blob: d461b23a27a1176decc9229edf45c8ae1c388472 [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;
Mel Gorman0e0b8642006-09-27 01:49:56 -0700106static unsigned long __initdata 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
129 struct node_active_region __initdata early_node_map[MAX_ACTIVE_REGIONS];
130 int __initdata nr_nodemap_entries;
131 unsigned long __initdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
132 unsigned long __initdata 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
Nick Piggin13e74442006-01-06 00:10:58 -0800139#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700140static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700142 int ret = 0;
143 unsigned seq;
144 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700145
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700146 do {
147 seq = zone_span_seqbegin(zone);
148 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
149 ret = 1;
150 else if (pfn < zone->zone_start_pfn)
151 ret = 1;
152 } while (zone_span_seqretry(zone, seq));
153
154 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700155}
156
157static int page_is_consistent(struct zone *zone, struct page *page)
158{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159#ifdef CONFIG_HOLES_IN_ZONE
160 if (!pfn_valid(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700161 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162#endif
163 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700164 return 0;
165
166 return 1;
167}
168/*
169 * Temporary debugging check for pages not lying within a given zone.
170 */
171static int bad_range(struct zone *zone, struct page *page)
172{
173 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700175 if (!page_is_consistent(zone, page))
176 return 1;
177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 return 0;
179}
Nick Piggin13e74442006-01-06 00:10:58 -0800180#else
181static inline int bad_range(struct zone *zone, struct page *page)
182{
183 return 0;
184}
185#endif
186
Nick Piggin224abf92006-01-06 00:11:11 -0800187static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
Nick Piggin224abf92006-01-06 00:11:11 -0800189 printk(KERN_EMERG "Bad page state in process '%s'\n"
Hugh Dickins7365f3d2006-01-11 12:17:18 -0800190 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
191 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
192 KERN_EMERG "Backtrace:\n",
Nick Piggin224abf92006-01-06 00:11:11 -0800193 current->comm, page, (int)(2*sizeof(unsigned long)),
194 (unsigned long)page->flags, page->mapping,
195 page_mapcount(page), page_count(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 dump_stack();
Hugh Dickins334795e2005-06-21 17:15:08 -0700197 page->flags &= ~(1 << PG_lru |
198 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 1 << PG_active |
201 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700202 1 << PG_reclaim |
203 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 1 << PG_swapcache |
Nick Piggin676165a2006-04-10 11:21:48 +1000205 1 << PG_writeback |
206 1 << PG_buddy );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 set_page_count(page, 0);
208 reset_page_mapcount(page);
209 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700210 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211}
212
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213/*
214 * Higher-order pages are called "compound pages". They are structured thusly:
215 *
216 * The first PAGE_SIZE page is called the "head page".
217 *
218 * The remaining PAGE_SIZE pages are called "tail pages".
219 *
220 * All pages have PG_compound set. All pages have their ->private pointing at
221 * the head page (even the head page has this).
222 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800223 * The first tail page's ->lru.next holds the address of the compound page's
224 * put_page() function. Its ->lru.prev holds the order of allocation.
225 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800227
228static void free_compound_page(struct page *page)
229{
230 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
231}
232
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233static void prep_compound_page(struct page *page, unsigned long order)
234{
235 int i;
236 int nr_pages = 1 << order;
237
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800238 set_compound_page_dtor(page, free_compound_page);
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800239 page[1].lru.prev = (void *)order;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 for (i = 0; i < nr_pages; i++) {
241 struct page *p = page + i;
242
Nick Piggin5e9dace2006-03-22 00:08:01 -0800243 __SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700244 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 }
246}
247
248static void destroy_compound_page(struct page *page, unsigned long order)
249{
250 int i;
251 int nr_pages = 1 << order;
252
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800253 if (unlikely((unsigned long)page[1].lru.prev != order))
Nick Piggin224abf92006-01-06 00:11:11 -0800254 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
256 for (i = 0; i < nr_pages; i++) {
257 struct page *p = page + i;
258
Nick Piggin224abf92006-01-06 00:11:11 -0800259 if (unlikely(!PageCompound(p) |
260 (page_private(p) != (unsigned long)page)))
261 bad_page(page);
Nick Piggin5e9dace2006-03-22 00:08:01 -0800262 __ClearPageCompound(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 }
264}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Nick Piggin17cf4402006-03-22 00:08:41 -0800266static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
267{
268 int i;
269
Nick Piggin725d7042006-09-25 23:30:55 -0700270 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
Andrew Morton6626c5d2006-03-22 00:08:42 -0800271 /*
272 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
273 * and __GFP_HIGHMEM from hard or soft interrupt context.
274 */
Nick Piggin725d7042006-09-25 23:30:55 -0700275 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800276 for (i = 0; i < (1 << order); i++)
277 clear_highpage(page + i);
278}
279
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280/*
281 * function for dealing with page's order in buddy system.
282 * zone->lock is already acquired when we use these.
283 * So, we don't need atomic page->flags operations here.
284 */
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700285static inline unsigned long page_order(struct page *page)
286{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700287 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288}
289
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700290static inline void set_page_order(struct page *page, int order)
291{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700292 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000293 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294}
295
296static inline void rmv_page_order(struct page *page)
297{
Nick Piggin676165a2006-04-10 11:21:48 +1000298 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700299 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300}
301
302/*
303 * Locate the struct page for both the matching buddy in our
304 * pair (buddy1) and the combined O(n+1) page they form (page).
305 *
306 * 1) Any buddy B1 will have an order O twin B2 which satisfies
307 * the following equation:
308 * B2 = B1 ^ (1 << O)
309 * For example, if the starting buddy (buddy2) is #8 its order
310 * 1 buddy is #10:
311 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
312 *
313 * 2) Any buddy B will have an order O+1 parent P which
314 * satisfies the following equation:
315 * P = B & ~(1 << O)
316 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200317 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 */
319static inline struct page *
320__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
321{
322 unsigned long buddy_idx = page_idx ^ (1 << order);
323
324 return page + (buddy_idx - page_idx);
325}
326
327static inline unsigned long
328__find_combined_index(unsigned long page_idx, unsigned int order)
329{
330 return (page_idx & ~(1 << order));
331}
332
333/*
334 * This function checks whether a page is free && is the buddy
335 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800336 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000337 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700338 * (c) a page and its buddy have the same order &&
339 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 *
Nick Piggin676165a2006-04-10 11:21:48 +1000341 * For recording whether a page is in the buddy system, we use PG_buddy.
342 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
343 *
344 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700346static inline int page_is_buddy(struct page *page, struct page *buddy,
347 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348{
Nick Piggin13e74442006-01-06 00:10:58 -0800349#ifdef CONFIG_HOLES_IN_ZONE
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700350 if (!pfn_valid(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800351 return 0;
352#endif
353
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700354 if (page_zone_id(page) != page_zone_id(buddy))
355 return 0;
356
357 if (PageBuddy(buddy) && page_order(buddy) == order) {
358 BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700359 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000360 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700361 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362}
363
364/*
365 * Freeing function for a buddy system allocator.
366 *
367 * The concept of a buddy system is to maintain direct-mapped table
368 * (containing bit values) for memory blocks of various "orders".
369 * The bottom level table contains the map for the smallest allocatable
370 * units of memory (here, pages), and each level above it describes
371 * pairs of units from the levels below, hence, "buddies".
372 * At a high level, all that happens here is marking the table entry
373 * at the bottom level available, and propagating the changes upward
374 * as necessary, plus some accounting needed to play nicely with other
375 * parts of the VM system.
376 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000377 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700378 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 * So when we are allocating or freeing one, we can derive the state of the
380 * other. That is, if we allocate a small block, and both were
381 * free, the remainder of the region must be split into blocks.
382 * If a block is freed, and its buddy is also free, then this
383 * triggers coalescing into a block of larger size.
384 *
385 * -- wli
386 */
387
Nick Piggin48db57f2006-01-08 01:00:42 -0800388static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 struct zone *zone, unsigned int order)
390{
391 unsigned long page_idx;
392 int order_size = 1 << order;
393
Nick Piggin224abf92006-01-06 00:11:11 -0800394 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 destroy_compound_page(page, order);
396
397 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
398
Nick Piggin725d7042006-09-25 23:30:55 -0700399 VM_BUG_ON(page_idx & (order_size - 1));
400 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401
Christoph Lameterd23ad422007-02-10 01:43:02 -0800402 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 while (order < MAX_ORDER-1) {
404 unsigned long combined_idx;
405 struct free_area *area;
406 struct page *buddy;
407
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700409 if (!page_is_buddy(page, buddy, order))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800411
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 list_del(&buddy->lru);
413 area = zone->free_area + order;
414 area->nr_free--;
415 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800416 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 page = page + (combined_idx - page_idx);
418 page_idx = combined_idx;
419 order++;
420 }
421 set_page_order(page, order);
422 list_add(&page->lru, &zone->free_area[order].free_list);
423 zone->free_area[order].nr_free++;
424}
425
Nick Piggin224abf92006-01-06 00:11:11 -0800426static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427{
Nick Piggin92be2e332006-01-06 00:10:57 -0800428 if (unlikely(page_mapcount(page) |
429 (page->mapping != NULL) |
430 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 (page->flags & (
432 1 << PG_lru |
433 1 << PG_private |
434 1 << PG_locked |
435 1 << PG_active |
436 1 << PG_reclaim |
437 1 << PG_slab |
438 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700439 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000440 1 << PG_reserved |
441 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800442 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700444 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800445 /*
446 * For now, we report if PG_reserved was found set, but do not
447 * clear it, and do not free the page. But we shall soon need
448 * to do more, for when the ZERO_PAGE count wraps negative.
449 */
450 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451}
452
453/*
454 * Frees a list of pages.
455 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700456 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 *
458 * If the zone was previously in an "all pages pinned" state then look to
459 * see if this freeing clears that state.
460 *
461 * And clear the zone's pages_scanned counter, to hold off the "all pages are
462 * pinned" detection logic.
463 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800464static void free_pages_bulk(struct zone *zone, int count,
465 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466{
Nick Pigginc54ad302006-01-06 00:10:56 -0800467 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 zone->all_unreclaimable = 0;
469 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800470 while (count--) {
471 struct page *page;
472
Nick Piggin725d7042006-09-25 23:30:55 -0700473 VM_BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800475 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800477 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800479 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480}
481
Nick Piggin48db57f2006-01-08 01:00:42 -0800482static void free_one_page(struct zone *zone, struct page *page, int order)
483{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700484 spin_lock(&zone->lock);
485 zone->all_unreclaimable = 0;
486 zone->pages_scanned = 0;
Paul Jackson0798e512006-12-06 20:31:38 -0800487 __free_one_page(page, zone, order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700488 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800489}
490
491static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492{
Nick Pigginc54ad302006-01-06 00:10:56 -0800493 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800495 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800498 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800499 if (reserved)
500 return;
501
Nick Piggin9858db52006-10-11 01:21:30 -0700502 if (!PageHighMem(page))
503 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Nick Piggindafb1362006-10-11 01:21:30 -0700504 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800505 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700506
Nick Pigginc54ad302006-01-06 00:10:56 -0800507 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700508 __count_vm_events(PGFREE, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800509 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800510 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511}
512
David Howellsa226f6c2006-01-06 00:11:08 -0800513/*
514 * permit the bootmem allocator to evade page validation on high-order frees
515 */
516void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
517{
518 if (order == 0) {
519 __ClearPageReserved(page);
520 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800521 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800522 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800523 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800524 int loop;
525
Nick Piggin545b1ea2006-03-22 00:08:07 -0800526 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800527 for (loop = 0; loop < BITS_PER_LONG; loop++) {
528 struct page *p = &page[loop];
529
Nick Piggin545b1ea2006-03-22 00:08:07 -0800530 if (loop + 1 < BITS_PER_LONG)
531 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800532 __ClearPageReserved(p);
533 set_page_count(p, 0);
534 }
535
Nick Piggin7835e982006-03-22 00:08:40 -0800536 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800537 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800538 }
539}
540
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
542/*
543 * The order of subdivision here is critical for the IO subsystem.
544 * Please do not alter this order without good reasons and regression
545 * testing. Specifically, as large blocks of memory are subdivided,
546 * the order in which smaller blocks are delivered depends on the order
547 * they're subdivided in this function. This is the primary factor
548 * influencing the order in which pages are delivered to the IO
549 * subsystem according to empirical testing, and this is also justified
550 * by considering the behavior of a buddy system containing a single
551 * large block of memory acted on by a series of small allocations.
