
Mel Gorman56fd56b2007-10-16 01:25:58 -0700799 /* Use MIGRATE_RESERVE rather than fail an allocation */
800 return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700801}
802
Mel Gorman56fd56b2007-10-16 01:25:58 -0700803/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 * Do the hard work of removing an element from the buddy allocator.
805 * Call me with the zone->lock already held.
806 */
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700807static struct page *__rmqueue(struct zone *zone, unsigned int order,
808 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 struct page *page;
811
Mel Gorman56fd56b2007-10-16 01:25:58 -0700812 page = __rmqueue_smallest(zone, order, migratetype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
Mel Gorman56fd56b2007-10-16 01:25:58 -0700814 if (unlikely(!page))
815 page = __rmqueue_fallback(zone, order, migratetype);
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700816
817 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818}
819
820/*
821 * Obtain a specified number of elements from the buddy allocator, all under
822 * a single hold of the lock, for efficiency. Add them to the supplied list.
823 * Returns the number of new pages which were placed at *list.
824 */
825static int rmqueue_bulk(struct zone *zone, unsigned int order,
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700826 unsigned long count, struct list_head *list,
827 int migratetype)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
Nick Pigginc54ad302006-01-06 00:10:56 -0800831 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 for (i = 0; i < count; ++i) {
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700833 struct page *page = __rmqueue(zone, order, migratetype);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800834 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 break;
Mel Gorman81eabcb2007-12-17 16:20:05 -0800836
837 /*
838 * Split buddy pages returned by expand() are received here
839 * in physical page order. The page is added to the callers and
840 * list and the list head then moves forward. From the callers
841 * perspective, the linked list is ordered by page number in
842 * some conditions. This is useful for IO devices that can
843 * merge IO requests if the physical pages are ordered
844 * properly.
845 */
Mel Gorman535131e62007-10-16 01:25:49 -0700846 list_add(&page->lru, list);
847 set_page_private(page, migratetype);
Mel Gorman81eabcb2007-12-17 16:20:05 -0800848 list = &page->lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800850 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800851 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852}
853
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700854#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800855/*
Christoph Lameter4037d452007-05-09 02:35:14 -0700856 * Called from the vmstat counter updater to drain pagesets of this
857 * currently executing processor on remote nodes after they have
858 * expired.
859 *
Christoph Lameter879336c2006-03-22 00:09:08 -0800860 * Note that this function must be called with the thread pinned to
861 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800862 */
Christoph Lameter4037d452007-05-09 02:35:14 -0700863void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700864{
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700865 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -0700866 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700867
Christoph Lameter4037d452007-05-09 02:35:14 -0700868 local_irq_save(flags);
869 if (pcp->count >= pcp->batch)
870 to_drain = pcp->batch;
871 else
872 to_drain = pcp->count;
873 free_pages_bulk(zone, to_drain, &pcp->list, 0);
874 pcp->count -= to_drain;
875 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700876}
877#endif
878
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800879/*
880 * Drain pages of the indicated processor.
881 *
882 * The processor must either be the current processor and the
883 * thread pinned to the current processor or a processor that
884 * is not online.
885 */
886static void drain_pages(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887{
Nick Pigginc54ad302006-01-06 00:10:56 -0800888 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
891 for_each_zone(zone) {
892 struct per_cpu_pageset *pset;
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800893 struct per_cpu_pages *pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
Christoph Lameterf2e12bb2007-01-05 16:37:02 -0800895 if (!populated_zone(zone))
896 continue;
897
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700898 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800900 pcp = &pset->pcp;
901 local_irq_save(flags);
902 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
903 pcp->count = 0;
904 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 }
906}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800908/*
909 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
910 */
911void drain_local_pages(void *arg)
912{
913 drain_pages(smp_processor_id());
914}
915
916/*
917 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
918 */
919void drain_all_pages(void)
920{
Jens Axboe15c8b6c2008-05-09 09:39:44 +0200921 on_each_cpu(drain_local_pages, NULL, 1);
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800922}
923
Rafael J. Wysocki296699d2007-07-29 23:27:18 +0200924#ifdef CONFIG_HIBERNATION
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
926void mark_free_pages(struct zone *zone)
927{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700928 unsigned long pfn, max_zone_pfn;
929 unsigned long flags;
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700930 int order, t;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 struct list_head *curr;
932
933 if (!zone->spanned_pages)
934 return;
935
936 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700937
938 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
939 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
940 if (pfn_valid(pfn)) {
941 struct page *page = pfn_to_page(pfn);
942
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700943 if (!swsusp_page_is_forbidden(page))
944 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700945 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700947 for_each_migratetype_order(order, t) {
948 list_for_each(curr, &zone->free_area[order].free_list[t]) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700949 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700951 pfn = page_to_pfn(list_entry(curr, struct page, lru));
952 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700953 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700954 }
Mel Gormanb2a0ac82007-10-16 01:25:48 -0700955 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 spin_unlock_irqrestore(&zone->lock, flags);
957}
Mel Gormane2c55dc2007-10-16 01:25:50 -0700958#endif /* CONFIG_PM */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
960/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 * Free a 0-order page
962 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800963static void free_hot_cold_page(struct page *page, int cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964{
965 struct zone *zone = page_zone(page);
966 struct per_cpu_pages *pcp;
967 unsigned long flags;
968
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 if (PageAnon(page))
970 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800971 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800972 return;
973
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700974 if (!PageHighMem(page)) {
Nick Piggin9858db52006-10-11 01:21:30 -0700975 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700976 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
977 }
Nick Piggindafb1362006-10-11 01:21:30 -0700978 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800979 kernel_map_pages(page, 1, 0);
980
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800981 pcp = &zone_pcp(zone, get_cpu())->pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700983 __count_vm_event(PGFREE);
Christoph Lameter3dfa5722008-02-04 22:29:19 -0800984 if (cold)
985 list_add_tail(&page->lru, &pcp->list);
986 else
987 list_add(&page->lru, &pcp->list);
Mel Gorman535131e62007-10-16 01:25:49 -0700988 set_page_private(page, get_pageblock_migratetype(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800990 if (pcp->count >= pcp->high) {
991 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
992 pcp->count -= pcp->batch;
993 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 local_irq_restore(flags);
995 put_cpu();
996}
997
Harvey Harrison920c7a52008-02-04 22:29:26 -0800998void free_hot_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999{
1000 free_hot_cold_page(page, 0);
1001}
1002
Harvey Harrison920c7a52008-02-04 22:29:26 -08001003void free_cold_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004{
1005 free_hot_cold_page(page, 1);
1006}
1007
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001008/*
1009 * split_page takes a non-compound higher-order page, and splits it into
1010 * n (1<<order) sub-pages: page[0..n]
1011 * Each sub-page must be freed individually.
1012 *
1013 * Note: this is probably too low level an operation for use in drivers.
1014 * Please consult with lkml before using this in your driver.
1015 */
1016void split_page(struct page *page, unsigned int order)
1017{
1018 int i;
1019
Nick Piggin725d7042006-09-25 23:30:55 -07001020 VM_BUG_ON(PageCompound(page));
1021 VM_BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -08001022 for (i = 1; i < (1 << order); i++)
1023 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001024}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001025
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026/*
1027 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1028 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1029 * or two.
1030 */
Mel Gorman18ea7e72008-04-28 02:12:14 -07001031static struct page *buffered_rmqueue(struct zone *preferred_zone,
Nick Piggina74609f2006-01-06 00:11:20 -08001032 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033{
1034 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -08001035 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -08001037 int cpu;
Mel Gorman64c5e132007-10-16 01:25:59 -07001038 int migratetype = allocflags_to_migratetype(gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
Hugh Dickins689bceb2005-11-21 21:32:20 -08001040again:
Nick Piggina74609f2006-01-06 00:11:20 -08001041 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -08001042 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 struct per_cpu_pages *pcp;
1044
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001045 pcp = &zone_pcp(zone, cpu)->pcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -08001047 if (!pcp->count) {
nkalmala941c7102006-11-02 22:07:04 -08001048 pcp->count = rmqueue_bulk(zone, 0,
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001049 pcp->batch, &pcp->list, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001050 if (unlikely(!pcp->count))
1051 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 }
Mel Gorman535131e62007-10-16 01:25:49 -07001053
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001054 /* Find a page of the appropriate migrate type */
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001055 if (cold) {
1056 list_for_each_entry_reverse(page, &pcp->list, lru)
1057 if (page_private(page) == migratetype)
1058 break;
1059 } else {
1060 list_for_each_entry(page, &pcp->list, lru)
1061 if (page_private(page) == migratetype)
1062 break;
1063 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001064
1065 /* Allocate more to the pcp list if necessary */
1066 if (unlikely(&page->lru == &pcp->list)) {
Mel Gorman535131e62007-10-16 01:25:49 -07001067 pcp->count += rmqueue_bulk(zone, 0,
1068 pcp->batch, &pcp->list, migratetype);
1069 page = list_entry(pcp->list.next, struct page, lru);
Mel Gorman535131e62007-10-16 01:25:49 -07001070 }
Mel Gormanb92a6ed2007-10-16 01:25:50 -07001071
1072 list_del(&page->lru);
1073 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001074 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 spin_lock_irqsave(&zone->lock, flags);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07001076 page = __rmqueue(zone, order, migratetype);
Nick Piggina74609f2006-01-06 00:11:20 -08001077 spin_unlock(&zone->lock);
1078 if (!page)
1079 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 }
1081
Christoph Lameterf8891e52006-06-30 01:55:45 -07001082 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Mel Gorman18ea7e72008-04-28 02:12:14 -07001083 zone_statistics(preferred_zone, zone);
Nick Piggina74609f2006-01-06 00:11:20 -08001084 local_irq_restore(flags);
1085 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
Nick Piggin725d7042006-09-25 23:30:55 -07001087 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -08001088 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -08001089 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 return page;
Nick Piggina74609f2006-01-06 00:11:20 -08001091
1092failed:
1093 local_irq_restore(flags);
1094 put_cpu();
1095 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096}
1097
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001098#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -08001099#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1100#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1101#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1102#define ALLOC_HARDER 0x10 /* try to alloc harder */
1103#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1104#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001105
Akinobu Mita933e3122006-12-08 02:39:45 -08001106#ifdef CONFIG_FAIL_PAGE_ALLOC
1107
1108static struct fail_page_alloc_attr {
1109 struct fault_attr attr;
1110
1111 u32 ignore_gfp_highmem;
1112 u32 ignore_gfp_wait;
Akinobu Mita54114992007-07-15 23:40:23 -07001113 u32 min_order;
Akinobu Mita933e3122006-12-08 02:39:45 -08001114
1115#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1116
1117 struct dentry *ignore_gfp_highmem_file;
1118 struct dentry *ignore_gfp_wait_file;
Akinobu Mita54114992007-07-15 23:40:23 -07001119 struct dentry *min_order_file;
Akinobu Mita933e3122006-12-08 02:39:45 -08001120
1121#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1122
1123} fail_page_alloc = {
1124 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -08001125 .ignore_gfp_wait = 1,
1126 .ignore_gfp_highmem = 1,
Akinobu Mita54114992007-07-15 23:40:23 -07001127 .min_order = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -08001128};
1129
1130static int __init setup_fail_page_alloc(char *str)
1131{
1132 return setup_fault_attr(&fail_page_alloc.attr, str);
1133}
1134__setup("fail_page_alloc=", setup_fail_page_alloc);
1135
1136static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1137{
Akinobu Mita54114992007-07-15 23:40:23 -07001138 if (order < fail_page_alloc.min_order)
1139 return 0;
Akinobu Mita933e3122006-12-08 02:39:45 -08001140 if (gfp_mask & __GFP_NOFAIL)
1141 return 0;
1142 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1143 return 0;
1144 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1145 return 0;
1146
1147 return should_fail(&fail_page_alloc.attr, 1 << order);
1148}
1149
1150#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1151
1152static int __init fail_page_alloc_debugfs(void)
1153{
1154 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1155 struct dentry *dir;
1156 int err;
1157
1158 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1159 "fail_page_alloc");
1160 if (err)
1161 return err;
1162 dir = fail_page_alloc.attr.dentries.dir;
1163
1164 fail_page_alloc.ignore_gfp_wait_file =
1165 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1166 &fail_page_alloc.ignore_gfp_wait);
1167
1168 fail_page_alloc.ignore_gfp_highmem_file =
1169 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1170 &fail_page_alloc.ignore_gfp_highmem);
Akinobu Mita54114992007-07-15 23:40:23 -07001171 fail_page_alloc.min_order_file =
1172 debugfs_create_u32("min-order", mode, dir,
1173 &fail_page_alloc.min_order);
Akinobu Mita933e3122006-12-08 02:39:45 -08001174
1175 if (!fail_page_alloc.ignore_gfp_wait_file ||
Akinobu Mita54114992007-07-15 23:40:23 -07001176 !fail_page_alloc.ignore_gfp_highmem_file ||
1177 !fail_page_alloc.min_order_file) {
Akinobu Mita933e3122006-12-08 02:39:45 -08001178 err = -ENOMEM;
1179 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1180 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
Akinobu Mita54114992007-07-15 23:40:23 -07001181 debugfs_remove(fail_page_alloc.min_order_file);
Akinobu Mita933e3122006-12-08 02:39:45 -08001182 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1183 }
1184
1185 return err;
1186}
1187
1188late_initcall(fail_page_alloc_debugfs);
1189
1190#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1191
1192#else /* CONFIG_FAIL_PAGE_ALLOC */
1193
1194static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1195{
1196 return 0;
1197}
1198
1199#endif /* CONFIG_FAIL_PAGE_ALLOC */
1200
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201/*
1202 * Return 1 if free pages are above 'mark'. This takes into account the order
1203 * of the allocation.
1204 */
1205int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001206 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207{
1208 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001209 long min = mark;
1210 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 int o;
1212
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001213 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001215 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 min -= min / 4;
1217
1218 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1219 return 0;
1220 for (o = 0; o < order; o++) {
1221 /* At the next order, this order's pages become unavailable */
1222 free_pages -= z->free_area[o].nr_free << o;
1223
1224 /* Require fewer higher order pages to be free */
1225 min >>= 1;
1226
1227 if (free_pages <= min)
1228 return 0;
1229 }
1230 return 1;
1231}
1232
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001233#ifdef CONFIG_NUMA
1234/*
1235 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1236 * skip over zones that are not allowed by the cpuset, or that have
1237 * been recently (in last second) found to be nearly full. See further
1238 * comments in mmzone.h. Reduces cache footprint of zonelist scans
Simon Arlott183ff222007-10-20 01:27:18 +02001239 * that have to skip over a lot of full or unallowed zones.
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001240 *
1241 * If the zonelist cache is present in the passed in zonelist, then
1242 * returns a pointer to the allowed node mask (either the current
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001243 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001244 *
1245 * If the zonelist cache is not available for this zonelist, does
1246 * nothing and returns NULL.
1247 *
1248 * If the fullzones BITMAP in the zonelist cache is stale (more than
1249 * a second since last zap'd) then we zap it out (clear its bits.)
1250 *
1251 * We hold off even calling zlc_setup, until after we've checked the
1252 * first zone in the zonelist, on the theory that most allocations will
1253 * be satisfied from that first zone, so best to examine that zone as
1254 * quickly as we can.
1255 */
1256static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1257{
1258 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1259 nodemask_t *allowednodes; /* zonelist_cache approximation */
1260
1261 zlc = zonelist->zlcache_ptr;
1262 if (!zlc)
1263 return NULL;
1264
S.Caglar Onurf05111f2008-04-28 02:12:38 -07001265 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001266 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1267 zlc->last_full_zap = jiffies;
1268 }
1269
1270 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1271 &cpuset_current_mems_allowed :
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001272 &node_states[N_HIGH_MEMORY];
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001273 return allowednodes;
1274}
1275
1276/*
1277 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1278 * if it is worth looking at further for free memory:
1279 * 1) Check that the zone isn't thought to be full (doesn't have its
1280 * bit set in the zonelist_cache fullzones BITMAP).
1281 * 2) Check that the zones node (obtained from the zonelist_cache
1282 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1283 * Return true (non-zero) if zone is worth looking at further, or
1284 * else return false (zero) if it is not.
1285 *
1286 * This check -ignores- the distinction between various watermarks,
1287 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1288 * found to be full for any variation of these watermarks, it will
1289 * be considered full for up to one second by all requests, unless
1290 * we are so low on memory on all allowed nodes that we are forced
1291 * into the second scan of the zonelist.
1292 *
1293 * In the second scan we ignore this zonelist cache and exactly
1294 * apply the watermarks to all zones, even it is slower to do so.
