blob: b55bb358b832975265043df8375dc523c2b3d807 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
44#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070045#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include "internal.h"
47
48/*
49 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
50 * initializer cleaner
51 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070052nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
Dean Nelson7223a932005-03-23 19:00:00 -070053EXPORT_SYMBOL(node_online_map);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070054nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
Dean Nelson7223a932005-03-23 19:00:00 -070055EXPORT_SYMBOL(node_possible_map);
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070056unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070057unsigned long totalreserve_pages __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058long nr_swap_pages;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080059int percpu_pagelist_fraction;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080061static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080062
Linus Torvalds1da177e2005-04-16 15:20:36 -070063/*
64 * results with 256, 32 in the lowmem_reserve sysctl:
65 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
66 * 1G machine -> (16M dma, 784M normal, 224M high)
67 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
68 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
69 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010070 *
71 * TBD: should special case ZONE_DMA32 machines here - in those we normally
72 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -070074int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
75 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -070076#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -070077 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -070078#endif
Christoph Lametere53ef382006-09-25 23:31:14 -070079#ifdef CONFIG_HIGHMEM
Christoph Lameter2f1b6242006-09-25 23:31:13 -070080 32
Christoph Lametere53ef382006-09-25 23:31:14 -070081#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -070082};
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
84EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
86/*
87 * Used by page_zone() to look up the address of the struct zone whose
88 * id is encoded in the upper bits of page->flags
89 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070090struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070091EXPORT_SYMBOL(zone_table);
92
Christoph Lameter2f1b6242006-09-25 23:31:13 -070093static char *zone_names[MAX_NR_ZONES] = {
94 "DMA",
Christoph Lameterfb0e7942006-09-25 23:31:13 -070095#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -070096 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -070097#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -070098 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -070099#ifdef CONFIG_HIGHMEM
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700100 "HighMem"
Christoph Lametere53ef382006-09-25 23:31:14 -0700101#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700102};
103
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104int min_free_kbytes = 1024;
105
Yasunori Goto86356ab2006-06-23 02:03:09 -0700106unsigned long __meminitdata nr_kernel_pages;
107unsigned long __meminitdata nr_all_pages;
Mel Gorman0e0b8642006-09-27 01:49:56 -0700108static unsigned long __initdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
Mel Gormanc7132162006-09-27 01:49:43 -0700110#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
111 /*
112 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
113 * ranges of memory (RAM) that may be registered with add_active_range().
114 * Ranges passed to add_active_range() will be merged if possible
115 * so the number of times add_active_range() can be called is
116 * related to the number of nodes and the number of holes
117 */
118 #ifdef CONFIG_MAX_ACTIVE_REGIONS
119 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
120 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
121 #else
122 #if MAX_NUMNODES >= 32
123 /* If there can be many nodes, allow up to 50 holes per node */
124 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
125 #else
126 /* By default, allow up to 256 distinct regions */
127 #define MAX_ACTIVE_REGIONS 256
128 #endif
129 #endif
130
131 struct node_active_region __initdata early_node_map[MAX_ACTIVE_REGIONS];
132 int __initdata nr_nodemap_entries;
133 unsigned long __initdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
134 unsigned long __initdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Mel Gormanfb014392006-09-27 01:49:59 -0700135#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
136 unsigned long __initdata node_boundary_start_pfn[MAX_NUMNODES];
137 unsigned long __initdata node_boundary_end_pfn[MAX_NUMNODES];
138#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -0700139#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
140
Nick Piggin13e74442006-01-06 00:10:58 -0800141#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700142static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700144 int ret = 0;
145 unsigned seq;
146 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700147
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700148 do {
149 seq = zone_span_seqbegin(zone);
150 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
151 ret = 1;
152 else if (pfn < zone->zone_start_pfn)
153 ret = 1;
154 } while (zone_span_seqretry(zone, seq));
155
156 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700157}
158
159static int page_is_consistent(struct zone *zone, struct page *page)
160{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161#ifdef CONFIG_HOLES_IN_ZONE
162 if (!pfn_valid(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700163 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164#endif
165 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700166 return 0;
167
168 return 1;
169}
170/*
171 * Temporary debugging check for pages not lying within a given zone.