552 * This behavior is a critical factor in sglist merging's success.
553 *
554 * -- wli
555 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800556static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 int low, int high, struct free_area *area)
558{
559 unsigned long size = 1 << high;
560
561 while (high > low) {
562 area--;
563 high--;
564 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700565 VM_BUG_ON(bad_range(zone, &page[size]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 list_add(&page[size].lru, &area->free_list);
567 area->nr_free++;
568 set_page_order(&page[size], high);
569 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570}
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572/*
573 * This page is about to be returned from the page allocator
574 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800575static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576{
Nick Piggin92be2e332006-01-06 00:10:57 -0800577 if (unlikely(page_mapcount(page) |
578 (page->mapping != NULL) |
579 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700580 (page->flags & (
581 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 1 << PG_private |
583 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 1 << PG_active |
585 1 << PG_dirty |
586 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700587 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700589 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000590 1 << PG_reserved |
591 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800592 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
Hugh Dickins689bceb2005-11-21 21:32:20 -0800594 /*
595 * For now, we report if PG_reserved was found set, but do not
596 * clear it, and do not allocate the page: as a safety net.
597 */
598 if (PageReserved(page))
599 return 1;
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
602 1 << PG_referenced | 1 << PG_arch_1 |
603 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700604 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800605 set_page_refcounted(page);
Nick Piggincc1025092006-12-06 20:32:00 -0800606
607 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800609
610 if (gfp_flags & __GFP_ZERO)
611 prep_zero_page(page, order, gfp_flags);
612
613 if (order && (gfp_flags & __GFP_COMP))
614 prep_compound_page(page, order);
615
Hugh Dickins689bceb2005-11-21 21:32:20 -0800616 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617}
618
619/*
620 * Do the hard work of removing an element from the buddy allocator.
621 * Call me with the zone->lock already held.
622 */
623static struct page *__rmqueue(struct zone *zone, unsigned int order)
624{
625 struct free_area * area;
626 unsigned int current_order;
627 struct page *page;
628
629 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
630 area = zone->free_area + current_order;
631 if (list_empty(&area->free_list))
632 continue;
633
634 page = list_entry(area->free_list.next, struct page, lru);
635 list_del(&page->lru);
636 rmv_page_order(page);
637 area->nr_free--;
Christoph Lameterd23ad422007-02-10 01:43:02 -0800638 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
Nick Piggin085cc7d2006-01-06 00:11:01 -0800639 expand(zone, page, order, current_order, area);
640 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 }
642
643 return NULL;
644}
645
646/*
647 * Obtain a specified number of elements from the buddy allocator, all under
648 * a single hold of the lock, for efficiency. Add them to the supplied list.
649 * Returns the number of new pages which were placed at *list.
650 */
651static int rmqueue_bulk(struct zone *zone, unsigned int order,
652 unsigned long count, struct list_head *list)
653{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
Nick Pigginc54ad302006-01-06 00:10:56 -0800656 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 for (i = 0; i < count; ++i) {
Nick Piggin085cc7d2006-01-06 00:11:01 -0800658 struct page *page = __rmqueue(zone, order);
659 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 list_add_tail(&page->lru, list);
662 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800663 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800664 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665}
666
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700667#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800668/*
669 * Called from the slab reaper to drain pagesets on a particular node that
Christoph Lameter39bbcb82006-09-25 23:31:49 -0700670 * belongs to the currently executing processor.
Christoph Lameter879336c2006-03-22 00:09:08 -0800671 * Note that this function must be called with the thread pinned to
672 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800673 */
674void drain_node_pages(int nodeid)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700675{
Christoph Lameter2f6726e2006-09-25 23:31:18 -0700676 int i;
677 enum zone_type z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700678 unsigned long flags;
679
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800680 for (z = 0; z < MAX_NR_ZONES; z++) {
681 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700682 struct per_cpu_pageset *pset;
683
Christoph Lameter39bbcb82006-09-25 23:31:49 -0700684 if (!populated_zone(zone))
685 continue;
686
Nick Piggin23316bc2006-01-08 01:00:41 -0800687 pset = zone_pcp(zone, smp_processor_id());
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700688 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
689 struct per_cpu_pages *pcp;
690
691 pcp = &pset->pcp[i];
Christoph Lameter879336c2006-03-22 00:09:08 -0800692 if (pcp->count) {
Christoph Lameterbc4ba392006-12-06 20:33:02 -0800693 int to_drain;
694
Christoph Lameter879336c2006-03-22 00:09:08 -0800695 local_irq_save(flags);
Christoph Lameterbc4ba392006-12-06 20:33:02 -0800696 if (pcp->count >= pcp->batch)
697 to_drain = pcp->batch;
698 else
699 to_drain = pcp->count;
700 free_pages_bulk(zone, to_drain, &pcp->list, 0);
701 pcp->count -= to_drain;
Christoph Lameter879336c2006-03-22 00:09:08 -0800702 local_irq_restore(flags);
703 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700704 }
705 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700706}
707#endif
708
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709static void __drain_pages(unsigned int cpu)
710{
Nick Pigginc54ad302006-01-06 00:10:56 -0800711 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 struct zone *zone;
713 int i;
714
715 for_each_zone(zone) {
716 struct per_cpu_pageset *pset;
717
Christoph Lameterf2e12bb2007-01-05 16:37:02 -0800718 if (!populated_zone(zone))
719 continue;
720
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700721 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
723 struct per_cpu_pages *pcp;
724
725 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800726 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800727 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
728 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800729 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 }
731 }
732}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
734#ifdef CONFIG_PM
735
736void mark_free_pages(struct zone *zone)
737{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700738 unsigned long pfn, max_zone_pfn;
739 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 int order;
741 struct list_head *curr;
742
743 if (!zone->spanned_pages)
744 return;
745
746 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700747
748 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
749 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
750 if (pfn_valid(pfn)) {
751 struct page *page = pfn_to_page(pfn);
752
753 if (!PageNosave(page))
754 ClearPageNosaveFree(page);
755 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
757 for (order = MAX_ORDER - 1; order >= 0; --order)
758 list_for_each(curr, &zone->free_area[order].free_list) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700759 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700761 pfn = page_to_pfn(list_entry(curr, struct page, lru));
762 for (i = 0; i < (1UL << order); i++)
763 SetPageNosaveFree(pfn_to_page(pfn + i));
764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 spin_unlock_irqrestore(&zone->lock, flags);
767}
768
769/*
770 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
771 */
772void drain_local_pages(void)
773{
774 unsigned long flags;
775
776 local_irq_save(flags);
777 __drain_pages(smp_processor_id());
778 local_irq_restore(flags);
779}
780#endif /* CONFIG_PM */
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782/*
783 * Free a 0-order page
784 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785static void fastcall free_hot_cold_page(struct page *page, int cold)
786{
787 struct zone *zone = page_zone(page);
788 struct per_cpu_pages *pcp;
789 unsigned long flags;
790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 if (PageAnon(page))
792 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800793 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800794 return;
795
Nick Piggin9858db52006-10-11 01:21:30 -0700796 if (!PageHighMem(page))
797 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Nick Piggindafb1362006-10-11 01:21:30 -0700798 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800799 kernel_map_pages(page, 1, 0);
800
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700801 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700803 __count_vm_event(PGFREE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 list_add(&page->lru, &pcp->list);
805 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800806 if (pcp->count >= pcp->high) {
807 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
808 pcp->count -= pcp->batch;
809 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 local_irq_restore(flags);
811 put_cpu();
812}
813
814void fastcall free_hot_page(struct page *page)
815{
816 free_hot_cold_page(page, 0);
817}
818
819void fastcall free_cold_page(struct page *page)
820{
821 free_hot_cold_page(page, 1);
822}
823
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800824/*
825 * split_page takes a non-compound higher-order page, and splits it into
826 * n (1<<order) sub-pages: page[0..n]
827 * Each sub-page must be freed individually.
828 *
829 * Note: this is probably too low level an operation for use in drivers.
830 * Please consult with lkml before using this in your driver.
831 */
832void split_page(struct page *page, unsigned int order)
833{
834 int i;
835
Nick Piggin725d7042006-09-25 23:30:55 -0700836 VM_BUG_ON(PageCompound(page));
837 VM_BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -0800838 for (i = 1; i < (1 << order); i++)
839 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800840}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800841
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842/*
843 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
844 * we cheat by calling it from here, in the order > 0 path. Saves a branch
845 * or two.
846 */
Nick Piggina74609f2006-01-06 00:11:20 -0800847static struct page *buffered_rmqueue(struct zonelist *zonelist,
848 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849{
850 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800851 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800853 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
Hugh Dickins689bceb2005-11-21 21:32:20 -0800855again:
Nick Piggina74609f2006-01-06 00:11:20 -0800856 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800857 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 struct per_cpu_pages *pcp;
859
Nick Piggina74609f2006-01-06 00:11:20 -0800860 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800862 if (!pcp->count) {
nkalmala941c7102006-11-02 22:07:04 -0800863 pcp->count = rmqueue_bulk(zone, 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800865 if (unlikely(!pcp->count))
866 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 }
Nick Piggina74609f2006-01-06 00:11:20 -0800868 page = list_entry(pcp->list.next, struct page, lru);
869 list_del(&page->lru);
870 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800871 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 spin_lock_irqsave(&zone->lock, flags);
873 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800874 spin_unlock(&zone->lock);
875 if (!page)
876 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 }
878
Christoph Lameterf8891e52006-06-30 01:55:45 -0700879 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Christoph Lameterca889e62006-06-30 01:55:44 -0700880 zone_statistics(zonelist, zone);
Nick Piggina74609f2006-01-06 00:11:20 -0800881 local_irq_restore(flags);
882 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
Nick Piggin725d7042006-09-25 23:30:55 -0700884 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -0800885 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -0800886 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800888
889failed:
890 local_irq_restore(flags);
891 put_cpu();
892 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893}
894
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800895#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800896#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
897#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
898#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
899#define ALLOC_HARDER 0x10 /* try to alloc harder */
900#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
901#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800902
Akinobu Mita933e3122006-12-08 02:39:45 -0800903#ifdef CONFIG_FAIL_PAGE_ALLOC
904
905static struct fail_page_alloc_attr {
906 struct fault_attr attr;
907
908 u32 ignore_gfp_highmem;
909 u32 ignore_gfp_wait;
910
911#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
912
913 struct dentry *ignore_gfp_highmem_file;
914 struct dentry *ignore_gfp_wait_file;
915
916#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
917
918} fail_page_alloc = {
919 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -0800920 .ignore_gfp_wait = 1,
921 .ignore_gfp_highmem = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -0800922};
923
924static int __init setup_fail_page_alloc(char *str)
925{
926 return setup_fault_attr(&fail_page_alloc.attr, str);
927}
928__setup("fail_page_alloc=", setup_fail_page_alloc);
929
930static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
931{
932 if (gfp_mask & __GFP_NOFAIL)
933 return 0;
934 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
935 return 0;
936 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
937 return 0;
938
939 return should_fail(&fail_page_alloc.attr, 1 << order);
940}
941
942#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
943
944static int __init fail_page_alloc_debugfs(void)
945{
946 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
947 struct dentry *dir;
948 int err;
949
950 err = init_fault_attr_dentries(&fail_page_alloc.attr,
951 "fail_page_alloc");
952 if (err)
953 return err;
954 dir = fail_page_alloc.attr.dentries.dir;
955
956 fail_page_alloc.ignore_gfp_wait_file =
957 debugfs_create_bool("ignore-gfp-wait", mode, dir,
958 &fail_page_alloc.ignore_gfp_wait);
959
960 fail_page_alloc.ignore_gfp_highmem_file =
961 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
962 &fail_page_alloc.ignore_gfp_highmem);
963
964 if (!fail_page_alloc.ignore_gfp_wait_file ||
965 !fail_page_alloc.ignore_gfp_highmem_file) {
966 err = -ENOMEM;
967 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
968 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
969 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
970 }
971
972 return err;
973}
974
975late_initcall(fail_page_alloc_debugfs);
976
977#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
978
979#else /* CONFIG_FAIL_PAGE_ALLOC */
980
981static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
982{
983 return 0;
984}
985
986#endif /* CONFIG_FAIL_PAGE_ALLOC */
987
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988/*
989 * Return 1 if free pages are above 'mark'. This takes into account the order
990 * of the allocation.