1295 * We are low on memory in the second scan, and should leave no stone
1296 * unturned looking for a free page.
1297 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001298static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001299 nodemask_t *allowednodes)
1300{
1301 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1302 int i; /* index of *z in zonelist zones */
1303 int n; /* node that zone *z is on */
1304
1305 zlc = zonelist->zlcache_ptr;
1306 if (!zlc)
1307 return 1;
1308
Mel Gormandd1a2392008-04-28 02:12:17 -07001309 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001310 n = zlc->z_to_n[i];
1311
1312 /* This zone is worth trying if it is allowed but not full */
1313 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1314}
1315
1316/*
1317 * Given 'z' scanning a zonelist, set the corresponding bit in
1318 * zlc->fullzones, so that subsequent attempts to allocate a page
1319 * from that zone don't waste time re-examining it.
1320 */
Mel Gormandd1a2392008-04-28 02:12:17 -07001321static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001322{
1323 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1324 int i; /* index of *z in zonelist zones */
1325
1326 zlc = zonelist->zlcache_ptr;
1327 if (!zlc)
1328 return;
1329
Mel Gormandd1a2392008-04-28 02:12:17 -07001330 i = z - zonelist->_zonerefs;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001331
1332 set_bit(i, zlc->fullzones);
1333}
1334
1335#else /* CONFIG_NUMA */
1336
1337static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1338{
1339 return NULL;
1340}
1341
Mel Gormandd1a2392008-04-28 02:12:17 -07001342static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001343 nodemask_t *allowednodes)
1344{
1345 return 1;
1346}
1347
Mel Gormandd1a2392008-04-28 02:12:17 -07001348static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001349{
1350}
1351#endif /* CONFIG_NUMA */
1352
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001353/*
Paul Jackson0798e512006-12-06 20:31:38 -08001354 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001355 * a page.
1356 */
1357static struct page *
Mel Gorman19770b32008-04-28 02:12:18 -07001358get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001359 struct zonelist *zonelist, int high_zoneidx, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -07001360{
Mel Gormandd1a2392008-04-28 02:12:17 -07001361 struct zoneref *z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001362 struct page *page = NULL;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001363 int classzone_idx;
Mel Gorman18ea7e72008-04-28 02:12:14 -07001364 struct zone *zone, *preferred_zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001365 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1366 int zlc_active = 0; /* set if using zonelist_cache */
1367 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Mel Gorman54a6eb52008-04-28 02:12:16 -07001368
Mel Gorman19770b32008-04-28 02:12:18 -07001369 (void)first_zones_zonelist(zonelist, high_zoneidx, nodemask,
1370 &preferred_zone);
Andy Whitcroft7eb54822008-05-23 13:04:50 -07001371 if (!preferred_zone)
1372 return NULL;
1373
Mel Gorman19770b32008-04-28 02:12:18 -07001374 classzone_idx = zone_idx(preferred_zone);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001375
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001376zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001377 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001378 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001379 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1380 */
Mel Gorman19770b32008-04-28 02:12:18 -07001381 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1382 high_zoneidx, nodemask) {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001383 if (NUMA_BUILD && zlc_active &&
1384 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1385 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001386 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001387 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001388 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001389
1390 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001391 unsigned long mark;
1392 if (alloc_flags & ALLOC_WMARK_MIN)
Christoph Lameter1192d522006-09-25 23:31:45 -07001393 mark = zone->pages_min;
Nick Piggin31488902005-11-28 13:44:03 -08001394 else if (alloc_flags & ALLOC_WMARK_LOW)
Christoph Lameter1192d522006-09-25 23:31:45 -07001395 mark = zone->pages_low;
Nick Piggin31488902005-11-28 13:44:03 -08001396 else
Christoph Lameter1192d522006-09-25 23:31:45 -07001397 mark = zone->pages_high;
Paul Jackson0798e512006-12-06 20:31:38 -08001398 if (!zone_watermark_ok(zone, order, mark,
1399 classzone_idx, alloc_flags)) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001400 if (!zone_reclaim_mode ||
Christoph Lameter1192d522006-09-25 23:31:45 -07001401 !zone_reclaim(zone, gfp_mask, order))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001402 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001403 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001404 }
1405
Mel Gorman18ea7e72008-04-28 02:12:14 -07001406 page = buffered_rmqueue(preferred_zone, zone, order, gfp_mask);
Paul Jackson0798e512006-12-06 20:31:38 -08001407 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001408 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001409this_zone_full:
1410 if (NUMA_BUILD)
1411 zlc_mark_zone_full(zonelist, z);
1412try_next_zone:
1413 if (NUMA_BUILD && !did_zlc_setup) {
1414 /* we do zlc_setup after the first zone is tried */
1415 allowednodes = zlc_setup(zonelist, alloc_flags);
1416 zlc_active = 1;
1417 did_zlc_setup = 1;
1418 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07001419 }
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001420
1421 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1422 /* Disable zlc cache for second zonelist scan */
1423 zlc_active = 0;
1424 goto zonelist_scan;
1425 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001426 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001427}
1428
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429/*
1430 * This is the 'heart' of the zoned buddy allocator.
1431 */
Mel Gorman19770b32008-04-28 02:12:18 -07001432static struct page *
1433__alloc_pages_internal(gfp_t gfp_mask, unsigned int order,
1434 struct zonelist *zonelist, nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435{
Al Viro260b2362005-10-21 03:22:44 -04001436 const gfp_t wait = gfp_mask & __GFP_WAIT;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001437 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Mel Gormandd1a2392008-04-28 02:12:17 -07001438 struct zoneref *z;
1439 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 struct page *page;
1441 struct reclaim_state reclaim_state;
1442 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001444 int alloc_flags;
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001445 unsigned long did_some_progress;
1446 unsigned long pages_reclaimed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
1448 might_sleep_if(wait);
1449
Akinobu Mita933e3122006-12-08 02:39:45 -08001450 if (should_fail_alloc_page(gfp_mask, order))
1451 return NULL;
1452
Jens Axboe6b1de912005-11-17 21:35:02 +01001453restart:
Mel Gormandd1a2392008-04-28 02:12:17 -07001454 z = zonelist->_zonerefs; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
Mel Gormandd1a2392008-04-28 02:12:17 -07001456 if (unlikely(!z->zone)) {
Christoph Lameter523b9452007-10-16 01:25:37 -07001457 /*
1458 * Happens if we have an empty zonelist as a result of
1459 * GFP_THISNODE being used on a memoryless node
1460 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 return NULL;
1462 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001463
Mel Gorman19770b32008-04-28 02:12:18 -07001464 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001465 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001466 if (page)
1467 goto got_pg;
1468
Christoph Lameter952f3b52006-12-06 20:33:26 -08001469 /*
1470 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1471 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1472 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1473 * using a larger set of nodes after it has established that the
1474 * allowed per node queues are empty and that nodes are
1475 * over allocated.
1476 */
1477 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1478 goto nopage;
1479
Mel Gormandd1a2392008-04-28 02:12:17 -07001480 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1481 wakeup_kswapd(zone, order);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001482
Paul Jackson9bf22292005-09-06 15:18:12 -07001483 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001484 * OK, we're below the kswapd watermark and have kicked background
1485 * reclaim. Now things get more complex, so set up alloc_flags according
1486 * to how we want to proceed.
1487 *
1488 * The caller may dip into page reserves a bit more if the caller
1489 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -08001490 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1491 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -07001492 */
Nick Piggin31488902005-11-28 13:44:03 -08001493 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001494 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1495 alloc_flags |= ALLOC_HARDER;
1496 if (gfp_mask & __GFP_HIGH)
1497 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -07001498 if (wait)
1499 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500
1501 /*
1502 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001503 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 *
1505 * This is the last chance, in general, before the goto nopage.
1506 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -07001507 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 */
Mel Gorman19770b32008-04-28 02:12:18 -07001509 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001510 high_zoneidx, alloc_flags);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001511 if (page)
1512 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
1514 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -07001515
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001516rebalance:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001517 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1518 && !in_interrupt()) {
1519 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -08001520nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001521 /* go through the zonelist yet again, ignoring mins */
Mel Gorman19770b32008-04-28 02:12:18 -07001522 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001523 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001524 if (page)
1525 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -08001526 if (gfp_mask & __GFP_NOFAIL) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001527 congestion_wait(WRITE, HZ/50);
Kirill Korotaev885036d2005-11-13 16:06:41 -08001528 goto nofail_alloc;
1529 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 }
1531 goto nopage;
1532 }
1533
1534 /* Atomic allocations - we can't balance anything */
1535 if (!wait)
1536 goto nopage;
1537
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 cond_resched();
1539
1540 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001541 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 p->flags |= PF_MEMALLOC;
1543 reclaim_state.reclaimed_slab = 0;
1544 p->reclaim_state = &reclaim_state;
1545
Mel Gormandac1d272008-04-28 02:12:12 -07001546 did_some_progress = try_to_free_pages(zonelist, order, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
1548 p->reclaim_state = NULL;
1549 p->flags &= ~PF_MEMALLOC;
1550
1551 cond_resched();
1552
Mel Gormane2c55dc2007-10-16 01:25:50 -07001553 if (order != 0)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08001554 drain_all_pages();
Mel Gormane2c55dc2007-10-16 01:25:50 -07001555
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 if (likely(did_some_progress)) {
Mel Gorman19770b32008-04-28 02:12:18 -07001557 page = get_page_from_freelist(gfp_mask, nodemask, order,
Mel Gorman54a6eb52008-04-28 02:12:16 -07001558 zonelist, high_zoneidx, alloc_flags);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001559 if (page)
1560 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07001562 if (!try_set_zone_oom(zonelist, gfp_mask)) {
David Rientjesff0ceb92007-10-16 23:25:56 -07001563 schedule_timeout_uninterruptible(1);
1564 goto restart;
1565 }
1566
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 /*
1568 * Go through the zonelist yet one more time, keep
1569 * very high watermark here, this is only to catch
1570 * a parallel oom killing, we must fail if we're still
1571 * under heavy pressure.
1572 */
Mel Gorman19770b32008-04-28 02:12:18 -07001573 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1574 order, zonelist, high_zoneidx,
1575 ALLOC_WMARK_HIGH|ALLOC_CPUSET);
David Rientjesff0ceb92007-10-16 23:25:56 -07001576 if (page) {
Mel Gormandd1a2392008-04-28 02:12:17 -07001577 clear_zonelist_oom(zonelist, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001578 goto got_pg;
David Rientjesff0ceb92007-10-16 23:25:56 -07001579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580
Mel Gormana8bbf722007-07-31 00:37:30 -07001581 /* The OOM killer will not help higher order allocs so fail */
David Rientjesff0ceb92007-10-16 23:25:56 -07001582 if (order > PAGE_ALLOC_COSTLY_ORDER) {
Mel Gormandd1a2392008-04-28 02:12:17 -07001583 clear_zonelist_oom(zonelist, gfp_mask);
Mel Gormana8bbf722007-07-31 00:37:30 -07001584 goto nopage;
David Rientjesff0ceb92007-10-16 23:25:56 -07001585 }
Mel Gormana8bbf722007-07-31 00:37:30 -07001586
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001587 out_of_memory(zonelist, gfp_mask, order);
Mel Gormandd1a2392008-04-28 02:12:17 -07001588 clear_zonelist_oom(zonelist, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 goto restart;
1590 }
1591
1592 /*
1593 * Don't let big-order allocations loop unless the caller explicitly
1594 * requests that. Wait for some write requests to complete then retry.
1595 *
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001596 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1597 * means __GFP_NOFAIL, but that may not be true in other
Nishanth Aravamudanab857d02008-04-29 00:58:23 -07001598 * implementations.
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001599 *
1600 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1601 * specified, then we retry until we no longer reclaim any pages
1602 * (above), or we've reclaimed an order of pages at least as
1603 * large as the allocation's order. In both cases, if the
1604 * allocation still fails, we stop retrying.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 */
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001606 pages_reclaimed += did_some_progress;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 do_retry = 0;
1608 if (!(gfp_mask & __GFP_NORETRY)) {
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001609 if (order <= PAGE_ALLOC_COSTLY_ORDER) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 do_retry = 1;
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001611 } else {
1612 if (gfp_mask & __GFP_REPEAT &&
1613 pages_reclaimed < (1 << order))
1614 do_retry = 1;
1615 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 if (gfp_mask & __GFP_NOFAIL)
1617 do_retry = 1;
1618 }
1619 if (do_retry) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001620 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 goto rebalance;
1622 }
1623
1624nopage:
1625 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1626 printk(KERN_WARNING "%s: page allocation failure."
1627 " order:%d, mode:0x%x\n",
1628 p->comm, order, gfp_mask);
1629 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001630 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 return page;
1634}
1635
Mel Gorman19770b32008-04-28 02:12:18 -07001636struct page *
1637__alloc_pages(gfp_t gfp_mask, unsigned int order,
1638 struct zonelist *zonelist)
1639{
1640 return __alloc_pages_internal(gfp_mask, order, zonelist, NULL);
1641}
1642
1643struct page *
1644__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
1645 struct zonelist *zonelist, nodemask_t *nodemask)
1646{
1647 return __alloc_pages_internal(gfp_mask, order, zonelist, nodemask);
1648}
1649
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650EXPORT_SYMBOL(__alloc_pages);
1651
1652/*
1653 * Common helper functions.
1654 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001655unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656{
1657 struct page * page;
1658 page = alloc_pages(gfp_mask, order);
1659 if (!page)
1660 return 0;
1661 return (unsigned long) page_address(page);
1662}
1663
1664EXPORT_SYMBOL(__get_free_pages);
1665
Harvey Harrison920c7a52008-02-04 22:29:26 -08001666unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667{
1668 struct page * page;
1669
1670 /*
1671 * get_zeroed_page() returns a 32-bit address, which cannot represent
1672 * a highmem page
1673 */
Nick Piggin725d7042006-09-25 23:30:55 -07001674 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
1676 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1677 if (page)
1678 return (unsigned long) page_address(page);
1679 return 0;
1680}
1681
1682EXPORT_SYMBOL(get_zeroed_page);
1683
1684void __pagevec_free(struct pagevec *pvec)
1685{
1686 int i = pagevec_count(pvec);
1687
1688 while (--i >= 0)
1689 free_hot_cold_page(pvec->pages[i], pvec->cold);
1690}
1691
Harvey Harrison920c7a52008-02-04 22:29:26 -08001692void __free_pages(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693{
Nick Pigginb5810032005-10-29 18:16:12 -07001694 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 if (order == 0)
1696 free_hot_page(page);
1697 else
1698 __free_pages_ok(page, order);
1699 }
1700}
1701
1702EXPORT_SYMBOL(__free_pages);
1703
Harvey Harrison920c7a52008-02-04 22:29:26 -08001704void free_pages(unsigned long addr, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705{
1706 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001707 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 __free_pages(virt_to_page((void *)addr), order);
1709 }
1710}
1711
1712EXPORT_SYMBOL(free_pages);
1713
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714static unsigned int nr_free_zone_pages(int offset)
1715{
Mel Gormandd1a2392008-04-28 02:12:17 -07001716 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001717 struct zone *zone;
1718
Martin J. Blighe310fd42005-07-29 22:59:18 -07001719 /* Just pick one node, since fallback list is circular */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 unsigned int sum = 0;
1721
Mel Gorman0e884602008-04-28 02:12:14 -07001722 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
Mel Gorman54a6eb52008-04-28 02:12:16 -07001724 for_each_zone_zonelist(zone, z, zonelist, offset) {
Martin J. Blighe310fd42005-07-29 22:59:18 -07001725 unsigned long size = zone->present_pages;
1726 unsigned long high = zone->pages_high;
1727 if (size > high)
1728 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 }
1730
1731 return sum;
1732}
1733
1734/*
1735 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1736 */
1737unsigned int nr_free_buffer_pages(void)
1738{
Al Viroaf4ca452005-10-21 02:55:38 -04001739 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740}
Meelap Shahc2f1a552007-07-17 04:04:39 -07001741EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
1743/*
1744 * Amount of free RAM allocatable within all zones
1745 */
1746unsigned int nr_free_pagecache_pages(void)
1747{
Mel Gorman2a1e2742007-07-17 04:03:12 -07001748 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001750
1751static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001753 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08001754 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757void si_meminfo(struct sysinfo *val)
1758{
1759 val->totalram = totalram_pages;
1760 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001761 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 val->totalhigh = totalhigh_pages;
1764 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 val->mem_unit = PAGE_SIZE;
1766}
1767
1768EXPORT_SYMBOL(si_meminfo);
1769
1770#ifdef CONFIG_NUMA
1771void si_meminfo_node(struct sysinfo *val, int nid)
1772{
1773 pg_data_t *pgdat = NODE_DATA(nid);
1774
1775 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001776 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001777#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001779 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1780 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001781#else
1782 val->totalhigh = 0;
1783 val->freehigh = 0;
1784#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 val->mem_unit = PAGE_SIZE;
1786}
1787#endif
1788
1789#define K(x) ((x) << (PAGE_SHIFT-10))
1790
1791/*
1792 * Show free area list (used inside shift_scroll-lock stuff)
1793 * We also calculate the percentage fragmentation. We do this by counting the
1794 * memory on each free list with the exception of the first item on the list.
