blob: 05ace44852eb88c9f2c6fa50b98d3ecd569d202c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
22#include <linux/bootmem.h>
23#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070024#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/module.h>
26#include <linux/suspend.h>
27#include <linux/pagevec.h>
28#include <linux/blkdev.h>
29#include <linux/slab.h>
30#include <linux/notifier.h>
31#include <linux/topology.h>
32#include <linux/sysctl.h>
33#include <linux/cpu.h>
34#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070035#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/nodemask.h>
37#include <linux/vmalloc.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080038#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070039#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070040#include <linux/sort.h>
41#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070042#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080043#include <linux/fault-inject.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070046#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include "internal.h"
48
49/*
50 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
51 * initializer cleaner
52 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070053nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
Dean Nelson7223a932005-03-23 19:00:00 -070054EXPORT_SYMBOL(node_online_map);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070055nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
Dean Nelson7223a932005-03-23 19:00:00 -070056EXPORT_SYMBOL(node_possible_map);
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070057unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070058unsigned long totalreserve_pages __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059long nr_swap_pages;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080060int percpu_pagelist_fraction;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080062static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080063
Linus Torvalds1da177e2005-04-16 15:20:36 -070064/*
65 * results with 256, 32 in the lowmem_reserve sysctl:
66 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
67 * 1G machine -> (16M dma, 784M normal, 224M high)
68 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
69 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
70 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010071 *
72 * TBD: should special case ZONE_DMA32 machines here - in those we normally
73 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -070075int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -080076#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -070077 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -080078#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -070079#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -070080 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -070081#endif
Christoph Lametere53ef382006-09-25 23:31:14 -070082#ifdef CONFIG_HIGHMEM
Christoph Lameter2f1b6242006-09-25 23:31:13 -070083 32
Christoph Lametere53ef382006-09-25 23:31:14 -070084#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -070085};
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
87EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Helge Deller15ad7cd2006-12-06 20:40:36 -080089static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -080090#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -070091 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -080092#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -070093#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -070094 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -070095#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -070096 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -070097#ifdef CONFIG_HIGHMEM
Christoph Lameter2f1b6242006-09-25 23:31:13 -070098 "HighMem"
Christoph Lametere53ef382006-09-25 23:31:14 -070099#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700100};
101
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102int min_free_kbytes = 1024;
103
Yasunori Goto86356ab2006-06-23 02:03:09 -0700104unsigned long __meminitdata nr_kernel_pages;
105unsigned long __meminitdata nr_all_pages;
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700106static unsigned long __meminitdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
Mel Gormanc7132162006-09-27 01:49:43 -0700108#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
109 /*
110 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
111 * ranges of memory (RAM) that may be registered with add_active_range().
112 * Ranges passed to add_active_range() will be merged if possible
113 * so the number of times add_active_range() can be called is
114 * related to the number of nodes and the number of holes
115 */
116 #ifdef CONFIG_MAX_ACTIVE_REGIONS
117 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
118 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
119 #else
120 #if MAX_NUMNODES >= 32
121 /* If there can be many nodes, allow up to 50 holes per node */
122 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
123 #else
124 /* By default, allow up to 256 distinct regions */
125 #define MAX_ACTIVE_REGIONS 256
126 #endif
127 #endif
128
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700129 struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
130 int __meminitdata nr_nodemap_entries;
131 unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
132 unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Mel Gormanfb014392006-09-27 01:49:59 -0700133#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
134 unsigned long __initdata node_boundary_start_pfn[MAX_NUMNODES];
135 unsigned long __initdata node_boundary_end_pfn[MAX_NUMNODES];
136#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -0700137#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
138
Miklos Szeredi418508c2007-05-23 13:57:55 -0700139#if MAX_NUMNODES > 1
140int nr_node_ids __read_mostly = MAX_NUMNODES;
141EXPORT_SYMBOL(nr_node_ids);
142#endif
143
Nick Piggin13e74442006-01-06 00:10:58 -0800144#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700145static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700147 int ret = 0;
148 unsigned seq;
149 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700150
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700151 do {
152 seq = zone_span_seqbegin(zone);
153 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
154 ret = 1;
155 else if (pfn < zone->zone_start_pfn)
156 ret = 1;
157 } while (zone_span_seqretry(zone, seq));
158
159 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700160}
161
162static int page_is_consistent(struct zone *zone, struct page *page)
163{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700164 if (!pfn_valid_within(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700165 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700167 return 0;
168
169 return 1;
170}
171/*
172 * Temporary debugging check for pages not lying within a given zone.
173 */
174static int bad_range(struct zone *zone, struct page *page)
175{
176 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700178 if (!page_is_consistent(zone, page))
179 return 1;
180
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 return 0;
182}
Nick Piggin13e74442006-01-06 00:10:58 -0800183#else
184static inline int bad_range(struct zone *zone, struct page *page)
185{
186 return 0;
187}
188#endif
189
Nick Piggin224abf92006-01-06 00:11:11 -0800190static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191{
Nick Piggin224abf92006-01-06 00:11:11 -0800192 printk(KERN_EMERG "Bad page state in process '%s'\n"
Hugh Dickins7365f3d2006-01-11 12:17:18 -0800193 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
194 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
195 KERN_EMERG "Backtrace:\n",
Nick Piggin224abf92006-01-06 00:11:11 -0800196 current->comm, page, (int)(2*sizeof(unsigned long)),
197 (unsigned long)page->flags, page->mapping,
198 page_mapcount(page), page_count(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 dump_stack();
Hugh Dickins334795e2005-06-21 17:15:08 -0700200 page->flags &= ~(1 << PG_lru |
201 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 1 << PG_active |
204 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700205 1 << PG_reclaim |
206 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 1 << PG_swapcache |
Nick Piggin676165a2006-04-10 11:21:48 +1000208 1 << PG_writeback |
209 1 << PG_buddy );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 set_page_count(page, 0);
211 reset_page_mapcount(page);
212 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700213 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214}
215
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216/*
217 * Higher-order pages are called "compound pages". They are structured thusly:
218 *
219 * The first PAGE_SIZE page is called the "head page".
220 *
221 * The remaining PAGE_SIZE pages are called "tail pages".
222 *
223 * All pages have PG_compound set. All pages have their ->private pointing at
224 * the head page (even the head page has this).
225 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800226 * The first tail page's ->lru.next holds the address of the compound page's
227 * put_page() function. Its ->lru.prev holds the order of allocation.
228 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800230
231static void free_compound_page(struct page *page)
232{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700233 __free_pages_ok(page, compound_order(page));
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800234}
235
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236static void prep_compound_page(struct page *page, unsigned long order)
237{
238 int i;
239 int nr_pages = 1 << order;
240
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800241 set_compound_page_dtor(page, free_compound_page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700242 set_compound_order(page, order);
Christoph Lameter6d777952007-05-06 14:49:40 -0700243 __SetPageHead(page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700244 for (i = 1; i < nr_pages; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 struct page *p = page + i;
246
Christoph Lameterd85f3382007-05-06 14:49:39 -0700247 __SetPageTail(p);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700248 p->first_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 }
250}
251
252static void destroy_compound_page(struct page *page, unsigned long order)
253{
254 int i;
255 int nr_pages = 1 << order;
256
Christoph Lameterd85f3382007-05-06 14:49:39 -0700257 if (unlikely(compound_order(page) != order))
Nick Piggin224abf92006-01-06 00:11:11 -0800258 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Christoph Lameter6d777952007-05-06 14:49:40 -0700260 if (unlikely(!PageHead(page)))
Christoph Lameterd85f3382007-05-06 14:49:39 -0700261 bad_page(page);
Christoph Lameter6d777952007-05-06 14:49:40 -0700262 __ClearPageHead(page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700263 for (i = 1; i < nr_pages; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 struct page *p = page + i;
265
Christoph Lameter6d777952007-05-06 14:49:40 -0700266 if (unlikely(!PageTail(p) |
Christoph Lameterd85f3382007-05-06 14:49:39 -0700267 (p->first_page != page)))
Nick Piggin224abf92006-01-06 00:11:11 -0800268 bad_page(page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700269 __ClearPageTail(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 }
271}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Nick Piggin17cf4402006-03-22 00:08:41 -0800273static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
274{
275 int i;
276
Nick Piggin725d7042006-09-25 23:30:55 -0700277 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
Andrew Morton6626c5d2006-03-22 00:08:42 -0800278 /*
279 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
280 * and __GFP_HIGHMEM from hard or soft interrupt context.
281 */
Nick Piggin725d7042006-09-25 23:30:55 -0700282 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800283 for (i = 0; i < (1 << order); i++)
284 clear_highpage(page + i);
285}
286
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287/*
288 * function for dealing with page's order in buddy system.
289 * zone->lock is already acquired when we use these.
290 * So, we don't need atomic page->flags operations here.
291 */
Andrew Morton6aa30012006-04-18 22:20:52 -0700292static inline unsigned long page_order(struct page *page)
293{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700294 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
296
Andrew Morton6aa30012006-04-18 22:20:52 -0700297static inline void set_page_order(struct page *page, int order)
298{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700299 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000300 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301}
302
303static inline void rmv_page_order(struct page *page)
304{
Nick Piggin676165a2006-04-10 11:21:48 +1000305 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700306 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307}
308
309/*
310 * Locate the struct page for both the matching buddy in our
311 * pair (buddy1) and the combined O(n+1) page they form (page).
312 *
313 * 1) Any buddy B1 will have an order O twin B2 which satisfies
314 * the following equation:
315 * B2 = B1 ^ (1 << O)
316 * For example, if the starting buddy (buddy2) is #8 its order
317 * 1 buddy is #10:
318 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
319 *
320 * 2) Any buddy B will have an order O+1 parent P which
321 * satisfies the following equation:
322 * P = B & ~(1 << O)
323 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200324 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 */
326static inline struct page *
327__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
328{
329 unsigned long buddy_idx = page_idx ^ (1 << order);
330
331 return page + (buddy_idx - page_idx);
332}
333
334static inline unsigned long
335__find_combined_index(unsigned long page_idx, unsigned int order)
336{
337 return (page_idx & ~(1 << order));
338}
339
340/*
341 * This function checks whether a page is free && is the buddy
342 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800343 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000344 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700345 * (c) a page and its buddy have the same order &&
346 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 *
Nick Piggin676165a2006-04-10 11:21:48 +1000348 * For recording whether a page is in the buddy system, we use PG_buddy.
349 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
350 *
351 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700353static inline int page_is_buddy(struct page *page, struct page *buddy,
354 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355{
Andy Whitcroft14e07292007-05-06 14:49:14 -0700356 if (!pfn_valid_within(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800357 return 0;
Nick Piggin13e74442006-01-06 00:10:58 -0800358
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700359 if (page_zone_id(page) != page_zone_id(buddy))
360 return 0;
361
362 if (PageBuddy(buddy) && page_order(buddy) == order) {
363 BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700364 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000365 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700366 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367}
368
369/*
370 * Freeing function for a buddy system allocator.
371 *
372 * The concept of a buddy system is to maintain direct-mapped table
373 * (containing bit values) for memory blocks of various "orders".
374 * The bottom level table contains the map for the smallest allocatable
375 * units of memory (here, pages), and each level above it describes
376 * pairs of units from the levels below, hence, "buddies".
377 * At a high level, all that happens here is marking the table entry
378 * at the bottom level available, and propagating the changes upward
379 * as necessary, plus some accounting needed to play nicely with other
380 * parts of the VM system.
381 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000382 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700383 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 * So when we are allocating or freeing one, we can derive the state of the
385 * other. That is, if we allocate a small block, and both were
386 * free, the remainder of the region must be split into blocks.
387 * If a block is freed, and its buddy is also free, then this
388 * triggers coalescing into a block of larger size.
389 *
390 * -- wli
391 */
392
Nick Piggin48db57f2006-01-08 01:00:42 -0800393static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 struct zone *zone, unsigned int order)
395{
396 unsigned long page_idx;
397 int order_size = 1 << order;
398
Nick Piggin224abf92006-01-06 00:11:11 -0800399 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 destroy_compound_page(page, order);
401
402 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
403
Nick Piggin725d7042006-09-25 23:30:55 -0700404 VM_BUG_ON(page_idx & (order_size - 1));
405 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Christoph Lameterd23ad422007-02-10 01:43:02 -0800407 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 while (order < MAX_ORDER-1) {
409 unsigned long combined_idx;
410 struct free_area *area;
411 struct page *buddy;
412
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700414 if (!page_is_buddy(page, buddy, order))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800416
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 list_del(&buddy->lru);
418 area = zone->free_area + order;
419 area->nr_free--;
420 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800421 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 page = page + (combined_idx - page_idx);
423 page_idx = combined_idx;
424 order++;
425 }
426 set_page_order(page, order);
427 list_add(&page->lru, &zone->free_area[order].free_list);
428 zone->free_area[order].nr_free++;
429}
430
Nick Piggin224abf92006-01-06 00:11:11 -0800431static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
Nick Piggin92be2e32006-01-06 00:10:57 -0800433 if (unlikely(page_mapcount(page) |
434 (page->mapping != NULL) |
435 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 (page->flags & (
437 1 << PG_lru |
438 1 << PG_private |
439 1 << PG_locked |
440 1 << PG_active |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 1 << PG_slab |
442 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700443 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000444 1 << PG_reserved |
445 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800446 bad_page(page);
Christoph Lameterd85f3382007-05-06 14:49:39 -0700447 /*
448 * PageReclaim == PageTail. It is only an error
449 * for PageReclaim to be set if PageCompound is clear.