991 */
992int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800993 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
995 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -0800996 long min = mark;
997 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 int o;
999
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001000 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001002 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 min -= min / 4;
1004
1005 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1006 return 0;
1007 for (o = 0; o < order; o++) {
1008 /* At the next order, this order's pages become unavailable */
1009 free_pages -= z->free_area[o].nr_free << o;
1010
1011 /* Require fewer higher order pages to be free */
1012 min >>= 1;
1013
1014 if (free_pages <= min)
1015 return 0;
1016 }
1017 return 1;
1018}
1019
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001020#ifdef CONFIG_NUMA
1021/*
1022 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1023 * skip over zones that are not allowed by the cpuset, or that have
1024 * been recently (in last second) found to be nearly full. See further
1025 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1026 * that have to skip over alot of full or unallowed zones.
1027 *
1028 * If the zonelist cache is present in the passed in zonelist, then
1029 * returns a pointer to the allowed node mask (either the current
1030 * tasks mems_allowed, or node_online_map.)
1031 *
1032 * If the zonelist cache is not available for this zonelist, does
1033 * nothing and returns NULL.
1034 *
1035 * If the fullzones BITMAP in the zonelist cache is stale (more than
1036 * a second since last zap'd) then we zap it out (clear its bits.)
1037 *
1038 * We hold off even calling zlc_setup, until after we've checked the
1039 * first zone in the zonelist, on the theory that most allocations will
1040 * be satisfied from that first zone, so best to examine that zone as
1041 * quickly as we can.
1042 */
1043static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1044{
1045 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1046 nodemask_t *allowednodes; /* zonelist_cache approximation */
1047
1048 zlc = zonelist->zlcache_ptr;
1049 if (!zlc)
1050 return NULL;
1051
1052 if (jiffies - zlc->last_full_zap > 1 * HZ) {
1053 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1054 zlc->last_full_zap = jiffies;
1055 }
1056
1057 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1058 &cpuset_current_mems_allowed :
1059 &node_online_map;
1060 return allowednodes;
1061}
1062
1063/*
1064 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1065 * if it is worth looking at further for free memory:
1066 * 1) Check that the zone isn't thought to be full (doesn't have its
1067 * bit set in the zonelist_cache fullzones BITMAP).
1068 * 2) Check that the zones node (obtained from the zonelist_cache
1069 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1070 * Return true (non-zero) if zone is worth looking at further, or
1071 * else return false (zero) if it is not.
1072 *
1073 * This check -ignores- the distinction between various watermarks,
1074 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1075 * found to be full for any variation of these watermarks, it will
1076 * be considered full for up to one second by all requests, unless
1077 * we are so low on memory on all allowed nodes that we are forced
1078 * into the second scan of the zonelist.
1079 *
1080 * In the second scan we ignore this zonelist cache and exactly
1081 * apply the watermarks to all zones, even it is slower to do so.
1082 * We are low on memory in the second scan, and should leave no stone
1083 * unturned looking for a free page.
1084 */
1085static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1086 nodemask_t *allowednodes)
1087{
1088 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1089 int i; /* index of *z in zonelist zones */
1090 int n; /* node that zone *z is on */
1091
1092 zlc = zonelist->zlcache_ptr;
1093 if (!zlc)
1094 return 1;
1095
1096 i = z - zonelist->zones;
1097 n = zlc->z_to_n[i];
1098
1099 /* This zone is worth trying if it is allowed but not full */
1100 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1101}
1102
1103/*
1104 * Given 'z' scanning a zonelist, set the corresponding bit in
1105 * zlc->fullzones, so that subsequent attempts to allocate a page
1106 * from that zone don't waste time re-examining it.
1107 */
1108static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1109{
1110 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1111 int i; /* index of *z in zonelist zones */
1112
1113 zlc = zonelist->zlcache_ptr;
1114 if (!zlc)
1115 return;
1116
1117 i = z - zonelist->zones;
1118
1119 set_bit(i, zlc->fullzones);
1120}
1121
1122#else /* CONFIG_NUMA */
1123
1124static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1125{
1126 return NULL;
1127}
1128
1129static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1130 nodemask_t *allowednodes)
1131{
1132 return 1;
1133}
1134
1135static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1136{
1137}
1138#endif /* CONFIG_NUMA */
1139
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001140/*
Paul Jackson0798e512006-12-06 20:31:38 -08001141 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001142 * a page.
1143 */
1144static struct page *
1145get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1146 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -07001147{
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001148 struct zone **z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001149 struct page *page = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001150 int classzone_idx = zone_idx(zonelist->zones[0]);
Christoph Lameter1192d522006-09-25 23:31:45 -07001151 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001152 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1153 int zlc_active = 0; /* set if using zonelist_cache */
1154 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001155
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001156zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001157 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001158 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001159 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1160 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001161 z = zonelist->zones;
1162
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001163 do {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001164 if (NUMA_BUILD && zlc_active &&
1165 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1166 continue;
Christoph Lameter1192d522006-09-25 23:31:45 -07001167 zone = *z;
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001168 if (unlikely(NUMA_BUILD && (gfp_mask & __GFP_THISNODE) &&
Christoph Lameter1192d522006-09-25 23:31:45 -07001169 zone->zone_pgdat != zonelist->zones[0]->zone_pgdat))
Christoph Lameter9b819d22006-09-25 23:31:40 -07001170 break;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001171 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001172 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001173 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001174
1175 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001176 unsigned long mark;
1177 if (alloc_flags & ALLOC_WMARK_MIN)
Christoph Lameter1192d522006-09-25 23:31:45 -07001178 mark = zone->pages_min;
Nick Piggin31488902005-11-28 13:44:03 -08001179 else if (alloc_flags & ALLOC_WMARK_LOW)
Christoph Lameter1192d522006-09-25 23:31:45 -07001180 mark = zone->pages_low;
Nick Piggin31488902005-11-28 13:44:03 -08001181 else
Christoph Lameter1192d522006-09-25 23:31:45 -07001182 mark = zone->pages_high;
Paul Jackson0798e512006-12-06 20:31:38 -08001183 if (!zone_watermark_ok(zone, order, mark,
1184 classzone_idx, alloc_flags)) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001185 if (!zone_reclaim_mode ||
Christoph Lameter1192d522006-09-25 23:31:45 -07001186 !zone_reclaim(zone, gfp_mask, order))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001187 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001188 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001189 }
1190
Christoph Lameter1192d522006-09-25 23:31:45 -07001191 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
Paul Jackson0798e512006-12-06 20:31:38 -08001192 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001193 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001194this_zone_full:
1195 if (NUMA_BUILD)
1196 zlc_mark_zone_full(zonelist, z);
1197try_next_zone:
1198 if (NUMA_BUILD && !did_zlc_setup) {
1199 /* we do zlc_setup after the first zone is tried */
1200 allowednodes = zlc_setup(zonelist, alloc_flags);
1201 zlc_active = 1;
1202 did_zlc_setup = 1;
1203 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001204 } while (*(++z) != NULL);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001205
1206 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1207 /* Disable zlc cache for second zonelist scan */
1208 zlc_active = 0;
1209 goto zonelist_scan;
1210 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001211 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001212}
1213
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214/*
1215 * This is the 'heart' of the zoned buddy allocator.
1216 */
1217struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +01001218__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 struct zonelist *zonelist)
1220{
Al Viro260b2362005-10-21 03:22:44 -04001221 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001222 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 struct page *page;
1224 struct reclaim_state reclaim_state;
1225 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001227 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 int did_some_progress;
1229
1230 might_sleep_if(wait);
1231
Akinobu Mita933e3122006-12-08 02:39:45 -08001232 if (should_fail_alloc_page(gfp_mask, order))
1233 return NULL;
1234
Jens Axboe6b1de912005-11-17 21:35:02 +01001235restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001236 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001238 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 /* Should this ever happen?? */
1240 return NULL;
1241 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001242
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001243 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001244 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001245 if (page)
1246 goto got_pg;
1247
Christoph Lameter952f3b52006-12-06 20:33:26 -08001248 /*
1249 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1250 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1251 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1252 * using a larger set of nodes after it has established that the
1253 * allowed per node queues are empty and that nodes are
1254 * over allocated.
1255 */
1256 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1257 goto nopage;
1258
Paul Jackson0798e512006-12-06 20:31:38 -08001259 for (z = zonelist->zones; *z; z++)
Chris Wright43b0bc02006-06-25 05:47:55 -07001260 wakeup_kswapd(*z, order);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001261
Paul Jackson9bf22292005-09-06 15:18:12 -07001262 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001263 * OK, we're below the kswapd watermark and have kicked background
1264 * reclaim. Now things get more complex, so set up alloc_flags according
1265 * to how we want to proceed.
1266 *
1267 * The caller may dip into page reserves a bit more if the caller
1268 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -08001269 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1270 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -07001271 */
Nick Piggin31488902005-11-28 13:44:03 -08001272 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001273 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1274 alloc_flags |= ALLOC_HARDER;
1275 if (gfp_mask & __GFP_HIGH)
1276 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -07001277 if (wait)
1278 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279
1280 /*
1281 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001282 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 *
1284 * This is the last chance, in general, before the goto nopage.
1285 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -07001286 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001288 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1289 if (page)
1290 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291
1292 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -07001293
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001294rebalance:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001295 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1296 && !in_interrupt()) {
1297 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -08001298nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001299 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001300 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -08001301 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001302 if (page)
1303 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -08001304 if (gfp_mask & __GFP_NOFAIL) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001305 congestion_wait(WRITE, HZ/50);
Kirill Korotaev885036d2005-11-13 16:06:41 -08001306 goto nofail_alloc;
1307 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 }
1309 goto nopage;
1310 }
1311
1312 /* Atomic allocations - we can't balance anything */
1313 if (!wait)
1314 goto nopage;
1315
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 cond_resched();
1317
1318 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001319 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 p->flags |= PF_MEMALLOC;
1321 reclaim_state.reclaimed_slab = 0;
1322 p->reclaim_state = &reclaim_state;
1323
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001324 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
1326 p->reclaim_state = NULL;
1327 p->flags &= ~PF_MEMALLOC;
1328
1329 cond_resched();
1330
1331 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001332 page = get_page_from_freelist(gfp_mask, order,
1333 zonelist, alloc_flags);
1334 if (page)
1335 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1337 /*
1338 * Go through the zonelist yet one more time, keep
1339 * very high watermark here, this is only to catch
1340 * a parallel oom killing, we must fail if we're still
1341 * under heavy pressure.
1342 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001343 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001344 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001345 if (page)
1346 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001348 out_of_memory(zonelist, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 goto restart;
1350 }
1351
1352 /*
1353 * Don't let big-order allocations loop unless the caller explicitly
1354 * requests that. Wait for some write requests to complete then retry.
1355 *
1356 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1357 * <= 3, but that may not be true in other implementations.
1358 */
1359 do_retry = 0;
1360 if (!(gfp_mask & __GFP_NORETRY)) {
1361 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1362 do_retry = 1;
1363 if (gfp_mask & __GFP_NOFAIL)
1364 do_retry = 1;
1365 }
1366 if (do_retry) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001367 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 goto rebalance;
1369 }
1370
1371nopage:
1372 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1373 printk(KERN_WARNING "%s: page allocation failure."
1374 " order:%d, mode:0x%x\n",
1375 p->comm, order, gfp_mask);
1376 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001377 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 return page;
1381}
1382
1383EXPORT_SYMBOL(__alloc_pages);
1384
1385/*
1386 * Common helper functions.