1795 */
1796void show_free_areas(void)
1797{
Jes Sorensenc7241912006-09-27 01:50:05 -07001798 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 struct zone *zone;
1800
1801 for_each_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07001802 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07001804
1805 show_node(zone);
1806 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
Dave Jones6b482c62005-11-10 15:45:56 -05001808 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 struct per_cpu_pageset *pageset;
1810
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001811 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
Christoph Lameter3dfa5722008-02-04 22:29:19 -08001813 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1814 cpu, pageset->pcp.high,
1815 pageset->pcp.batch, pageset->pcp.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 }
1817 }
1818
Andrew Mortona25700a2007-02-08 14:20:40 -08001819 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
Christoph Lameterd23ad422007-02-10 01:43:02 -08001820 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
Christoph Lameter65e458d42007-02-10 01:43:05 -08001821 global_page_state(NR_ACTIVE),
1822 global_page_state(NR_INACTIVE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07001823 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07001824 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001825 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08001826 global_page_state(NR_FREE_PAGES),
Christoph Lameter972d1a72006-09-25 23:31:51 -07001827 global_page_state(NR_SLAB_RECLAIMABLE) +
1828 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001829 global_page_state(NR_FILE_MAPPED),
Andrew Mortona25700a2007-02-08 14:20:40 -08001830 global_page_state(NR_PAGETABLE),
1831 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832
1833 for_each_zone(zone) {
1834 int i;
1835
Jes Sorensenc7241912006-09-27 01:50:05 -07001836 if (!populated_zone(zone))
1837 continue;
1838
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 show_node(zone);
1840 printk("%s"
1841 " free:%lukB"
1842 " min:%lukB"
1843 " low:%lukB"
1844 " high:%lukB"
1845 " active:%lukB"
1846 " inactive:%lukB"
1847 " present:%lukB"
1848 " pages_scanned:%lu"
1849 " all_unreclaimable? %s"
1850 "\n",
1851 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08001852 K(zone_page_state(zone, NR_FREE_PAGES)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 K(zone->pages_min),
1854 K(zone->pages_low),
1855 K(zone->pages_high),
Christoph Lameterc8785382007-02-10 01:43:01 -08001856 K(zone_page_state(zone, NR_ACTIVE)),
1857 K(zone_page_state(zone, NR_INACTIVE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 K(zone->present_pages),
1859 zone->pages_scanned,
David Rientjese815af92007-10-16 23:25:54 -07001860 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 );
1862 printk("lowmem_reserve[]:");
1863 for (i = 0; i < MAX_NR_ZONES; i++)
1864 printk(" %lu", zone->lowmem_reserve[i]);
1865 printk("\n");
1866 }
1867
1868 for_each_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001869 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
Jes Sorensenc7241912006-09-27 01:50:05 -07001871 if (!populated_zone(zone))
1872 continue;
1873
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 show_node(zone);
1875 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876
1877 spin_lock_irqsave(&zone->lock, flags);
1878 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001879 nr[order] = zone->free_area[order].nr_free;
1880 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 }
1882 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001883 for (order = 0; order < MAX_ORDER; order++)
1884 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 printk("= %lukB\n", K(total));
1886 }
1887
Larry Woodmane6f36022008-02-04 22:29:30 -08001888 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
1889
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 show_swap_cache_info();
1891}
1892
Mel Gorman19770b32008-04-28 02:12:18 -07001893static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
1894{
1895 zoneref->zone = zone;
1896 zoneref->zone_idx = zone_idx(zone);
1897}
1898
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899/*
1900 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001901 *
1902 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001904static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
1905 int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906{
Christoph Lameter1a932052006-01-06 00:11:16 -08001907 struct zone *zone;
1908
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001909 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001910 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001911
1912 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001913 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08001914 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001915 if (populated_zone(zone)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07001916 zoneref_set_zone(zone,
1917 &zonelist->_zonerefs[nr_zones++]);
Christoph Lameter070f8032006-01-06 00:11:19 -08001918 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08001920
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001921 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08001922 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923}
1924
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001925
1926/*
1927 * zonelist_order:
1928 * 0 = automatic detection of better ordering.
1929 * 1 = order by ([node] distance, -zonetype)
1930 * 2 = order by (-zonetype, [node] distance)
1931 *
1932 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1933 * the same zonelist. So only NUMA can configure this param.
1934 */
1935#define ZONELIST_ORDER_DEFAULT 0
1936#define ZONELIST_ORDER_NODE 1
1937#define ZONELIST_ORDER_ZONE 2
1938
1939/* zonelist order in the kernel.
1940 * set_zonelist_order() will set this to NODE or ZONE.
1941 */
1942static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
1943static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
1944
1945
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946#ifdef CONFIG_NUMA
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07001947/* The value user specified ....changed by config */
1948static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1949/* string for sysctl */
1950#define NUMA_ZONELIST_ORDER_LEN 16
1951char numa_zonelist_order[16] = "default";
1952
1953/*
1954 * interface for configure zonelist ordering.
1955 * command line option "numa_zonelist_order"
1956 * = "[dD]efault - default, automatic configuration.
1957 * = "[nN]ode - order by node locality, then by zone within node
1958 * = "[zZ]one - order by zone, then by locality within zone
1959 */
1960
1961static int __parse_numa_zonelist_order(char *s)
1962{
1963 if (*s == 'd' || *s == 'D') {
1964 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1965 } else if (*s == 'n' || *s == 'N') {
1966 user_zonelist_order = ZONELIST_ORDER_NODE;
1967 } else if (*s == 'z' || *s == 'Z') {
1968 user_zonelist_order = ZONELIST_ORDER_ZONE;
1969 } else {
1970 printk(KERN_WARNING
1971 "Ignoring invalid numa_zonelist_order value: "
1972 "%s\n", s);
1973 return -EINVAL;
1974 }
1975 return 0;
1976}
1977
1978static __init int setup_numa_zonelist_order(char *s)
1979{
1980 if (s)
1981 return __parse_numa_zonelist_order(s);
1982 return 0;
1983}
1984early_param("numa_zonelist_order", setup_numa_zonelist_order);
1985
1986/*
1987 * sysctl handler for numa_zonelist_order
1988 */
1989int numa_zonelist_order_handler(ctl_table *table, int write,
1990 struct file *file, void __user *buffer, size_t *length,
1991 loff_t *ppos)
1992{
1993 char saved_string[NUMA_ZONELIST_ORDER_LEN];
1994 int ret;
1995
1996 if (write)
1997 strncpy(saved_string, (char*)table->data,
1998 NUMA_ZONELIST_ORDER_LEN);
1999 ret = proc_dostring(table, write, file, buffer, length, ppos);
2000 if (ret)
2001 return ret;
2002 if (write) {
2003 int oldval = user_zonelist_order;
2004 if (__parse_numa_zonelist_order((char*)table->data)) {
2005 /*
2006 * bogus value. restore saved string
2007 */
2008 strncpy((char*)table->data, saved_string,
2009 NUMA_ZONELIST_ORDER_LEN);
2010 user_zonelist_order = oldval;
2011 } else if (oldval != user_zonelist_order)
2012 build_all_zonelists();
2013 }
2014 return 0;
2015}
2016
2017
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018#define MAX_NODE_LOAD (num_online_nodes())
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002019static int node_load[MAX_NUMNODES];
2020
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07002022 * 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 -07002023 * @node: node whose fallback list we're appending
2024 * @used_node_mask: nodemask_t of already used nodes
2025 *
2026 * We use a number of factors to determine which is the next node that should
2027 * appear on a given node's fallback list. The node should not have appeared
2028 * already in @node's fallback list, and it should be the next closest node
2029 * according to the distance array (which contains arbitrary distance values
2030 * from each node to each node in the system), and should also prefer nodes
2031 * with no CPUs, since presumably they'll have very little allocation pressure
2032 * on them otherwise.
2033 * It returns -1 if no node is found.
2034 */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002035static int find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002037 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 int min_val = INT_MAX;
2039 int best_node = -1;
Mike Travisc5f59f02008-04-04 18:11:10 -07002040 node_to_cpumask_ptr(tmp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002042 /* Use the local node if we haven't already */
2043 if (!node_isset(node, *used_node_mask)) {
2044 node_set(node, *used_node_mask);
2045 return node;
2046 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002048 for_each_node_state(n, N_HIGH_MEMORY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
2050 /* Don't want a node to appear more than once */
2051 if (node_isset(n, *used_node_mask))
2052 continue;
2053
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 /* Use the distance array to find the distance */
2055 val = node_distance(node, n);
2056
Linus Torvalds4cf808e2006-02-17 20:38:21 +01002057 /* Penalize nodes under us ("prefer the next node") */
2058 val += (n < node);
2059
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 /* Give preference to headless and unused nodes */
Mike Travisc5f59f02008-04-04 18:11:10 -07002061 node_to_cpumask_ptr_next(tmp, n);
2062 if (!cpus_empty(*tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 val += PENALTY_FOR_NODE_WITH_CPUS;
2064
2065 /* Slight preference for less loaded node */
2066 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2067 val += node_load[n];
2068
2069 if (val < min_val) {
2070 min_val = val;
2071 best_node = n;
2072 }
2073 }
2074
2075 if (best_node >= 0)
2076 node_set(best_node, *used_node_mask);
2077
2078 return best_node;
2079}
2080
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002081
2082/*
2083 * Build zonelists ordered by node and zones within node.
2084 * This results in maximum locality--normal zone overflows into local
2085 * DMA zone, if any--but risks exhausting DMA zone.
2086 */
2087static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002089 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 struct zonelist *zonelist;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002091
Mel Gorman54a6eb52008-04-28 02:12:16 -07002092 zonelist = &pgdat->node_zonelists[0];
Mel Gormandd1a2392008-04-28 02:12:17 -07002093 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
Mel Gorman54a6eb52008-04-28 02:12:16 -07002094 ;
2095 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2096 MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002097 zonelist->_zonerefs[j].zone = NULL;
2098 zonelist->_zonerefs[j].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002099}
2100
2101/*
Christoph Lameter523b9452007-10-16 01:25:37 -07002102 * Build gfp_thisnode zonelists
2103 */
2104static void build_thisnode_zonelists(pg_data_t *pgdat)
2105{
Christoph Lameter523b9452007-10-16 01:25:37 -07002106 int j;
2107 struct zonelist *zonelist;
2108
Mel Gorman54a6eb52008-04-28 02:12:16 -07002109 zonelist = &pgdat->node_zonelists[1];
2110 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Mel Gormandd1a2392008-04-28 02:12:17 -07002111 zonelist->_zonerefs[j].zone = NULL;
2112 zonelist->_zonerefs[j].zone_idx = 0;
Christoph Lameter523b9452007-10-16 01:25:37 -07002113}
2114
2115/*
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002116 * Build zonelists ordered by zone and nodes within zones.
2117 * This results in conserving DMA zone[s] until all Normal memory is
2118 * exhausted, but results in overflowing to remote node while memory
2119 * may still exist in local DMA zone.
2120 */
2121static int node_order[MAX_NUMNODES];
2122
2123static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2124{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002125 int pos, j, node;
2126 int zone_type; /* needs to be signed */
2127 struct zone *z;
2128 struct zonelist *zonelist;
2129
Mel Gorman54a6eb52008-04-28 02:12:16 -07002130 zonelist = &pgdat->node_zonelists[0];
2131 pos = 0;
2132 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2133 for (j = 0; j < nr_nodes; j++) {
2134 node = node_order[j];
2135 z = &NODE_DATA(node)->node_zones[zone_type];
2136 if (populated_zone(z)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07002137 zoneref_set_zone(z,
2138 &zonelist->_zonerefs[pos++]);
Mel Gorman54a6eb52008-04-28 02:12:16 -07002139 check_highest_zone(zone_type);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002140 }
2141 }
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002142 }
Mel Gormandd1a2392008-04-28 02:12:17 -07002143 zonelist->_zonerefs[pos].zone = NULL;
2144 zonelist->_zonerefs[pos].zone_idx = 0;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002145}
2146
2147static int default_zonelist_order(void)
2148{
2149 int nid, zone_type;
2150 unsigned long low_kmem_size,total_size;
2151 struct zone *z;
2152 int average_size;
2153 /*
2154 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2155 * If they are really small and used heavily, the system can fall
2156 * into OOM very easily.
2157 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2158 */
2159 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2160 low_kmem_size = 0;
2161 total_size = 0;
2162 for_each_online_node(nid) {
2163 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2164 z = &NODE_DATA(nid)->node_zones[zone_type];
2165 if (populated_zone(z)) {
2166 if (zone_type < ZONE_NORMAL)
2167 low_kmem_size += z->present_pages;
2168 total_size += z->present_pages;
2169 }
2170 }
2171 }
2172 if (!low_kmem_size || /* there are no DMA area. */
2173 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2174 return ZONELIST_ORDER_NODE;
2175 /*
2176 * look into each node's config.
2177 * If there is a node whose DMA/DMA32 memory is very big area on
2178 * local memory, NODE_ORDER may be suitable.
2179 */
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002180 average_size = total_size /
2181 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002182 for_each_online_node(nid) {
2183 low_kmem_size = 0;
2184 total_size = 0;
2185 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2186 z = &NODE_DATA(nid)->node_zones[zone_type];
2187 if (populated_zone(z)) {
2188 if (zone_type < ZONE_NORMAL)
2189 low_kmem_size += z->present_pages;
2190 total_size += z->present_pages;
2191 }
2192 }
2193 if (low_kmem_size &&
2194 total_size > average_size && /* ignore small node */
2195 low_kmem_size > total_size * 70/100)
2196 return ZONELIST_ORDER_NODE;
2197 }
2198 return ZONELIST_ORDER_ZONE;
2199}
2200
2201static void set_zonelist_order(void)
2202{
2203 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2204 current_zonelist_order = default_zonelist_order();
2205 else
2206 current_zonelist_order = user_zonelist_order;
2207}
2208
2209static void build_zonelists(pg_data_t *pgdat)
2210{
2211 int j, node, load;
2212 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 nodemask_t used_mask;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002214 int local_node, prev_node;
2215 struct zonelist *zonelist;
2216 int order = current_zonelist_order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
2218 /* initialize zonelists */
Christoph Lameter523b9452007-10-16 01:25:37 -07002219 for (i = 0; i < MAX_ZONELISTS; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 zonelist = pgdat->node_zonelists + i;
Mel Gormandd1a2392008-04-28 02:12:17 -07002221 zonelist->_zonerefs[0].zone = NULL;
2222 zonelist->_zonerefs[0].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 }
2224
2225 /* NUMA-aware ordering of nodes */
2226 local_node = pgdat->node_id;
2227 load = num_online_nodes();
2228 prev_node = local_node;
2229 nodes_clear(used_mask);
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002230
2231 memset(node_load, 0, sizeof(node_load));
2232 memset(node_order, 0, sizeof(node_order));
2233 j = 0;
2234
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08002236 int distance = node_distance(local_node, node);
2237
2238 /*
2239 * If another node is sufficiently far away then it is better
2240 * to reclaim pages in a zone before going off node.
2241 */
2242 if (distance > RECLAIM_DISTANCE)
2243 zone_reclaim_mode = 1;
2244
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 /*
2246 * We don't want to pressure a particular node.
2247 * So adding penalty to the first node in same
2248 * distance group to make it round-robin.