450 */
451 if (unlikely(!PageCompound(page) && PageReclaim(page)))
452 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700454 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800455 /*
456 * For now, we report if PG_reserved was found set, but do not
457 * clear it, and do not free the page. But we shall soon need
458 * to do more, for when the ZERO_PAGE count wraps negative.
459 */
460 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461}
462
463/*
464 * Frees a list of pages.
465 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700466 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 *
468 * If the zone was previously in an "all pages pinned" state then look to
469 * see if this freeing clears that state.
470 *
471 * And clear the zone's pages_scanned counter, to hold off the "all pages are
472 * pinned" detection logic.
473 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800474static void free_pages_bulk(struct zone *zone, int count,
475 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
Nick Pigginc54ad302006-01-06 00:10:56 -0800477 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 zone->all_unreclaimable = 0;
479 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800480 while (count--) {
481 struct page *page;
482
Nick Piggin725d7042006-09-25 23:30:55 -0700483 VM_BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800485 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800487 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800489 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490}
491
Nick Piggin48db57f2006-01-08 01:00:42 -0800492static void free_one_page(struct zone *zone, struct page *page, int order)
493{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700494 spin_lock(&zone->lock);
495 zone->all_unreclaimable = 0;
496 zone->pages_scanned = 0;
Paul Jackson0798e512006-12-06 20:31:38 -0800497 __free_one_page(page, zone, order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700498 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800499}
500
501static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
Nick Pigginc54ad302006-01-06 00:10:56 -0800503 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800505 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800508 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800509 if (reserved)
510 return;
511
Nick Piggin9858db52006-10-11 01:21:30 -0700512 if (!PageHighMem(page))
513 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Nick Piggindafb1362006-10-11 01:21:30 -0700514 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800515 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700516
Nick Pigginc54ad302006-01-06 00:10:56 -0800517 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700518 __count_vm_events(PGFREE, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800519 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800520 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521}
522
David Howellsa226f6c2006-01-06 00:11:08 -0800523/*
524 * permit the bootmem allocator to evade page validation on high-order frees
525 */
526void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
527{
528 if (order == 0) {
529 __ClearPageReserved(page);
530 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800531 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800532 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800533 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800534 int loop;
535
Nick Piggin545b1ea2006-03-22 00:08:07 -0800536 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800537 for (loop = 0; loop < BITS_PER_LONG; loop++) {
538 struct page *p = &page[loop];
539
Nick Piggin545b1ea2006-03-22 00:08:07 -0800540 if (loop + 1 < BITS_PER_LONG)
541 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800542 __ClearPageReserved(p);
543 set_page_count(p, 0);
544 }
545
Nick Piggin7835e982006-03-22 00:08:40 -0800546 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800547 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800548 }
549}
550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
552/*
553 * The order of subdivision here is critical for the IO subsystem.
554 * Please do not alter this order without good reasons and regression
555 * testing. Specifically, as large blocks of memory are subdivided,
556 * the order in which smaller blocks are delivered depends on the order
557 * they're subdivided in this function. This is the primary factor
558 * influencing the order in which pages are delivered to the IO
559 * subsystem according to empirical testing, and this is also justified
560 * by considering the behavior of a buddy system containing a single
561 * large block of memory acted on by a series of small allocations.
562 * This behavior is a critical factor in sglist merging's success.
563 *
564 * -- wli
565 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800566static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 int low, int high, struct free_area *area)
568{
569 unsigned long size = 1 << high;
570
571 while (high > low) {
572 area--;
573 high--;
574 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700575 VM_BUG_ON(bad_range(zone, &page[size]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 list_add(&page[size].lru, &area->free_list);
577 area->nr_free++;
578 set_page_order(&page[size], high);
579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580}
581
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582/*
583 * This page is about to be returned from the page allocator
584 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800585static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586{
Nick Piggin92be2e32006-01-06 00:10:57 -0800587 if (unlikely(page_mapcount(page) |
588 (page->mapping != NULL) |
589 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700590 (page->flags & (
591 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 1 << PG_private |
593 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 1 << PG_active |
595 1 << PG_dirty |
596 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700597 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700599 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000600 1 << PG_reserved |
601 1 << PG_buddy ))))
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
612 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 Piggincc102502006-12-06 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
629/*
630 * Do the hard work of removing an element from the buddy allocator.
631 * Call me with the zone->lock already held.
632 */
633static struct page *__rmqueue(struct zone *zone, unsigned int order)
634{
635 struct free_area * area;
636 unsigned int current_order;
637 struct page *page;
638
639 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
640 area = zone->free_area + current_order;
641 if (list_empty(&area->free_list))
642 continue;
643
644 page = list_entry(area->free_list.next, struct page, lru);
645 list_del(&page->lru);
646 rmv_page_order(page);
647 area->nr_free--;
Christoph Lameterd23ad422007-02-10 01:43:02 -0800648 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
Nick Piggin085cc7d2006-01-06 00:11:01 -0800649 expand(zone, page, order, current_order, area);
650 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 }
652
653 return NULL;
654}
655
656/*
657 * Obtain a specified number of elements from the buddy allocator, all under
658 * a single hold of the lock, for efficiency. Add them to the supplied list.
659 * Returns the number of new pages which were placed at *list.
660 */
661static int rmqueue_bulk(struct zone *zone, unsigned int order,
662 unsigned long count, struct list_head *list)
663{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
Nick Pigginc54ad302006-01-06 00:10:56 -0800666 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 for (i = 0; i < count; ++i) {
Nick Piggin085cc7d2006-01-06 00:11:01 -0800668 struct page *page = __rmqueue(zone, order);
669 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 list_add_tail(&page->lru, list);
672 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800673 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800674 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675}
676
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700677#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800678/*
Christoph Lameter4037d452007-05-09 02:35:14 -0700679 * Called from the vmstat counter updater to drain pagesets of this
680 * currently executing processor on remote nodes after they have
681 * expired.
682 *
Christoph Lameter879336c2006-03-22 00:09:08 -0800683 * Note that this function must be called with the thread pinned to
684 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800685 */
Christoph Lameter4037d452007-05-09 02:35:14 -0700686void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700687{
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700688 unsigned long flags;
Christoph Lameter4037d452007-05-09 02:35:14 -0700689 int to_drain;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700690
Christoph Lameter4037d452007-05-09 02:35:14 -0700691 local_irq_save(flags);
692 if (pcp->count >= pcp->batch)
693 to_drain = pcp->batch;
694 else
695 to_drain = pcp->count;
696 free_pages_bulk(zone, to_drain, &pcp->list, 0);
697 pcp->count -= to_drain;
698 local_irq_restore(flags);
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700699}
700#endif
701
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702static void __drain_pages(unsigned int cpu)
703{
Nick Pigginc54ad302006-01-06 00:10:56 -0800704 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 struct zone *zone;
706 int i;
707
708 for_each_zone(zone) {
709 struct per_cpu_pageset *pset;
710
Christoph Lameterf2e12bb2007-01-05 16:37:02 -0800711 if (!populated_zone(zone))
712 continue;
713
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700714 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
716 struct per_cpu_pages *pcp;
717
718 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800719 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800720 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
721 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800722 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 }
724 }
725}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
727#ifdef CONFIG_PM
728
729void mark_free_pages(struct zone *zone)
730{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700731 unsigned long pfn, max_zone_pfn;
732 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 int order;
734 struct list_head *curr;
735
736 if (!zone->spanned_pages)
737 return;
738
739 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700740
741 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
742 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
743 if (pfn_valid(pfn)) {
744 struct page *page = pfn_to_page(pfn);
745
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700746 if (!swsusp_page_is_forbidden(page))
747 swsusp_unset_page_free(page);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700748 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
750 for (order = MAX_ORDER - 1; order >= 0; --order)
751 list_for_each(curr, &zone->free_area[order].free_list) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700752 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700754 pfn = page_to_pfn(list_entry(curr, struct page, lru));
755 for (i = 0; i < (1UL << order); i++)
Rafael J. Wysocki7be98232007-05-06 14:50:42 -0700756 swsusp_set_page_free(pfn_to_page(pfn + i));
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700757 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 spin_unlock_irqrestore(&zone->lock, flags);
760}
761
762/*
763 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
764 */
765void drain_local_pages(void)
766{
767 unsigned long flags;
768
769 local_irq_save(flags);
770 __drain_pages(smp_processor_id());
771 local_irq_restore(flags);
772}
773#endif /* CONFIG_PM */
774
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775/*
776 * Free a 0-order page
777 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778static void fastcall free_hot_cold_page(struct page *page, int cold)
779{
780 struct zone *zone = page_zone(page);
781 struct per_cpu_pages *pcp;
782 unsigned long flags;
783
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 if (PageAnon(page))
785 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800786 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800787 return;
788
Nick Piggin9858db52006-10-11 01:21:30 -0700789 if (!PageHighMem(page))
790 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Nick Piggindafb1362006-10-11 01:21:30 -0700791 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800792 kernel_map_pages(page, 1, 0);
793
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700794 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700796 __count_vm_event(PGFREE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 list_add(&page->lru, &pcp->list);
798 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800799 if (pcp->count >= pcp->high) {
800 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
801 pcp->count -= pcp->batch;
802 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 local_irq_restore(flags);
804 put_cpu();
805}
806
807void fastcall free_hot_page(struct page *page)
808{
809 free_hot_cold_page(page, 0);
810}
811
812void fastcall free_cold_page(struct page *page)
813{
814 free_hot_cold_page(page, 1);
815}
816
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800817/*
818 * split_page takes a non-compound higher-order page, and splits it into
819 * n (1<<order) sub-pages: page[0..n]
820 * Each sub-page must be freed individually.
821 *
822 * Note: this is probably too low level an operation for use in drivers.
823 * Please consult with lkml before using this in your driver.
824 */
825void split_page(struct page *page, unsigned int order)
826{
827 int i;
828
Nick Piggin725d7042006-09-25 23:30:55 -0700829 VM_BUG_ON(PageCompound(page));
830 VM_BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -0800831 for (i = 1; i < (1 << order); i++)
832 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800833}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800834
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835/*
836 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
837 * we cheat by calling it from here, in the order > 0 path. Saves a branch
838 * or two.