1387 */
Al Virodd0fc662005-10-07 07:46:04 +01001388fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389{
1390 struct page * page;
1391 page = alloc_pages(gfp_mask, order);
1392 if (!page)
1393 return 0;
1394 return (unsigned long) page_address(page);
1395}
1396
1397EXPORT_SYMBOL(__get_free_pages);
1398
Al Virodd0fc662005-10-07 07:46:04 +01001399fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
1401 struct page * page;
1402
1403 /*
1404 * get_zeroed_page() returns a 32-bit address, which cannot represent
1405 * a highmem page
1406 */
Nick Piggin725d7042006-09-25 23:30:55 -07001407 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
1409 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1410 if (page)
1411 return (unsigned long) page_address(page);
1412 return 0;
1413}
1414
1415EXPORT_SYMBOL(get_zeroed_page);
1416
1417void __pagevec_free(struct pagevec *pvec)
1418{
1419 int i = pagevec_count(pvec);
1420
1421 while (--i >= 0)
1422 free_hot_cold_page(pvec->pages[i], pvec->cold);
1423}
1424
1425fastcall void __free_pages(struct page *page, unsigned int order)
1426{
Nick Pigginb5810032005-10-29 18:16:12 -07001427 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 if (order == 0)
1429 free_hot_page(page);
1430 else
1431 __free_pages_ok(page, order);
1432 }
1433}
1434
1435EXPORT_SYMBOL(__free_pages);
1436
1437fastcall void free_pages(unsigned long addr, unsigned int order)
1438{
1439 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001440 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 __free_pages(virt_to_page((void *)addr), order);
1442 }
1443}
1444
1445EXPORT_SYMBOL(free_pages);
1446
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447static unsigned int nr_free_zone_pages(int offset)
1448{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001449 /* Just pick one node, since fallback list is circular */
1450 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 unsigned int sum = 0;
1452
Martin J. Blighe310fd42005-07-29 22:59:18 -07001453 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1454 struct zone **zonep = zonelist->zones;
1455 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
Martin J. Blighe310fd42005-07-29 22:59:18 -07001457 for (zone = *zonep++; zone; zone = *zonep++) {
1458 unsigned long size = zone->present_pages;
1459 unsigned long high = zone->pages_high;
1460 if (size > high)
1461 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 }
1463
1464 return sum;
1465}
1466
1467/*
1468 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1469 */
1470unsigned int nr_free_buffer_pages(void)
1471{
Al Viroaf4ca452005-10-21 02:55:38 -04001472 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473}
1474
1475/*
1476 * Amount of free RAM allocatable within all zones
1477 */
1478unsigned int nr_free_pagecache_pages(void)
1479{
Al Viroaf4ca452005-10-21 02:55:38 -04001480 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001482
1483static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001485 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08001486 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489void si_meminfo(struct sysinfo *val)
1490{
1491 val->totalram = totalram_pages;
1492 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001493 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 val->totalhigh = totalhigh_pages;
1496 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 val->mem_unit = PAGE_SIZE;
1498}
1499
1500EXPORT_SYMBOL(si_meminfo);
1501
1502#ifdef CONFIG_NUMA
1503void si_meminfo_node(struct sysinfo *val, int nid)
1504{
1505 pg_data_t *pgdat = NODE_DATA(nid);
1506
1507 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001508 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001509#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001511 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1512 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001513#else
1514 val->totalhigh = 0;
1515 val->freehigh = 0;
1516#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 val->mem_unit = PAGE_SIZE;
1518}
1519#endif
1520
1521#define K(x) ((x) << (PAGE_SHIFT-10))
1522
1523/*
1524 * Show free area list (used inside shift_scroll-lock stuff)
1525 * We also calculate the percentage fragmentation. We do this by counting the
1526 * memory on each free list with the exception of the first item on the list.
1527 */
1528void show_free_areas(void)
1529{
Jes Sorensenc7241912006-09-27 01:50:05 -07001530 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 struct zone *zone;
1532
1533 for_each_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07001534 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07001536
1537 show_node(zone);
1538 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539
Dave Jones6b482c62005-11-10 15:45:56 -05001540 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 struct per_cpu_pageset *pageset;
1542
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001543 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
Jes Sorensenc7241912006-09-27 01:50:05 -07001545 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1546 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1547 cpu, pageset->pcp[0].high,
1548 pageset->pcp[0].batch, pageset->pcp[0].count,
1549 pageset->pcp[1].high, pageset->pcp[1].batch,
1550 pageset->pcp[1].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 }
1552 }
1553
Andrew Mortona25700a2007-02-08 14:20:40 -08001554 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
Christoph Lameterd23ad422007-02-10 01:43:02 -08001555 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
Christoph Lameter65e458d42007-02-10 01:43:05 -08001556 global_page_state(NR_ACTIVE),
1557 global_page_state(NR_INACTIVE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07001558 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07001559 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001560 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08001561 global_page_state(NR_FREE_PAGES),
Christoph Lameter972d1a72006-09-25 23:31:51 -07001562 global_page_state(NR_SLAB_RECLAIMABLE) +
1563 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001564 global_page_state(NR_FILE_MAPPED),
Andrew Mortona25700a2007-02-08 14:20:40 -08001565 global_page_state(NR_PAGETABLE),
1566 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
1568 for_each_zone(zone) {
1569 int i;
1570
Jes Sorensenc7241912006-09-27 01:50:05 -07001571 if (!populated_zone(zone))
1572 continue;
1573
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 show_node(zone);
1575 printk("%s"
1576 " free:%lukB"
1577 " min:%lukB"
1578 " low:%lukB"
1579 " high:%lukB"
1580 " active:%lukB"
1581 " inactive:%lukB"
1582 " present:%lukB"
1583 " pages_scanned:%lu"
1584 " all_unreclaimable? %s"
1585 "\n",
1586 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08001587 K(zone_page_state(zone, NR_FREE_PAGES)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 K(zone->pages_min),
1589 K(zone->pages_low),
1590 K(zone->pages_high),
Christoph Lameterc8785382007-02-10 01:43:01 -08001591 K(zone_page_state(zone, NR_ACTIVE)),
1592 K(zone_page_state(zone, NR_INACTIVE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 K(zone->present_pages),
1594 zone->pages_scanned,
1595 (zone->all_unreclaimable ? "yes" : "no")
1596 );
1597 printk("lowmem_reserve[]:");
1598 for (i = 0; i < MAX_NR_ZONES; i++)
1599 printk(" %lu", zone->lowmem_reserve[i]);
1600 printk("\n");
1601 }
1602
1603 for_each_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001604 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
Jes Sorensenc7241912006-09-27 01:50:05 -07001606 if (!populated_zone(zone))
1607 continue;
1608
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 show_node(zone);
1610 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
1612 spin_lock_irqsave(&zone->lock, flags);
1613 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001614 nr[order] = zone->free_area[order].nr_free;
1615 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 }
1617 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001618 for (order = 0; order < MAX_ORDER; order++)
1619 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 printk("= %lukB\n", K(total));
1621 }
1622
1623 show_swap_cache_info();
1624}
1625
1626/*
1627 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001628 *
1629 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001631static int __meminit build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001632 struct zonelist *zonelist, int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633{
Christoph Lameter1a932052006-01-06 00:11:16 -08001634 struct zone *zone;
1635
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001636 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001637 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001638
1639 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001640 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08001641 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001642 if (populated_zone(zone)) {
Christoph Lameter070f8032006-01-06 00:11:19 -08001643 zonelist->zones[nr_zones++] = zone;
1644 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08001646
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001647 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08001648 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649}
1650
1651#ifdef CONFIG_NUMA
1652#define MAX_NODE_LOAD (num_online_nodes())
Yasunori Goto86356ab2006-06-23 02:03:09 -07001653static int __meminitdata node_load[MAX_NUMNODES];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001655 * 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 -07001656 * @node: node whose fallback list we're appending
1657 * @used_node_mask: nodemask_t of already used nodes
1658 *
1659 * We use a number of factors to determine which is the next node that should
1660 * appear on a given node's fallback list. The node should not have appeared
1661 * already in @node's fallback list, and it should be the next closest node
1662 * according to the distance array (which contains arbitrary distance values
1663 * from each node to each node in the system), and should also prefer nodes
1664 * with no CPUs, since presumably they'll have very little allocation pressure
1665 * on them otherwise.
1666 * It returns -1 if no node is found.
1667 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001668static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669{
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01001670 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 int min_val = INT_MAX;
1672 int best_node = -1;
1673
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01001674 /* Use the local node if we haven't already */
1675 if (!node_isset(node, *used_node_mask)) {
1676 node_set(node, *used_node_mask);
1677 return node;
1678 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01001680 for_each_online_node(n) {
1681 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
1683 /* Don't want a node to appear more than once */
1684 if (node_isset(n, *used_node_mask))
1685 continue;
1686
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 /* Use the distance array to find the distance */
1688 val = node_distance(node, n);
1689
Linus Torvalds4cf808eb2006-02-17 20:38:21 +01001690 /* Penalize nodes under us ("prefer the next node") */
1691 val += (n < node);
1692
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 /* Give preference to headless and unused nodes */
1694 tmp = node_to_cpumask(n);
1695 if (!cpus_empty(tmp))
1696 val += PENALTY_FOR_NODE_WITH_CPUS;
1697
1698 /* Slight preference for less loaded node */
1699 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1700 val += node_load[n];
1701
1702 if (val < min_val) {
1703 min_val = val;
1704 best_node = n;
1705 }
1706 }
1707
1708 if (best_node >= 0)
1709 node_set(best_node, *used_node_mask);
1710
1711 return best_node;
1712}
1713
Yasunori Goto86356ab2006-06-23 02:03:09 -07001714static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715{
Christoph Lameter19655d32006-09-25 23:31:19 -07001716 int j, node, local_node;
1717 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 int prev_node, load;
1719 struct zonelist *zonelist;
1720 nodemask_t used_mask;
1721
1722 /* initialize zonelists */
Christoph Lameter19655d32006-09-25 23:31:19 -07001723 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 zonelist = pgdat->node_zonelists + i;
1725 zonelist->zones[0] = NULL;
1726 }
1727
1728 /* NUMA-aware ordering of nodes */
1729 local_node = pgdat->node_id;
1730 load = num_online_nodes();
1731 prev_node = local_node;
1732 nodes_clear(used_mask);
1733 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001734 int distance = node_distance(local_node, node);
1735
1736 /*
1737 * If another node is sufficiently far away then it is better
1738 * to reclaim pages in a zone before going off node.
1739 */
1740 if (distance > RECLAIM_DISTANCE)
1741 zone_reclaim_mode = 1;
1742
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 /*
1744 * We don't want to pressure a particular node.
1745 * So adding penalty to the first node in same
1746 * distance group to make it round-robin.
1747 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08001748
1749 if (distance != node_distance(local_node, prev_node))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 node_load[node] += load;
1751 prev_node = node;
1752 load--;
Christoph Lameter19655d32006-09-25 23:31:19 -07001753 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 zonelist = pgdat->node_zonelists + i;
1755 for (j = 0; zonelist->zones[j] != NULL; j++);
1756
Christoph Lameter19655d32006-09-25 23:31:19 -07001757 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 zonelist->zones[j] = NULL;
1759 }
1760 }
1761}
1762
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001763/* Construct the zonelist performance cache - see further mmzone.h */
1764static void __meminit build_zonelist_cache(pg_data_t *pgdat)
1765{
1766 int i;
1767
1768 for (i = 0; i < MAX_NR_ZONES; i++) {
1769 struct zonelist *zonelist;
1770 struct zonelist_cache *zlc;
1771 struct zone **z;
1772
1773 zonelist = pgdat->node_zonelists + i;
1774 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
1775 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1776 for (z = zonelist->zones; *z; z++)
1777 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
1778 }
1779}
1780
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781#else /* CONFIG_NUMA */
1782
Yasunori Goto86356ab2006-06-23 02:03:09 -07001783static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784{
Christoph Lameter19655d32006-09-25 23:31:19 -07001785 int node, local_node;
1786 enum zone_type i,j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787
1788 local_node = pgdat->node_id;
Christoph Lameter19655d32006-09-25 23:31:19 -07001789 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 struct zonelist *zonelist;
1791
1792 zonelist = pgdat->node_zonelists + i;
1793
Christoph Lameter19655d32006-09-25 23:31:19 -07001794 j = build_zonelists_node(pgdat, zonelist, 0, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 /*
1796 * Now we build the zonelist so that it contains the zones
1797 * of all the other nodes.