2249 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08002250 if (distance != node_distance(local_node, prev_node))
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002251 node_load[node] = load;
2252
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 prev_node = node;
2254 load--;
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002255 if (order == ZONELIST_ORDER_NODE)
2256 build_zonelists_in_node_order(pgdat, node);
2257 else
2258 node_order[j++] = node; /* remember order */
2259 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002261 if (order == ZONELIST_ORDER_ZONE) {
2262 /* calculate node order -- i.e., DMA last! */
2263 build_zonelists_in_zone_order(pgdat, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 }
Christoph Lameter523b9452007-10-16 01:25:37 -07002265
2266 build_thisnode_zonelists(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267}
2268
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002269/* Construct the zonelist performance cache - see further mmzone.h */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002270static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002271{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002272 struct zonelist *zonelist;
2273 struct zonelist_cache *zlc;
Mel Gormandd1a2392008-04-28 02:12:17 -07002274 struct zoneref *z;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002275
Mel Gorman54a6eb52008-04-28 02:12:16 -07002276 zonelist = &pgdat->node_zonelists[0];
2277 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2278 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
Mel Gormandd1a2392008-04-28 02:12:17 -07002279 for (z = zonelist->_zonerefs; z->zone; z++)
2280 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002281}
2282
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002283
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284#else /* CONFIG_NUMA */
2285
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002286static void set_zonelist_order(void)
2287{
2288 current_zonelist_order = ZONELIST_ORDER_ZONE;
2289}
2290
2291static void build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292{
Christoph Lameter19655d32006-09-25 23:31:19 -07002293 int node, local_node;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002294 enum zone_type j;
2295 struct zonelist *zonelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296
2297 local_node = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
Mel Gorman54a6eb52008-04-28 02:12:16 -07002299 zonelist = &pgdat->node_zonelists[0];
2300 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
Mel Gorman54a6eb52008-04-28 02:12:16 -07002302 /*
2303 * Now we build the zonelist so that it contains the zones
2304 * of all the other nodes.
2305 * We don't want to pressure a particular node, so when
2306 * building the zones for node N, we make sure that the
2307 * zones coming right after the local ones are those from
2308 * node N+1 (modulo N)
2309 */
2310 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2311 if (!node_online(node))
2312 continue;
2313 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2314 MAX_NR_ZONES - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 }
Mel Gorman54a6eb52008-04-28 02:12:16 -07002316 for (node = 0; node < local_node; node++) {
2317 if (!node_online(node))
2318 continue;
2319 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2320 MAX_NR_ZONES - 1);
2321 }
2322
Mel Gormandd1a2392008-04-28 02:12:17 -07002323 zonelist->_zonerefs[j].zone = NULL;
2324 zonelist->_zonerefs[j].zone_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325}
2326
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002327/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002328static void build_zonelist_cache(pg_data_t *pgdat)
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002329{
Mel Gorman54a6eb52008-04-28 02:12:16 -07002330 pgdat->node_zonelists[0].zlcache_ptr = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002331}
2332
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333#endif /* CONFIG_NUMA */
2334
Yasunori Goto68113782006-06-23 02:03:11 -07002335/* return values int ....just for stop_machine_run() */
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002336static int __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337{
Yasunori Goto68113782006-06-23 02:03:11 -07002338 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002339
2340 for_each_online_node(nid) {
Christoph Lameter7ea15302007-10-16 01:25:29 -07002341 pg_data_t *pgdat = NODE_DATA(nid);
2342
2343 build_zonelists(pgdat);
2344 build_zonelist_cache(pgdat);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08002345 }
Yasunori Goto68113782006-06-23 02:03:11 -07002346 return 0;
2347}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002349void build_all_zonelists(void)
Yasunori Goto68113782006-06-23 02:03:11 -07002350{
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002351 set_zonelist_order();
2352
Yasunori Goto68113782006-06-23 02:03:11 -07002353 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07002354 __build_all_zonelists(NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07002355 cpuset_init_current_mems_allowed();
2356 } else {
Simon Arlott183ff222007-10-20 01:27:18 +02002357 /* we have to stop all cpus to guarantee there is no user
Yasunori Goto68113782006-06-23 02:03:11 -07002358 of zonelist */
2359 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
2360 /* cpuset refresh routine should be here */
2361 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07002362 vm_total_pages = nr_free_pagecache_pages();
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002363 /*
2364 * Disable grouping by mobility if the number of pages in the
2365 * system is too low to allow the mechanism to work. It would be
2366 * more accurate, but expensive to check per-zone. This check is
2367 * made on memory-hotadd so a system can start with mobility
2368 * disabled and enable it later
2369 */
Mel Gormand9c23402007-10-16 01:26:01 -07002370 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002371 page_group_by_mobility_disabled = 1;
2372 else
2373 page_group_by_mobility_disabled = 0;
2374
2375 printk("Built %i zonelists in %s order, mobility grouping %s. "
2376 "Total pages: %ld\n",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002377 num_online_nodes(),
2378 zonelist_order_name[current_zonelist_order],
Mel Gorman9ef9acb2007-10-16 01:25:54 -07002379 page_group_by_mobility_disabled ? "off" : "on",
KAMEZAWA Hiroyukif0c0b2b2007-07-15 23:38:01 -07002380 vm_total_pages);
2381#ifdef CONFIG_NUMA
2382 printk("Policy zone: %s\n", zone_names[policy_zone]);
2383#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384}
2385
2386/*
2387 * Helper functions to size the waitqueue hash table.
2388 * Essentially these want to choose hash table sizes sufficiently
2389 * large so that collisions trying to wait on pages are rare.
2390 * But in fact, the number of active page waitqueues on typical
2391 * systems is ridiculously low, less than 200. So this is even
2392 * conservative, even though it seems large.
2393 *
2394 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2395 * waitqueues, i.e. the size of the waitq table given the number of pages.
2396 */
2397#define PAGES_PER_WAITQUEUE 256
2398
Yasunori Gotocca448f2006-06-23 02:03:10 -07002399#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07002400static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401{
2402 unsigned long size = 1;
2403
2404 pages /= PAGES_PER_WAITQUEUE;
2405
2406 while (size < pages)
2407 size <<= 1;
2408
2409 /*
2410 * Once we have dozens or even hundreds of threads sleeping
2411 * on IO we've got bigger problems than wait queue collision.
2412 * Limit the size of the wait table to a reasonable size.
2413 */
2414 size = min(size, 4096UL);
2415
2416 return max(size, 4UL);
2417}
Yasunori Gotocca448f2006-06-23 02:03:10 -07002418#else
2419/*
2420 * A zone's size might be changed by hot-add, so it is not possible to determine
2421 * a suitable size for its wait_table. So we use the maximum size now.
2422 *
2423 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2424 *
2425 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2426 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2427 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2428 *
2429 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2430 * or more by the traditional way. (See above). It equals:
2431 *
2432 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2433 * ia64(16K page size) : = ( 8G + 4M)byte.
2434 * powerpc (64K page size) : = (32G +16M)byte.
2435 */
2436static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2437{
2438 return 4096UL;
2439}
2440#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
2442/*
2443 * This is an integer logarithm so that shifts can be used later
2444 * to extract the more random high bits from the multiplicative
2445 * hash function before the remainder is taken.
2446 */
2447static inline unsigned long wait_table_bits(unsigned long size)
2448{
2449 return ffz(~size);
2450}
2451
2452#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2453
Mel Gorman56fd56b2007-10-16 01:25:58 -07002454/*
Mel Gormand9c23402007-10-16 01:26:01 -07002455 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
Mel Gorman56fd56b2007-10-16 01:25:58 -07002456 * of blocks reserved is based on zone->pages_min. The memory within the
2457 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2458 * higher will lead to a bigger reserve which will get freed as contiguous
2459 * blocks as reclaim kicks in
2460 */
2461static void setup_zone_migrate_reserve(struct zone *zone)
2462{
2463 unsigned long start_pfn, pfn, end_pfn;
2464 struct page *page;
2465 unsigned long reserve, block_migratetype;
2466
2467 /* Get the start pfn, end pfn and the number of blocks to reserve */
2468 start_pfn = zone->zone_start_pfn;
2469 end_pfn = start_pfn + zone->spanned_pages;
Mel Gormand9c23402007-10-16 01:26:01 -07002470 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2471 pageblock_order;
Mel Gorman56fd56b2007-10-16 01:25:58 -07002472
Mel Gormand9c23402007-10-16 01:26:01 -07002473 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
Mel Gorman56fd56b2007-10-16 01:25:58 -07002474 if (!pfn_valid(pfn))
2475 continue;
2476 page = pfn_to_page(pfn);
2477
2478 /* Blocks with reserved pages will never free, skip them. */
2479 if (PageReserved(page))
2480 continue;
2481
2482 block_migratetype = get_pageblock_migratetype(page);
2483
2484 /* If this block is reserved, account for it */
2485 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2486 reserve--;
2487 continue;
2488 }
2489
2490 /* Suitable for reserving if this block is movable */
2491 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2492 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2493 move_freepages_block(zone, page, MIGRATE_RESERVE);
2494 reserve--;
2495 continue;
2496 }
2497
2498 /*
2499 * If the reserve is met and this is a previous reserved block,
2500 * take it back
2501 */
2502 if (block_migratetype == MIGRATE_RESERVE) {
2503 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2504 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2505 }
2506 }
2507}
Mel Gormanac0e5b72007-10-16 01:25:58 -07002508
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509/*
2510 * Initially all pages are reserved - free ones are freed
2511 * up by free_all_bootmem() once the early boot process is
2512 * done. Non-atomic initialization, single-pass.
2513 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002514void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08002515 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07002518 unsigned long end_pfn = start_pfn + size;
2519 unsigned long pfn;
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002520 struct zone *z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002522 z = &NODE_DATA(nid)->node_zones[zone];
Greg Ungerercbe8dd42006-01-12 01:05:24 -08002523 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08002524 /*
2525 * There can be holes in boot-time mem_map[]s
2526 * handed to this function. They do not
2527 * exist on hotplugged memory.
2528 */
2529 if (context == MEMMAP_EARLY) {
2530 if (!early_pfn_valid(pfn))
2531 continue;
2532 if (!early_pfn_in_nid(pfn, nid))
2533 continue;
2534 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002535 page = pfn_to_page(pfn);
2536 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08002537 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 reset_page_mapcount(page);
2539 SetPageReserved(page);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002540 /*
2541 * Mark the block movable so that blocks are reserved for
2542 * movable at startup. This will force kernel allocations
2543 * to reserve their blocks rather than leaking throughout
2544 * the address space during boot when many long-lived
Mel Gorman56fd56b2007-10-16 01:25:58 -07002545 * kernel allocations are made. Later some blocks near
2546 * the start are marked MIGRATE_RESERVE by
2547 * setup_zone_migrate_reserve()
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002548 *
2549 * bitmap is created for zone's valid pfn range. but memmap
2550 * can be created for invalid pages (for alignment)
2551 * check here not to call set_pageblock_migratetype() against
2552 * pfn out of zone.
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002553 */
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07002554 if ((z->zone_start_pfn <= pfn)
2555 && (pfn < z->zone_start_pfn + z->spanned_pages)
2556 && !(pfn & (pageblock_nr_pages - 1)))
Mel Gorman56fd56b2007-10-16 01:25:58 -07002557 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002558
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 INIT_LIST_HEAD(&page->lru);
2560#ifdef WANT_PAGE_VIRTUAL
2561 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2562 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07002563 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 }
2566}
2567
Andi Kleen1e548de2008-02-04 22:29:26 -08002568static void __meminit zone_init_free_lists(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569{
Mel Gormanb2a0ac82007-10-16 01:25:48 -07002570 int order, t;
2571 for_each_migratetype_order(order, t) {
2572 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 zone->free_area[order].nr_free = 0;
2574 }
2575}
2576
2577#ifndef __HAVE_ARCH_MEMMAP_INIT
2578#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08002579 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580#endif
2581
Sam Ravnborg1d6f4e62008-01-17 15:21:12 -08002582static int zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002583{
2584 int batch;
2585
2586 /*
2587 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07002588 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002589 *
2590 * OK, so we don't know how big the cache is. So guess.
2591 */
2592 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07002593 if (batch * PAGE_SIZE > 512 * 1024)
2594 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002595 batch /= 4; /* We effectively *= 4 below */
2596 if (batch < 1)
2597 batch = 1;
2598
2599 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002600 * Clamp the batch to a 2^n - 1 value. Having a power
2601 * of 2 value was found to be more likely to have
2602 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002603 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002604 * For example if 2 tasks are alternately allocating
2605 * batches of pages, one task can end up with a lot
2606 * of pages of one half of the possible page colors
2607 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002608 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002609 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07002610
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002611 return batch;
2612}
2613
Christoph Lameter2caaad42005-06-21 17:15:00 -07002614inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2615{
2616 struct per_cpu_pages *pcp;
2617
Magnus Damm1c6fe942005-10-26 01:58:59 -07002618 memset(p, 0, sizeof(*p));
2619
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002620 pcp = &p->pcp;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002621 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002622 pcp->high = 6 * batch;
2623 pcp->batch = max(1UL, 1 * batch);
2624 INIT_LIST_HEAD(&pcp->list);
Christoph Lameter2caaad42005-06-21 17:15:00 -07002625}
2626
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002627/*
2628 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2629 * to the value high for the pageset p.
2630 */
2631
2632static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2633 unsigned long high)
2634{
2635 struct per_cpu_pages *pcp;
2636
Christoph Lameter3dfa5722008-02-04 22:29:19 -08002637 pcp = &p->pcp;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002638 pcp->high = high;
2639 pcp->batch = max(1UL, high/4);
2640 if ((high/4) > (PAGE_SHIFT * 8))
2641 pcp->batch = PAGE_SHIFT * 8;
2642}
2643
2644
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002645#ifdef CONFIG_NUMA
2646/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07002647 * Boot pageset table. One per cpu which is going to be used for all
2648 * zones and all nodes. The parameters will be set in such a way
2649 * that an item put on a list will immediately be handed over to
2650 * the buddy list. This is safe since pageset manipulation is done
2651 * with interrupts disabled.
2652 *
2653 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07002654 *
2655 * The boot_pagesets must be kept even after bootup is complete for
2656 * unused processors and/or zones. They do play a role for bootstrapping
2657 * hotplugged processors.
2658 *
2659 * zoneinfo_show() and maybe other functions do
2660 * not check if the processor is online before following the pageset pointer.
2661 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07002662 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08002663static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07002664
2665/*
2666 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002667 * per cpu pageset array in struct zone.
2668 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08002669static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002670{
2671 struct zone *zone, *dzone;
Christoph Lameter37c07082007-10-16 01:25:36 -07002672 int node = cpu_to_node(cpu);
2673
2674 node_set_state(node, N_CPU); /* this node has a cpu */
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002675
2676 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002677
Christoph Lameter66a55032006-09-27 01:50:09 -07002678 if (!populated_zone(zone))
2679 continue;
2680
Nick Piggin23316bc2006-01-08 01:00:41 -08002681 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lameter37c07082007-10-16 01:25:36 -07002682 GFP_KERNEL, node);
Nick Piggin23316bc2006-01-08 01:00:41 -08002683 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002684 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002685
Nick Piggin23316bc2006-01-08 01:00:41 -08002686 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002687
2688 if (percpu_pagelist_fraction)
2689 setup_pagelist_highmark(zone_pcp(zone, cpu),
2690 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002691 }
2692
2693 return 0;
2694bad:
2695 for_each_zone(dzone) {
Andrew Morton64191682007-08-30 23:56:17 -07002696 if (!populated_zone(dzone))
2697 continue;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002698 if (dzone == zone)
2699 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08002700 kfree(zone_pcp(dzone, cpu));
2701 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002702 }
2703 return -ENOMEM;
2704}
2705
2706static inline void free_zone_pagesets(int cpu)
2707{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002708 struct zone *zone;
2709
2710 for_each_zone(zone) {
2711 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2712
David Rientjesf3ef9ea2006-09-25 16:24:57 -07002713 /* Free per_cpu_pageset if it is slab allocated */
2714 if (pset != &boot_pageset[cpu])
2715 kfree(pset);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002716 zone_pcp(zone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002717 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002718}
2719
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07002720static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002721 unsigned long action,
2722 void *hcpu)
2723{
2724 int cpu = (long)hcpu;
2725 int ret = NOTIFY_OK;
2726
2727 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08002728 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002729 case CPU_UP_PREPARE_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002730 if (process_zones(cpu))
2731 ret = NOTIFY_BAD;
2732 break;
2733 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002734 case CPU_UP_CANCELED_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002735 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002736 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002737 free_zone_pagesets(cpu);
2738 break;
2739 default:
2740 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002741 }
2742 return ret;
2743}
2744
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07002745static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002746 { &pageset_cpuup_callback, NULL, 0 };
2747
Al Viro78d99552005-12-15 09:18:25 +00002748void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002749{
2750 int err;
2751
2752 /* Initialize per_cpu_pageset for cpu 0.