839 */
Nick Piggina74609f2006-01-06 00:11:20 -0800840static struct page *buffered_rmqueue(struct zonelist *zonelist,
841 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842{
843 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800844 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800846 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
Hugh Dickins689bceb2005-11-21 21:32:20 -0800848again:
Nick Piggina74609f2006-01-06 00:11:20 -0800849 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800850 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 struct per_cpu_pages *pcp;
852
Nick Piggina74609f2006-01-06 00:11:20 -0800853 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800855 if (!pcp->count) {
nkalmala941c7102006-11-02 22:07:04 -0800856 pcp->count = rmqueue_bulk(zone, 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800858 if (unlikely(!pcp->count))
859 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 }
Nick Piggina74609f2006-01-06 00:11:20 -0800861 page = list_entry(pcp->list.next, struct page, lru);
862 list_del(&page->lru);
863 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800864 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 spin_lock_irqsave(&zone->lock, flags);
866 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800867 spin_unlock(&zone->lock);
868 if (!page)
869 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 }
871
Christoph Lameterf8891e52006-06-30 01:55:45 -0700872 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Christoph Lameterca889e62006-06-30 01:55:44 -0700873 zone_statistics(zonelist, zone);
Nick Piggina74609f2006-01-06 00:11:20 -0800874 local_irq_restore(flags);
875 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876
Nick Piggin725d7042006-09-25 23:30:55 -0700877 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -0800878 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -0800879 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800881
882failed:
883 local_irq_restore(flags);
884 put_cpu();
885 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886}
887
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800888#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800889#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
890#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
891#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
892#define ALLOC_HARDER 0x10 /* try to alloc harder */
893#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
894#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800895
Akinobu Mita933e3122006-12-08 02:39:45 -0800896#ifdef CONFIG_FAIL_PAGE_ALLOC
897
898static struct fail_page_alloc_attr {
899 struct fault_attr attr;
900
901 u32 ignore_gfp_highmem;
902 u32 ignore_gfp_wait;
903
904#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
905
906 struct dentry *ignore_gfp_highmem_file;
907 struct dentry *ignore_gfp_wait_file;
908
909#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
910
911} fail_page_alloc = {
912 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -0800913 .ignore_gfp_wait = 1,
914 .ignore_gfp_highmem = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -0800915};
916
917static int __init setup_fail_page_alloc(char *str)
918{
919 return setup_fault_attr(&fail_page_alloc.attr, str);
920}
921__setup("fail_page_alloc=", setup_fail_page_alloc);
922
923static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
924{
925 if (gfp_mask & __GFP_NOFAIL)
926 return 0;
927 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
928 return 0;
929 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
930 return 0;
931
932 return should_fail(&fail_page_alloc.attr, 1 << order);
933}
934
935#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
936
937static int __init fail_page_alloc_debugfs(void)
938{
939 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
940 struct dentry *dir;
941 int err;
942
943 err = init_fault_attr_dentries(&fail_page_alloc.attr,
944 "fail_page_alloc");
945 if (err)
946 return err;
947 dir = fail_page_alloc.attr.dentries.dir;
948
949 fail_page_alloc.ignore_gfp_wait_file =
950 debugfs_create_bool("ignore-gfp-wait", mode, dir,
951 &fail_page_alloc.ignore_gfp_wait);
952
953 fail_page_alloc.ignore_gfp_highmem_file =
954 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
955 &fail_page_alloc.ignore_gfp_highmem);
956
957 if (!fail_page_alloc.ignore_gfp_wait_file ||
958 !fail_page_alloc.ignore_gfp_highmem_file) {
959 err = -ENOMEM;
960 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
961 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
962 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
963 }
964
965 return err;
966}
967
968late_initcall(fail_page_alloc_debugfs);
969
970#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
971
972#else /* CONFIG_FAIL_PAGE_ALLOC */
973
974static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
975{
976 return 0;
977}
978
979#endif /* CONFIG_FAIL_PAGE_ALLOC */
980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981/*
982 * Return 1 if free pages are above 'mark'. This takes into account the order
983 * of the allocation.
984 */
985int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800986 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987{
988 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -0800989 long min = mark;
990 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 int o;
992
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800993 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800995 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 min -= min / 4;
997
998 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
999 return 0;
1000 for (o = 0; o < order; o++) {
1001 /* At the next order, this order's pages become unavailable */
1002 free_pages -= z->free_area[o].nr_free << o;
1003
1004 /* Require fewer higher order pages to be free */
1005 min >>= 1;
1006
1007 if (free_pages <= min)
1008 return 0;
1009 }
1010 return 1;
1011}
1012
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001013#ifdef CONFIG_NUMA
1014/*
1015 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1016 * skip over zones that are not allowed by the cpuset, or that have
1017 * been recently (in last second) found to be nearly full. See further
1018 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1019 * that have to skip over alot of full or unallowed zones.
1020 *
1021 * If the zonelist cache is present in the passed in zonelist, then
1022 * returns a pointer to the allowed node mask (either the current
1023 * tasks mems_allowed, or node_online_map.)
1024 *
1025 * If the zonelist cache is not available for this zonelist, does
1026 * nothing and returns NULL.
1027 *
1028 * If the fullzones BITMAP in the zonelist cache is stale (more than
1029 * a second since last zap'd) then we zap it out (clear its bits.)
1030 *
1031 * We hold off even calling zlc_setup, until after we've checked the
1032 * first zone in the zonelist, on the theory that most allocations will
1033 * be satisfied from that first zone, so best to examine that zone as
1034 * quickly as we can.
1035 */
1036static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1037{
1038 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1039 nodemask_t *allowednodes; /* zonelist_cache approximation */
1040
1041 zlc = zonelist->zlcache_ptr;
1042 if (!zlc)
1043 return NULL;
1044
1045 if (jiffies - zlc->last_full_zap > 1 * HZ) {
1046 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1047 zlc->last_full_zap = jiffies;
1048 }
1049
1050 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1051 &cpuset_current_mems_allowed :
1052 &node_online_map;
1053 return allowednodes;
1054}
1055
1056/*
1057 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1058 * if it is worth looking at further for free memory:
1059 * 1) Check that the zone isn't thought to be full (doesn't have its
1060 * bit set in the zonelist_cache fullzones BITMAP).
1061 * 2) Check that the zones node (obtained from the zonelist_cache
1062 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1063 * Return true (non-zero) if zone is worth looking at further, or
1064 * else return false (zero) if it is not.
1065 *
1066 * This check -ignores- the distinction between various watermarks,
1067 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1068 * found to be full for any variation of these watermarks, it will
1069 * be considered full for up to one second by all requests, unless
1070 * we are so low on memory on all allowed nodes that we are forced
1071 * into the second scan of the zonelist.
1072 *
1073 * In the second scan we ignore this zonelist cache and exactly
1074 * apply the watermarks to all zones, even it is slower to do so.
1075 * We are low on memory in the second scan, and should leave no stone
1076 * unturned looking for a free page.
1077 */
1078static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1079 nodemask_t *allowednodes)
1080{
1081 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1082 int i; /* index of *z in zonelist zones */
1083 int n; /* node that zone *z is on */
1084
1085 zlc = zonelist->zlcache_ptr;
1086 if (!zlc)
1087 return 1;
1088
1089 i = z - zonelist->zones;
1090 n = zlc->z_to_n[i];
1091
1092 /* This zone is worth trying if it is allowed but not full */
1093 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1094}
1095
1096/*
1097 * Given 'z' scanning a zonelist, set the corresponding bit in
1098 * zlc->fullzones, so that subsequent attempts to allocate a page
1099 * from that zone don't waste time re-examining it.
1100 */
1101static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1102{
1103 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1104 int i; /* index of *z in zonelist zones */
1105
1106 zlc = zonelist->zlcache_ptr;
1107 if (!zlc)
1108 return;
1109
1110 i = z - zonelist->zones;
1111
1112 set_bit(i, zlc->fullzones);
1113}
1114
1115#else /* CONFIG_NUMA */
1116
1117static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1118{
1119 return NULL;
1120}
1121
1122static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1123 nodemask_t *allowednodes)
1124{
1125 return 1;
1126}
1127
1128static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1129{
1130}
1131#endif /* CONFIG_NUMA */
1132
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001133/*
Paul Jackson0798e512006-12-06 20:31:38 -08001134 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001135 * a page.
1136 */
1137static struct page *
1138get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1139 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -07001140{
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001141 struct zone **z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001142 struct page *page = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001143 int classzone_idx = zone_idx(zonelist->zones[0]);
Christoph Lameter1192d522006-09-25 23:31:45 -07001144 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001145 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1146 int zlc_active = 0; /* set if using zonelist_cache */
1147 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001148
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001149zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001150 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001151 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001152 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1153 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001154 z = zonelist->zones;
1155
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001156 do {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001157 if (NUMA_BUILD && zlc_active &&
1158 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1159 continue;
Christoph Lameter1192d522006-09-25 23:31:45 -07001160 zone = *z;
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001161 if (unlikely(NUMA_BUILD && (gfp_mask & __GFP_THISNODE) &&
Christoph Lameter1192d522006-09-25 23:31:45 -07001162 zone->zone_pgdat != zonelist->zones[0]->zone_pgdat))
Christoph Lameter9b819d22006-09-25 23:31:40 -07001163 break;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001164 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001165 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001166 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001167
1168 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001169 unsigned long mark;
1170 if (alloc_flags & ALLOC_WMARK_MIN)
Christoph Lameter1192d522006-09-25 23:31:45 -07001171 mark = zone->pages_min;
Nick Piggin31488902005-11-28 13:44:03 -08001172 else if (alloc_flags & ALLOC_WMARK_LOW)
Christoph Lameter1192d522006-09-25 23:31:45 -07001173 mark = zone->pages_low;
Nick Piggin31488902005-11-28 13:44:03 -08001174 else
Christoph Lameter1192d522006-09-25 23:31:45 -07001175 mark = zone->pages_high;
Paul Jackson0798e512006-12-06 20:31:38 -08001176 if (!zone_watermark_ok(zone, order, mark,
1177 classzone_idx, alloc_flags)) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001178 if (!zone_reclaim_mode ||
Christoph Lameter1192d522006-09-25 23:31:45 -07001179 !zone_reclaim(zone, gfp_mask, order))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001180 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001181 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001182 }
1183
Christoph Lameter1192d522006-09-25 23:31:45 -07001184 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
Paul Jackson0798e512006-12-06 20:31:38 -08001185 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001186 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001187this_zone_full:
1188 if (NUMA_BUILD)
1189 zlc_mark_zone_full(zonelist, z);
1190try_next_zone:
1191 if (NUMA_BUILD && !did_zlc_setup) {
1192 /* we do zlc_setup after the first zone is tried */
1193 allowednodes = zlc_setup(zonelist, alloc_flags);
1194 zlc_active = 1;
1195 did_zlc_setup = 1;
1196 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001197 } while (*(++z) != NULL);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001198
1199 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1200 /* Disable zlc cache for second zonelist scan */
1201 zlc_active = 0;
1202 goto zonelist_scan;
1203 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001204 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001205}
1206
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207/*
1208 * This is the 'heart' of the zoned buddy allocator.
1209 */
1210struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +01001211__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 struct zonelist *zonelist)
1213{
Al Viro260b2362005-10-21 03:22:44 -04001214 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001215 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 struct page *page;
1217 struct reclaim_state reclaim_state;
1218 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001220 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 int did_some_progress;
1222
1223 might_sleep_if(wait);
1224
Akinobu Mita933e3122006-12-08 02:39:45 -08001225 if (should_fail_alloc_page(gfp_mask, order))
1226 return NULL;
1227
Jens Axboe6b1de912005-11-17 21:35:02 +01001228restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001229 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001231 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 /* Should this ever happen?? */
1233 return NULL;
1234 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001235
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001236 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001237 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001238 if (page)
1239 goto got_pg;
1240
Christoph Lameter952f3b52006-12-06 20:33:26 -08001241 /*
1242 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1243 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1244 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1245 * using a larger set of nodes after it has established that the
1246 * allowed per node queues are empty and that nodes are
1247 * over allocated.
1248 */
1249 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1250 goto nopage;
1251
Paul Jackson0798e512006-12-06 20:31:38 -08001252 for (z = zonelist->zones; *z; z++)
Chris Wright43b0bc02006-06-25 05:47:55 -07001253 wakeup_kswapd(*z, order);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001254
Paul Jackson9bf22292005-09-06 15:18:12 -07001255 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001256 * OK, we're below the kswapd watermark and have kicked background
1257 * reclaim. Now things get more complex, so set up alloc_flags according
1258 * to how we want to proceed.
1259 *
1260 * The caller may dip into page reserves a bit more if the caller
1261 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -08001262 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1263 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -07001264 */
Nick Piggin31488902005-11-28 13:44:03 -08001265 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001266 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1267 alloc_flags |= ALLOC_HARDER;
1268 if (gfp_mask & __GFP_HIGH)
1269 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -07001270 if (wait)
1271 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
1273 /*
1274 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001275 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 *
1277 * This is the last chance, in general, before the goto nopage.
1278 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -07001279 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001281 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1282 if (page)
1283 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
1285 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -07001286
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001287rebalance:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001288 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1289 && !in_interrupt()) {
1290 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -08001291nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001292 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001293 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -08001294 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001295 if (page)
1296 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -08001297 if (gfp_mask & __GFP_NOFAIL) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001298 congestion_wait(WRITE, HZ/50);
Kirill Korotaev885036d2005-11-13 16:06:41 -08001299 goto nofail_alloc;
1300 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 }
1302 goto nopage;
1303 }
1304
1305 /* Atomic allocations - we can't balance anything */
1306 if (!wait)
1307 goto nopage;
1308
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 cond_resched();
1310
1311 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001312 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 p->flags |= PF_MEMALLOC;
1314 reclaim_state.reclaimed_slab = 0;
1315 p->reclaim_state = &reclaim_state;
1316
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001317 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318
1319 p->reclaim_state = NULL;
1320 p->flags &= ~PF_MEMALLOC;
1321
1322 cond_resched();
1323
1324 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001325 page = get_page_from_freelist(gfp_mask, order,
1326 zonelist, alloc_flags);
1327 if (page)
1328 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1330 /*
1331 * Go through the zonelist yet one more time, keep
1332 * very high watermark here, this is only to catch
1333 * a parallel oom killing, we must fail if we're still
1334 * under heavy pressure.
1335 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001336 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001337 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001338 if (page)
1339 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001341 out_of_memory(zonelist, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 goto restart;
1343 }
1344
1345 /*
1346 * Don't let big-order allocations loop unless the caller explicitly
1347 * requests that. Wait for some write requests to complete then retry.