1798 * We don't want to pressure a particular node, so when
1799 * building the zones for node N, we make sure that the
1800 * zones coming right after the local ones are those from
1801 * node N+1 (modulo N)
1802 */
1803 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1804 if (!node_online(node))
1805 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07001806 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 }
1808 for (node = 0; node < local_node; node++) {
1809 if (!node_online(node))
1810 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07001811 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 }
1813
1814 zonelist->zones[j] = NULL;
1815 }
1816}
1817
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001818/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
1819static void __meminit build_zonelist_cache(pg_data_t *pgdat)
1820{
1821 int i;
1822
1823 for (i = 0; i < MAX_NR_ZONES; i++)
1824 pgdat->node_zonelists[i].zlcache_ptr = NULL;
1825}
1826
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827#endif /* CONFIG_NUMA */
1828
Yasunori Goto68113782006-06-23 02:03:11 -07001829/* return values int ....just for stop_machine_run() */
1830static int __meminit __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831{
Yasunori Goto68113782006-06-23 02:03:11 -07001832 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001833
1834 for_each_online_node(nid) {
Yasunori Goto68113782006-06-23 02:03:11 -07001835 build_zonelists(NODE_DATA(nid));
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001836 build_zonelist_cache(NODE_DATA(nid));
1837 }
Yasunori Goto68113782006-06-23 02:03:11 -07001838 return 0;
1839}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840
Yasunori Goto68113782006-06-23 02:03:11 -07001841void __meminit build_all_zonelists(void)
1842{
1843 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07001844 __build_all_zonelists(NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07001845 cpuset_init_current_mems_allowed();
1846 } else {
1847 /* we have to stop all cpus to guaranntee there is no user
1848 of zonelist */
1849 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
1850 /* cpuset refresh routine should be here */
1851 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07001852 vm_total_pages = nr_free_pagecache_pages();
1853 printk("Built %i zonelists. Total pages: %ld\n",
1854 num_online_nodes(), vm_total_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855}
1856
1857/*
1858 * Helper functions to size the waitqueue hash table.
1859 * Essentially these want to choose hash table sizes sufficiently
1860 * large so that collisions trying to wait on pages are rare.
1861 * But in fact, the number of active page waitqueues on typical
1862 * systems is ridiculously low, less than 200. So this is even
1863 * conservative, even though it seems large.
1864 *
1865 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1866 * waitqueues, i.e. the size of the waitq table given the number of pages.
1867 */
1868#define PAGES_PER_WAITQUEUE 256
1869
Yasunori Gotocca448f2006-06-23 02:03:10 -07001870#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07001871static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872{
1873 unsigned long size = 1;
1874
1875 pages /= PAGES_PER_WAITQUEUE;
1876
1877 while (size < pages)
1878 size <<= 1;
1879
1880 /*
1881 * Once we have dozens or even hundreds of threads sleeping
1882 * on IO we've got bigger problems than wait queue collision.
1883 * Limit the size of the wait table to a reasonable size.
1884 */
1885 size = min(size, 4096UL);
1886
1887 return max(size, 4UL);
1888}
Yasunori Gotocca448f2006-06-23 02:03:10 -07001889#else
1890/*
1891 * A zone's size might be changed by hot-add, so it is not possible to determine
1892 * a suitable size for its wait_table. So we use the maximum size now.
1893 *
1894 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1895 *
1896 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1897 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1898 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1899 *
1900 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1901 * or more by the traditional way. (See above). It equals:
1902 *
1903 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1904 * ia64(16K page size) : = ( 8G + 4M)byte.
1905 * powerpc (64K page size) : = (32G +16M)byte.
1906 */
1907static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1908{
1909 return 4096UL;
1910}
1911#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912
1913/*
1914 * This is an integer logarithm so that shifts can be used later
1915 * to extract the more random high bits from the multiplicative
1916 * hash function before the remainder is taken.
1917 */
1918static inline unsigned long wait_table_bits(unsigned long size)
1919{
1920 return ffz(~size);
1921}
1922
1923#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1924
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925/*
1926 * Initially all pages are reserved - free ones are freed
1927 * up by free_all_bootmem() once the early boot process is
1928 * done. Non-atomic initialization, single-pass.
1929 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001930void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa022007-01-10 23:15:30 -08001931 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001934 unsigned long end_pfn = start_pfn + size;
1935 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001937 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa022007-01-10 23:15:30 -08001938 /*
1939 * There can be holes in boot-time mem_map[]s
1940 * handed to this function. They do not
1941 * exist on hotplugged memory.
1942 */
1943 if (context == MEMMAP_EARLY) {
1944 if (!early_pfn_valid(pfn))
1945 continue;
1946 if (!early_pfn_in_nid(pfn, nid))
1947 continue;
1948 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001949 page = pfn_to_page(pfn);
1950 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08001951 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 reset_page_mapcount(page);
1953 SetPageReserved(page);
1954 INIT_LIST_HEAD(&page->lru);
1955#ifdef WANT_PAGE_VIRTUAL
1956 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1957 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001958 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 }
1961}
1962
1963void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1964 unsigned long size)
1965{
1966 int order;
1967 for (order = 0; order < MAX_ORDER ; order++) {
1968 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1969 zone->free_area[order].nr_free = 0;
1970 }
1971}
1972
1973#ifndef __HAVE_ARCH_MEMMAP_INIT
1974#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa022007-01-10 23:15:30 -08001975 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976#endif
1977
Ashok Raj6292d9a2006-02-01 03:04:44 -08001978static int __cpuinit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001979{
1980 int batch;
1981
1982 /*
1983 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001984 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001985 *
1986 * OK, so we don't know how big the cache is. So guess.
1987 */
1988 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001989 if (batch * PAGE_SIZE > 512 * 1024)
1990 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001991 batch /= 4; /* We effectively *= 4 below */
1992 if (batch < 1)
1993 batch = 1;
1994
1995 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001996 * Clamp the batch to a 2^n - 1 value. Having a power
1997 * of 2 value was found to be more likely to have
1998 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001999 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002000 * For example if 2 tasks are alternately allocating
2001 * batches of pages, one task can end up with a lot
2002 * of pages of one half of the possible page colors
2003 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002004 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002005 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07002006
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002007 return batch;
2008}
2009
Christoph Lameter2caaad42005-06-21 17:15:00 -07002010inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2011{
2012 struct per_cpu_pages *pcp;
2013
Magnus Damm1c6fe942005-10-26 01:58:59 -07002014 memset(p, 0, sizeof(*p));
2015
Christoph Lameter2caaad42005-06-21 17:15:00 -07002016 pcp = &p->pcp[0]; /* hot */
2017 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002018 pcp->high = 6 * batch;
2019 pcp->batch = max(1UL, 1 * batch);
2020 INIT_LIST_HEAD(&pcp->list);
2021
2022 pcp = &p->pcp[1]; /* cold*/
2023 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002024 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07002025 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07002026 INIT_LIST_HEAD(&pcp->list);
2027}
2028
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002029/*
2030 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2031 * to the value high for the pageset p.
2032 */
2033
2034static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2035 unsigned long high)
2036{
2037 struct per_cpu_pages *pcp;
2038
2039 pcp = &p->pcp[0]; /* hot list */
2040 pcp->high = high;
2041 pcp->batch = max(1UL, high/4);
2042 if ((high/4) > (PAGE_SHIFT * 8))
2043 pcp->batch = PAGE_SHIFT * 8;
2044}
2045
2046
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002047#ifdef CONFIG_NUMA
2048/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07002049 * Boot pageset table. One per cpu which is going to be used for all
2050 * zones and all nodes. The parameters will be set in such a way
2051 * that an item put on a list will immediately be handed over to
2052 * the buddy list. This is safe since pageset manipulation is done
2053 * with interrupts disabled.
2054 *
2055 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07002056 *
2057 * The boot_pagesets must be kept even after bootup is complete for
2058 * unused processors and/or zones. They do play a role for bootstrapping
2059 * hotplugged processors.
2060 *
2061 * zoneinfo_show() and maybe other functions do
2062 * not check if the processor is online before following the pageset pointer.
2063 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07002064 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08002065static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07002066
2067/*
2068 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002069 * per cpu pageset array in struct zone.
2070 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08002071static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002072{
2073 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002074
2075 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002076
Christoph Lameter66a55032006-09-27 01:50:09 -07002077 if (!populated_zone(zone))
2078 continue;
2079
Nick Piggin23316bc2006-01-08 01:00:41 -08002080 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002081 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08002082 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002083 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002084
Nick Piggin23316bc2006-01-08 01:00:41 -08002085 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002086
2087 if (percpu_pagelist_fraction)
2088 setup_pagelist_highmark(zone_pcp(zone, cpu),
2089 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002090 }
2091
2092 return 0;
2093bad:
2094 for_each_zone(dzone) {
2095 if (dzone == zone)
2096 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08002097 kfree(zone_pcp(dzone, cpu));
2098 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002099 }
2100 return -ENOMEM;
2101}
2102
2103static inline void free_zone_pagesets(int cpu)
2104{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002105 struct zone *zone;
2106
2107 for_each_zone(zone) {
2108 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2109
David Rientjesf3ef9ea2006-09-25 16:24:57 -07002110 /* Free per_cpu_pageset if it is slab allocated */
2111 if (pset != &boot_pageset[cpu])
2112 kfree(pset);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002113 zone_pcp(zone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002114 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002115}
2116
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07002117static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002118 unsigned long action,
2119 void *hcpu)
2120{
2121 int cpu = (long)hcpu;
2122 int ret = NOTIFY_OK;
2123
2124 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08002125 case CPU_UP_PREPARE:
2126 if (process_zones(cpu))
2127 ret = NOTIFY_BAD;
2128 break;
2129 case CPU_UP_CANCELED:
2130 case CPU_DEAD:
2131 free_zone_pagesets(cpu);
2132 break;
2133 default:
2134 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002135 }
2136 return ret;
2137}
2138
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07002139static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002140 { &pageset_cpuup_callback, NULL, 0 };
2141
Al Viro78d99552005-12-15 09:18:25 +00002142void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002143{
2144 int err;
2145
2146 /* Initialize per_cpu_pageset for cpu 0.
2147 * A cpuup callback will do this for every cpu
2148 * as it comes online
2149 */
2150 err = process_zones(smp_processor_id());
2151 BUG_ON(err);
2152 register_cpu_notifier(&pageset_notifier);
2153}
2154
2155#endif
2156
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002157static __meminit
Yasunori Gotocca448f2006-06-23 02:03:10 -07002158int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07002159{
2160 int i;
2161 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002162 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07002163
2164 /*
2165 * The per-page waitqueue mechanism uses hashed waitqueues
2166 * per zone.
2167 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07002168 zone->wait_table_hash_nr_entries =
2169 wait_table_hash_nr_entries(zone_size_pages);
2170 zone->wait_table_bits =
2171 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002172 alloc_size = zone->wait_table_hash_nr_entries
2173 * sizeof(wait_queue_head_t);
2174
2175 if (system_state == SYSTEM_BOOTING) {
2176 zone->wait_table = (wait_queue_head_t *)
2177 alloc_bootmem_node(pgdat, alloc_size);
2178 } else {
2179 /*
2180 * This case means that a zone whose size was 0 gets new memory
2181 * via memory hot-add.
2182 * But it may be the case that a new node was hot-added. In
2183 * this case vmalloc() will not be able to use this new node's
2184 * memory - this wait_table must be initialized to use this new
2185 * node itself as well.
2186 * To use this new node's memory, further consideration will be
2187 * necessary.