2753 * A cpuup callback will do this for every cpu
2754 * as it comes online
2755 */
2756 err = process_zones(smp_processor_id());
2757 BUG_ON(err);
2758 register_cpu_notifier(&pageset_notifier);
2759}
2760
2761#endif
2762
Sam Ravnborg577a32f2007-05-17 23:29:25 +02002763static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07002764int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07002765{
2766 int i;
2767 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002768 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07002769
2770 /*
2771 * The per-page waitqueue mechanism uses hashed waitqueues
2772 * per zone.
2773 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07002774 zone->wait_table_hash_nr_entries =
2775 wait_table_hash_nr_entries(zone_size_pages);
2776 zone->wait_table_bits =
2777 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002778 alloc_size = zone->wait_table_hash_nr_entries
2779 * sizeof(wait_queue_head_t);
2780
Heiko Carstenscd94b9d2008-05-23 13:04:52 -07002781 if (!slab_is_available()) {
Yasunori Gotocca448f2006-06-23 02:03:10 -07002782 zone->wait_table = (wait_queue_head_t *)
2783 alloc_bootmem_node(pgdat, alloc_size);
2784 } else {
2785 /*
2786 * This case means that a zone whose size was 0 gets new memory
2787 * via memory hot-add.
2788 * But it may be the case that a new node was hot-added. In
2789 * this case vmalloc() will not be able to use this new node's
2790 * memory - this wait_table must be initialized to use this new
2791 * node itself as well.
2792 * To use this new node's memory, further consideration will be
2793 * necessary.
2794 */
Jesper Juhl8691f3a2007-10-16 01:24:49 -07002795 zone->wait_table = vmalloc(alloc_size);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002796 }
2797 if (!zone->wait_table)
2798 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07002799
Yasunori Goto02b694d2006-06-23 02:03:08 -07002800 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07002801 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002802
2803 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002804}
2805
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002806static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07002807{
2808 int cpu;
2809 unsigned long batch = zone_batchsize(zone);
2810
2811 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2812#ifdef CONFIG_NUMA
2813 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002814 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002815 setup_pageset(&boot_pageset[cpu],0);
2816#else
2817 setup_pageset(zone_pcp(zone,cpu), batch);
2818#endif
2819 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08002820 if (zone->present_pages)
2821 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2822 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07002823}
2824
Yasunori Goto718127c2006-06-23 02:03:10 -07002825__meminit int init_currently_empty_zone(struct zone *zone,
2826 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08002827 unsigned long size,
2828 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07002829{
2830 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002831 int ret;
2832 ret = zone_wait_table_init(zone, size);
2833 if (ret)
2834 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07002835 pgdat->nr_zones = zone_idx(zone) + 1;
2836
Dave Hansened8ece22005-10-29 18:16:50 -07002837 zone->zone_start_pfn = zone_start_pfn;
2838
Andi Kleen1e548de2008-02-04 22:29:26 -08002839 zone_init_free_lists(zone);
Yasunori Goto718127c2006-06-23 02:03:10 -07002840
2841 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002842}
2843
Mel Gormanc7132162006-09-27 01:49:43 -07002844#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2845/*
2846 * Basic iterator support. Return the first range of PFNs for a node
2847 * Note: nid == MAX_NUMNODES returns first region regardless of node
2848 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002849static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07002850{
2851 int i;
2852
2853 for (i = 0; i < nr_nodemap_entries; i++)
2854 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2855 return i;
2856
2857 return -1;
2858}
2859
2860/*
2861 * Basic iterator support. Return the next active range of PFNs for a node
Simon Arlott183ff222007-10-20 01:27:18 +02002862 * Note: nid == MAX_NUMNODES returns next region regardless of node
Mel Gormanc7132162006-09-27 01:49:43 -07002863 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002864static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07002865{
2866 for (index = index + 1; index < nr_nodemap_entries; index++)
2867 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2868 return index;
2869
2870 return -1;
2871}
2872
2873#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2874/*
2875 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2876 * Architectures may implement their own version but if add_active_range()
2877 * was used and there are no special requirements, this is a convenient
2878 * alternative
2879 */
Stephen Rothwell6f076f52007-05-10 03:15:27 -07002880int __meminit early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07002881{
2882 int i;
2883
2884 for (i = 0; i < nr_nodemap_entries; i++) {
2885 unsigned long start_pfn = early_node_map[i].start_pfn;
2886 unsigned long end_pfn = early_node_map[i].end_pfn;
2887
2888 if (start_pfn <= pfn && pfn < end_pfn)
2889 return early_node_map[i].nid;
2890 }
2891
2892 return 0;
2893}
2894#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2895
2896/* Basic iterator support to walk early_node_map[] */
2897#define for_each_active_range_index_in_nid(i, nid) \
2898 for (i = first_active_region_index_in_nid(nid); i != -1; \
2899 i = next_active_region_index_in_nid(i, nid))
2900
2901/**
2902 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002903 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2904 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07002905 *
2906 * If an architecture guarantees that all ranges registered with
2907 * add_active_ranges() contain no holes and may be freed, this
2908 * this function may be used instead of calling free_bootmem() manually.
2909 */
2910void __init free_bootmem_with_active_regions(int nid,
2911 unsigned long max_low_pfn)
2912{
2913 int i;
2914
2915 for_each_active_range_index_in_nid(i, nid) {
2916 unsigned long size_pages = 0;
2917 unsigned long end_pfn = early_node_map[i].end_pfn;
2918
2919 if (early_node_map[i].start_pfn >= max_low_pfn)
2920 continue;
2921
2922 if (end_pfn > max_low_pfn)
2923 end_pfn = max_low_pfn;
2924
2925 size_pages = end_pfn - early_node_map[i].start_pfn;
2926 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2927 PFN_PHYS(early_node_map[i].start_pfn),
2928 size_pages << PAGE_SHIFT);
2929 }
2930}
2931
Yinghai Lub5bc6c02008-06-14 18:32:52 -07002932void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
2933{
2934 int i;
Yinghai Lud52d53b2008-06-16 20:10:55 -07002935 int ret;
Yinghai Lub5bc6c02008-06-14 18:32:52 -07002936
Yinghai Lud52d53b2008-06-16 20:10:55 -07002937 for_each_active_range_index_in_nid(i, nid) {
2938 ret = work_fn(early_node_map[i].start_pfn,
2939 early_node_map[i].end_pfn, data);
2940 if (ret)
2941 break;
2942 }
Yinghai Lub5bc6c02008-06-14 18:32:52 -07002943}
Mel Gormanc7132162006-09-27 01:49:43 -07002944/**
2945 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002946 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07002947 *
2948 * If an architecture guarantees that all ranges registered with
2949 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002950 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07002951 */
2952void __init sparse_memory_present_with_active_regions(int nid)
2953{
2954 int i;
2955
2956 for_each_active_range_index_in_nid(i, nid)
2957 memory_present(early_node_map[i].nid,
2958 early_node_map[i].start_pfn,
2959 early_node_map[i].end_pfn);
2960}
2961
2962/**
Mel Gormanfb014392006-09-27 01:49:59 -07002963 * push_node_boundaries - Push node boundaries to at least the requested boundary
2964 * @nid: The nid of the node to push the boundary for
2965 * @start_pfn: The start pfn of the node
2966 * @end_pfn: The end pfn of the node
2967 *
2968 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2969 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2970 * be hotplugged even though no physical memory exists. This function allows
2971 * an arch to push out the node boundaries so mem_map is allocated that can
2972 * be used later.
2973 */
2974#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2975void __init push_node_boundaries(unsigned int nid,
2976 unsigned long start_pfn, unsigned long end_pfn)
2977{
2978 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2979 nid, start_pfn, end_pfn);
2980
2981 /* Initialise the boundary for this node if necessary */
2982 if (node_boundary_end_pfn[nid] == 0)
2983 node_boundary_start_pfn[nid] = -1UL;
2984
2985 /* Update the boundaries */
2986 if (node_boundary_start_pfn[nid] > start_pfn)
2987 node_boundary_start_pfn[nid] = start_pfn;
2988 if (node_boundary_end_pfn[nid] < end_pfn)
2989 node_boundary_end_pfn[nid] = end_pfn;
2990}
2991
2992/* If necessary, push the node boundary out for reserve hotadd */
Jan Beulich98011f52007-07-15 23:38:17 -07002993static void __meminit account_node_boundary(unsigned int nid,
Mel Gormanfb014392006-09-27 01:49:59 -07002994 unsigned long *start_pfn, unsigned long *end_pfn)
2995{
2996 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2997 nid, *start_pfn, *end_pfn);
2998
2999 /* Return if boundary information has not been provided */
3000 if (node_boundary_end_pfn[nid] == 0)
3001 return;
3002
3003 /* Check the boundaries and update if necessary */
3004 if (node_boundary_start_pfn[nid] < *start_pfn)
3005 *start_pfn = node_boundary_start_pfn[nid];
3006 if (node_boundary_end_pfn[nid] > *end_pfn)
3007 *end_pfn = node_boundary_end_pfn[nid];
3008}
3009#else
3010void __init push_node_boundaries(unsigned int nid,
3011 unsigned long start_pfn, unsigned long end_pfn) {}
3012
Jan Beulich98011f52007-07-15 23:38:17 -07003013static void __meminit account_node_boundary(unsigned int nid,
Mel Gormanfb014392006-09-27 01:49:59 -07003014 unsigned long *start_pfn, unsigned long *end_pfn) {}
3015#endif
3016
3017
3018/**
Mel Gormanc7132162006-09-27 01:49:43 -07003019 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003020 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3021 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3022 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07003023 *
3024 * It returns the start and end page frame of a node based on information
3025 * provided by an arch calling add_active_range(). If called for a node
3026 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003027 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07003028 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003029void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003030 unsigned long *start_pfn, unsigned long *end_pfn)
3031{
3032 int i;
3033 *start_pfn = -1UL;
3034 *end_pfn = 0;
3035
3036 for_each_active_range_index_in_nid(i, nid) {
3037 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3038 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3039 }
3040
Christoph Lameter633c0662007-10-16 01:25:37 -07003041 if (*start_pfn == -1UL)
Mel Gormanc7132162006-09-27 01:49:43 -07003042 *start_pfn = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07003043
3044 /* Push the node boundaries out if requested */
3045 account_node_boundary(nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003046}
3047
3048/*
Mel Gorman2a1e2742007-07-17 04:03:12 -07003049 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3050 * assumption is made that zones within a node are ordered in monotonic
3051 * increasing memory addresses so that the "highest" populated zone is used
3052 */
3053void __init find_usable_zone_for_movable(void)
3054{
3055 int zone_index;
3056 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3057 if (zone_index == ZONE_MOVABLE)
3058 continue;
3059
3060 if (arch_zone_highest_possible_pfn[zone_index] >
3061 arch_zone_lowest_possible_pfn[zone_index])
3062 break;
3063 }
3064
3065 VM_BUG_ON(zone_index == -1);
3066 movable_zone = zone_index;
3067}
3068
3069/*
3070 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3071 * because it is sized independant of architecture. Unlike the other zones,
3072 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3073 * in each node depending on the size of each node and how evenly kernelcore
3074 * is distributed. This helper function adjusts the zone ranges
3075 * provided by the architecture for a given node by using the end of the
3076 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3077 * zones within a node are in order of monotonic increases memory addresses
3078 */
3079void __meminit adjust_zone_range_for_zone_movable(int nid,
3080 unsigned long zone_type,
3081 unsigned long node_start_pfn,
3082 unsigned long node_end_pfn,
3083 unsigned long *zone_start_pfn,
3084 unsigned long *zone_end_pfn)
3085{
3086 /* Only adjust if ZONE_MOVABLE is on this node */
3087 if (zone_movable_pfn[nid]) {
3088 /* Size ZONE_MOVABLE */
3089 if (zone_type == ZONE_MOVABLE) {
3090 *zone_start_pfn = zone_movable_pfn[nid];
3091 *zone_end_pfn = min(node_end_pfn,
3092 arch_zone_highest_possible_pfn[movable_zone]);
3093
3094 /* Adjust for ZONE_MOVABLE starting within this range */
3095 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3096 *zone_end_pfn > zone_movable_pfn[nid]) {
3097 *zone_end_pfn = zone_movable_pfn[nid];
3098
3099 /* Check if this whole range is within ZONE_MOVABLE */
3100 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3101 *zone_start_pfn = *zone_end_pfn;
3102 }
3103}
3104
3105/*
Mel Gormanc7132162006-09-27 01:49:43 -07003106 * Return the number of pages a zone spans in a node, including holes
3107 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3108 */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003109static unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003110 unsigned long zone_type,
3111 unsigned long *ignored)
3112{
3113 unsigned long node_start_pfn, node_end_pfn;
3114 unsigned long zone_start_pfn, zone_end_pfn;
3115
3116 /* Get the start and end of the node and zone */
3117 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3118 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3119 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
Mel Gorman2a1e2742007-07-17 04:03:12 -07003120 adjust_zone_range_for_zone_movable(nid, zone_type,
3121 node_start_pfn, node_end_pfn,
3122 &zone_start_pfn, &zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003123
3124 /* Check that this node has pages within the zone's required range */
3125 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3126 return 0;
3127
3128 /* Move the zone boundaries inside the node if necessary */
3129 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3130 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3131
3132 /* Return the spanned pages */
3133 return zone_end_pfn - zone_start_pfn;
3134}
3135
3136/*
3137 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003138 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07003139 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003140unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003141 unsigned long range_start_pfn,
3142 unsigned long range_end_pfn)
3143{
3144 int i = 0;
3145 unsigned long prev_end_pfn = 0, hole_pages = 0;
3146 unsigned long start_pfn;
3147
3148 /* Find the end_pfn of the first active range of pfns in the node */
3149 i = first_active_region_index_in_nid(nid);
3150 if (i == -1)
3151 return 0;
3152
Mel Gormanb5445f92007-07-26 10:41:18 -07003153 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3154
Mel Gorman9c7cd682006-09-27 01:49:58 -07003155 /* Account for ranges before physical memory on this node */
3156 if (early_node_map[i].start_pfn > range_start_pfn)
Mel Gormanb5445f92007-07-26 10:41:18 -07003157 hole_pages = prev_end_pfn - range_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003158
3159 /* Find all holes for the zone within the node */
3160 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3161
3162 /* No need to continue if prev_end_pfn is outside the zone */
3163 if (prev_end_pfn >= range_end_pfn)
3164 break;
3165
3166 /* Make sure the end of the zone is not within the hole */
3167 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3168 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3169
3170 /* Update the hole size cound and move on */
3171 if (start_pfn > range_start_pfn) {
3172 BUG_ON(prev_end_pfn > start_pfn);
3173 hole_pages += start_pfn - prev_end_pfn;
3174 }
3175 prev_end_pfn = early_node_map[i].end_pfn;
3176 }
3177
Mel Gorman9c7cd682006-09-27 01:49:58 -07003178 /* Account for ranges past physical memory on this node */
3179 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07003180 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07003181 max(range_start_pfn, prev_end_pfn);
3182
Mel Gormanc7132162006-09-27 01:49:43 -07003183 return hole_pages;
3184}
3185
3186/**
3187 * absent_pages_in_range - Return number of page frames in holes within a range
3188 * @start_pfn: The start PFN to start searching for holes
3189 * @end_pfn: The end PFN to stop searching for holes
3190 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003191 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07003192 */
3193unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3194 unsigned long end_pfn)
3195{
3196 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3197}
3198
3199/* Return the number of page frames in holes in a zone on a node */
Paul Mundt6ea6e682007-07-15 23:38:20 -07003200static unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003201 unsigned long zone_type,
3202 unsigned long *ignored)
3203{
Mel Gorman9c7cd682006-09-27 01:49:58 -07003204 unsigned long node_start_pfn, node_end_pfn;
3205 unsigned long zone_start_pfn, zone_end_pfn;
3206
3207 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3208 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3209 node_start_pfn);
3210 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3211 node_end_pfn);
3212
Mel Gorman2a1e2742007-07-17 04:03:12 -07003213 adjust_zone_range_for_zone_movable(nid, zone_type,
3214 node_start_pfn, node_end_pfn,
3215 &zone_start_pfn, &zone_end_pfn);
Mel Gorman9c7cd682006-09-27 01:49:58 -07003216 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003217}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003218
Mel Gormanc7132162006-09-27 01:49:43 -07003219#else
Paul Mundt6ea6e682007-07-15 23:38:20 -07003220static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003221 unsigned long zone_type,
3222 unsigned long *zones_size)
3223{
3224 return zones_size[zone_type];
3225}
3226
Paul Mundt6ea6e682007-07-15 23:38:20 -07003227static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003228 unsigned long zone_type,
3229 unsigned long *zholes_size)
3230{
3231 if (!zholes_size)
3232 return 0;
3233
3234 return zholes_size[zone_type];
3235}
Mel Gorman0e0b8642006-09-27 01:49:56 -07003236
Mel Gormanc7132162006-09-27 01:49:43 -07003237#endif
3238
Yasunori Gotoa3142c82007-05-08 00:23:07 -07003239static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07003240 unsigned long *zones_size, unsigned long *zholes_size)
3241{
3242 unsigned long realtotalpages, totalpages = 0;
3243 enum zone_type i;
3244
3245 for (i = 0; i < MAX_NR_ZONES; i++)
3246 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3247 zones_size);
3248 pgdat->node_spanned_pages = totalpages;
3249
3250 realtotalpages = totalpages;
3251 for (i = 0; i < MAX_NR_ZONES; i++)
3252 realtotalpages -=
3253 zone_absent_pages_in_node(pgdat->node_id, i,
3254 zholes_size);
3255 pgdat->node_present_pages = realtotalpages;
3256 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3257 realtotalpages);
3258}
3259
Mel Gorman835c1342007-10-16 01:25:47 -07003260#ifndef CONFIG_SPARSEMEM
3261/*
3262 * Calculate the size of the zone->blockflags rounded to an unsigned long
Mel Gormand9c23402007-10-16 01:26:01 -07003263 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3264 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
Mel Gorman835c1342007-10-16 01:25:47 -07003265 * round what is now in bits to nearest long in bits, then return it in
3266 * bytes.