172 */
173static int bad_range(struct zone *zone, struct page *page)
174{
175 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700177 if (!page_is_consistent(zone, page))
178 return 1;
179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 return 0;
181}
Nick Piggin13e74442006-01-06 00:10:58 -0800182#else
183static inline int bad_range(struct zone *zone, struct page *page)
184{
185 return 0;
186}
187#endif
188
Nick Piggin224abf92006-01-06 00:11:11 -0800189static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190{
Nick Piggin224abf92006-01-06 00:11:11 -0800191 printk(KERN_EMERG "Bad page state in process '%s'\n"
Hugh Dickins7365f3d2006-01-11 12:17:18 -0800192 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
193 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
194 KERN_EMERG "Backtrace:\n",
Nick Piggin224abf92006-01-06 00:11:11 -0800195 current->comm, page, (int)(2*sizeof(unsigned long)),
196 (unsigned long)page->flags, page->mapping,
197 page_mapcount(page), page_count(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 dump_stack();
Hugh Dickins334795e2005-06-21 17:15:08 -0700199 page->flags &= ~(1 << PG_lru |
200 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 1 << PG_active |
203 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700204 1 << PG_reclaim |
205 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 1 << PG_swapcache |
Nick Piggin676165a2006-04-10 11:21:48 +1000207 1 << PG_writeback |
208 1 << PG_buddy );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 set_page_count(page, 0);
210 reset_page_mapcount(page);
211 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700212 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213}
214
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215/*
216 * Higher-order pages are called "compound pages". They are structured thusly:
217 *
218 * The first PAGE_SIZE page is called the "head page".
219 *
220 * The remaining PAGE_SIZE pages are called "tail pages".
221 *
222 * All pages have PG_compound set. All pages have their ->private pointing at
223 * the head page (even the head page has this).
224 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800225 * The first tail page's ->lru.next holds the address of the compound page's
226 * put_page() function. Its ->lru.prev holds the order of allocation.
227 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800229
230static void free_compound_page(struct page *page)
231{
232 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
233}
234
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235static void prep_compound_page(struct page *page, unsigned long order)
236{
237 int i;
238 int nr_pages = 1 << order;
239
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800240 page[1].lru.next = (void *)free_compound_page; /* set dtor */
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800241 page[1].lru.prev = (void *)order;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 for (i = 0; i < nr_pages; i++) {
243 struct page *p = page + i;
244
Nick Piggin5e9dace2006-03-22 00:08:01 -0800245 __SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700246 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 }
248}
249
250static void destroy_compound_page(struct page *page, unsigned long order)
251{
252 int i;
253 int nr_pages = 1 << order;
254
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800255 if (unlikely((unsigned long)page[1].lru.prev != order))
Nick Piggin224abf92006-01-06 00:11:11 -0800256 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
258 for (i = 0; i < nr_pages; i++) {
259 struct page *p = page + i;
260
Nick Piggin224abf92006-01-06 00:11:11 -0800261 if (unlikely(!PageCompound(p) |
262 (page_private(p) != (unsigned long)page)))
263 bad_page(page);
Nick Piggin5e9dace2006-03-22 00:08:01 -0800264 __ClearPageCompound(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 }
266}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Nick Piggin17cf4402006-03-22 00:08:41 -0800268static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
269{
270 int i;
271
Nick Piggin725d7042006-09-25 23:30:55 -0700272 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
Andrew Morton6626c5d2006-03-22 00:08:42 -0800273 /*
274 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
275 * and __GFP_HIGHMEM from hard or soft interrupt context.
276 */
Nick Piggin725d7042006-09-25 23:30:55 -0700277 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800278 for (i = 0; i < (1 << order); i++)
279 clear_highpage(page + i);
280}
281
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282/*
283 * function for dealing with page's order in buddy system.
284 * zone->lock is already acquired when we use these.
285 * So, we don't need atomic page->flags operations here.
286 */
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700287static inline unsigned long page_order(struct page *page)
288{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700289 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290}
291
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700292static inline void set_page_order(struct page *page, int order)
293{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700294 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000295 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296}
297
298static inline void rmv_page_order(struct page *page)
299{
Nick Piggin676165a2006-04-10 11:21:48 +1000300 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700301 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302}
303
304/*
305 * Locate the struct page for both the matching buddy in our
306 * pair (buddy1) and the combined O(n+1) page they form (page).