2188 */
2189 zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size);
2190 }
2191 if (!zone->wait_table)
2192 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07002193
Yasunori Goto02b694d2006-06-23 02:03:08 -07002194 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07002195 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002196
2197 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002198}
2199
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002200static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07002201{
2202 int cpu;
2203 unsigned long batch = zone_batchsize(zone);
2204
2205 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2206#ifdef CONFIG_NUMA
2207 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002208 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002209 setup_pageset(&boot_pageset[cpu],0);
2210#else
2211 setup_pageset(zone_pcp(zone,cpu), batch);
2212#endif
2213 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08002214 if (zone->present_pages)
2215 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2216 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07002217}
2218
Yasunori Goto718127c2006-06-23 02:03:10 -07002219__meminit int init_currently_empty_zone(struct zone *zone,
2220 unsigned long zone_start_pfn,
Dave Hansena2f3aa022007-01-10 23:15:30 -08002221 unsigned long size,
2222 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07002223{
2224 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002225 int ret;
2226 ret = zone_wait_table_init(zone, size);
2227 if (ret)
2228 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07002229 pgdat->nr_zones = zone_idx(zone) + 1;
2230
Dave Hansened8ece22005-10-29 18:16:50 -07002231 zone->zone_start_pfn = zone_start_pfn;
2232
2233 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2234
2235 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
Yasunori Goto718127c2006-06-23 02:03:10 -07002236
2237 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002238}
2239
Mel Gormanc7132162006-09-27 01:49:43 -07002240#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2241/*
2242 * Basic iterator support. Return the first range of PFNs for a node
2243 * Note: nid == MAX_NUMNODES returns first region regardless of node
2244 */
2245static int __init first_active_region_index_in_nid(int nid)
2246{
2247 int i;
2248
2249 for (i = 0; i < nr_nodemap_entries; i++)
2250 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2251 return i;
2252
2253 return -1;
2254}
2255
2256/*
2257 * Basic iterator support. Return the next active range of PFNs for a node
2258 * Note: nid == MAX_NUMNODES returns next region regardles of node
2259 */
2260static int __init next_active_region_index_in_nid(int index, int nid)
2261{
2262 for (index = index + 1; index < nr_nodemap_entries; index++)
2263 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2264 return index;
2265
2266 return -1;
2267}
2268
2269#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2270/*
2271 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2272 * Architectures may implement their own version but if add_active_range()
2273 * was used and there are no special requirements, this is a convenient
2274 * alternative
2275 */
2276int __init early_pfn_to_nid(unsigned long pfn)
2277{
2278 int i;
2279
2280 for (i = 0; i < nr_nodemap_entries; i++) {
2281 unsigned long start_pfn = early_node_map[i].start_pfn;
2282 unsigned long end_pfn = early_node_map[i].end_pfn;
2283
2284 if (start_pfn <= pfn && pfn < end_pfn)
2285 return early_node_map[i].nid;
2286 }
2287
2288 return 0;
2289}
2290#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2291
2292/* Basic iterator support to walk early_node_map[] */
2293#define for_each_active_range_index_in_nid(i, nid) \
2294 for (i = first_active_region_index_in_nid(nid); i != -1; \
2295 i = next_active_region_index_in_nid(i, nid))
2296
2297/**
2298 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002299 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2300 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07002301 *
2302 * If an architecture guarantees that all ranges registered with
2303 * add_active_ranges() contain no holes and may be freed, this
2304 * this function may be used instead of calling free_bootmem() manually.
2305 */
2306void __init free_bootmem_with_active_regions(int nid,
2307 unsigned long max_low_pfn)
2308{
2309 int i;
2310
2311 for_each_active_range_index_in_nid(i, nid) {
2312 unsigned long size_pages = 0;
2313 unsigned long end_pfn = early_node_map[i].end_pfn;
2314
2315 if (early_node_map[i].start_pfn >= max_low_pfn)
2316 continue;
2317
2318 if (end_pfn > max_low_pfn)
2319 end_pfn = max_low_pfn;
2320
2321 size_pages = end_pfn - early_node_map[i].start_pfn;
2322 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2323 PFN_PHYS(early_node_map[i].start_pfn),
2324 size_pages << PAGE_SHIFT);
2325 }
2326}
2327
2328/**
2329 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002330 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07002331 *
2332 * If an architecture guarantees that all ranges registered with
2333 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002334 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07002335 */
2336void __init sparse_memory_present_with_active_regions(int nid)
2337{
2338 int i;
2339
2340 for_each_active_range_index_in_nid(i, nid)
2341 memory_present(early_node_map[i].nid,
2342 early_node_map[i].start_pfn,
2343 early_node_map[i].end_pfn);
2344}
2345
2346/**
Mel Gormanfb014392006-09-27 01:49:59 -07002347 * push_node_boundaries - Push node boundaries to at least the requested boundary
2348 * @nid: The nid of the node to push the boundary for
2349 * @start_pfn: The start pfn of the node
2350 * @end_pfn: The end pfn of the node
2351 *
2352 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2353 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2354 * be hotplugged even though no physical memory exists. This function allows
2355 * an arch to push out the node boundaries so mem_map is allocated that can
2356 * be used later.
2357 */
2358#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2359void __init push_node_boundaries(unsigned int nid,
2360 unsigned long start_pfn, unsigned long end_pfn)
2361{
2362 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2363 nid, start_pfn, end_pfn);
2364
2365 /* Initialise the boundary for this node if necessary */
2366 if (node_boundary_end_pfn[nid] == 0)
2367 node_boundary_start_pfn[nid] = -1UL;
2368
2369 /* Update the boundaries */
2370 if (node_boundary_start_pfn[nid] > start_pfn)
2371 node_boundary_start_pfn[nid] = start_pfn;
2372 if (node_boundary_end_pfn[nid] < end_pfn)
2373 node_boundary_end_pfn[nid] = end_pfn;
2374}
2375
2376/* If necessary, push the node boundary out for reserve hotadd */
2377static void __init account_node_boundary(unsigned int nid,
2378 unsigned long *start_pfn, unsigned long *end_pfn)
2379{
2380 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2381 nid, *start_pfn, *end_pfn);
2382
2383 /* Return if boundary information has not been provided */
2384 if (node_boundary_end_pfn[nid] == 0)
2385 return;
2386
2387 /* Check the boundaries and update if necessary */
2388 if (node_boundary_start_pfn[nid] < *start_pfn)
2389 *start_pfn = node_boundary_start_pfn[nid];
2390 if (node_boundary_end_pfn[nid] > *end_pfn)
2391 *end_pfn = node_boundary_end_pfn[nid];
2392}
2393#else
2394void __init push_node_boundaries(unsigned int nid,
2395 unsigned long start_pfn, unsigned long end_pfn) {}
2396
2397static void __init account_node_boundary(unsigned int nid,
2398 unsigned long *start_pfn, unsigned long *end_pfn) {}
2399#endif
2400
2401
2402/**
Mel Gormanc7132162006-09-27 01:49:43 -07002403 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002404 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2405 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2406 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07002407 *
2408 * It returns the start and end page frame of a node based on information
2409 * provided by an arch calling add_active_range(). If called for a node
2410 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002411 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07002412 */
2413void __init get_pfn_range_for_nid(unsigned int nid,
2414 unsigned long *start_pfn, unsigned long *end_pfn)
2415{
2416 int i;
2417 *start_pfn = -1UL;
2418 *end_pfn = 0;
2419
2420 for_each_active_range_index_in_nid(i, nid) {
2421 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
2422 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
2423 }
2424
2425 if (*start_pfn == -1UL) {
2426 printk(KERN_WARNING "Node %u active with no memory\n", nid);
2427 *start_pfn = 0;
2428 }
Mel Gormanfb014392006-09-27 01:49:59 -07002429
2430 /* Push the node boundaries out if requested */
2431 account_node_boundary(nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07002432}
2433
2434/*
2435 * Return the number of pages a zone spans in a node, including holes
2436 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2437 */
2438unsigned long __init zone_spanned_pages_in_node(int nid,
2439 unsigned long zone_type,
2440 unsigned long *ignored)
2441{
2442 unsigned long node_start_pfn, node_end_pfn;
2443 unsigned long zone_start_pfn, zone_end_pfn;
2444
2445 /* Get the start and end of the node and zone */
2446 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2447 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
2448 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
2449
2450 /* Check that this node has pages within the zone's required range */
2451 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
2452 return 0;
2453
2454 /* Move the zone boundaries inside the node if necessary */
2455 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
2456 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
2457
2458 /* Return the spanned pages */
2459 return zone_end_pfn - zone_start_pfn;
2460}
2461
2462/*
2463 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002464 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07002465 */
2466unsigned long __init __absent_pages_in_range(int nid,
2467 unsigned long range_start_pfn,
2468 unsigned long range_end_pfn)
2469{
2470 int i = 0;
2471 unsigned long prev_end_pfn = 0, hole_pages = 0;
2472 unsigned long start_pfn;
2473
2474 /* Find the end_pfn of the first active range of pfns in the node */
2475 i = first_active_region_index_in_nid(nid);
2476 if (i == -1)
2477 return 0;
2478
Mel Gorman9c7cd682006-09-27 01:49:58 -07002479 /* Account for ranges before physical memory on this node */
2480 if (early_node_map[i].start_pfn > range_start_pfn)
2481 hole_pages = early_node_map[i].start_pfn - range_start_pfn;
2482
Mel Gormanc7132162006-09-27 01:49:43 -07002483 prev_end_pfn = early_node_map[i].start_pfn;
2484
2485 /* Find all holes for the zone within the node */
2486 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
2487
2488 /* No need to continue if prev_end_pfn is outside the zone */
2489 if (prev_end_pfn >= range_end_pfn)
2490 break;
2491
2492 /* Make sure the end of the zone is not within the hole */
2493 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
2494 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
2495
2496 /* Update the hole size cound and move on */
2497 if (start_pfn > range_start_pfn) {
2498 BUG_ON(prev_end_pfn > start_pfn);
2499 hole_pages += start_pfn - prev_end_pfn;
2500 }
2501 prev_end_pfn = early_node_map[i].end_pfn;
2502 }
2503
Mel Gorman9c7cd682006-09-27 01:49:58 -07002504 /* Account for ranges past physical memory on this node */
2505 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07002506 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07002507 max(range_start_pfn, prev_end_pfn);
2508
Mel Gormanc7132162006-09-27 01:49:43 -07002509 return hole_pages;
2510}
2511
2512/**
2513 * absent_pages_in_range - Return number of page frames in holes within a range
2514 * @start_pfn: The start PFN to start searching for holes
2515 * @end_pfn: The end PFN to stop searching for holes
2516 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002517 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07002518 */
2519unsigned long __init absent_pages_in_range(unsigned long start_pfn,
2520 unsigned long end_pfn)
2521{
2522 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
2523}
2524
2525/* Return the number of page frames in holes in a zone on a node */
2526unsigned long __init zone_absent_pages_in_node(int nid,
2527 unsigned long zone_type,
2528 unsigned long *ignored)
2529{
Mel Gorman9c7cd682006-09-27 01:49:58 -07002530 unsigned long node_start_pfn, node_end_pfn;
2531 unsigned long zone_start_pfn, zone_end_pfn;
2532
2533 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2534 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
2535 node_start_pfn);
2536 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
2537 node_end_pfn);
2538
2539 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07002540}
Mel Gorman0e0b8642006-09-27 01:49:56 -07002541
Mel Gormanc7132162006-09-27 01:49:43 -07002542#else
2543static inline unsigned long zone_spanned_pages_in_node(int nid,
2544 unsigned long zone_type,
2545 unsigned long *zones_size)
2546{
2547 return zones_size[zone_type];
2548}
2549
2550static inline unsigned long zone_absent_pages_in_node(int nid,
2551 unsigned long zone_type,
2552 unsigned long *zholes_size)
2553{
2554 if (!zholes_size)
2555 return 0;
2556
2557 return zholes_size[zone_type];
2558}
Mel Gorman0e0b8642006-09-27 01:49:56 -07002559
Mel Gormanc7132162006-09-27 01:49:43 -07002560#endif
2561
2562static void __init calculate_node_totalpages(struct pglist_data *pgdat,
2563 unsigned long *zones_size, unsigned long *zholes_size)
2564{
2565 unsigned long realtotalpages, totalpages = 0;
2566 enum zone_type i;
2567
2568 for (i = 0; i < MAX_NR_ZONES; i++)
2569 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
2570 zones_size);
2571 pgdat->node_spanned_pages = totalpages;
2572
2573 realtotalpages = totalpages;
2574 for (i = 0; i < MAX_NR_ZONES; i++)
2575 realtotalpages -=
2576 zone_absent_pages_in_node(pgdat->node_id, i,
2577 zholes_size);
2578 pgdat->node_present_pages = realtotalpages;
2579 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
2580 realtotalpages);
2581}
2582
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583/*
2584 * Set up the zone data structures:
2585 * - mark all pages reserved
2586 * - mark all memory queues empty
2587 * - clear the memory bitmaps
2588 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07002589static void __meminit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 unsigned long *zones_size, unsigned long *zholes_size)
2591{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07002592 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07002593 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07002595 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596
Dave Hansen208d54e2005-10-29 18:16:52 -07002597 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 pgdat->nr_zones = 0;
2599 init_waitqueue_head(&pgdat->kswapd_wait);
2600 pgdat->kswapd_max_order = 0;
2601
2602 for (j = 0; j < MAX_NR_ZONES; j++) {
2603 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07002604 unsigned long size, realsize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605
Mel Gormanc7132162006-09-27 01:49:43 -07002606 size = zone_spanned_pages_in_node(nid, j, zones_size);
2607 realsize = size - zone_absent_pages_in_node(nid, j,
2608 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609
Mel Gorman0e0b8642006-09-27 01:49:56 -07002610 /*
2611 * Adjust realsize so that it accounts for how much memory
2612 * is used by this zone for memmap. This affects the watermark
2613 * and per-cpu initialisations
2614 */
2615 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
2616 if (realsize >= memmap_pages) {
2617 realsize -= memmap_pages;
2618 printk(KERN_DEBUG
2619 " %s zone: %lu pages used for memmap\n",
2620 zone_names[j], memmap_pages);
2621 } else
2622 printk(KERN_WARNING
2623 " %s zone: %lu pages exceeds realsize %lu\n",
2624 zone_names[j], memmap_pages, realsize);
2625
Christoph Lameter62672762007-02-10 01:43:07 -08002626 /* Account for reserved pages */
2627 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07002628 realsize -= dma_reserve;
Christoph Lameter62672762007-02-10 01:43:07 -08002629 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
2630 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07002631 }
2632
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002633 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 nr_kernel_pages += realsize;
2635 nr_all_pages += realsize;
2636
2637 zone->spanned_pages = size;
2638 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07002639#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07002640 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07002641 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07002642 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07002643 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07002644#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 zone->name = zone_names[j];
2646 spin_lock_init(&zone->lock);
2647 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002648 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650
Martin Bligh3bb1a8522006-10-28 10:38:24 -07002651 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652
Dave Hansened8ece22005-10-29 18:16:50 -07002653 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 INIT_LIST_HEAD(&zone->active_list);
2655 INIT_LIST_HEAD(&zone->inactive_list);
2656 zone->nr_scan_active = 0;
2657 zone->nr_scan_inactive = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07002658 zap_zone_vm_stats(zone);
Martin Hicks53e9a612005-09-03 15:54:51 -07002659 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 if (!size)
2661 continue;
2662
Dave Hansena2f3aa022007-01-10 23:15:30 -08002663 ret = init_currently_empty_zone(zone, zone_start_pfn,
2664 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07002665 BUG_ON(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 }
2668}
2669
2670static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2671{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 /* Skip empty nodes */
2673 if (!pgdat->node_spanned_pages)
2674 return;
2675
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002676#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 /* ia64 gets its own node_mem_map, before this, without bootmem */
2678 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07002679 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002680 struct page *map;
2681
Bob Piccoe984bb42006-05-20 15:00:31 -07002682 /*
2683 * The zone's endpoints aren't required to be MAX_ORDER
2684 * aligned but the node_mem_map endpoints must be in order
2685 * for the buddy allocator to function correctly.