1348 *
1349 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1350 * <= 3, but that may not be true in other implementations.
1351 */
1352 do_retry = 0;
1353 if (!(gfp_mask & __GFP_NORETRY)) {
1354 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1355 do_retry = 1;
1356 if (gfp_mask & __GFP_NOFAIL)
1357 do_retry = 1;
1358 }
1359 if (do_retry) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001360 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 goto rebalance;
1362 }
1363
1364nopage:
1365 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1366 printk(KERN_WARNING "%s: page allocation failure."
1367 " order:%d, mode:0x%x\n",
1368 p->comm, order, gfp_mask);
1369 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001370 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 return page;
1374}
1375
1376EXPORT_SYMBOL(__alloc_pages);
1377
1378/*
1379 * Common helper functions.
1380 */
Al Virodd0fc662005-10-07 07:46:04 +01001381fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382{
1383 struct page * page;
1384 page = alloc_pages(gfp_mask, order);
1385 if (!page)
1386 return 0;
1387 return (unsigned long) page_address(page);
1388}
1389
1390EXPORT_SYMBOL(__get_free_pages);
1391
Al Virodd0fc662005-10-07 07:46:04 +01001392fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393{
1394 struct page * page;
1395
1396 /*
1397 * get_zeroed_page() returns a 32-bit address, which cannot represent
1398 * a highmem page
1399 */
Nick Piggin725d7042006-09-25 23:30:55 -07001400 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
1402 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1403 if (page)
1404 return (unsigned long) page_address(page);
1405 return 0;
1406}
1407
1408EXPORT_SYMBOL(get_zeroed_page);
1409
1410void __pagevec_free(struct pagevec *pvec)
1411{
1412 int i = pagevec_count(pvec);
1413
1414 while (--i >= 0)
1415 free_hot_cold_page(pvec->pages[i], pvec->cold);
1416}
1417
1418fastcall void __free_pages(struct page *page, unsigned int order)
1419{
Nick Pigginb5810032005-10-29 18:16:12 -07001420 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 if (order == 0)
1422 free_hot_page(page);
1423 else
1424 __free_pages_ok(page, order);
1425 }
1426}
1427
1428EXPORT_SYMBOL(__free_pages);
1429
1430fastcall void free_pages(unsigned long addr, unsigned int order)
1431{
1432 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001433 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 __free_pages(virt_to_page((void *)addr), order);
1435 }
1436}
1437
1438EXPORT_SYMBOL(free_pages);
1439
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440static unsigned int nr_free_zone_pages(int offset)
1441{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001442 /* Just pick one node, since fallback list is circular */
1443 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 unsigned int sum = 0;
1445
Martin J. Blighe310fd42005-07-29 22:59:18 -07001446 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1447 struct zone **zonep = zonelist->zones;
1448 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
Martin J. Blighe310fd42005-07-29 22:59:18 -07001450 for (zone = *zonep++; zone; zone = *zonep++) {
1451 unsigned long size = zone->present_pages;
1452 unsigned long high = zone->pages_high;
1453 if (size > high)
1454 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 }
1456
1457 return sum;
1458}
1459
1460/*
1461 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1462 */
1463unsigned int nr_free_buffer_pages(void)
1464{
Al Viroaf4ca452005-10-21 02:55:38 -04001465 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466}
1467
1468/*
1469 * Amount of free RAM allocatable within all zones
1470 */
1471unsigned int nr_free_pagecache_pages(void)
1472{
Al Viroaf4ca452005-10-21 02:55:38 -04001473 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001475
1476static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001478 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08001479 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482void si_meminfo(struct sysinfo *val)
1483{
1484 val->totalram = totalram_pages;
1485 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001486 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 val->totalhigh = totalhigh_pages;
1489 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 val->mem_unit = PAGE_SIZE;
1491}
1492
1493EXPORT_SYMBOL(si_meminfo);
1494
1495#ifdef CONFIG_NUMA
1496void si_meminfo_node(struct sysinfo *val, int nid)
1497{
1498 pg_data_t *pgdat = NODE_DATA(nid);
1499
1500 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001501 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001502#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001504 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1505 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001506#else
1507 val->totalhigh = 0;
1508 val->freehigh = 0;
1509#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 val->mem_unit = PAGE_SIZE;
1511}
1512#endif
1513
1514#define K(x) ((x) << (PAGE_SHIFT-10))
1515
1516/*
1517 * Show free area list (used inside shift_scroll-lock stuff)
1518 * We also calculate the percentage fragmentation. We do this by counting the
1519 * memory on each free list with the exception of the first item on the list.
1520 */
1521void show_free_areas(void)
1522{
Jes Sorensenc7241912006-09-27 01:50:05 -07001523 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 struct zone *zone;
1525
1526 for_each_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07001527 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07001529
1530 show_node(zone);
1531 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
Dave Jones6b482c62005-11-10 15:45:56 -05001533 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 struct per_cpu_pageset *pageset;
1535
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001536 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
Jes Sorensenc7241912006-09-27 01:50:05 -07001538 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1539 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1540 cpu, pageset->pcp[0].high,
1541 pageset->pcp[0].batch, pageset->pcp[0].count,
1542 pageset->pcp[1].high, pageset->pcp[1].batch,
1543 pageset->pcp[1].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 }
1545 }
1546
Andrew Mortona25700a2007-02-08 14:20:40 -08001547 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
Christoph Lameterd23ad422007-02-10 01:43:02 -08001548 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
Christoph Lameter65e458d42007-02-10 01:43:05 -08001549 global_page_state(NR_ACTIVE),
1550 global_page_state(NR_INACTIVE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07001551 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07001552 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001553 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08001554 global_page_state(NR_FREE_PAGES),
Christoph Lameter972d1a72006-09-25 23:31:51 -07001555 global_page_state(NR_SLAB_RECLAIMABLE) +
1556 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001557 global_page_state(NR_FILE_MAPPED),
Andrew Mortona25700a2007-02-08 14:20:40 -08001558 global_page_state(NR_PAGETABLE),
1559 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560
1561 for_each_zone(zone) {
1562 int i;
1563
Jes Sorensenc7241912006-09-27 01:50:05 -07001564 if (!populated_zone(zone))
1565 continue;
1566
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 show_node(zone);
1568 printk("%s"
1569 " free:%lukB"
1570 " min:%lukB"
1571 " low:%lukB"
1572 " high:%lukB"
1573 " active:%lukB"
1574 " inactive:%lukB"
1575 " present:%lukB"
1576 " pages_scanned:%lu"
1577 " all_unreclaimable? %s"
1578 "\n",
1579 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08001580 K(zone_page_state(zone, NR_FREE_PAGES)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 K(zone->pages_min),
1582 K(zone->pages_low),
1583 K(zone->pages_high),
Christoph Lameterc8785382007-02-10 01:43:01 -08001584 K(zone_page_state(zone, NR_ACTIVE)),
1585 K(zone_page_state(zone, NR_INACTIVE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 K(zone->present_pages),
1587 zone->pages_scanned,
1588 (zone->all_unreclaimable ? "yes" : "no")
1589 );
1590 printk("lowmem_reserve[]:");
1591 for (i = 0; i < MAX_NR_ZONES; i++)
1592 printk(" %lu", zone->lowmem_reserve[i]);
1593 printk("\n");
1594 }
1595
1596 for_each_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001597 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598
Jes Sorensenc7241912006-09-27 01:50:05 -07001599 if (!populated_zone(zone))
1600 continue;
1601
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 show_node(zone);
1603 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604
1605 spin_lock_irqsave(&zone->lock, flags);
1606 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001607 nr[order] = zone->free_area[order].nr_free;
1608 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 }
1610 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001611 for (order = 0; order < MAX_ORDER; order++)
1612 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 printk("= %lukB\n", K(total));
1614 }
1615
1616 show_swap_cache_info();
1617}
1618
1619/*
1620 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001621 *
1622 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001624static int __meminit build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001625 struct zonelist *zonelist, int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626{
Christoph Lameter1a932052006-01-06 00:11:16 -08001627 struct zone *zone;
1628
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001629 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001630 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001631
1632 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001633 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08001634 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001635 if (populated_zone(zone)) {
Christoph Lameter070f8032006-01-06 00:11:19 -08001636 zonelist->zones[nr_zones++] = zone;
1637 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08001639
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001640 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08001641 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642}
1643
1644#ifdef CONFIG_NUMA
1645#define MAX_NODE_LOAD (num_online_nodes())
Yasunori Goto86356ab2006-06-23 02:03:09 -07001646static int __meminitdata node_load[MAX_NUMNODES];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001648 * 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 -07001649 * @node: node whose fallback list we're appending
1650 * @used_node_mask: nodemask_t of already used nodes
1651 *
1652 * We use a number of factors to determine which is the next node that should
1653 * appear on a given node's fallback list. The node should not have appeared
1654 * already in @node's fallback list, and it should be the next closest node
1655 * according to the distance array (which contains arbitrary distance values
1656 * from each node to each node in the system), and should also prefer nodes
1657 * with no CPUs, since presumably they'll have very little allocation pressure
1658 * on them otherwise.
1659 * It returns -1 if no node is found.
1660 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001661static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001663 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 int min_val = INT_MAX;
1665 int best_node = -1;
1666
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001667 /* Use the local node if we haven't already */
1668 if (!node_isset(node, *used_node_mask)) {
1669 node_set(node, *used_node_mask);
1670 return node;
1671 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001673 for_each_online_node(n) {
1674 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
1676 /* Don't want a node to appear more than once */
1677 if (node_isset(n, *used_node_mask))
1678 continue;
1679
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 /* Use the distance array to find the distance */
1681 val = node_distance(node, n);
1682
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001683 /* Penalize nodes under us ("prefer the next node") */
1684 val += (n < node);
1685
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 /* Give preference to headless and unused nodes */
1687 tmp = node_to_cpumask(n);
1688 if (!cpus_empty(tmp))
1689 val += PENALTY_FOR_NODE_WITH_CPUS;
1690
1691 /* Slight preference for less loaded node */
1692 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1693 val += node_load[n];
1694
1695 if (val < min_val) {
1696 min_val = val;
1697 best_node = n;
1698 }
1699 }
1700
1701 if (best_node >= 0)
1702 node_set(best_node, *used_node_mask);
1703
1704 return best_node;
1705}
1706
Yasunori Goto86356ab2006-06-23 02:03:09 -07001707static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708{
Christoph Lameter19655d32006-09-25 23:31:19 -07001709 int j, node, local_node;
1710 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 int prev_node, load;
1712 struct zonelist *zonelist;
1713 nodemask_t used_mask;
1714
1715 /* initialize zonelists */
Christoph Lameter19655d32006-09-25 23:31:19 -07001716 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 zonelist = pgdat->node_zonelists + i;
1718 zonelist->zones[0] = NULL;
1719 }
1720
1721 /* NUMA-aware ordering of nodes */
1722 local_node = pgdat->node_id;
1723 load = num_online_nodes();
1724 prev_node = local_node;
1725 nodes_clear(used_mask);
1726 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001727 int distance = node_distance(local_node, node);
1728
1729 /*
1730 * If another node is sufficiently far away then it is better
1731 * to reclaim pages in a zone before going off node.
1732 */
1733 if (distance > RECLAIM_DISTANCE)
1734 zone_reclaim_mode = 1;
1735
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 /*
1737 * We don't want to pressure a particular node.
1738 * So adding penalty to the first node in same
1739 * distance group to make it round-robin.
1740 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08001741
1742 if (distance != node_distance(local_node, prev_node))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 node_load[node] += load;
1744 prev_node = node;
1745 load--;
Christoph Lameter19655d32006-09-25 23:31:19 -07001746 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 zonelist = pgdat->node_zonelists + i;
1748 for (j = 0; zonelist->zones[j] != NULL; j++);
1749
Christoph Lameter19655d32006-09-25 23:31:19 -07001750 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 zonelist->zones[j] = NULL;
1752 }
1753 }
1754}
1755
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001756/* Construct the zonelist performance cache - see further mmzone.h */
1757static void __meminit build_zonelist_cache(pg_data_t *pgdat)
1758{
1759 int i;
1760
1761 for (i = 0; i < MAX_NR_ZONES; i++) {
1762 struct zonelist *zonelist;
1763 struct zonelist_cache *zlc;
1764 struct zone **z;
1765
1766 zonelist = pgdat->node_zonelists + i;
1767 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
1768 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1769 for (z = zonelist->zones; *z; z++)
1770 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
1771 }
1772}
1773
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774#else /* CONFIG_NUMA */
1775
Yasunori Goto86356ab2006-06-23 02:03:09 -07001776static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777{
Christoph Lameter19655d32006-09-25 23:31:19 -07001778 int node, local_node;
1779 enum zone_type i,j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780
1781 local_node = pgdat->node_id;
Christoph Lameter19655d32006-09-25 23:31:19 -07001782 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 struct zonelist *zonelist;
1784
1785 zonelist = pgdat->node_zonelists + i;
1786
Christoph Lameter19655d32006-09-25 23:31:19 -07001787 j = build_zonelists_node(pgdat, zonelist, 0, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 /*
1789 * Now we build the zonelist so that it contains the zones
1790 * of all the other nodes.