307 *
308 * 1) Any buddy B1 will have an order O twin B2 which satisfies
309 * the following equation:
310 * B2 = B1 ^ (1 << O)
311 * For example, if the starting buddy (buddy2) is #8 its order
312 * 1 buddy is #10:
313 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
314 *
315 * 2) Any buddy B will have an order O+1 parent P which
316 * satisfies the following equation:
317 * P = B & ~(1 << O)
318 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200319 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 */
321static inline struct page *
322__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
323{
324 unsigned long buddy_idx = page_idx ^ (1 << order);
325
326 return page + (buddy_idx - page_idx);
327}
328
329static inline unsigned long
330__find_combined_index(unsigned long page_idx, unsigned int order)
331{
332 return (page_idx & ~(1 << order));
333}
334
335/*
336 * This function checks whether a page is free && is the buddy
337 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800338 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000339 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700340 * (c) a page and its buddy have the same order &&
341 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 *
Nick Piggin676165a2006-04-10 11:21:48 +1000343 * For recording whether a page is in the buddy system, we use PG_buddy.
344 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
345 *
346 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700348static inline int page_is_buddy(struct page *page, struct page *buddy,
349 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350{
Nick Piggin13e74442006-01-06 00:10:58 -0800351#ifdef CONFIG_HOLES_IN_ZONE
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700352 if (!pfn_valid(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800353 return 0;
354#endif
355
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700356 if (page_zone_id(page) != page_zone_id(buddy))
357 return 0;
358
359 if (PageBuddy(buddy) && page_order(buddy) == order) {
360 BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700361 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000362 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700363 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364}
365
366/*
367 * Freeing function for a buddy system allocator.
368 *
369 * The concept of a buddy system is to maintain direct-mapped table
370 * (containing bit values) for memory blocks of various "orders".
371 * The bottom level table contains the map for the smallest allocatable
372 * units of memory (here, pages), and each level above it describes
373 * pairs of units from the levels below, hence, "buddies".
374 * At a high level, all that happens here is marking the table entry
375 * at the bottom level available, and propagating the changes upward
376 * as necessary, plus some accounting needed to play nicely with other
377 * parts of the VM system.
378 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000379 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700380 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 * So when we are allocating or freeing one, we can derive the state of the
382 * other. That is, if we allocate a small block, and both were
383 * free, the remainder of the region must be split into blocks.
384 * If a block is freed, and its buddy is also free, then this
385 * triggers coalescing into a block of larger size.
386 *
387 * -- wli
388 */
389
Nick Piggin48db57f2006-01-08 01:00:42 -0800390static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 struct zone *zone, unsigned int order)
392{
393 unsigned long page_idx;
394 int order_size = 1 << order;
395
Nick Piggin224abf92006-01-06 00:11:11 -0800396 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 destroy_compound_page(page, order);
398
399 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
400
Nick Piggin725d7042006-09-25 23:30:55 -0700401 VM_BUG_ON(page_idx & (order_size - 1));
402 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403
404 zone->free_pages += order_size;
405 while (order < MAX_ORDER-1) {
406 unsigned long combined_idx;
407 struct free_area *area;
408 struct page *buddy;
409
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700411 if (!page_is_buddy(page, buddy, order))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800413
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 list_del(&buddy->lru);
415 area = zone->free_area + order;
416 area->nr_free--;
417 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800418 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 page = page + (combined_idx - page_idx);
420 page_idx = combined_idx;
421 order++;
422 }
423 set_page_order(page, order);
424 list_add(&page->lru, &zone->free_area[order].free_list);
425 zone->free_area[order].nr_free++;
426}
427
Nick Piggin224abf92006-01-06 00:11:11 -0800428static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429{
Nick Piggin92be2e332006-01-06 00:10:57 -0800430 if (unlikely(page_mapcount(page) |
431 (page->mapping != NULL) |
432 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 (page->flags & (
434 1 << PG_lru |
435 1 << PG_private |
436 1 << PG_locked |
437 1 << PG_active |
438 1 << PG_reclaim |
439 1 << PG_slab |
440 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700441 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000442 1 << PG_reserved |
443 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800444 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700446 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800447 /*
448 * For now, we report if PG_reserved was found set, but do not
449 * clear it, and do not free the page. But we shall soon need
450 * to do more, for when the ZERO_PAGE count wraps negative.