2686 */
2687 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
2688 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
2689 end = ALIGN(end, MAX_ORDER_NR_PAGES);
2690 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002691 map = alloc_remap(pgdat->node_id, size);
2692 if (!map)
2693 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07002694 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002696#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 /*
2698 * With no DISCONTIG, the global mem_map is just set as node 0's
2699 */
Mel Gormanc7132162006-09-27 01:49:43 -07002700 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07002702#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2703 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
2704 mem_map -= pgdat->node_start_pfn;
2705#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2706 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002708#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709}
2710
Yasunori Goto86356ab2006-06-23 02:03:09 -07002711void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 unsigned long *zones_size, unsigned long node_start_pfn,
2713 unsigned long *zholes_size)
2714{
2715 pgdat->node_id = nid;
2716 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07002717 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718
2719 alloc_node_mem_map(pgdat);
2720
2721 free_area_init_core(pgdat, zones_size, zholes_size);
2722}
2723
Mel Gormanc7132162006-09-27 01:49:43 -07002724#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2725/**
2726 * add_active_range - Register a range of PFNs backed by physical memory
2727 * @nid: The node ID the range resides on
2728 * @start_pfn: The start PFN of the available physical memory
2729 * @end_pfn: The end PFN of the available physical memory
2730 *
2731 * These ranges are stored in an early_node_map[] and later used by
2732 * free_area_init_nodes() to calculate zone sizes and holes. If the
2733 * range spans a memory hole, it is up to the architecture to ensure
2734 * the memory is not freed by the bootmem allocator. If possible
2735 * the range being registered will be merged with existing ranges.
2736 */
2737void __init add_active_range(unsigned int nid, unsigned long start_pfn,
2738 unsigned long end_pfn)
2739{
2740 int i;
2741
2742 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
2743 "%d entries of %d used\n",
2744 nid, start_pfn, end_pfn,
2745 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
2746
2747 /* Merge with existing active regions if possible */
2748 for (i = 0; i < nr_nodemap_entries; i++) {
2749 if (early_node_map[i].nid != nid)
2750 continue;
2751
2752 /* Skip if an existing region covers this new one */
2753 if (start_pfn >= early_node_map[i].start_pfn &&
2754 end_pfn <= early_node_map[i].end_pfn)
2755 return;
2756
2757 /* Merge forward if suitable */
2758 if (start_pfn <= early_node_map[i].end_pfn &&
2759 end_pfn > early_node_map[i].end_pfn) {
2760 early_node_map[i].end_pfn = end_pfn;
2761 return;
2762 }
2763
2764 /* Merge backward if suitable */
2765 if (start_pfn < early_node_map[i].end_pfn &&
2766 end_pfn >= early_node_map[i].start_pfn) {
2767 early_node_map[i].start_pfn = start_pfn;
2768 return;
2769 }
2770 }
2771
2772 /* Check that early_node_map is large enough */
2773 if (i >= MAX_ACTIVE_REGIONS) {
2774 printk(KERN_CRIT "More than %d memory regions, truncating\n",
2775 MAX_ACTIVE_REGIONS);
2776 return;
2777 }
2778
2779 early_node_map[i].nid = nid;
2780 early_node_map[i].start_pfn = start_pfn;
2781 early_node_map[i].end_pfn = end_pfn;
2782 nr_nodemap_entries = i + 1;
2783}
2784
2785/**
2786 * shrink_active_range - Shrink an existing registered range of PFNs
2787 * @nid: The node id the range is on that should be shrunk
2788 * @old_end_pfn: The old end PFN of the range
2789 * @new_end_pfn: The new PFN of the range
2790 *
2791 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
2792 * The map is kept at the end physical page range that has already been
2793 * registered with add_active_range(). This function allows an arch to shrink
2794 * an existing registered range.
2795 */
2796void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
2797 unsigned long new_end_pfn)
2798{
2799 int i;
2800
2801 /* Find the old active region end and shrink */
2802 for_each_active_range_index_in_nid(i, nid)
2803 if (early_node_map[i].end_pfn == old_end_pfn) {
2804 early_node_map[i].end_pfn = new_end_pfn;
2805 break;
2806 }
2807}
2808
2809/**
2810 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002811 *
Mel Gormanc7132162006-09-27 01:49:43 -07002812 * During discovery, it may be found that a table like SRAT is invalid
2813 * and an alternative discovery method must be used. This function removes
2814 * all currently registered regions.
2815 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002816void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07002817{
2818 memset(early_node_map, 0, sizeof(early_node_map));
2819 nr_nodemap_entries = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07002820#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2821 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
2822 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
2823#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -07002824}
2825
2826/* Compare two active node_active_regions */
2827static int __init cmp_node_active_region(const void *a, const void *b)
2828{
2829 struct node_active_region *arange = (struct node_active_region *)a;
2830 struct node_active_region *brange = (struct node_active_region *)b;
2831
2832 /* Done this way to avoid overflows */
2833 if (arange->start_pfn > brange->start_pfn)
2834 return 1;
2835 if (arange->start_pfn < brange->start_pfn)
2836 return -1;
2837
2838 return 0;
2839}
2840
2841/* sort the node_map by start_pfn */
2842static void __init sort_node_map(void)
2843{
2844 sort(early_node_map, (size_t)nr_nodemap_entries,
2845 sizeof(struct node_active_region),
2846 cmp_node_active_region, NULL);
2847}
2848
Mel Gormana6af2bc2007-02-10 01:42:57 -08002849/* Find the lowest pfn for a node */
Mel Gormanc7132162006-09-27 01:49:43 -07002850unsigned long __init find_min_pfn_for_node(unsigned long nid)
2851{
2852 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08002853 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00002854
Mel Gormanc7132162006-09-27 01:49:43 -07002855 /* Assuming a sorted map, the first range found has the starting pfn */
2856 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08002857 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07002858
Mel Gormana6af2bc2007-02-10 01:42:57 -08002859 if (min_pfn == ULONG_MAX) {
2860 printk(KERN_WARNING
2861 "Could not find start_pfn for node %lu\n", nid);
2862 return 0;
2863 }
2864
2865 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07002866}
2867
2868/**
2869 * find_min_pfn_with_active_regions - Find the minimum PFN registered
2870 *
2871 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002872 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07002873 */
2874unsigned long __init find_min_pfn_with_active_regions(void)
2875{
2876 return find_min_pfn_for_node(MAX_NUMNODES);
2877}
2878
2879/**
2880 * find_max_pfn_with_active_regions - Find the maximum PFN registered
2881 *
2882 * It returns the maximum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002883 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07002884 */
2885unsigned long __init find_max_pfn_with_active_regions(void)
2886{
2887 int i;
2888 unsigned long max_pfn = 0;
2889
2890 for (i = 0; i < nr_nodemap_entries; i++)
2891 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
2892
2893 return max_pfn;
2894}
2895
2896/**
2897 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002898 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07002899 *
2900 * This will call free_area_init_node() for each active node in the system.
2901 * Using the page ranges provided by add_active_range(), the size of each
2902 * zone in each node and their holes is calculated. If the maximum PFN
2903 * between two adjacent zones match, it is assumed that the zone is empty.
2904 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
2905 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
2906 * starts where the previous one ended. For example, ZONE_DMA32 starts
2907 * at arch_max_dma_pfn.
2908 */
2909void __init free_area_init_nodes(unsigned long *max_zone_pfn)
2910{
2911 unsigned long nid;
2912 enum zone_type i;
2913
Mel Gormana6af2bc2007-02-10 01:42:57 -08002914 /* Sort early_node_map as initialisation assumes it is sorted */
2915 sort_node_map();
2916
Mel Gormanc7132162006-09-27 01:49:43 -07002917 /* Record where the zone boundaries are */
2918 memset(arch_zone_lowest_possible_pfn, 0,
2919 sizeof(arch_zone_lowest_possible_pfn));
2920 memset(arch_zone_highest_possible_pfn, 0,
2921 sizeof(arch_zone_highest_possible_pfn));
2922 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
2923 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
2924 for (i = 1; i < MAX_NR_ZONES; i++) {
2925 arch_zone_lowest_possible_pfn[i] =
2926 arch_zone_highest_possible_pfn[i-1];
2927 arch_zone_highest_possible_pfn[i] =
2928 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
2929 }
2930
Mel Gormanc7132162006-09-27 01:49:43 -07002931 /* Print out the zone ranges */
2932 printk("Zone PFN ranges:\n");
2933 for (i = 0; i < MAX_NR_ZONES; i++)
2934 printk(" %-8s %8lu -> %8lu\n",
2935 zone_names[i],
2936 arch_zone_lowest_possible_pfn[i],
2937 arch_zone_highest_possible_pfn[i]);
2938
2939 /* Print out the early_node_map[] */
2940 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
2941 for (i = 0; i < nr_nodemap_entries; i++)
2942 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
2943 early_node_map[i].start_pfn,
2944 early_node_map[i].end_pfn);
2945
2946 /* Initialise every node */
2947 for_each_online_node(nid) {
2948 pg_data_t *pgdat = NODE_DATA(nid);
2949 free_area_init_node(nid, pgdat, NULL,
2950 find_min_pfn_for_node(nid), NULL);
2951 }
2952}
2953#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2954
Mel Gorman0e0b8642006-09-27 01:49:56 -07002955/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002956 * set_dma_reserve - set the specified number of pages reserved in the first zone
2957 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07002958 *
2959 * The per-cpu batchsize and zone watermarks are determined by present_pages.