1791 * We don't want to pressure a particular node, so when
1792 * building the zones for node N, we make sure that the
1793 * zones coming right after the local ones are those from
1794 * node N+1 (modulo N)
1795 */
1796 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1797 if (!node_online(node))
1798 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07001799 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 }
1801 for (node = 0; node < local_node; node++) {
1802 if (!node_online(node))
1803 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07001804 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 }
1806
1807 zonelist->zones[j] = NULL;
1808 }
1809}
1810
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001811/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
1812static void __meminit build_zonelist_cache(pg_data_t *pgdat)
1813{
1814 int i;
1815
1816 for (i = 0; i < MAX_NR_ZONES; i++)
1817 pgdat->node_zonelists[i].zlcache_ptr = NULL;
1818}
1819
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820#endif /* CONFIG_NUMA */
1821
Yasunori Goto68113782006-06-23 02:03:11 -07001822/* return values int ....just for stop_machine_run() */
1823static int __meminit __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824{
Yasunori Goto68113782006-06-23 02:03:11 -07001825 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001826
1827 for_each_online_node(nid) {
Yasunori Goto68113782006-06-23 02:03:11 -07001828 build_zonelists(NODE_DATA(nid));
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001829 build_zonelist_cache(NODE_DATA(nid));
1830 }
Yasunori Goto68113782006-06-23 02:03:11 -07001831 return 0;
1832}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833
Yasunori Goto68113782006-06-23 02:03:11 -07001834void __meminit build_all_zonelists(void)
1835{
1836 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07001837 __build_all_zonelists(NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07001838 cpuset_init_current_mems_allowed();
1839 } else {
1840 /* we have to stop all cpus to guaranntee there is no user
1841 of zonelist */
1842 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
1843 /* cpuset refresh routine should be here */
1844 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07001845 vm_total_pages = nr_free_pagecache_pages();
1846 printk("Built %i zonelists. Total pages: %ld\n",
1847 num_online_nodes(), vm_total_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848}
1849
1850/*
1851 * Helper functions to size the waitqueue hash table.
1852 * Essentially these want to choose hash table sizes sufficiently
1853 * large so that collisions trying to wait on pages are rare.
1854 * But in fact, the number of active page waitqueues on typical
1855 * systems is ridiculously low, less than 200. So this is even
1856 * conservative, even though it seems large.
1857 *
1858 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1859 * waitqueues, i.e. the size of the waitq table given the number of pages.
1860 */
1861#define PAGES_PER_WAITQUEUE 256
1862
Yasunori Gotocca448f2006-06-23 02:03:10 -07001863#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07001864static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865{
1866 unsigned long size = 1;
1867
1868 pages /= PAGES_PER_WAITQUEUE;
1869
1870 while (size < pages)
1871 size <<= 1;
1872
1873 /*
1874 * Once we have dozens or even hundreds of threads sleeping
1875 * on IO we've got bigger problems than wait queue collision.
1876 * Limit the size of the wait table to a reasonable size.
1877 */
1878 size = min(size, 4096UL);
1879
1880 return max(size, 4UL);
1881}
Yasunori Gotocca448f2006-06-23 02:03:10 -07001882#else
1883/*
1884 * A zone's size might be changed by hot-add, so it is not possible to determine
1885 * a suitable size for its wait_table. So we use the maximum size now.
1886 *
1887 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1888 *
1889 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1890 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1891 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1892 *
1893 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1894 * or more by the traditional way. (See above). It equals:
1895 *
1896 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1897 * ia64(16K page size) : = ( 8G + 4M)byte.
1898 * powerpc (64K page size) : = (32G +16M)byte.
1899 */
1900static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1901{
1902 return 4096UL;
1903}
1904#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
1906/*
1907 * This is an integer logarithm so that shifts can be used later
1908 * to extract the more random high bits from the multiplicative
1909 * hash function before the remainder is taken.
1910 */
1911static inline unsigned long wait_table_bits(unsigned long size)
1912{
1913 return ffz(~size);
1914}
1915
1916#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1917
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918/*
1919 * Initially all pages are reserved - free ones are freed
1920 * up by free_all_bootmem() once the early boot process is
1921 * done. Non-atomic initialization, single-pass.
1922 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001923void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa02007-01-10 23:15:30 -08001924 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001927 unsigned long end_pfn = start_pfn + size;
1928 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001930 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa02007-01-10 23:15:30 -08001931 /*
1932 * There can be holes in boot-time mem_map[]s
1933 * handed to this function. They do not
1934 * exist on hotplugged memory.
1935 */
1936 if (context == MEMMAP_EARLY) {
1937 if (!early_pfn_valid(pfn))
1938 continue;
1939 if (!early_pfn_in_nid(pfn, nid))
1940 continue;
1941 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001942 page = pfn_to_page(pfn);
1943 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08001944 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 reset_page_mapcount(page);
1946 SetPageReserved(page);
1947 INIT_LIST_HEAD(&page->lru);
1948#ifdef WANT_PAGE_VIRTUAL
1949 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1950 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001951 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 }
1954}
1955
1956void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1957 unsigned long size)
1958{
1959 int order;
1960 for (order = 0; order < MAX_ORDER ; order++) {
1961 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1962 zone->free_area[order].nr_free = 0;
1963 }
1964}
1965
1966#ifndef __HAVE_ARCH_MEMMAP_INIT
1967#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa02007-01-10 23:15:30 -08001968 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969#endif
1970
Paul Mundtd09c6b82007-06-14 15:13:16 +09001971static int __devinit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001972{
1973 int batch;
1974
1975 /*
1976 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001977 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001978 *
1979 * OK, so we don't know how big the cache is. So guess.
1980 */
1981 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001982 if (batch * PAGE_SIZE > 512 * 1024)
1983 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001984 batch /= 4; /* We effectively *= 4 below */
1985 if (batch < 1)
1986 batch = 1;
1987
1988 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001989 * Clamp the batch to a 2^n - 1 value. Having a power
1990 * of 2 value was found to be more likely to have
1991 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001992 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001993 * For example if 2 tasks are alternately allocating
1994 * batches of pages, one task can end up with a lot
1995 * of pages of one half of the possible page colors
1996 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001997 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001998 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001999
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002000 return batch;
2001}
2002
Christoph Lameter2caaad42005-06-21 17:15:00 -07002003inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2004{
2005 struct per_cpu_pages *pcp;
2006
Magnus Damm1c6fe942005-10-26 01:58:59 -07002007 memset(p, 0, sizeof(*p));
2008
Christoph Lameter2caaad42005-06-21 17:15:00 -07002009 pcp = &p->pcp[0]; /* hot */
2010 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002011 pcp->high = 6 * batch;
2012 pcp->batch = max(1UL, 1 * batch);
2013 INIT_LIST_HEAD(&pcp->list);
2014
2015 pcp = &p->pcp[1]; /* cold*/
2016 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002017 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07002018 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07002019 INIT_LIST_HEAD(&pcp->list);
2020}
2021
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002022/*
2023 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2024 * to the value high for the pageset p.
2025 */
2026
2027static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2028 unsigned long high)
2029{
2030 struct per_cpu_pages *pcp;
2031
2032 pcp = &p->pcp[0]; /* hot list */
2033 pcp->high = high;
2034 pcp->batch = max(1UL, high/4);
2035 if ((high/4) > (PAGE_SHIFT * 8))
2036 pcp->batch = PAGE_SHIFT * 8;
2037}
2038
2039
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002040#ifdef CONFIG_NUMA
2041/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07002042 * Boot pageset table. One per cpu which is going to be used for all
2043 * zones and all nodes. The parameters will be set in such a way
2044 * that an item put on a list will immediately be handed over to
2045 * the buddy list. This is safe since pageset manipulation is done
2046 * with interrupts disabled.
2047 *
2048 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07002049 *
2050 * The boot_pagesets must be kept even after bootup is complete for
2051 * unused processors and/or zones. They do play a role for bootstrapping
2052 * hotplugged processors.
2053 *
2054 * zoneinfo_show() and maybe other functions do
2055 * not check if the processor is online before following the pageset pointer.
2056 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07002057 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08002058static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07002059
2060/*
2061 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002062 * per cpu pageset array in struct zone.
2063 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08002064static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002065{
2066 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002067
2068 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002069
Christoph Lameter66a55032006-09-27 01:50:09 -07002070 if (!populated_zone(zone))
2071 continue;
2072
Nick Piggin23316bc2006-01-08 01:00:41 -08002073 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002074 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08002075 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002076 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002077
Nick Piggin23316bc2006-01-08 01:00:41 -08002078 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002079
2080 if (percpu_pagelist_fraction)
2081 setup_pagelist_highmark(zone_pcp(zone, cpu),
2082 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002083 }
2084
2085 return 0;
2086bad:
2087 for_each_zone(dzone) {
2088 if (dzone == zone)
2089 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08002090 kfree(zone_pcp(dzone, cpu));
2091 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002092 }
2093 return -ENOMEM;
2094}
2095
2096static inline void free_zone_pagesets(int cpu)
2097{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002098 struct zone *zone;
2099
2100 for_each_zone(zone) {
2101 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2102
David Rientjesf3ef9ea2006-09-25 16:24:57 -07002103 /* Free per_cpu_pageset if it is slab allocated */
2104 if (pset != &boot_pageset[cpu])
2105 kfree(pset);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002106 zone_pcp(zone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002107 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002108}
2109
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07002110static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002111 unsigned long action,
2112 void *hcpu)
2113{
2114 int cpu = (long)hcpu;
2115 int ret = NOTIFY_OK;
2116
2117 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08002118 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002119 case CPU_UP_PREPARE_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002120 if (process_zones(cpu))
2121 ret = NOTIFY_BAD;
2122 break;
2123 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002124 case CPU_UP_CANCELED_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002125 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002126 case CPU_DEAD_FROZEN:
Andy Whitcroftce421c72006-12-06 20:33:08 -08002127 free_zone_pagesets(cpu);
2128 break;
2129 default:
2130 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002131 }
2132 return ret;
2133}
2134
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07002135static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002136 { &pageset_cpuup_callback, NULL, 0 };
2137
Al Viro78d99552005-12-15 09:18:25 +00002138void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002139{
2140 int err;
2141
2142 /* Initialize per_cpu_pageset for cpu 0.
2143 * A cpuup callback will do this for every cpu
2144 * as it comes online
2145 */
2146 err = process_zones(smp_processor_id());
2147 BUG_ON(err);
2148 register_cpu_notifier(&pageset_notifier);
2149}
2150
2151#endif
2152
Sam Ravnborg577a32f2007-05-17 23:29:25 +02002153static noinline __init_refok
Yasunori Gotocca448f2006-06-23 02:03:10 -07002154int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07002155{
2156 int i;
2157 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002158 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07002159
2160 /*
2161 * The per-page waitqueue mechanism uses hashed waitqueues
2162 * per zone.
2163 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07002164 zone->wait_table_hash_nr_entries =
2165 wait_table_hash_nr_entries(zone_size_pages);
2166 zone->wait_table_bits =
2167 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002168 alloc_size = zone->wait_table_hash_nr_entries
2169 * sizeof(wait_queue_head_t);
2170
2171 if (system_state == SYSTEM_BOOTING) {
2172 zone->wait_table = (wait_queue_head_t *)
2173 alloc_bootmem_node(pgdat, alloc_size);
2174 } else {
2175 /*
2176 * This case means that a zone whose size was 0 gets new memory
2177 * via memory hot-add.
2178 * But it may be the case that a new node was hot-added. In
2179 * this case vmalloc() will not be able to use this new node's
2180 * memory - this wait_table must be initialized to use this new
2181 * node itself as well.
2182 * To use this new node's memory, further consideration will be
2183 * necessary.