451 */
452 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453}
454
455/*
456 * Frees a list of pages.
457 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700458 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 *
460 * If the zone was previously in an "all pages pinned" state then look to
461 * see if this freeing clears that state.
462 *
463 * And clear the zone's pages_scanned counter, to hold off the "all pages are
464 * pinned" detection logic.
465 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800466static void free_pages_bulk(struct zone *zone, int count,
467 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468{
Nick Pigginc54ad302006-01-06 00:10:56 -0800469 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 zone->all_unreclaimable = 0;
471 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800472 while (count--) {
473 struct page *page;
474
Nick Piggin725d7042006-09-25 23:30:55 -0700475 VM_BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800477 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800479 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800481 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
483
Nick Piggin48db57f2006-01-08 01:00:42 -0800484static void free_one_page(struct zone *zone, struct page *page, int order)
485{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700486 spin_lock(&zone->lock);
487 zone->all_unreclaimable = 0;
488 zone->pages_scanned = 0;
489 __free_one_page(page, zone ,order);
490 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800491}
492
493static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494{
Nick Pigginc54ad302006-01-06 00:10:56 -0800495 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800497 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800500 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800501 if (reserved)
502 return;
503
Nick Piggin9858db52006-10-11 01:21:30 -0700504 if (!PageHighMem(page))
505 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Nick Piggindafb1362006-10-11 01:21:30 -0700506 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800507 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700508
Nick Pigginc54ad302006-01-06 00:10:56 -0800509 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700510 __count_vm_events(PGFREE, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800511 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800512 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513}
514
David Howellsa226f6c2006-01-06 00:11:08 -0800515/*
516 * permit the bootmem allocator to evade page validation on high-order frees
517 */
518void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
519{
520 if (order == 0) {
521 __ClearPageReserved(page);
522 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800523 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800524 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800525 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800526 int loop;
527
Nick Piggin545b1ea2006-03-22 00:08:07 -0800528 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800529 for (loop = 0; loop < BITS_PER_LONG; loop++) {
530 struct page *p = &page[loop];
531
Nick Piggin545b1ea2006-03-22 00:08:07 -0800532 if (loop + 1 < BITS_PER_LONG)
533 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800534 __ClearPageReserved(p);
535 set_page_count(p, 0);
536 }
537
Nick Piggin7835e982006-03-22 00:08:40 -0800538 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800539 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800540 }
541}
542
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
544/*
545 * The order of subdivision here is critical for the IO subsystem.
546 * Please do not alter this order without good reasons and regression
547 * testing. Specifically, as large blocks of memory are subdivided,
548 * the order in which smaller blocks are delivered depends on the order
549 * they're subdivided in this function. This is the primary factor
550 * influencing the order in which pages are delivered to the IO
551 * subsystem according to empirical testing, and this is also justified
552 * by considering the behavior of a buddy system containing a single
553 * large block of memory acted on by a series of small allocations.
554 * This behavior is a critical factor in sglist merging's success.
555 *
556 * -- wli
557 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800558static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 int low, int high, struct free_area *area)
560{
561 unsigned long size = 1 << high;
562
563 while (high > low) {
564 area--;
565 high--;
566 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700567 VM_BUG_ON(bad_range(zone, &page[size]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 list_add(&page[size].lru, &area->free_list);
569 area->nr_free++;
570 set_page_order(&page[size], high);
571 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572}
573
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574/*
575 * This page is about to be returned from the page allocator
576 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800577static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578{
Nick Piggin92be2e332006-01-06 00:10:57 -0800579 if (unlikely(page_mapcount(page) |
580 (page->mapping != NULL) |
581 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700582 (page->flags & (
583 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 1 << PG_private |
585 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 1 << PG_active |
587 1 << PG_dirty |
588 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700589 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700591 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000592 1 << PG_reserved |
593 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800594 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
Hugh Dickins689bceb2005-11-21 21:32:20 -0800596 /*
597 * For now, we report if PG_reserved was found set, but do not
598 * clear it, and do not allocate the page: as a safety net.