3267 */
3268static unsigned long __init usemap_size(unsigned long zonesize)
3269{
3270 unsigned long usemapsize;
3271
Mel Gormand9c23402007-10-16 01:26:01 -07003272 usemapsize = roundup(zonesize, pageblock_nr_pages);
3273 usemapsize = usemapsize >> pageblock_order;
Mel Gorman835c1342007-10-16 01:25:47 -07003274 usemapsize *= NR_PAGEBLOCK_BITS;
3275 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3276
3277 return usemapsize / 8;
3278}
3279
3280static void __init setup_usemap(struct pglist_data *pgdat,
3281 struct zone *zone, unsigned long zonesize)
3282{
3283 unsigned long usemapsize = usemap_size(zonesize);
3284 zone->pageblock_flags = NULL;
3285 if (usemapsize) {
3286 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3287 memset(zone->pageblock_flags, 0, usemapsize);
3288 }
3289}
3290#else
3291static void inline setup_usemap(struct pglist_data *pgdat,
3292 struct zone *zone, unsigned long zonesize) {}
3293#endif /* CONFIG_SPARSEMEM */
3294
Mel Gormand9c23402007-10-16 01:26:01 -07003295#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
Mel Gormanba72cb82007-11-28 16:21:13 -08003296
3297/* Return a sensible default order for the pageblock size. */
3298static inline int pageblock_default_order(void)
3299{
3300 if (HPAGE_SHIFT > PAGE_SHIFT)
3301 return HUGETLB_PAGE_ORDER;
3302
3303 return MAX_ORDER-1;
3304}
3305
Mel Gormand9c23402007-10-16 01:26:01 -07003306/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3307static inline void __init set_pageblock_order(unsigned int order)
3308{
3309 /* Check that pageblock_nr_pages has not already been setup */
3310 if (pageblock_order)
3311 return;
3312
3313 /*
3314 * Assume the largest contiguous order of interest is a huge page.
3315 * This value may be variable depending on boot parameters on IA64
3316 */
3317 pageblock_order = order;
3318}
3319#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3320
Mel Gormanba72cb82007-11-28 16:21:13 -08003321/*
3322 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3323 * and pageblock_default_order() are unused as pageblock_order is set
3324 * at compile-time. See include/linux/pageblock-flags.h for the values of
3325 * pageblock_order based on the kernel config
3326 */
3327static inline int pageblock_default_order(unsigned int order)
3328{
3329 return MAX_ORDER-1;
3330}
Mel Gormand9c23402007-10-16 01:26:01 -07003331#define set_pageblock_order(x) do {} while (0)
3332
3333#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3334
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335/*
3336 * Set up the zone data structures:
3337 * - mark all pages reserved
3338 * - mark all memory queues empty
3339 * - clear the memory bitmaps
3340 */
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08003341static void __paginginit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342 unsigned long *zones_size, unsigned long *zholes_size)
3343{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07003344 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07003345 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07003347 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348
Dave Hansen208d54e2005-10-29 18:16:52 -07003349 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 pgdat->nr_zones = 0;
3351 init_waitqueue_head(&pgdat->kswapd_wait);
3352 pgdat->kswapd_max_order = 0;
3353
3354 for (j = 0; j < MAX_NR_ZONES; j++) {
3355 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003356 unsigned long size, realsize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003357
Mel Gormanc7132162006-09-27 01:49:43 -07003358 size = zone_spanned_pages_in_node(nid, j, zones_size);
3359 realsize = size - zone_absent_pages_in_node(nid, j,
3360 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003361
Mel Gorman0e0b8642006-09-27 01:49:56 -07003362 /*
3363 * Adjust realsize so that it accounts for how much memory
3364 * is used by this zone for memmap. This affects the watermark
3365 * and per-cpu initialisations
3366 */
Johannes Weinerf7232152008-05-23 13:04:21 -07003367 memmap_pages =
3368 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
Mel Gorman0e0b8642006-09-27 01:49:56 -07003369 if (realsize >= memmap_pages) {
3370 realsize -= memmap_pages;
3371 printk(KERN_DEBUG
3372 " %s zone: %lu pages used for memmap\n",
3373 zone_names[j], memmap_pages);
3374 } else
3375 printk(KERN_WARNING
3376 " %s zone: %lu pages exceeds realsize %lu\n",
3377 zone_names[j], memmap_pages, realsize);
3378
Christoph Lameter62672762007-02-10 01:43:07 -08003379 /* Account for reserved pages */
3380 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07003381 realsize -= dma_reserve;
Christoph Lameter62672762007-02-10 01:43:07 -08003382 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3383 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07003384 }
3385
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07003386 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387 nr_kernel_pages += realsize;
3388 nr_all_pages += realsize;
3389
3390 zone->spanned_pages = size;
3391 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07003392#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07003393 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07003394 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07003395 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07003396 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07003397#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398 zone->name = zone_names[j];
3399 spin_lock_init(&zone->lock);
3400 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07003401 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003403
Martin Bligh3bb1a8522006-10-28 10:38:24 -07003404 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405
Dave Hansened8ece22005-10-29 18:16:50 -07003406 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407 INIT_LIST_HEAD(&zone->active_list);
3408 INIT_LIST_HEAD(&zone->inactive_list);
3409 zone->nr_scan_active = 0;
3410 zone->nr_scan_inactive = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07003411 zap_zone_vm_stats(zone);
David Rientjese815af92007-10-16 23:25:54 -07003412 zone->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413 if (!size)
3414 continue;
3415
Mel Gormanba72cb82007-11-28 16:21:13 -08003416 set_pageblock_order(pageblock_default_order());
Mel Gorman835c1342007-10-16 01:25:47 -07003417 setup_usemap(pgdat, zone, size);
Dave Hansena2f3aa022007-01-10 23:15:30 -08003418 ret = init_currently_empty_zone(zone, zone_start_pfn,
3419 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07003420 BUG_ON(ret);
Heiko Carstens76cdd582008-05-14 16:05:52 -07003421 memmap_init(size, nid, j, zone_start_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423 }
3424}
3425
Sam Ravnborg577a32f2007-05-17 23:29:25 +02003426static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428 /* Skip empty nodes */
3429 if (!pgdat->node_spanned_pages)
3430 return;
3431
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003432#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07003433 /* ia64 gets its own node_mem_map, before this, without bootmem */
3434 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07003435 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003436 struct page *map;
3437
Bob Piccoe984bb42006-05-20 15:00:31 -07003438 /*
3439 * The zone's endpoints aren't required to be MAX_ORDER
3440 * aligned but the node_mem_map endpoints must be in order
3441 * for the buddy allocator to function correctly.
3442 */
3443 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3444 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3445 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3446 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07003447 map = alloc_remap(pgdat->node_id, size);
3448 if (!map)
3449 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07003450 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003451 }
Roman Zippel12d810c2007-05-31 00:40:54 -07003452#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 /*
3454 * With no DISCONTIG, the global mem_map is just set as node 0's
3455 */
Mel Gormanc7132162006-09-27 01:49:43 -07003456 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07003458#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3459 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
Thomas Bogendoerfer467bc462008-01-08 15:33:11 -08003460 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
Mel Gormanc7132162006-09-27 01:49:43 -07003461#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3462 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003463#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07003464#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003465}
3466
Alexander van Heukelumb5a0e012008-02-23 15:24:06 -08003467void __paginginit free_area_init_node(int nid, struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003468 unsigned long *zones_size, unsigned long node_start_pfn,
3469 unsigned long *zholes_size)
3470{
3471 pgdat->node_id = nid;
3472 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003473 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003474
3475 alloc_node_mem_map(pgdat);
Yinghai Lue8c27ac2008-06-01 13:15:22 -07003476#ifdef CONFIG_FLAT_NODE_MEM_MAP
3477 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3478 nid, (unsigned long)pgdat,
3479 (unsigned long)pgdat->node_mem_map);
3480#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003481
3482 free_area_init_core(pgdat, zones_size, zholes_size);
3483}
3484
Mel Gormanc7132162006-09-27 01:49:43 -07003485#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07003486
3487#if MAX_NUMNODES > 1
3488/*
3489 * Figure out the number of possible node ids.
3490 */
3491static void __init setup_nr_node_ids(void)
3492{
3493 unsigned int node;
3494 unsigned int highest = 0;
3495
3496 for_each_node_mask(node, node_possible_map)
3497 highest = node;
3498 nr_node_ids = highest + 1;
3499}
3500#else
3501static inline void setup_nr_node_ids(void)
3502{
3503}
3504#endif
3505
Mel Gormanc7132162006-09-27 01:49:43 -07003506/**
3507 * add_active_range - Register a range of PFNs backed by physical memory
3508 * @nid: The node ID the range resides on
3509 * @start_pfn: The start PFN of the available physical memory
3510 * @end_pfn: The end PFN of the available physical memory
3511 *
3512 * These ranges are stored in an early_node_map[] and later used by
3513 * free_area_init_nodes() to calculate zone sizes and holes. If the
3514 * range spans a memory hole, it is up to the architecture to ensure
3515 * the memory is not freed by the bootmem allocator. If possible
3516 * the range being registered will be merged with existing ranges.
3517 */
3518void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3519 unsigned long end_pfn)
3520{
3521 int i;
3522
Paul Jackson5dab8ec2008-06-25 05:44:40 -07003523 printk(KERN_DEBUG "Entering add_active_range(%d, %#lx, %#lx) "
Mel Gormanc7132162006-09-27 01:49:43 -07003524 "%d entries of %d used\n",
3525 nid, start_pfn, end_pfn,
3526 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3527
3528 /* Merge with existing active regions if possible */
3529 for (i = 0; i < nr_nodemap_entries; i++) {
3530 if (early_node_map[i].nid != nid)
3531 continue;
3532
3533 /* Skip if an existing region covers this new one */
3534 if (start_pfn >= early_node_map[i].start_pfn &&
3535 end_pfn <= early_node_map[i].end_pfn)
3536 return;
3537
3538 /* Merge forward if suitable */
3539 if (start_pfn <= early_node_map[i].end_pfn &&
3540 end_pfn > early_node_map[i].end_pfn) {
3541 early_node_map[i].end_pfn = end_pfn;
3542 return;
3543 }
3544
3545 /* Merge backward if suitable */
3546 if (start_pfn < early_node_map[i].end_pfn &&
3547 end_pfn >= early_node_map[i].start_pfn) {
3548 early_node_map[i].start_pfn = start_pfn;
3549 return;
3550 }
3551 }
3552
3553 /* Check that early_node_map is large enough */
3554 if (i >= MAX_ACTIVE_REGIONS) {
3555 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3556 MAX_ACTIVE_REGIONS);
3557 return;
3558 }
3559
3560 early_node_map[i].nid = nid;
3561 early_node_map[i].start_pfn = start_pfn;
3562 early_node_map[i].end_pfn = end_pfn;
3563 nr_nodemap_entries = i + 1;
3564}
3565
3566/**
Yinghai Lucc1050b2008-06-13 19:08:52 -07003567 * remove_active_range - Shrink an existing registered range of PFNs
Mel Gormanc7132162006-09-27 01:49:43 -07003568 * @nid: The node id the range is on that should be shrunk
Yinghai Lucc1050b2008-06-13 19:08:52 -07003569 * @start_pfn: The new PFN of the range
3570 * @end_pfn: The new PFN of the range
Mel Gormanc7132162006-09-27 01:49:43 -07003571 *
3572 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003573 * The map is kept near the end physical page range that has already been
3574 * registered. This function allows an arch to shrink an existing registered
3575 * range.
Mel Gormanc7132162006-09-27 01:49:43 -07003576 */
Yinghai Lucc1050b2008-06-13 19:08:52 -07003577void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3578 unsigned long end_pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07003579{
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003580 int i, j;
3581 int removed = 0;
Mel Gormanc7132162006-09-27 01:49:43 -07003582
Yinghai Lucc1050b2008-06-13 19:08:52 -07003583 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3584 nid, start_pfn, end_pfn);
3585
Mel Gormanc7132162006-09-27 01:49:43 -07003586 /* Find the old active region end and shrink */
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003587 for_each_active_range_index_in_nid(i, nid) {
Yinghai Lucc1050b2008-06-13 19:08:52 -07003588 if (early_node_map[i].start_pfn >= start_pfn &&
3589 early_node_map[i].end_pfn <= end_pfn) {
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003590 /* clear it */
Yinghai Lucc1050b2008-06-13 19:08:52 -07003591 early_node_map[i].start_pfn = 0;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003592 early_node_map[i].end_pfn = 0;
3593 removed = 1;
3594 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07003595 }
Yinghai Lucc1050b2008-06-13 19:08:52 -07003596 if (early_node_map[i].start_pfn < start_pfn &&
3597 early_node_map[i].end_pfn > start_pfn) {
3598 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3599 early_node_map[i].end_pfn = start_pfn;
3600 if (temp_end_pfn > end_pfn)
3601 add_active_range(nid, end_pfn, temp_end_pfn);
3602 continue;
3603 }
3604 if (early_node_map[i].start_pfn >= start_pfn &&
3605 early_node_map[i].end_pfn > end_pfn &&
3606 early_node_map[i].start_pfn < end_pfn) {
3607 early_node_map[i].start_pfn = end_pfn;
Yinghai Lucc1a9d82008-06-08 19:39:16 -07003608 continue;
3609 }
3610 }
3611
3612 if (!removed)
3613 return;
3614
3615 /* remove the blank ones */
3616 for (i = nr_nodemap_entries - 1; i > 0; i--) {
3617 if (early_node_map[i].nid != nid)
3618 continue;
3619 if (early_node_map[i].end_pfn)
3620 continue;
3621 /* we found it, get rid of it */
3622 for (j = i; j < nr_nodemap_entries - 1; j++)
3623 memcpy(&early_node_map[j], &early_node_map[j+1],
3624 sizeof(early_node_map[j]));
3625 j = nr_nodemap_entries - 1;
3626 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3627 nr_nodemap_entries--;
3628 }
Mel Gormanc7132162006-09-27 01:49:43 -07003629}
3630
3631/**
3632 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003633 *
Mel Gormanc7132162006-09-27 01:49:43 -07003634 * During discovery, it may be found that a table like SRAT is invalid
3635 * and an alternative discovery method must be used. This function removes
3636 * all currently registered regions.