2184 */
2185 zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size);
2186 }
2187 if (!zone->wait_table)
2188 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07002189
Yasunori Goto02b694d2006-06-23 02:03:08 -07002190 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07002191 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002192
2193 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002194}
2195
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002196static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07002197{
2198 int cpu;
2199 unsigned long batch = zone_batchsize(zone);
2200
2201 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2202#ifdef CONFIG_NUMA
2203 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002204 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002205 setup_pageset(&boot_pageset[cpu],0);
2206#else
2207 setup_pageset(zone_pcp(zone,cpu), batch);
2208#endif
2209 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08002210 if (zone->present_pages)
2211 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2212 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07002213}
2214
Yasunori Goto718127c2006-06-23 02:03:10 -07002215__meminit int init_currently_empty_zone(struct zone *zone,
2216 unsigned long zone_start_pfn,
Dave Hansena2f3aa02007-01-10 23:15:30 -08002217 unsigned long size,
2218 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07002219{
2220 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002221 int ret;
2222 ret = zone_wait_table_init(zone, size);
2223 if (ret)
2224 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07002225 pgdat->nr_zones = zone_idx(zone) + 1;
2226
Dave Hansened8ece22005-10-29 18:16:50 -07002227 zone->zone_start_pfn = zone_start_pfn;
2228
2229 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2230
2231 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
Yasunori Goto718127c2006-06-23 02:03:10 -07002232
2233 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002234}
2235
Mel Gormanc7132162006-09-27 01:49:43 -07002236#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2237/*
2238 * Basic iterator support. Return the first range of PFNs for a node
2239 * Note: nid == MAX_NUMNODES returns first region regardless of node
2240 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002241static int __meminit first_active_region_index_in_nid(int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07002242{
2243 int i;
2244
2245 for (i = 0; i < nr_nodemap_entries; i++)
2246 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2247 return i;
2248
2249 return -1;
2250}
2251
2252/*
2253 * Basic iterator support. Return the next active range of PFNs for a node
2254 * Note: nid == MAX_NUMNODES returns next region regardles of node
2255 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002256static int __meminit next_active_region_index_in_nid(int index, int nid)
Mel Gormanc7132162006-09-27 01:49:43 -07002257{
2258 for (index = index + 1; index < nr_nodemap_entries; index++)
2259 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2260 return index;
2261
2262 return -1;
2263}
2264
2265#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2266/*
2267 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2268 * Architectures may implement their own version but if add_active_range()
2269 * was used and there are no special requirements, this is a convenient
2270 * alternative
2271 */
Stephen Rothwell6f076f52007-05-10 03:15:27 -07002272int __meminit early_pfn_to_nid(unsigned long pfn)
Mel Gormanc7132162006-09-27 01:49:43 -07002273{
2274 int i;
2275
2276 for (i = 0; i < nr_nodemap_entries; i++) {
2277 unsigned long start_pfn = early_node_map[i].start_pfn;
2278 unsigned long end_pfn = early_node_map[i].end_pfn;
2279
2280 if (start_pfn <= pfn && pfn < end_pfn)
2281 return early_node_map[i].nid;
2282 }
2283
2284 return 0;
2285}
2286#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2287
2288/* Basic iterator support to walk early_node_map[] */
2289#define for_each_active_range_index_in_nid(i, nid) \
2290 for (i = first_active_region_index_in_nid(nid); i != -1; \
2291 i = next_active_region_index_in_nid(i, nid))
2292
2293/**
2294 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002295 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2296 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07002297 *
2298 * If an architecture guarantees that all ranges registered with
2299 * add_active_ranges() contain no holes and may be freed, this
2300 * this function may be used instead of calling free_bootmem() manually.
2301 */
2302void __init free_bootmem_with_active_regions(int nid,
2303 unsigned long max_low_pfn)
2304{
2305 int i;
2306
2307 for_each_active_range_index_in_nid(i, nid) {
2308 unsigned long size_pages = 0;
2309 unsigned long end_pfn = early_node_map[i].end_pfn;
2310
2311 if (early_node_map[i].start_pfn >= max_low_pfn)
2312 continue;
2313
2314 if (end_pfn > max_low_pfn)
2315 end_pfn = max_low_pfn;
2316
2317 size_pages = end_pfn - early_node_map[i].start_pfn;
2318 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2319 PFN_PHYS(early_node_map[i].start_pfn),
2320 size_pages << PAGE_SHIFT);
2321 }
2322}
2323
2324/**
2325 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002326 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07002327 *
2328 * If an architecture guarantees that all ranges registered with
2329 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002330 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07002331 */
2332void __init sparse_memory_present_with_active_regions(int nid)
2333{
2334 int i;
2335
2336 for_each_active_range_index_in_nid(i, nid)
2337 memory_present(early_node_map[i].nid,
2338 early_node_map[i].start_pfn,
2339 early_node_map[i].end_pfn);
2340}
2341
2342/**
Mel Gormanfb014392006-09-27 01:49:59 -07002343 * push_node_boundaries - Push node boundaries to at least the requested boundary
2344 * @nid: The nid of the node to push the boundary for
2345 * @start_pfn: The start pfn of the node
2346 * @end_pfn: The end pfn of the node
2347 *
2348 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2349 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2350 * be hotplugged even though no physical memory exists. This function allows
2351 * an arch to push out the node boundaries so mem_map is allocated that can
2352 * be used later.
2353 */
2354#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2355void __init push_node_boundaries(unsigned int nid,
2356 unsigned long start_pfn, unsigned long end_pfn)
2357{
2358 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2359 nid, start_pfn, end_pfn);
2360
2361 /* Initialise the boundary for this node if necessary */
2362 if (node_boundary_end_pfn[nid] == 0)
2363 node_boundary_start_pfn[nid] = -1UL;
2364
2365 /* Update the boundaries */
2366 if (node_boundary_start_pfn[nid] > start_pfn)
2367 node_boundary_start_pfn[nid] = start_pfn;
2368 if (node_boundary_end_pfn[nid] < end_pfn)
2369 node_boundary_end_pfn[nid] = end_pfn;
2370}
2371
2372/* If necessary, push the node boundary out for reserve hotadd */
2373static void __init account_node_boundary(unsigned int nid,
2374 unsigned long *start_pfn, unsigned long *end_pfn)
2375{
2376 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2377 nid, *start_pfn, *end_pfn);
2378
2379 /* Return if boundary information has not been provided */
2380 if (node_boundary_end_pfn[nid] == 0)
2381 return;
2382
2383 /* Check the boundaries and update if necessary */
2384 if (node_boundary_start_pfn[nid] < *start_pfn)
2385 *start_pfn = node_boundary_start_pfn[nid];
2386 if (node_boundary_end_pfn[nid] > *end_pfn)
2387 *end_pfn = node_boundary_end_pfn[nid];
2388}
2389#else
2390void __init push_node_boundaries(unsigned int nid,
2391 unsigned long start_pfn, unsigned long end_pfn) {}
2392
2393static void __init account_node_boundary(unsigned int nid,
2394 unsigned long *start_pfn, unsigned long *end_pfn) {}
2395#endif
2396
2397
2398/**
Mel Gormanc7132162006-09-27 01:49:43 -07002399 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002400 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2401 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2402 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07002403 *
2404 * It returns the start and end page frame of a node based on information
2405 * provided by an arch calling add_active_range(). If called for a node
2406 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002407 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07002408 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002409void __meminit get_pfn_range_for_nid(unsigned int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07002410 unsigned long *start_pfn, unsigned long *end_pfn)
2411{
2412 int i;
2413 *start_pfn = -1UL;
2414 *end_pfn = 0;
2415
2416 for_each_active_range_index_in_nid(i, nid) {
2417 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
2418 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
2419 }
2420
2421 if (*start_pfn == -1UL) {
2422 printk(KERN_WARNING "Node %u active with no memory\n", nid);
2423 *start_pfn = 0;
2424 }
Mel Gormanfb014392006-09-27 01:49:59 -07002425
2426 /* Push the node boundaries out if requested */
2427 account_node_boundary(nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07002428}
2429
2430/*
2431 * Return the number of pages a zone spans in a node, including holes
2432 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2433 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002434unsigned long __meminit zone_spanned_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07002435 unsigned long zone_type,
2436 unsigned long *ignored)
2437{
2438 unsigned long node_start_pfn, node_end_pfn;
2439 unsigned long zone_start_pfn, zone_end_pfn;
2440
2441 /* Get the start and end of the node and zone */
2442 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2443 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
2444 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
2445
2446 /* Check that this node has pages within the zone's required range */
2447 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
2448 return 0;
2449
2450 /* Move the zone boundaries inside the node if necessary */
2451 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
2452 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
2453
2454 /* Return the spanned pages */
2455 return zone_end_pfn - zone_start_pfn;
2456}
2457
2458/*
2459 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002460 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07002461 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002462unsigned long __meminit __absent_pages_in_range(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07002463 unsigned long range_start_pfn,
2464 unsigned long range_end_pfn)
2465{
2466 int i = 0;
2467 unsigned long prev_end_pfn = 0, hole_pages = 0;
2468 unsigned long start_pfn;
2469
2470 /* Find the end_pfn of the first active range of pfns in the node */
2471 i = first_active_region_index_in_nid(nid);
2472 if (i == -1)
2473 return 0;
2474
Mel Gorman9c7cd682006-09-27 01:49:58 -07002475 /* Account for ranges before physical memory on this node */
2476 if (early_node_map[i].start_pfn > range_start_pfn)
2477 hole_pages = early_node_map[i].start_pfn - range_start_pfn;
2478
Mel Gormanc7132162006-09-27 01:49:43 -07002479 prev_end_pfn = early_node_map[i].start_pfn;
2480
2481 /* Find all holes for the zone within the node */
2482 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
2483
2484 /* No need to continue if prev_end_pfn is outside the zone */
2485 if (prev_end_pfn >= range_end_pfn)
2486 break;
2487
2488 /* Make sure the end of the zone is not within the hole */
2489 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
2490 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
2491
2492 /* Update the hole size cound and move on */
2493 if (start_pfn > range_start_pfn) {
2494 BUG_ON(prev_end_pfn > start_pfn);
2495 hole_pages += start_pfn - prev_end_pfn;
2496 }
2497 prev_end_pfn = early_node_map[i].end_pfn;
2498 }
2499
Mel Gorman9c7cd682006-09-27 01:49:58 -07002500 /* Account for ranges past physical memory on this node */
2501 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07002502 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07002503 max(range_start_pfn, prev_end_pfn);
2504
Mel Gormanc7132162006-09-27 01:49:43 -07002505 return hole_pages;
2506}
2507
2508/**
2509 * absent_pages_in_range - Return number of page frames in holes within a range
2510 * @start_pfn: The start PFN to start searching for holes
2511 * @end_pfn: The end PFN to stop searching for holes
2512 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002513 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07002514 */
2515unsigned long __init absent_pages_in_range(unsigned long start_pfn,
2516 unsigned long end_pfn)
2517{
2518 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
2519}
2520
2521/* Return the number of page frames in holes in a zone on a node */
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002522unsigned long __meminit zone_absent_pages_in_node(int nid,
Mel Gormanc7132162006-09-27 01:49:43 -07002523 unsigned long zone_type,
2524 unsigned long *ignored)
2525{
Mel Gorman9c7cd682006-09-27 01:49:58 -07002526 unsigned long node_start_pfn, node_end_pfn;
2527 unsigned long zone_start_pfn, zone_end_pfn;
2528
2529 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2530 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
2531 node_start_pfn);
2532 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
2533 node_end_pfn);
2534
2535 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07002536}
Mel Gorman0e0b8642006-09-27 01:49:56 -07002537
Mel Gormanc7132162006-09-27 01:49:43 -07002538#else
2539static inline unsigned long zone_spanned_pages_in_node(int nid,
2540 unsigned long zone_type,
2541 unsigned long *zones_size)
2542{
2543 return zones_size[zone_type];
2544}
2545
2546static inline unsigned long zone_absent_pages_in_node(int nid,
2547 unsigned long zone_type,
2548 unsigned long *zholes_size)
2549{
2550 if (!zholes_size)
2551 return 0;
2552
2553 return zholes_size[zone_type];
2554}
Mel Gorman0e0b8642006-09-27 01:49:56 -07002555
Mel Gormanc7132162006-09-27 01:49:43 -07002556#endif
2557
Yasunori Gotoa3142c82007-05-08 00:23:07 -07002558static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
Mel Gormanc7132162006-09-27 01:49:43 -07002559 unsigned long *zones_size, unsigned long *zholes_size)
2560{
2561 unsigned long realtotalpages, totalpages = 0;
2562 enum zone_type i;
2563
2564 for (i = 0; i < MAX_NR_ZONES; i++)
2565 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
2566 zones_size);
2567 pgdat->node_spanned_pages = totalpages;
2568
2569 realtotalpages = totalpages;
2570 for (i = 0; i < MAX_NR_ZONES; i++)
2571 realtotalpages -=
2572 zone_absent_pages_in_node(pgdat->node_id, i,
2573 zholes_size);
2574 pgdat->node_present_pages = realtotalpages;
2575 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
2576 realtotalpages);
2577}
2578
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579/*
2580 * Set up the zone data structures:
2581 * - mark all pages reserved
2582 * - mark all memory queues empty
2583 * - clear the memory bitmaps
2584 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07002585static void __meminit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 unsigned long *zones_size, unsigned long *zholes_size)
2587{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07002588 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07002589 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07002591 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592
Dave Hansen208d54e2005-10-29 18:16:52 -07002593 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 pgdat->nr_zones = 0;
2595 init_waitqueue_head(&pgdat->kswapd_wait);
2596 pgdat->kswapd_max_order = 0;
2597
2598 for (j = 0; j < MAX_NR_ZONES; j++) {
2599 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07002600 unsigned long size, realsize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601
Mel Gormanc7132162006-09-27 01:49:43 -07002602 size = zone_spanned_pages_in_node(nid, j, zones_size);
2603 realsize = size - zone_absent_pages_in_node(nid, j,
2604 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605
Mel Gorman0e0b8642006-09-27 01:49:56 -07002606 /*
2607 * Adjust realsize so that it accounts for how much memory
2608 * is used by this zone for memmap. This affects the watermark
2609 * and per-cpu initialisations
2610 */
2611 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
2612 if (realsize >= memmap_pages) {
2613 realsize -= memmap_pages;
2614 printk(KERN_DEBUG
2615 " %s zone: %lu pages used for memmap\n",
2616 zone_names[j], memmap_pages);
2617 } else
2618 printk(KERN_WARNING
2619 " %s zone: %lu pages exceeds realsize %lu\n",
2620 zone_names[j], memmap_pages, realsize);
2621
Christoph Lameter62672762007-02-10 01:43:07 -08002622 /* Account for reserved pages */
2623 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07002624 realsize -= dma_reserve;
Christoph Lameter62672762007-02-10 01:43:07 -08002625 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
2626 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07002627 }
2628
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002629 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 nr_kernel_pages += realsize;
2631 nr_all_pages += realsize;
2632
2633 zone->spanned_pages = size;
2634 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07002635#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07002636 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07002637 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07002638 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07002639 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07002640#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 zone->name = zone_names[j];
2642 spin_lock_init(&zone->lock);
2643 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002644 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646
Martin Bligh3bb1a852006-10-28 10:38:24 -07002647 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648
Dave Hansened8ece22005-10-29 18:16:50 -07002649 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 INIT_LIST_HEAD(&zone->active_list);
2651 INIT_LIST_HEAD(&zone->inactive_list);
2652 zone->nr_scan_active = 0;
2653 zone->nr_scan_inactive = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07002654 zap_zone_vm_stats(zone);
Martin Hicks53e9a612005-09-03 15:54:51 -07002655 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 if (!size)
2657 continue;
2658
Dave Hansena2f3aa02007-01-10 23:15:30 -08002659 ret = init_currently_empty_zone(zone, zone_start_pfn,
2660 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07002661 BUG_ON(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 }
2664}
2665
Sam Ravnborg577a32f2007-05-17 23:29:25 +02002666static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 /* Skip empty nodes */
2669 if (!pgdat->node_spanned_pages)
2670 return;
2671
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002672#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 /* ia64 gets its own node_mem_map, before this, without bootmem */
2674 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07002675 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002676 struct page *map;
2677
Bob Piccoe984bb42006-05-20 15:00:31 -07002678 /*
2679 * The zone's endpoints aren't required to be MAX_ORDER
2680 * aligned but the node_mem_map endpoints must be in order
2681 * for the buddy allocator to function correctly.