599 */
600 if (PageReserved(page))
601 return 1;
602
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
604 1 << PG_referenced | 1 << PG_arch_1 |
605 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700606 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800607 set_page_refcounted(page);
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--;
638 zone->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) {
693 local_irq_save(flags);
694 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
695 pcp->count = 0;
696 local_irq_restore(flags);
697 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700698 }
699 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700700}
701#endif
702
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
704static void __drain_pages(unsigned int cpu)
705{
Nick Pigginc54ad302006-01-06 00:10:56 -0800706 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 struct zone *zone;
708 int i;
709
710 for_each_zone(zone) {
711 struct per_cpu_pageset *pset;
712
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700713 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
715 struct per_cpu_pages *pcp;
716
717 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800718 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800719 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
720 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800721 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 }
723 }
724}
725#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
726
727#ifdef CONFIG_PM
728
729void mark_free_pages(struct zone *zone)
730{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700731 unsigned long pfn, max_zone_pfn;
732 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 int order;
734 struct list_head *curr;
735
736 if (!zone->spanned_pages)
737 return;
738
739 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700740
741 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
742 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
743 if (pfn_valid(pfn)) {
744 struct page *page = pfn_to_page(pfn);
745
746 if (!PageNosave(page))
747 ClearPageNosaveFree(page);
748 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
750 for (order = MAX_ORDER - 1; order >= 0; --order)
751 list_for_each(curr, &zone->free_area[order].free_list) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700752 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700754 pfn = page_to_pfn(list_entry(curr, struct page, lru));
755 for (i = 0; i < (1UL << order); i++)
756 SetPageNosaveFree(pfn_to_page(pfn + i));
757 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 spin_unlock_irqrestore(&zone->lock, flags);
760}
761
762/*
763 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
764 */
765void drain_local_pages(void)
766{
767 unsigned long flags;
768
769 local_irq_save(flags);
770 __drain_pages(smp_processor_id());
771 local_irq_restore(flags);
772}
773#endif /* CONFIG_PM */
774
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775/*
776 * Free a 0-order page
777 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778static void fastcall free_hot_cold_page(struct page *page, int cold)
779{
780 struct zone *zone = page_zone(page);
781 struct per_cpu_pages *pcp;
782 unsigned long flags;
783
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 if (PageAnon(page))
785 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800786 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800787 return;
788
Nick Piggin9858db52006-10-11 01:21:30 -0700789 if (!PageHighMem(page))
790 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Nick Piggindafb1362006-10-11 01:21:30 -0700791 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800792 kernel_map_pages(page, 1, 0);
793
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700794 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700796 __count_vm_event(PGFREE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 list_add(&page->lru, &pcp->list);
798 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800799 if (pcp->count >= pcp->high) {
800 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
801 pcp->count -= pcp->batch;
802 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 local_irq_restore(flags);
804 put_cpu();
805}
806
807void fastcall free_hot_page(struct page *page)
808{
809 free_hot_cold_page(page, 0);
810}
811
812void fastcall free_cold_page(struct page *page)
813{
814 free_hot_cold_page(page, 1);
815}
816
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800817/*
818 * split_page takes a non-compound higher-order page, and splits it into
819 * n (1<<order) sub-pages: page[0..n]
820 * Each sub-page must be freed individually.
821 *
822 * Note: this is probably too low level an operation for use in drivers.
823 * Please consult with lkml before using this in your driver.
824 */
825void split_page(struct page *page, unsigned int order)
826{
827 int i;
828
Nick Piggin725d7042006-09-25 23:30:55 -0700829 VM_BUG_ON(PageCompound(page));
830 VM_BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -0800831 for (i = 1; i < (1 << order); i++)
832 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800833}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800834
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835/*
836 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
837 * we cheat by calling it from here, in the order > 0 path. Saves a branch
838 * or two.