3637 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003638void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07003639{
3640 memset(early_node_map, 0, sizeof(early_node_map));
3641 nr_nodemap_entries = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07003642#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3643 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3644 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3645#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -07003646}
3647
3648/* Compare two active node_active_regions */
3649static int __init cmp_node_active_region(const void *a, const void *b)
3650{
3651 struct node_active_region *arange = (struct node_active_region *)a;
3652 struct node_active_region *brange = (struct node_active_region *)b;
3653
3654 /* Done this way to avoid overflows */
3655 if (arange->start_pfn > brange->start_pfn)
3656 return 1;
3657 if (arange->start_pfn < brange->start_pfn)
3658 return -1;
3659
3660 return 0;
3661}
3662
3663/* sort the node_map by start_pfn */
3664static void __init sort_node_map(void)
3665{
3666 sort(early_node_map, (size_t)nr_nodemap_entries,
3667 sizeof(struct node_active_region),
3668 cmp_node_active_region, NULL);
3669}
3670
Mel Gormana6af2bc2007-02-10 01:42:57 -08003671/* Find the lowest pfn for a node */
Paul Jackson2bc0d262008-06-22 07:22:17 -07003672unsigned long __init find_min_pfn_for_node(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07003673{
3674 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08003675 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00003676
Mel Gormanc7132162006-09-27 01:49:43 -07003677 /* Assuming a sorted map, the first range found has the starting pfn */
3678 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08003679 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003680
Mel Gormana6af2bc2007-02-10 01:42:57 -08003681 if (min_pfn == ULONG_MAX) {
3682 printk(KERN_WARNING
Paul Jackson2bc0d262008-06-22 07:22:17 -07003683 "Could not find start_pfn for node %d\n", nid);
Mel Gormana6af2bc2007-02-10 01:42:57 -08003684 return 0;
3685 }
3686
3687 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07003688}
3689
3690/**
3691 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3692 *
3693 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003694 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07003695 */
3696unsigned long __init find_min_pfn_with_active_regions(void)
3697{
3698 return find_min_pfn_for_node(MAX_NUMNODES);
3699}
3700
3701/**
3702 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3703 *
3704 * It returns the maximum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003705 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07003706 */
3707unsigned long __init find_max_pfn_with_active_regions(void)
3708{
3709 int i;
3710 unsigned long max_pfn = 0;
3711
3712 for (i = 0; i < nr_nodemap_entries; i++)
3713 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
3714
3715 return max_pfn;
3716}
3717
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003718/*
3719 * early_calculate_totalpages()
3720 * Sum pages in active regions for movable zone.
3721 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3722 */
Adrian Bunk484f51f2007-10-16 01:26:03 -07003723static unsigned long __init early_calculate_totalpages(void)
Mel Gorman7e63efe2007-07-17 04:03:15 -07003724{
3725 int i;
3726 unsigned long totalpages = 0;
3727
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003728 for (i = 0; i < nr_nodemap_entries; i++) {
3729 unsigned long pages = early_node_map[i].end_pfn -
Mel Gorman7e63efe2007-07-17 04:03:15 -07003730 early_node_map[i].start_pfn;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003731 totalpages += pages;
3732 if (pages)
3733 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3734 }
3735 return totalpages;
Mel Gorman7e63efe2007-07-17 04:03:15 -07003736}
3737
Mel Gorman2a1e2742007-07-17 04:03:12 -07003738/*
3739 * Find the PFN the Movable zone begins in each node. Kernel memory
3740 * is spread evenly between nodes as long as the nodes have enough
3741 * memory. When they don't, some nodes will have more kernelcore than
3742 * others
3743 */
3744void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3745{
3746 int i, nid;
3747 unsigned long usable_startpfn;
3748 unsigned long kernelcore_node, kernelcore_remaining;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003749 unsigned long totalpages = early_calculate_totalpages();
3750 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07003751
Mel Gorman7e63efe2007-07-17 04:03:15 -07003752 /*
3753 * If movablecore was specified, calculate what size of
3754 * kernelcore that corresponds so that memory usable for
3755 * any allocation type is evenly spread. If both kernelcore
3756 * and movablecore are specified, then the value of kernelcore
3757 * will be used for required_kernelcore if it's greater than
3758 * what movablecore would have allowed.
3759 */
3760 if (required_movablecore) {
Mel Gorman7e63efe2007-07-17 04:03:15 -07003761 unsigned long corepages;
3762
3763 /*
3764 * Round-up so that ZONE_MOVABLE is at least as large as what
3765 * was requested by the user
3766 */
3767 required_movablecore =
3768 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3769 corepages = totalpages - required_movablecore;
3770
3771 required_kernelcore = max(required_kernelcore, corepages);
3772 }
3773
Mel Gorman2a1e2742007-07-17 04:03:12 -07003774 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3775 if (!required_kernelcore)
3776 return;
3777
3778 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3779 find_usable_zone_for_movable();
3780 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3781
3782restart:
3783 /* Spread kernelcore memory as evenly as possible throughout nodes */
3784 kernelcore_node = required_kernelcore / usable_nodes;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003785 for_each_node_state(nid, N_HIGH_MEMORY) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07003786 /*
3787 * Recalculate kernelcore_node if the division per node
3788 * now exceeds what is necessary to satisfy the requested
3789 * amount of memory for the kernel
3790 */
3791 if (required_kernelcore < kernelcore_node)
3792 kernelcore_node = required_kernelcore / usable_nodes;
3793
3794 /*
3795 * As the map is walked, we track how much memory is usable
3796 * by the kernel using kernelcore_remaining. When it is
3797 * 0, the rest of the node is usable by ZONE_MOVABLE
3798 */
3799 kernelcore_remaining = kernelcore_node;
3800
3801 /* Go through each range of PFNs within this node */
3802 for_each_active_range_index_in_nid(i, nid) {
3803 unsigned long start_pfn, end_pfn;
3804 unsigned long size_pages;
3805
3806 start_pfn = max(early_node_map[i].start_pfn,
3807 zone_movable_pfn[nid]);
3808 end_pfn = early_node_map[i].end_pfn;
3809 if (start_pfn >= end_pfn)
3810 continue;
3811
3812 /* Account for what is only usable for kernelcore */
3813 if (start_pfn < usable_startpfn) {
3814 unsigned long kernel_pages;
3815 kernel_pages = min(end_pfn, usable_startpfn)
3816 - start_pfn;
3817
3818 kernelcore_remaining -= min(kernel_pages,
3819 kernelcore_remaining);
3820 required_kernelcore -= min(kernel_pages,
3821 required_kernelcore);
3822
3823 /* Continue if range is now fully accounted */
3824 if (end_pfn <= usable_startpfn) {
3825
3826 /*
3827 * Push zone_movable_pfn to the end so
3828 * that if we have to rebalance
3829 * kernelcore across nodes, we will
3830 * not double account here
3831 */
3832 zone_movable_pfn[nid] = end_pfn;
3833 continue;
3834 }
3835 start_pfn = usable_startpfn;
3836 }
3837
3838 /*
3839 * The usable PFN range for ZONE_MOVABLE is from
3840 * start_pfn->end_pfn. Calculate size_pages as the
3841 * number of pages used as kernelcore
3842 */
3843 size_pages = end_pfn - start_pfn;
3844 if (size_pages > kernelcore_remaining)
3845 size_pages = kernelcore_remaining;
3846 zone_movable_pfn[nid] = start_pfn + size_pages;
3847
3848 /*
3849 * Some kernelcore has been met, update counts and
3850 * break if the kernelcore for this node has been
3851 * satisified
3852 */
3853 required_kernelcore -= min(required_kernelcore,
3854 size_pages);
3855 kernelcore_remaining -= size_pages;
3856 if (!kernelcore_remaining)
3857 break;
3858 }
3859 }
3860
3861 /*
3862 * If there is still required_kernelcore, we do another pass with one
3863 * less node in the count. This will push zone_movable_pfn[nid] further
3864 * along on the nodes that still have memory until kernelcore is
3865 * satisified
3866 */
3867 usable_nodes--;
3868 if (usable_nodes && required_kernelcore > usable_nodes)
3869 goto restart;
3870
3871 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3872 for (nid = 0; nid < MAX_NUMNODES; nid++)
3873 zone_movable_pfn[nid] =
3874 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3875}
3876
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003877/* Any regular memory on that node ? */
3878static void check_for_regular_memory(pg_data_t *pgdat)
3879{
3880#ifdef CONFIG_HIGHMEM
3881 enum zone_type zone_type;
3882
3883 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3884 struct zone *zone = &pgdat->node_zones[zone_type];
3885 if (zone->present_pages)
3886 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3887 }
3888#endif
3889}
3890
Mel Gormanc7132162006-09-27 01:49:43 -07003891/**
3892 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003893 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07003894 *
3895 * This will call free_area_init_node() for each active node in the system.
3896 * Using the page ranges provided by add_active_range(), the size of each
3897 * zone in each node and their holes is calculated. If the maximum PFN
3898 * between two adjacent zones match, it is assumed that the zone is empty.
3899 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3900 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3901 * starts where the previous one ended. For example, ZONE_DMA32 starts
3902 * at arch_max_dma_pfn.
3903 */
3904void __init free_area_init_nodes(unsigned long *max_zone_pfn)
3905{
3906 unsigned long nid;
3907 enum zone_type i;
3908
Mel Gormana6af2bc2007-02-10 01:42:57 -08003909 /* Sort early_node_map as initialisation assumes it is sorted */
3910 sort_node_map();
3911
Mel Gormanc7132162006-09-27 01:49:43 -07003912 /* Record where the zone boundaries are */
3913 memset(arch_zone_lowest_possible_pfn, 0,
3914 sizeof(arch_zone_lowest_possible_pfn));
3915 memset(arch_zone_highest_possible_pfn, 0,
3916 sizeof(arch_zone_highest_possible_pfn));
3917 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
3918 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
3919 for (i = 1; i < MAX_NR_ZONES; i++) {
Mel Gorman2a1e2742007-07-17 04:03:12 -07003920 if (i == ZONE_MOVABLE)
3921 continue;
Mel Gormanc7132162006-09-27 01:49:43 -07003922 arch_zone_lowest_possible_pfn[i] =
3923 arch_zone_highest_possible_pfn[i-1];
3924 arch_zone_highest_possible_pfn[i] =
3925 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
3926 }
Mel Gorman2a1e2742007-07-17 04:03:12 -07003927 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
3928 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
3929
3930 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3931 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
3932 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07003933
Mel Gormanc7132162006-09-27 01:49:43 -07003934 /* Print out the zone ranges */
3935 printk("Zone PFN ranges:\n");
Mel Gorman2a1e2742007-07-17 04:03:12 -07003936 for (i = 0; i < MAX_NR_ZONES; i++) {
3937 if (i == ZONE_MOVABLE)
3938 continue;
Paul Jackson5dab8ec2008-06-25 05:44:40 -07003939 printk(" %-8s %0#10lx -> %0#10lx\n",
Mel Gormanc7132162006-09-27 01:49:43 -07003940 zone_names[i],
3941 arch_zone_lowest_possible_pfn[i],
3942 arch_zone_highest_possible_pfn[i]);
Mel Gorman2a1e2742007-07-17 04:03:12 -07003943 }
3944
3945 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3946 printk("Movable zone start PFN for each node\n");
3947 for (i = 0; i < MAX_NUMNODES; i++) {
3948 if (zone_movable_pfn[i])
3949 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
3950 }
Mel Gormanc7132162006-09-27 01:49:43 -07003951
3952 /* Print out the early_node_map[] */
3953 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
3954 for (i = 0; i < nr_nodemap_entries; i++)
Paul Jackson5dab8ec2008-06-25 05:44:40 -07003955 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
Mel Gormanc7132162006-09-27 01:49:43 -07003956 early_node_map[i].start_pfn,
3957 early_node_map[i].end_pfn);
3958
3959 /* Initialise every node */
Christoph Lameter8ef82862007-02-20 13:57:52 -08003960 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07003961 for_each_online_node(nid) {
3962 pg_data_t *pgdat = NODE_DATA(nid);
3963 free_area_init_node(nid, pgdat, NULL,
3964 find_min_pfn_for_node(nid), NULL);
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07003965
3966 /* Any memory on that node */
3967 if (pgdat->node_present_pages)
3968 node_set_state(nid, N_HIGH_MEMORY);
3969 check_for_regular_memory(pgdat);
Mel Gormanc7132162006-09-27 01:49:43 -07003970 }
3971}
Mel Gorman2a1e2742007-07-17 04:03:12 -07003972
Mel Gorman7e63efe2007-07-17 04:03:15 -07003973static int __init cmdline_parse_core(char *p, unsigned long *core)
Mel Gorman2a1e2742007-07-17 04:03:12 -07003974{
3975 unsigned long long coremem;
3976 if (!p)
3977 return -EINVAL;
3978
3979 coremem = memparse(p, &p);
Mel Gorman7e63efe2007-07-17 04:03:15 -07003980 *core = coremem >> PAGE_SHIFT;
Mel Gorman2a1e2742007-07-17 04:03:12 -07003981
Mel Gorman7e63efe2007-07-17 04:03:15 -07003982 /* Paranoid check that UL is enough for the coremem value */
Mel Gorman2a1e2742007-07-17 04:03:12 -07003983 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
3984
3985 return 0;
3986}
Mel Gormaned7ed362007-07-17 04:03:14 -07003987
Mel Gorman7e63efe2007-07-17 04:03:15 -07003988/*
3989 * kernelcore=size sets the amount of memory for use for allocations that
3990 * cannot be reclaimed or migrated.
3991 */
3992static int __init cmdline_parse_kernelcore(char *p)
3993{
3994 return cmdline_parse_core(p, &required_kernelcore);
3995}
3996
3997/*
3998 * movablecore=size sets the amount of memory for use for allocations that
3999 * can be reclaimed or migrated.
4000 */
4001static int __init cmdline_parse_movablecore(char *p)
4002{
4003 return cmdline_parse_core(p, &required_movablecore);
4004}
4005
Mel Gormaned7ed362007-07-17 04:03:14 -07004006early_param("kernelcore", cmdline_parse_kernelcore);
Mel Gorman7e63efe2007-07-17 04:03:15 -07004007early_param("movablecore", cmdline_parse_movablecore);
Mel Gormaned7ed362007-07-17 04:03:14 -07004008
Mel Gormanc7132162006-09-27 01:49:43 -07004009#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4010
Mel Gorman0e0b8642006-09-27 01:49:56 -07004011/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004012 * set_dma_reserve - set the specified number of pages reserved in the first zone
4013 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07004014 *
4015 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4016 * In the DMA zone, a significant percentage may be consumed by kernel image
4017 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004018 * function may optionally be used to account for unfreeable pages in the
4019 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4020 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07004021 */
4022void __init set_dma_reserve(unsigned long new_dma_reserve)
4023{
4024 dma_reserve = new_dma_reserve;
4025}
4026
Dave Hansen93b75042005-06-23 00:07:47 -07004027#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028static bootmem_data_t contig_bootmem_data;
4029struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
4030
4031EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07004032#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004033
4034void __init free_area_init(unsigned long *zones_size)
4035{
Dave Hansen93b75042005-06-23 00:07:47 -07004036 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004037 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4038}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004039
Linus Torvalds1da177e2005-04-16 15:20:36 -07004040static int page_alloc_cpu_notify(struct notifier_block *self,
4041 unsigned long action, void *hcpu)
4042{
4043 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004044
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004045 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004046 drain_pages(cpu);
4047
4048 /*
4049 * Spill the event counters of the dead processor
4050 * into the current processors event counters.
4051 * This artificially elevates the count of the current
4052 * processor.
4053 */
Christoph Lameterf8891e52006-06-30 01:55:45 -07004054 vm_events_fold_cpu(cpu);
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004055
4056 /*
4057 * Zero the differential counters of the dead processor
4058 * so that the vm statistics are consistent.
4059 *
4060 * This is only okay since the processor is dead and cannot
4061 * race with what we are doing.
4062 */
Christoph Lameter2244b952006-06-30 01:55:33 -07004063 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004064 }
4065 return NOTIFY_OK;
4066}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067
4068void __init page_alloc_init(void)
4069{
4070 hotcpu_notifier(page_alloc_cpu_notify, 0);
4071}
4072
4073/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004074 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4075 * or min_free_kbytes changes.
4076 */
4077static void calculate_totalreserve_pages(void)
4078{
4079 struct pglist_data *pgdat;
4080 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004081 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004082
4083 for_each_online_pgdat(pgdat) {
4084 for (i = 0; i < MAX_NR_ZONES; i++) {
4085 struct zone *zone = pgdat->node_zones + i;
4086 unsigned long max = 0;
4087
4088 /* Find valid and maximum lowmem_reserve in the zone */
4089 for (j = i; j < MAX_NR_ZONES; j++) {
4090 if (zone->lowmem_reserve[j] > max)
4091 max = zone->lowmem_reserve[j];
4092 }
4093
4094 /* we treat pages_high as reserved pages. */
4095 max += zone->pages_high;
4096
4097 if (max > zone->present_pages)
4098 max = zone->present_pages;
4099 reserve_pages += max;
4100 }
4101 }
4102 totalreserve_pages = reserve_pages;
4103}
4104
4105/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004106 * setup_per_zone_lowmem_reserve - called whenever
4107 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4108 * has a correct pages reserved value, so an adequate number of
4109 * pages are left in the zone after a successful __alloc_pages().