2682 */
2683 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
2684 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
2685 end = ALIGN(end, MAX_ORDER_NR_PAGES);
2686 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002687 map = alloc_remap(pgdat->node_id, size);
2688 if (!map)
2689 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07002690 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 }
Roman Zippel12d810c2007-05-31 00:40:54 -07002692#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 /*
2694 * With no DISCONTIG, the global mem_map is just set as node 0's
2695 */
Mel Gormanc7132162006-09-27 01:49:43 -07002696 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07002698#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2699 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
2700 mem_map -= pgdat->node_start_pfn;
2701#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2702 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002704#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705}
2706
Yasunori Goto86356ab2006-06-23 02:03:09 -07002707void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 unsigned long *zones_size, unsigned long node_start_pfn,
2709 unsigned long *zholes_size)
2710{
2711 pgdat->node_id = nid;
2712 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07002713 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
2715 alloc_node_mem_map(pgdat);
2716
2717 free_area_init_core(pgdat, zones_size, zholes_size);
2718}
2719
Mel Gormanc7132162006-09-27 01:49:43 -07002720#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Miklos Szeredi418508c2007-05-23 13:57:55 -07002721
2722#if MAX_NUMNODES > 1
2723/*
2724 * Figure out the number of possible node ids.
2725 */
2726static void __init setup_nr_node_ids(void)
2727{
2728 unsigned int node;
2729 unsigned int highest = 0;
2730
2731 for_each_node_mask(node, node_possible_map)
2732 highest = node;
2733 nr_node_ids = highest + 1;
2734}
2735#else
2736static inline void setup_nr_node_ids(void)
2737{
2738}
2739#endif
2740
Mel Gormanc7132162006-09-27 01:49:43 -07002741/**
2742 * add_active_range - Register a range of PFNs backed by physical memory
2743 * @nid: The node ID the range resides on
2744 * @start_pfn: The start PFN of the available physical memory
2745 * @end_pfn: The end PFN of the available physical memory
2746 *
2747 * These ranges are stored in an early_node_map[] and later used by
2748 * free_area_init_nodes() to calculate zone sizes and holes. If the
2749 * range spans a memory hole, it is up to the architecture to ensure
2750 * the memory is not freed by the bootmem allocator. If possible
2751 * the range being registered will be merged with existing ranges.
2752 */
2753void __init add_active_range(unsigned int nid, unsigned long start_pfn,
2754 unsigned long end_pfn)
2755{
2756 int i;
2757
2758 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
2759 "%d entries of %d used\n",
2760 nid, start_pfn, end_pfn,
2761 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
2762
2763 /* Merge with existing active regions if possible */
2764 for (i = 0; i < nr_nodemap_entries; i++) {
2765 if (early_node_map[i].nid != nid)
2766 continue;
2767
2768 /* Skip if an existing region covers this new one */
2769 if (start_pfn >= early_node_map[i].start_pfn &&
2770 end_pfn <= early_node_map[i].end_pfn)
2771 return;
2772
2773 /* Merge forward if suitable */
2774 if (start_pfn <= early_node_map[i].end_pfn &&
2775 end_pfn > early_node_map[i].end_pfn) {
2776 early_node_map[i].end_pfn = end_pfn;
2777 return;
2778 }
2779
2780 /* Merge backward if suitable */
2781 if (start_pfn < early_node_map[i].end_pfn &&
2782 end_pfn >= early_node_map[i].start_pfn) {
2783 early_node_map[i].start_pfn = start_pfn;
2784 return;
2785 }
2786 }
2787
2788 /* Check that early_node_map is large enough */
2789 if (i >= MAX_ACTIVE_REGIONS) {
2790 printk(KERN_CRIT "More than %d memory regions, truncating\n",
2791 MAX_ACTIVE_REGIONS);
2792 return;
2793 }
2794
2795 early_node_map[i].nid = nid;
2796 early_node_map[i].start_pfn = start_pfn;
2797 early_node_map[i].end_pfn = end_pfn;
2798 nr_nodemap_entries = i + 1;
2799}
2800
2801/**
2802 * shrink_active_range - Shrink an existing registered range of PFNs
2803 * @nid: The node id the range is on that should be shrunk
2804 * @old_end_pfn: The old end PFN of the range
2805 * @new_end_pfn: The new PFN of the range
2806 *
2807 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
2808 * The map is kept at the end physical page range that has already been
2809 * registered with add_active_range(). This function allows an arch to shrink
2810 * an existing registered range.
2811 */
2812void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
2813 unsigned long new_end_pfn)
2814{
2815 int i;
2816
2817 /* Find the old active region end and shrink */
2818 for_each_active_range_index_in_nid(i, nid)
2819 if (early_node_map[i].end_pfn == old_end_pfn) {
2820 early_node_map[i].end_pfn = new_end_pfn;
2821 break;
2822 }
2823}
2824
2825/**
2826 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002827 *
Mel Gormanc7132162006-09-27 01:49:43 -07002828 * During discovery, it may be found that a table like SRAT is invalid
2829 * and an alternative discovery method must be used. This function removes
2830 * all currently registered regions.
2831 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002832void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07002833{
2834 memset(early_node_map, 0, sizeof(early_node_map));
2835 nr_nodemap_entries = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07002836#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2837 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
2838 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
2839#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -07002840}
2841
2842/* Compare two active node_active_regions */
2843static int __init cmp_node_active_region(const void *a, const void *b)
2844{
2845 struct node_active_region *arange = (struct node_active_region *)a;
2846 struct node_active_region *brange = (struct node_active_region *)b;
2847
2848 /* Done this way to avoid overflows */
2849 if (arange->start_pfn > brange->start_pfn)
2850 return 1;
2851 if (arange->start_pfn < brange->start_pfn)
2852 return -1;
2853
2854 return 0;
2855}
2856
2857/* sort the node_map by start_pfn */
2858static void __init sort_node_map(void)
2859{
2860 sort(early_node_map, (size_t)nr_nodemap_entries,
2861 sizeof(struct node_active_region),
2862 cmp_node_active_region, NULL);
2863}
2864
Mel Gormana6af2bc2007-02-10 01:42:57 -08002865/* Find the lowest pfn for a node */
Mel Gormanc7132162006-09-27 01:49:43 -07002866unsigned long __init find_min_pfn_for_node(unsigned long nid)
2867{
2868 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08002869 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00002870
Mel Gormanc7132162006-09-27 01:49:43 -07002871 /* Assuming a sorted map, the first range found has the starting pfn */
2872 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08002873 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07002874
Mel Gormana6af2bc2007-02-10 01:42:57 -08002875 if (min_pfn == ULONG_MAX) {
2876 printk(KERN_WARNING
2877 "Could not find start_pfn for node %lu\n", nid);
2878 return 0;
2879 }
2880
2881 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07002882}
2883
2884/**
2885 * find_min_pfn_with_active_regions - Find the minimum PFN registered
2886 *
2887 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002888 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07002889 */
2890unsigned long __init find_min_pfn_with_active_regions(void)
2891{
2892 return find_min_pfn_for_node(MAX_NUMNODES);
2893}
2894
2895/**
2896 * find_max_pfn_with_active_regions - Find the maximum PFN registered
2897 *
2898 * It returns the maximum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002899 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07002900 */
2901unsigned long __init find_max_pfn_with_active_regions(void)
2902{
2903 int i;
2904 unsigned long max_pfn = 0;
2905
2906 for (i = 0; i < nr_nodemap_entries; i++)
2907 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
2908
2909 return max_pfn;
2910}
2911
2912/**
2913 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002914 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07002915 *
2916 * This will call free_area_init_node() for each active node in the system.
2917 * Using the page ranges provided by add_active_range(), the size of each
2918 * zone in each node and their holes is calculated. If the maximum PFN
2919 * between two adjacent zones match, it is assumed that the zone is empty.
2920 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
2921 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
2922 * starts where the previous one ended. For example, ZONE_DMA32 starts
2923 * at arch_max_dma_pfn.
2924 */
2925void __init free_area_init_nodes(unsigned long *max_zone_pfn)
2926{
2927 unsigned long nid;
2928 enum zone_type i;
2929
Mel Gormana6af2bc2007-02-10 01:42:57 -08002930 /* Sort early_node_map as initialisation assumes it is sorted */
2931 sort_node_map();
2932
Mel Gormanc7132162006-09-27 01:49:43 -07002933 /* Record where the zone boundaries are */
2934 memset(arch_zone_lowest_possible_pfn, 0,
2935 sizeof(arch_zone_lowest_possible_pfn));
2936 memset(arch_zone_highest_possible_pfn, 0,
2937 sizeof(arch_zone_highest_possible_pfn));
2938 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
2939 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
2940 for (i = 1; i < MAX_NR_ZONES; i++) {
2941 arch_zone_lowest_possible_pfn[i] =
2942 arch_zone_highest_possible_pfn[i-1];
2943 arch_zone_highest_possible_pfn[i] =
2944 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
2945 }
2946
Mel Gormanc7132162006-09-27 01:49:43 -07002947 /* Print out the zone ranges */
2948 printk("Zone PFN ranges:\n");
2949 for (i = 0; i < MAX_NR_ZONES; i++)
2950 printk(" %-8s %8lu -> %8lu\n",
2951 zone_names[i],
2952 arch_zone_lowest_possible_pfn[i],
2953 arch_zone_highest_possible_pfn[i]);
2954
2955 /* Print out the early_node_map[] */
2956 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
2957 for (i = 0; i < nr_nodemap_entries; i++)
2958 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
2959 early_node_map[i].start_pfn,
2960 early_node_map[i].end_pfn);
2961
2962 /* Initialise every node */
Christoph Lameter8ef82862007-02-20 13:57:52 -08002963 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07002964 for_each_online_node(nid) {
2965 pg_data_t *pgdat = NODE_DATA(nid);
2966 free_area_init_node(nid, pgdat, NULL,
2967 find_min_pfn_for_node(nid), NULL);
2968 }
2969}
2970#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2971
Mel Gorman0e0b8642006-09-27 01:49:56 -07002972/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002973 * set_dma_reserve - set the specified number of pages reserved in the first zone
2974 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07002975 *
2976 * The per-cpu batchsize and zone watermarks are determined by present_pages.