839 */
Nick Piggina74609f2006-01-06 00:11:20 -0800840static struct page *buffered_rmqueue(struct zonelist *zonelist,
841 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842{
843 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800844 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800846 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
Hugh Dickins689bceb2005-11-21 21:32:20 -0800848again:
Nick Piggina74609f2006-01-06 00:11:20 -0800849 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800850 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 struct per_cpu_pages *pcp;
852
Nick Piggina74609f2006-01-06 00:11:20 -0800853 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800855 if (!pcp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 pcp->count += rmqueue_bulk(zone, 0,
857 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800858 if (unlikely(!pcp->count))
859 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 }
Nick Piggina74609f2006-01-06 00:11:20 -0800861 page = list_entry(pcp->list.next, struct page, lru);
862 list_del(&page->lru);
863 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800864 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 spin_lock_irqsave(&zone->lock, flags);
866 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800867 spin_unlock(&zone->lock);
868 if (!page)
869 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 }
871
Christoph Lameterf8891e52006-06-30 01:55:45 -0700872 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Christoph Lameterca889e62006-06-30 01:55:44 -0700873 zone_statistics(zonelist, zone);
Nick Piggina74609f2006-01-06 00:11:20 -0800874 local_irq_restore(flags);
875 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876
Nick Piggin725d7042006-09-25 23:30:55 -0700877 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -0800878 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -0800879 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800881
882failed:
883 local_irq_restore(flags);
884 put_cpu();
885 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886}
887
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800888#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800889#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
890#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
891#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
892#define ALLOC_HARDER 0x10 /* try to alloc harder */
893#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
894#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800895
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896/*
897 * Return 1 if free pages are above 'mark'. This takes into account the order
898 * of the allocation.
899 */
900int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800901 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902{
903 /* free_pages my go negative - that's OK */
Nick Piggine80ee882006-10-04 02:15:23 -0700904 unsigned long min = mark;
905 long free_pages = z->free_pages - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 int o;
907
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800908 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800910 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 min -= min / 4;
912
913 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
914 return 0;
915 for (o = 0; o < order; o++) {
916 /* At the next order, this order's pages become unavailable */
917 free_pages -= z->free_area[o].nr_free << o;
918
919 /* Require fewer higher order pages to be free */
920 min >>= 1;
921
922 if (free_pages <= min)
923 return 0;
924 }
925 return 1;
926}
927
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800928/*
929 * get_page_from_freeliest goes through the zonelist trying to allocate
930 * a page.
931 */
932static struct page *
933get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
934 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -0700935{
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800936 struct zone **z = zonelist->zones;
937 struct page *page = NULL;
938 int classzone_idx = zone_idx(*z);
Christoph Lameter1192d522006-09-25 23:31:45 -0700939 struct zone *zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800940
941 /*
942 * Go through the zonelist once, looking for a zone with enough free.
943 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
944 */
945 do {
Christoph Lameter1192d522006-09-25 23:31:45 -0700946 zone = *z;
Christoph Lameter08e0f6a2006-09-27 01:50:06 -0700947 if (unlikely(NUMA_BUILD && (gfp_mask & __GFP_THISNODE) &&
Christoph Lameter1192d522006-09-25 23:31:45 -0700948 zone->zone_pgdat != zonelist->zones[0]->zone_pgdat))
Christoph Lameter9b819d22006-09-25 23:31:40 -0700949 break;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800950 if ((alloc_flags & ALLOC_CPUSET) &&
Christoph Lameter1192d522006-09-25 23:31:45 -0700951 !cpuset_zone_allowed(zone, gfp_mask))
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800952 continue;
953
954 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -0800955 unsigned long mark;
956 if (alloc_flags & ALLOC_WMARK_MIN)
Christoph Lameter1192d522006-09-25 23:31:45 -0700957 mark = zone->pages_min;
Nick Piggin31488902005-11-28 13:44:03 -0800958 else if (alloc_flags & ALLOC_WMARK_LOW)
Christoph Lameter1192d522006-09-25 23:31:45 -0700959 mark = zone->pages_low;
Nick Piggin31488902005-11-28 13:44:03 -0800960 else
Christoph Lameter1192d522006-09-25 23:31:45 -0700961 mark = zone->pages_high;
962 if (!zone_watermark_ok(zone , order, mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800963 classzone_idx, alloc_flags))
Christoph Lameter9eeff232006-01-18 17:42:31 -0800964 if (!zone_reclaim_mode ||
Christoph Lameter1192d522006-09-25 23:31:45 -0700965 !zone_reclaim(zone, gfp_mask, order))
Christoph Lameter9eeff232006-01-18 17:42:31 -0800966 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800967 }
968
Christoph Lameter1192d522006-09-25 23:31:45 -0700969 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800970 if (page) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800971 break;
972 }
973 } while (*(++z) != NULL);
974 return page;
Martin Hicks753ee722005-06-21 17:14:41 -0700975}
976
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977/*
978 * This is the 'heart' of the zoned buddy allocator.