4110 */
4111static void setup_per_zone_lowmem_reserve(void)
4112{
4113 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004114 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004115
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08004116 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004117 for (j = 0; j < MAX_NR_ZONES; j++) {
4118 struct zone *zone = pgdat->node_zones + j;
4119 unsigned long present_pages = zone->present_pages;
4120
4121 zone->lowmem_reserve[j] = 0;
4122
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004123 idx = j;
4124 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004125 struct zone *lower_zone;
4126
Christoph Lameter2f6726e2006-09-25 23:31:18 -07004127 idx--;
4128
Linus Torvalds1da177e2005-04-16 15:20:36 -07004129 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4130 sysctl_lowmem_reserve_ratio[idx] = 1;
4131
4132 lower_zone = pgdat->node_zones + idx;
4133 lower_zone->lowmem_reserve[j] = present_pages /
4134 sysctl_lowmem_reserve_ratio[idx];
4135 present_pages += lower_zone->present_pages;
4136 }
4137 }
4138 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004139
4140 /* update totalreserve_pages */
4141 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004142}
4143
Randy Dunlap88ca3b92006-10-04 02:15:25 -07004144/**
4145 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4146 *
4147 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4148 * with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004149 */
Dave Hansen3947be12005-10-29 18:16:54 -07004150void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004151{
4152 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4153 unsigned long lowmem_pages = 0;
4154 struct zone *zone;
4155 unsigned long flags;
4156
4157 /* Calculate total number of !ZONE_HIGHMEM pages */
4158 for_each_zone(zone) {
4159 if (!is_highmem(zone))
4160 lowmem_pages += zone->present_pages;
4161 }
4162
4163 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07004164 u64 tmp;
4165
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07004167 tmp = (u64)pages_min * zone->present_pages;
4168 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004169 if (is_highmem(zone)) {
4170 /*
Nick Piggin669ed172005-11-13 16:06:45 -08004171 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4172 * need highmem pages, so cap pages_min to a small
4173 * value here.
4174 *
4175 * The (pages_high-pages_low) and (pages_low-pages_min)
4176 * deltas controls asynch page reclaim, and so should
4177 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004178 */
4179 int min_pages;
4180
4181 min_pages = zone->present_pages / 1024;
4182 if (min_pages < SWAP_CLUSTER_MAX)
4183 min_pages = SWAP_CLUSTER_MAX;
4184 if (min_pages > 128)
4185 min_pages = 128;
4186 zone->pages_min = min_pages;
4187 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08004188 /*
4189 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07004190 * proportionate to the zone's size.
4191 */
Nick Piggin669ed172005-11-13 16:06:45 -08004192 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004193 }
4194
Andrew Mortonac924c62006-05-15 09:43:59 -07004195 zone->pages_low = zone->pages_min + (tmp >> 2);
4196 zone->pages_high = zone->pages_min + (tmp >> 1);
Mel Gorman56fd56b2007-10-16 01:25:58 -07004197 setup_zone_migrate_reserve(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004198 spin_unlock_irqrestore(&zone->lru_lock, flags);
4199 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07004200
4201 /* update totalreserve_pages */
4202 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004203}
4204
4205/*
4206 * Initialise min_free_kbytes.
4207 *
4208 * For small machines we want it small (128k min). For large machines
4209 * we want it large (64MB max). But it is not linear, because network
4210 * bandwidth does not increase linearly with machine size. We use
4211 *
4212 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4213 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4214 *
4215 * which yields
4216 *
4217 * 16MB: 512k
4218 * 32MB: 724k
4219 * 64MB: 1024k
4220 * 128MB: 1448k
4221 * 256MB: 2048k
4222 * 512MB: 2896k
4223 * 1024MB: 4096k
4224 * 2048MB: 5792k
4225 * 4096MB: 8192k
4226 * 8192MB: 11584k
4227 * 16384MB: 16384k
4228 */
4229static int __init init_per_zone_pages_min(void)
4230{
4231 unsigned long lowmem_kbytes;
4232
4233 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4234
4235 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4236 if (min_free_kbytes < 128)
4237 min_free_kbytes = 128;
4238 if (min_free_kbytes > 65536)
4239 min_free_kbytes = 65536;
4240 setup_per_zone_pages_min();
4241 setup_per_zone_lowmem_reserve();
4242 return 0;
4243}
4244module_init(init_per_zone_pages_min)
4245
4246/*
4247 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4248 * that we can call two helper functions whenever min_free_kbytes
4249 * changes.
4250 */
4251int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4252 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4253{
4254 proc_dointvec(table, write, file, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07004255 if (write)
4256 setup_per_zone_pages_min();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004257 return 0;
4258}
4259
Christoph Lameter96146342006-07-03 00:24:13 -07004260#ifdef CONFIG_NUMA
4261int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4262 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4263{
4264 struct zone *zone;
4265 int rc;
4266
4267 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4268 if (rc)
4269 return rc;
4270
4271 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07004272 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07004273 sysctl_min_unmapped_ratio) / 100;
4274 return 0;
4275}
Christoph Lameter0ff38492006-09-25 23:31:52 -07004276
4277int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4278 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4279{
4280 struct zone *zone;
4281 int rc;
4282
4283 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4284 if (rc)
4285 return rc;
4286
4287 for_each_zone(zone)
4288 zone->min_slab_pages = (zone->present_pages *
4289 sysctl_min_slab_ratio) / 100;
4290 return 0;
4291}
Christoph Lameter96146342006-07-03 00:24:13 -07004292#endif
4293
Linus Torvalds1da177e2005-04-16 15:20:36 -07004294/*
4295 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4296 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4297 * whenever sysctl_lowmem_reserve_ratio changes.
4298 *
4299 * The reserve ratio obviously has absolutely no relation with the
4300 * pages_min watermarks. The lowmem reserve ratio can only make sense
4301 * if in function of the boot time zone sizes.
4302 */
4303int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4304 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4305{
4306 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4307 setup_per_zone_lowmem_reserve();
4308 return 0;
4309}
4310
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08004311/*
4312 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4313 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4314 * can have before it gets flushed back to buddy allocator.
4315 */
4316
4317int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4318 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4319{
4320 struct zone *zone;
4321 unsigned int cpu;
4322 int ret;
4323
4324 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4325 if (!write || (ret == -EINVAL))
4326 return ret;
4327 for_each_zone(zone) {
4328 for_each_online_cpu(cpu) {
4329 unsigned long high;
4330 high = zone->present_pages / percpu_pagelist_fraction;
4331 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4332 }
4333 }
4334 return 0;
4335}
4336
David S. Millerf034b5d2006-08-24 03:08:07 -07004337int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004338
4339#ifdef CONFIG_NUMA
4340static int __init set_hashdist(char *str)
4341{
4342 if (!str)
4343 return 0;
4344 hashdist = simple_strtoul(str, &str, 0);
4345 return 1;
4346}
4347__setup("hashdist=", set_hashdist);
4348#endif
4349
4350/*
4351 * allocate a large system hash table from bootmem
4352 * - it is assumed that the hash table must contain an exact power-of-2
4353 * quantity of entries
4354 * - limit is the number of hash buckets, not the total allocation size
4355 */
4356void *__init alloc_large_system_hash(const char *tablename,
4357 unsigned long bucketsize,
4358 unsigned long numentries,
4359 int scale,
4360 int flags,
4361 unsigned int *_hash_shift,
4362 unsigned int *_hash_mask,
4363 unsigned long limit)
4364{
4365 unsigned long long max = limit;
4366 unsigned long log2qty, size;
4367 void *table = NULL;
4368
4369 /* allow the kernel cmdline to have a say */
4370 if (!numentries) {
4371 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08004372 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004373 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4374 numentries >>= 20 - PAGE_SHIFT;
4375 numentries <<= 20 - PAGE_SHIFT;
4376
4377 /* limit to 1 bucket per 2^scale bytes of low memory */
4378 if (scale > PAGE_SHIFT)
4379 numentries >>= (scale - PAGE_SHIFT);
4380 else
4381 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08004382
4383 /* Make sure we've got at least a 0-order allocation.. */
4384 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4385 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004386 }
John Hawkes6e692ed2006-03-25 03:08:02 -08004387 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004388
4389 /* limit allocation size to 1/16 total memory by default */
4390 if (max == 0) {
4391 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4392 do_div(max, bucketsize);
4393 }
4394
4395 if (numentries > max)
4396 numentries = max;
4397
David Howellsf0d1b0b2006-12-08 02:37:49 -08004398 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004399
4400 do {
4401 size = bucketsize << log2qty;
4402 if (flags & HASH_EARLY)
4403 table = alloc_bootmem(size);
4404 else if (hashdist)
4405 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4406 else {
Pavel Machek2309f9e2008-04-28 02:13:35 -07004407 unsigned long order = get_order(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004408 table = (void*) __get_free_pages(GFP_ATOMIC, order);
Eric Dumazet1037b832007-07-15 23:38:05 -07004409 /*
4410 * If bucketsize is not a power-of-two, we may free
4411 * some pages at the end of hash table.
4412 */
4413 if (table) {
4414 unsigned long alloc_end = (unsigned long)table +
4415 (PAGE_SIZE << order);
4416 unsigned long used = (unsigned long)table +
4417 PAGE_ALIGN(size);
4418 split_page(virt_to_page(table), order);
4419 while (used < alloc_end) {
4420 free_page(used);
4421 used += PAGE_SIZE;
4422 }
4423 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004424 }
4425 } while (!table && size > PAGE_SIZE && --log2qty);
4426
4427 if (!table)
4428 panic("Failed to allocate %s hash table\n", tablename);
4429
Dan Alonib49ad482007-07-15 23:38:23 -07004430 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004431 tablename,
4432 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08004433 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004434 size);
4435
4436 if (_hash_shift)
4437 *_hash_shift = log2qty;
4438 if (_hash_mask)
4439 *_hash_mask = (1 << log2qty) - 1;
4440
4441 return table;
4442}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004443
4444#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004445struct page *pfn_to_page(unsigned long pfn)
4446{
Andy Whitcroft67de6482006-06-23 02:03:12 -07004447 return __pfn_to_page(pfn);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004448}
4449unsigned long page_to_pfn(struct page *page)
4450{
Andy Whitcroft67de6482006-06-23 02:03:12 -07004451 return __page_to_pfn(page);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004452}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08004453EXPORT_SYMBOL(pfn_to_page);
4454EXPORT_SYMBOL(page_to_pfn);
4455#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
Andrew Morton6220ec72006-10-19 23:29:05 -07004456
Mel Gorman835c1342007-10-16 01:25:47 -07004457/* Return a pointer to the bitmap storing bits affecting a block of pages */
4458static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4459 unsigned long pfn)
4460{
4461#ifdef CONFIG_SPARSEMEM
4462 return __pfn_to_section(pfn)->pageblock_flags;
4463#else
4464 return zone->pageblock_flags;
4465#endif /* CONFIG_SPARSEMEM */
4466}
Andrew Morton6220ec72006-10-19 23:29:05 -07004467
Mel Gorman835c1342007-10-16 01:25:47 -07004468static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4469{
4470#ifdef CONFIG_SPARSEMEM
4471 pfn &= (PAGES_PER_SECTION-1);
Mel Gormand9c23402007-10-16 01:26:01 -07004472 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004473#else
4474 pfn = pfn - zone->zone_start_pfn;
Mel Gormand9c23402007-10-16 01:26:01 -07004475 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
Mel Gorman835c1342007-10-16 01:25:47 -07004476#endif /* CONFIG_SPARSEMEM */
4477}
4478
4479/**
Mel Gormand9c23402007-10-16 01:26:01 -07004480 * 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 -07004481 * @page: The page within the block of interest
4482 * @start_bitidx: The first bit of interest to retrieve
4483 * @end_bitidx: The last bit of interest
4484 * returns pageblock_bits flags
4485 */
4486unsigned long get_pageblock_flags_group(struct page *page,
4487 int start_bitidx, int end_bitidx)
4488{
4489 struct zone *zone;
4490 unsigned long *bitmap;
4491 unsigned long pfn, bitidx;
4492 unsigned long flags = 0;
4493 unsigned long value = 1;
4494
4495 zone = page_zone(page);
4496 pfn = page_to_pfn(page);
4497 bitmap = get_pageblock_bitmap(zone, pfn);
4498 bitidx = pfn_to_bitidx(zone, pfn);
4499
4500 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4501 if (test_bit(bitidx + start_bitidx, bitmap))
4502 flags |= value;
4503
4504 return flags;
4505}
4506
4507/**
Mel Gormand9c23402007-10-16 01:26:01 -07004508 * 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 -07004509 * @page: The page within the block of interest
4510 * @start_bitidx: The first bit of interest
4511 * @end_bitidx: The last bit of interest
4512 * @flags: The flags to set
4513 */
4514void set_pageblock_flags_group(struct page *page, unsigned long flags,
4515 int start_bitidx, int end_bitidx)
4516{
4517 struct zone *zone;
4518 unsigned long *bitmap;
4519 unsigned long pfn, bitidx;
4520 unsigned long value = 1;
4521
4522 zone = page_zone(page);
4523 pfn = page_to_pfn(page);
4524 bitmap = get_pageblock_bitmap(zone, pfn);
4525 bitidx = pfn_to_bitidx(zone, pfn);
KAMEZAWA Hiroyuki86051ca2008-04-29 00:58:21 -07004526 VM_BUG_ON(pfn < zone->zone_start_pfn);
4527 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
Mel Gorman835c1342007-10-16 01:25:47 -07004528
4529 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4530 if (flags & value)
4531 __set_bit(bitidx + start_bitidx, bitmap);
4532 else
4533 __clear_bit(bitidx + start_bitidx, bitmap);
4534}
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004535
4536/*
4537 * This is designed as sub function...plz see page_isolation.c also.
4538 * set/clear page block's type to be ISOLATE.
4539 * page allocater never alloc memory from ISOLATE block.
4540 */
4541
4542int set_migratetype_isolate(struct page *page)
4543{
4544 struct zone *zone;
4545 unsigned long flags;
4546 int ret = -EBUSY;
4547
4548 zone = page_zone(page);
4549 spin_lock_irqsave(&zone->lock, flags);
4550 /*
4551 * In future, more migrate types will be able to be isolation target.
4552 */
4553 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4554 goto out;
4555 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4556 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4557 ret = 0;
4558out:
4559 spin_unlock_irqrestore(&zone->lock, flags);
4560 if (!ret)
Christoph Lameter9f8f2172008-02-04 22:29:11 -08004561 drain_all_pages();
KAMEZAWA Hiroyukia5d76b52007-10-16 01:26:11 -07004562 return ret;
4563}
4564
4565void unset_migratetype_isolate(struct page *page)
4566{
4567 struct zone *zone;
4568 unsigned long flags;
4569 zone = page_zone(page);
4570 spin_lock_irqsave(&zone->lock, flags);
4571 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4572 goto out;
4573 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4574 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4575out:
4576 spin_unlock_irqrestore(&zone->lock, flags);
4577}
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07004578
4579#ifdef CONFIG_MEMORY_HOTREMOVE
4580/*
4581 * All pages in the range must be isolated before calling this.
4582 */
4583void
4584__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4585{
4586 struct page *page;
4587 struct zone *zone;
4588 int order, i;
4589 unsigned long pfn;
4590 unsigned long flags;
4591 /* find the first valid pfn */
4592 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4593 if (pfn_valid(pfn))
4594 break;
4595 if (pfn == end_pfn)
4596 return;
4597 zone = page_zone(pfn_to_page(pfn));
4598 spin_lock_irqsave(&zone->lock, flags);
4599 pfn = start_pfn;
4600 while (pfn < end_pfn) {
4601 if (!pfn_valid(pfn)) {
4602 pfn++;
4603 continue;
4604 }
4605 page = pfn_to_page(pfn);
4606 BUG_ON(page_count(page));
4607 BUG_ON(!PageBuddy(page));
4608 order = page_order(page);
4609#ifdef CONFIG_DEBUG_VM
4610 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4611 pfn, 1 << order, end_pfn);
4612#endif
4613 list_del(&page->lru);
4614 rmv_page_order(page);
4615 zone->free_area[order].nr_free--;
4616 __mod_zone_page_state(zone, NR_FREE_PAGES,
4617 - (1UL << order));
4618 for (i = 0; i < (1 << order); i++)
4619 SetPageReserved((page+i));
4620 pfn += (1 << order);
4621 }
4622 spin_unlock_irqrestore(&zone->lock, flags);
4623}
4624#endif