2977 * In the DMA zone, a significant percentage may be consumed by kernel image
2978 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002979 * function may optionally be used to account for unfreeable pages in the
2980 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
2981 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07002982 */
2983void __init set_dma_reserve(unsigned long new_dma_reserve)
2984{
2985 dma_reserve = new_dma_reserve;
2986}
2987
Dave Hansen93b75042005-06-23 00:07:47 -07002988#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989static bootmem_data_t contig_bootmem_data;
2990struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2991
2992EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002993#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994
2995void __init free_area_init(unsigned long *zones_size)
2996{
Dave Hansen93b75042005-06-23 00:07:47 -07002997 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2999}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001static int page_alloc_cpu_notify(struct notifier_block *self,
3002 unsigned long action, void *hcpu)
3003{
3004 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003006 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 local_irq_disable();
3008 __drain_pages(cpu);
Christoph Lameterf8891e52006-06-30 01:55:45 -07003009 vm_events_fold_cpu(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 local_irq_enable();
Christoph Lameter2244b952006-06-30 01:55:33 -07003011 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012 }
3013 return NOTIFY_OK;
3014}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015
3016void __init page_alloc_init(void)
3017{
3018 hotcpu_notifier(page_alloc_cpu_notify, 0);
3019}
3020
3021/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003022 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3023 * or min_free_kbytes changes.
3024 */
3025static void calculate_totalreserve_pages(void)
3026{
3027 struct pglist_data *pgdat;
3028 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003029 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003030
3031 for_each_online_pgdat(pgdat) {
3032 for (i = 0; i < MAX_NR_ZONES; i++) {
3033 struct zone *zone = pgdat->node_zones + i;
3034 unsigned long max = 0;
3035
3036 /* Find valid and maximum lowmem_reserve in the zone */
3037 for (j = i; j < MAX_NR_ZONES; j++) {
3038 if (zone->lowmem_reserve[j] > max)
3039 max = zone->lowmem_reserve[j];
3040 }
3041
3042 /* we treat pages_high as reserved pages. */
3043 max += zone->pages_high;
3044
3045 if (max > zone->present_pages)
3046 max = zone->present_pages;
3047 reserve_pages += max;
3048 }
3049 }
3050 totalreserve_pages = reserve_pages;
3051}
3052
3053/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054 * setup_per_zone_lowmem_reserve - called whenever
3055 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
3056 * has a correct pages reserved value, so an adequate number of
3057 * pages are left in the zone after a successful __alloc_pages().
3058 */
3059static void setup_per_zone_lowmem_reserve(void)
3060{
3061 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003062 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08003064 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 for (j = 0; j < MAX_NR_ZONES; j++) {
3066 struct zone *zone = pgdat->node_zones + j;
3067 unsigned long present_pages = zone->present_pages;
3068
3069 zone->lowmem_reserve[j] = 0;
3070
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003071 idx = j;
3072 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073 struct zone *lower_zone;
3074
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003075 idx--;
3076
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077 if (sysctl_lowmem_reserve_ratio[idx] < 1)
3078 sysctl_lowmem_reserve_ratio[idx] = 1;
3079
3080 lower_zone = pgdat->node_zones + idx;
3081 lower_zone->lowmem_reserve[j] = present_pages /
3082 sysctl_lowmem_reserve_ratio[idx];
3083 present_pages += lower_zone->present_pages;
3084 }
3085 }
3086 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003087
3088 /* update totalreserve_pages */
3089 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090}
3091
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003092/**
3093 * setup_per_zone_pages_min - called when min_free_kbytes changes.
3094 *
3095 * Ensures that the pages_{min,low,high} values for each zone are set correctly
3096 * with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 */
Dave Hansen3947be12005-10-29 18:16:54 -07003098void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099{
3100 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
3101 unsigned long lowmem_pages = 0;
3102 struct zone *zone;
3103 unsigned long flags;
3104
3105 /* Calculate total number of !ZONE_HIGHMEM pages */
3106 for_each_zone(zone) {
3107 if (!is_highmem(zone))
3108 lowmem_pages += zone->present_pages;
3109 }
3110
3111 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07003112 u64 tmp;
3113
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07003115 tmp = (u64)pages_min * zone->present_pages;
3116 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117 if (is_highmem(zone)) {
3118 /*
Nick Piggin669ed172005-11-13 16:06:45 -08003119 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
3120 * need highmem pages, so cap pages_min to a small
3121 * value here.
3122 *
3123 * The (pages_high-pages_low) and (pages_low-pages_min)
3124 * deltas controls asynch page reclaim, and so should
3125 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 */
3127 int min_pages;
3128
3129 min_pages = zone->present_pages / 1024;
3130 if (min_pages < SWAP_CLUSTER_MAX)
3131 min_pages = SWAP_CLUSTER_MAX;
3132 if (min_pages > 128)
3133 min_pages = 128;
3134 zone->pages_min = min_pages;
3135 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08003136 /*
3137 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07003138 * proportionate to the zone's size.
3139 */
Nick Piggin669ed172005-11-13 16:06:45 -08003140 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141 }
3142
Andrew Mortonac924c62006-05-15 09:43:59 -07003143 zone->pages_low = zone->pages_min + (tmp >> 2);
3144 zone->pages_high = zone->pages_min + (tmp >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145 spin_unlock_irqrestore(&zone->lru_lock, flags);
3146 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003147
3148 /* update totalreserve_pages */
3149 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150}
3151
3152/*
3153 * Initialise min_free_kbytes.
3154 *
3155 * For small machines we want it small (128k min). For large machines
3156 * we want it large (64MB max). But it is not linear, because network
3157 * bandwidth does not increase linearly with machine size. We use
3158 *
3159 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
3160 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
3161 *
3162 * which yields
3163 *
3164 * 16MB: 512k
3165 * 32MB: 724k
3166 * 64MB: 1024k
3167 * 128MB: 1448k
3168 * 256MB: 2048k
3169 * 512MB: 2896k
3170 * 1024MB: 4096k
3171 * 2048MB: 5792k
3172 * 4096MB: 8192k
3173 * 8192MB: 11584k
3174 * 16384MB: 16384k
3175 */
3176static int __init init_per_zone_pages_min(void)
3177{
3178 unsigned long lowmem_kbytes;
3179
3180 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
3181
3182 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
3183 if (min_free_kbytes < 128)
3184 min_free_kbytes = 128;
3185 if (min_free_kbytes > 65536)
3186 min_free_kbytes = 65536;
3187 setup_per_zone_pages_min();
3188 setup_per_zone_lowmem_reserve();
3189 return 0;
3190}
3191module_init(init_per_zone_pages_min)
3192
3193/*
3194 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
3195 * that we can call two helper functions whenever min_free_kbytes
3196 * changes.
3197 */
3198int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
3199 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3200{
3201 proc_dointvec(table, write, file, buffer, length, ppos);
Mel Gorman3b1d92c2007-05-06 14:49:30 -07003202 if (write)
3203 setup_per_zone_pages_min();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204 return 0;
3205}
3206
Christoph Lameter96146342006-07-03 00:24:13 -07003207#ifdef CONFIG_NUMA
3208int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
3209 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3210{
3211 struct zone *zone;
3212 int rc;
3213
3214 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3215 if (rc)
3216 return rc;
3217
3218 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07003219 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07003220 sysctl_min_unmapped_ratio) / 100;
3221 return 0;
3222}
Christoph Lameter0ff38492006-09-25 23:31:52 -07003223
3224int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
3225 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3226{
3227 struct zone *zone;
3228 int rc;
3229
3230 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3231 if (rc)
3232 return rc;
3233
3234 for_each_zone(zone)
3235 zone->min_slab_pages = (zone->present_pages *
3236 sysctl_min_slab_ratio) / 100;
3237 return 0;
3238}
Christoph Lameter96146342006-07-03 00:24:13 -07003239#endif
3240
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241/*
3242 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
3243 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
3244 * whenever sysctl_lowmem_reserve_ratio changes.
3245 *
3246 * The reserve ratio obviously has absolutely no relation with the
3247 * pages_min watermarks. The lowmem reserve ratio can only make sense
3248 * if in function of the boot time zone sizes.
3249 */
3250int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
3251 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3252{
3253 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3254 setup_per_zone_lowmem_reserve();
3255 return 0;
3256}
3257
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003258/*
3259 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
3260 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
3261 * can have before it gets flushed back to buddy allocator.
3262 */
3263
3264int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
3265 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3266{
3267 struct zone *zone;
3268 unsigned int cpu;
3269 int ret;
3270
3271 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3272 if (!write || (ret == -EINVAL))
3273 return ret;
3274 for_each_zone(zone) {
3275 for_each_online_cpu(cpu) {
3276 unsigned long high;
3277 high = zone->present_pages / percpu_pagelist_fraction;
3278 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
3279 }
3280 }
3281 return 0;
3282}
3283
David S. Millerf034b5d2006-08-24 03:08:07 -07003284int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285
3286#ifdef CONFIG_NUMA
3287static int __init set_hashdist(char *str)
3288{
3289 if (!str)
3290 return 0;
3291 hashdist = simple_strtoul(str, &str, 0);
3292 return 1;
3293}
3294__setup("hashdist=", set_hashdist);
3295#endif
3296
3297/*
3298 * allocate a large system hash table from bootmem
3299 * - it is assumed that the hash table must contain an exact power-of-2
3300 * quantity of entries
3301 * - limit is the number of hash buckets, not the total allocation size
3302 */
3303void *__init alloc_large_system_hash(const char *tablename,
3304 unsigned long bucketsize,
3305 unsigned long numentries,
3306 int scale,
3307 int flags,
3308 unsigned int *_hash_shift,
3309 unsigned int *_hash_mask,
3310 unsigned long limit)
3311{
3312 unsigned long long max = limit;
3313 unsigned long log2qty, size;
3314 void *table = NULL;
3315
3316 /* allow the kernel cmdline to have a say */
3317 if (!numentries) {
3318 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08003319 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
3321 numentries >>= 20 - PAGE_SHIFT;
3322 numentries <<= 20 - PAGE_SHIFT;
3323
3324 /* limit to 1 bucket per 2^scale bytes of low memory */
3325 if (scale > PAGE_SHIFT)
3326 numentries >>= (scale - PAGE_SHIFT);
3327 else
3328 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08003329
3330 /* Make sure we've got at least a 0-order allocation.. */
3331 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
3332 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333 }
John Hawkes6e692ed2006-03-25 03:08:02 -08003334 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335
3336 /* limit allocation size to 1/16 total memory by default */
3337 if (max == 0) {
3338 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
3339 do_div(max, bucketsize);
3340 }
3341
3342 if (numentries > max)
3343 numentries = max;
3344
David Howellsf0d1b0b2006-12-08 02:37:49 -08003345 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346
3347 do {
3348 size = bucketsize << log2qty;
3349 if (flags & HASH_EARLY)
3350 table = alloc_bootmem(size);
3351 else if (hashdist)
3352 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
3353 else {
3354 unsigned long order;
3355 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
3356 ;
3357 table = (void*) __get_free_pages(GFP_ATOMIC, order);
3358 }
3359 } while (!table && size > PAGE_SIZE && --log2qty);
3360
3361 if (!table)
3362 panic("Failed to allocate %s hash table\n", tablename);
3363
3364 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
3365 tablename,
3366 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08003367 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368 size);
3369
3370 if (_hash_shift)
3371 *_hash_shift = log2qty;
3372 if (_hash_mask)
3373 *_hash_mask = (1 << log2qty) - 1;
3374
3375 return table;
3376}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003377
3378#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003379struct page *pfn_to_page(unsigned long pfn)
3380{
Andy Whitcroft67de6482006-06-23 02:03:12 -07003381 return __pfn_to_page(pfn);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003382}
3383unsigned long page_to_pfn(struct page *page)
3384{
Andy Whitcroft67de6482006-06-23 02:03:12 -07003385 return __page_to_pfn(page);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003386}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003387EXPORT_SYMBOL(pfn_to_page);
3388EXPORT_SYMBOL(page_to_pfn);
3389#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
Andrew Morton6220ec72006-10-19 23:29:05 -07003390
Andrew Morton6220ec72006-10-19 23:29:05 -07003391