979 */
980struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +0100981__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 struct zonelist *zonelist)
983{
Al Viro260b2362005-10-21 03:22:44 -0400984 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800985 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 struct page *page;
987 struct reclaim_state reclaim_state;
988 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800990 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 int did_some_progress;
992
993 might_sleep_if(wait);
994
Jens Axboe6b1de912005-11-17 21:35:02 +0100995restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800996 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800998 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 /* Should this ever happen?? */
1000 return NULL;
1001 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001002
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001003 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001004 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001005 if (page)
1006 goto got_pg;
1007
Jens Axboe6b1de912005-11-17 21:35:02 +01001008 do {
Chris Wright43b0bc02006-06-25 05:47:55 -07001009 wakeup_kswapd(*z, order);
Jens Axboe6b1de912005-11-17 21:35:02 +01001010 } while (*(++z));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001011
Paul Jackson9bf22292005-09-06 15:18:12 -07001012 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001013 * OK, we're below the kswapd watermark and have kicked background
1014 * reclaim. Now things get more complex, so set up alloc_flags according
1015 * to how we want to proceed.
1016 *
1017 * The caller may dip into page reserves a bit more if the caller
1018 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -08001019 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1020 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -07001021 */
Nick Piggin31488902005-11-28 13:44:03 -08001022 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001023 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1024 alloc_flags |= ALLOC_HARDER;
1025 if (gfp_mask & __GFP_HIGH)
1026 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -07001027 if (wait)
1028 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
1030 /*
1031 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001032 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 *
1034 * This is the last chance, in general, before the goto nopage.
1035 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -07001036 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001038 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1039 if (page)
1040 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
1042 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -07001043
1044 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1045 && !in_interrupt()) {
1046 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -08001047nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001048 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001049 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -08001050 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001051 if (page)
1052 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -08001053 if (gfp_mask & __GFP_NOFAIL) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001054 congestion_wait(WRITE, HZ/50);
Kirill Korotaev885036d2005-11-13 16:06:41 -08001055 goto nofail_alloc;
1056 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 }
1058 goto nopage;
1059 }
1060
1061 /* Atomic allocations - we can't balance anything */
1062 if (!wait)
1063 goto nopage;
1064
1065rebalance:
1066 cond_resched();
1067
1068 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001069 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 p->flags |= PF_MEMALLOC;
1071 reclaim_state.reclaimed_slab = 0;
1072 p->reclaim_state = &reclaim_state;
1073
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001074 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
1076 p->reclaim_state = NULL;
1077 p->flags &= ~PF_MEMALLOC;
1078
1079 cond_resched();
1080
1081 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001082 page = get_page_from_freelist(gfp_mask, order,
1083 zonelist, alloc_flags);
1084 if (page)
1085 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1087 /*
1088 * Go through the zonelist yet one more time, keep
1089 * very high watermark here, this is only to catch
1090 * a parallel oom killing, we must fail if we're still
1091 * under heavy pressure.
1092 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001093 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001094 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001095 if (page)
1096 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001098 out_of_memory(zonelist, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 goto restart;
1100 }
1101
1102 /*
1103 * Don't let big-order allocations loop unless the caller explicitly
1104 * requests that. Wait for some write requests to complete then retry.
1105 *
1106 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1107 * <= 3, but that may not be true in other implementations.
1108 */
1109 do_retry = 0;
1110 if (!(gfp_mask & __GFP_NORETRY)) {
1111 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1112 do_retry = 1;
1113 if (gfp_mask & __GFP_NOFAIL)
1114 do_retry = 1;
1115 }
1116 if (do_retry) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001117 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 goto rebalance;
1119 }
1120
1121nopage:
1122 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1123 printk(KERN_WARNING "%s: page allocation failure."