2960 * In the DMA zone, a significant percentage may be consumed by kernel image
2961 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002962 * function may optionally be used to account for unfreeable pages in the
2963 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
2964 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07002965 */
2966void __init set_dma_reserve(unsigned long new_dma_reserve)
2967{
2968 dma_reserve = new_dma_reserve;
2969}
2970
Dave Hansen93b75042005-06-23 00:07:47 -07002971#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972static bootmem_data_t contig_bootmem_data;
2973struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2974
2975EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002976#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977
2978void __init free_area_init(unsigned long *zones_size)
2979{
Dave Hansen93b75042005-06-23 00:07:47 -07002980 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2982}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984static int page_alloc_cpu_notify(struct notifier_block *self,
2985 unsigned long action, void *hcpu)
2986{
2987 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988
2989 if (action == CPU_DEAD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 local_irq_disable();
2991 __drain_pages(cpu);
Christoph Lameterf8891e52006-06-30 01:55:45 -07002992 vm_events_fold_cpu(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993 local_irq_enable();
Christoph Lameter2244b952006-06-30 01:55:33 -07002994 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 }
2996 return NOTIFY_OK;
2997}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998
2999void __init page_alloc_init(void)
3000{
3001 hotcpu_notifier(page_alloc_cpu_notify, 0);
3002}
3003
3004/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003005 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3006 * or min_free_kbytes changes.
3007 */
3008static void calculate_totalreserve_pages(void)
3009{
3010 struct pglist_data *pgdat;
3011 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003012 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003013
3014 for_each_online_pgdat(pgdat) {
3015 for (i = 0; i < MAX_NR_ZONES; i++) {
3016 struct zone *zone = pgdat->node_zones + i;
3017 unsigned long max = 0;
3018
3019 /* Find valid and maximum lowmem_reserve in the zone */
3020 for (j = i; j < MAX_NR_ZONES; j++) {
3021 if (zone->lowmem_reserve[j] > max)
3022 max = zone->lowmem_reserve[j];
3023 }
3024
3025 /* we treat pages_high as reserved pages. */
3026 max += zone->pages_high;
3027
3028 if (max > zone->present_pages)
3029 max = zone->present_pages;
3030 reserve_pages += max;
3031 }
3032 }
3033 totalreserve_pages = reserve_pages;
3034}
3035
3036/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037 * setup_per_zone_lowmem_reserve - called whenever
3038 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
3039 * has a correct pages reserved value, so an adequate number of
3040 * pages are left in the zone after a successful __alloc_pages().
3041 */
3042static void setup_per_zone_lowmem_reserve(void)
3043{
3044 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003045 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08003047 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048 for (j = 0; j < MAX_NR_ZONES; j++) {
3049 struct zone *zone = pgdat->node_zones + j;
3050 unsigned long present_pages = zone->present_pages;
3051
3052 zone->lowmem_reserve[j] = 0;
3053
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003054 idx = j;
3055 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056 struct zone *lower_zone;
3057
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003058 idx--;
3059
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 if (sysctl_lowmem_reserve_ratio[idx] < 1)
3061 sysctl_lowmem_reserve_ratio[idx] = 1;
3062
3063 lower_zone = pgdat->node_zones + idx;
3064 lower_zone->lowmem_reserve[j] = present_pages /
3065 sysctl_lowmem_reserve_ratio[idx];
3066 present_pages += lower_zone->present_pages;
3067 }
3068 }
3069 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003070
3071 /* update totalreserve_pages */
3072 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073}
3074
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003075/**
3076 * setup_per_zone_pages_min - called when min_free_kbytes changes.
3077 *
3078 * Ensures that the pages_{min,low,high} values for each zone are set correctly
3079 * with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080 */
Dave Hansen3947be12005-10-29 18:16:54 -07003081void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082{
3083 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
3084 unsigned long lowmem_pages = 0;
3085 struct zone *zone;
3086 unsigned long flags;
3087
3088 /* Calculate total number of !ZONE_HIGHMEM pages */
3089 for_each_zone(zone) {
3090 if (!is_highmem(zone))
3091 lowmem_pages += zone->present_pages;
3092 }
3093
3094 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07003095 u64 tmp;
3096
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07003098 tmp = (u64)pages_min * zone->present_pages;
3099 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100 if (is_highmem(zone)) {
3101 /*
Nick Piggin669ed172005-11-13 16:06:45 -08003102 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
3103 * need highmem pages, so cap pages_min to a small
3104 * value here.
3105 *
3106 * The (pages_high-pages_low) and (pages_low-pages_min)
3107 * deltas controls asynch page reclaim, and so should
3108 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109 */
3110 int min_pages;
3111
3112 min_pages = zone->present_pages / 1024;
3113 if (min_pages < SWAP_CLUSTER_MAX)
3114 min_pages = SWAP_CLUSTER_MAX;
3115 if (min_pages > 128)
3116 min_pages = 128;
3117 zone->pages_min = min_pages;
3118 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08003119 /*
3120 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121 * proportionate to the zone's size.
3122 */
Nick Piggin669ed172005-11-13 16:06:45 -08003123 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124 }
3125
Andrew Mortonac924c62006-05-15 09:43:59 -07003126 zone->pages_low = zone->pages_min + (tmp >> 2);
3127 zone->pages_high = zone->pages_min + (tmp >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128 spin_unlock_irqrestore(&zone->lru_lock, flags);
3129 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003130
3131 /* update totalreserve_pages */
3132 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133}
3134
3135/*
3136 * Initialise min_free_kbytes.
3137 *
3138 * For small machines we want it small (128k min). For large machines
3139 * we want it large (64MB max). But it is not linear, because network
3140 * bandwidth does not increase linearly with machine size. We use
3141 *
3142 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
3143 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
3144 *
3145 * which yields
3146 *
3147 * 16MB: 512k
3148 * 32MB: 724k
3149 * 64MB: 1024k
3150 * 128MB: 1448k
3151 * 256MB: 2048k
3152 * 512MB: 2896k
3153 * 1024MB: 4096k
3154 * 2048MB: 5792k
3155 * 4096MB: 8192k
3156 * 8192MB: 11584k
3157 * 16384MB: 16384k
3158 */
3159static int __init init_per_zone_pages_min(void)
3160{
3161 unsigned long lowmem_kbytes;
3162
3163 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
3164
3165 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
3166 if (min_free_kbytes < 128)
3167 min_free_kbytes = 128;
3168 if (min_free_kbytes > 65536)
3169 min_free_kbytes = 65536;
3170 setup_per_zone_pages_min();
3171 setup_per_zone_lowmem_reserve();
3172 return 0;
3173}
3174module_init(init_per_zone_pages_min)
3175
3176/*
3177 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
3178 * that we can call two helper functions whenever min_free_kbytes
3179 * changes.
3180 */
3181int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
3182 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3183{
3184 proc_dointvec(table, write, file, buffer, length, ppos);
3185 setup_per_zone_pages_min();
3186 return 0;
3187}
3188
Christoph Lameter96146342006-07-03 00:24:13 -07003189#ifdef CONFIG_NUMA
3190int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
3191 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3192{
3193 struct zone *zone;
3194 int rc;
3195
3196 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3197 if (rc)
3198 return rc;
3199
3200 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07003201 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07003202 sysctl_min_unmapped_ratio) / 100;
3203 return 0;
3204}
Christoph Lameter0ff38492006-09-25 23:31:52 -07003205
3206int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
3207 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3208{
3209 struct zone *zone;
3210 int rc;
3211
3212 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3213 if (rc)
3214 return rc;
3215
3216 for_each_zone(zone)
3217 zone->min_slab_pages = (zone->present_pages *
3218 sysctl_min_slab_ratio) / 100;
3219 return 0;
3220}
Christoph Lameter96146342006-07-03 00:24:13 -07003221#endif
3222
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223/*
3224 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
3225 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
3226 * whenever sysctl_lowmem_reserve_ratio changes.
3227 *
3228 * The reserve ratio obviously has absolutely no relation with the
3229 * pages_min watermarks. The lowmem reserve ratio can only make sense
3230 * if in function of the boot time zone sizes.
3231 */
3232int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
3233 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3234{
3235 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3236 setup_per_zone_lowmem_reserve();
3237 return 0;
3238}
3239
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003240/*
3241 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
3242 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
3243 * can have before it gets flushed back to buddy allocator.
3244 */
3245
3246int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
3247 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3248{
3249 struct zone *zone;
3250 unsigned int cpu;
3251 int ret;
3252
3253 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3254 if (!write || (ret == -EINVAL))
3255 return ret;
3256 for_each_zone(zone) {
3257 for_each_online_cpu(cpu) {
3258 unsigned long high;
3259 high = zone->present_pages / percpu_pagelist_fraction;
3260 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
3261 }
3262 }
3263 return 0;
3264}
3265
David S. Millerf034b5d2006-08-24 03:08:07 -07003266int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267
3268#ifdef CONFIG_NUMA
3269static int __init set_hashdist(char *str)
3270{
3271 if (!str)
3272 return 0;
3273 hashdist = simple_strtoul(str, &str, 0);
3274 return 1;
3275}
3276__setup("hashdist=", set_hashdist);
3277#endif
3278
3279/*
3280 * allocate a large system hash table from bootmem
3281 * - it is assumed that the hash table must contain an exact power-of-2
3282 * quantity of entries
3283 * - limit is the number of hash buckets, not the total allocation size
3284 */
3285void *__init alloc_large_system_hash(const char *tablename,
3286 unsigned long bucketsize,
3287 unsigned long numentries,
3288 int scale,
3289 int flags,
3290 unsigned int *_hash_shift,
3291 unsigned int *_hash_mask,
3292 unsigned long limit)
3293{
3294 unsigned long long max = limit;
3295 unsigned long log2qty, size;
3296 void *table = NULL;
3297
3298 /* allow the kernel cmdline to have a say */
3299 if (!numentries) {
3300 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08003301 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
3303 numentries >>= 20 - PAGE_SHIFT;
3304 numentries <<= 20 - PAGE_SHIFT;
3305
3306 /* limit to 1 bucket per 2^scale bytes of low memory */
3307 if (scale > PAGE_SHIFT)
3308 numentries >>= (scale - PAGE_SHIFT);
3309 else
3310 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08003311
3312 /* Make sure we've got at least a 0-order allocation.. */
3313 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
3314 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 }
John Hawkes6e692ed2006-03-25 03:08:02 -08003316 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317
3318 /* limit allocation size to 1/16 total memory by default */
3319 if (max == 0) {
3320 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
3321 do_div(max, bucketsize);
3322 }
3323
3324 if (numentries > max)
3325 numentries = max;
3326
David Howellsf0d1b0b2006-12-08 02:37:49 -08003327 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328
3329 do {
3330 size = bucketsize << log2qty;
3331 if (flags & HASH_EARLY)
3332 table = alloc_bootmem(size);
3333 else if (hashdist)
3334 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
3335 else {
3336 unsigned long order;
3337 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
3338 ;
3339 table = (void*) __get_free_pages(GFP_ATOMIC, order);
3340 }
3341 } while (!table && size > PAGE_SIZE && --log2qty);
3342
3343 if (!table)
3344 panic("Failed to allocate %s hash table\n", tablename);
3345
3346 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
3347 tablename,
3348 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08003349 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 size);
3351
3352 if (_hash_shift)
3353 *_hash_shift = log2qty;
3354 if (_hash_mask)
3355 *_hash_mask = (1 << log2qty) - 1;
3356
3357 return table;
3358}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003359
3360#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003361struct page *pfn_to_page(unsigned long pfn)
3362{
Andy Whitcroft67de6482006-06-23 02:03:12 -07003363 return __pfn_to_page(pfn);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003364}
3365unsigned long page_to_pfn(struct page *page)
3366{
Andy Whitcroft67de6482006-06-23 02:03:12 -07003367 return __page_to_pfn(page);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003368}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003369EXPORT_SYMBOL(pfn_to_page);
3370EXPORT_SYMBOL(page_to_pfn);
3371#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
Andrew Morton6220ec72006-10-19 23:29:05 -07003372
3373#if MAX_NUMNODES > 1
3374/*
3375 * Find the highest possible node id.
3376 */
3377int highest_possible_node_id(void)
3378{
3379 unsigned int node;
3380 unsigned int highest = 0;
3381
3382 for_each_node_mask(node, node_possible_map)
3383 highest = node;
3384 return highest;
3385}
3386EXPORT_SYMBOL(highest_possible_node_id);
3387#endif