1124 " order:%d, mode:0x%x\n",
1125 p->comm, order, gfp_mask);
1126 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001127 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 return page;
1131}
1132
1133EXPORT_SYMBOL(__alloc_pages);
1134
1135/*
1136 * Common helper functions.
1137 */
Al Virodd0fc662005-10-07 07:46:04 +01001138fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139{
1140 struct page * page;
1141 page = alloc_pages(gfp_mask, order);
1142 if (!page)
1143 return 0;
1144 return (unsigned long) page_address(page);
1145}
1146
1147EXPORT_SYMBOL(__get_free_pages);
1148
Al Virodd0fc662005-10-07 07:46:04 +01001149fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150{
1151 struct page * page;
1152
1153 /*
1154 * get_zeroed_page() returns a 32-bit address, which cannot represent
1155 * a highmem page
1156 */
Nick Piggin725d7042006-09-25 23:30:55 -07001157 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158
1159 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1160 if (page)
1161 return (unsigned long) page_address(page);
1162 return 0;
1163}
1164
1165EXPORT_SYMBOL(get_zeroed_page);
1166
1167void __pagevec_free(struct pagevec *pvec)
1168{
1169 int i = pagevec_count(pvec);
1170
1171 while (--i >= 0)
1172 free_hot_cold_page(pvec->pages[i], pvec->cold);
1173}
1174
1175fastcall void __free_pages(struct page *page, unsigned int order)
1176{
Nick Pigginb5810032005-10-29 18:16:12 -07001177 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 if (order == 0)
1179 free_hot_page(page);
1180 else
1181 __free_pages_ok(page, order);
1182 }
1183}
1184
1185EXPORT_SYMBOL(__free_pages);
1186
1187fastcall void free_pages(unsigned long addr, unsigned int order)
1188{
1189 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001190 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 __free_pages(virt_to_page((void *)addr), order);
1192 }
1193}
1194
1195EXPORT_SYMBOL(free_pages);
1196
1197/*
1198 * Total amount of free (allocatable) RAM:
1199 */
1200unsigned int nr_free_pages(void)
1201{
1202 unsigned int sum = 0;
1203 struct zone *zone;
1204
1205 for_each_zone(zone)
1206 sum += zone->free_pages;
1207
1208 return sum;
1209}
1210
1211EXPORT_SYMBOL(nr_free_pages);
1212
1213#ifdef CONFIG_NUMA
1214unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1215{
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001216 unsigned int sum = 0;
1217 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218
1219 for (i = 0; i < MAX_NR_ZONES; i++)
1220 sum += pgdat->node_zones[i].free_pages;
1221
1222 return sum;
1223}
1224#endif
1225
1226static unsigned int nr_free_zone_pages(int offset)
1227{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001228 /* Just pick one node, since fallback list is circular */
1229 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 unsigned int sum = 0;
1231
Martin J. Blighe310fd42005-07-29 22:59:18 -07001232 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1233 struct zone **zonep = zonelist->zones;
1234 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235
Martin J. Blighe310fd42005-07-29 22:59:18 -07001236 for (zone = *zonep++; zone; zone = *zonep++) {
1237 unsigned long size = zone->present_pages;
1238 unsigned long high = zone->pages_high;
1239 if (size > high)
1240 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 }
1242
1243 return sum;
1244}
1245
1246/*
1247 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1248 */
1249unsigned int nr_free_buffer_pages(void)
1250{
Al Viroaf4ca452005-10-21 02:55:38 -04001251 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252}
1253
1254/*
1255 * Amount of free RAM allocatable within all zones
1256 */
1257unsigned int nr_free_pagecache_pages(void)
1258{
Al Viroaf4ca452005-10-21 02:55:38 -04001259 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001261
1262static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001264 if (NUMA_BUILD)
1265 printk("Node %ld ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268void si_meminfo(struct sysinfo *val)
1269{
1270 val->totalram = totalram_pages;
1271 val->sharedram = 0;
1272 val->freeram = nr_free_pages();
1273 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 val->totalhigh = totalhigh_pages;
1275 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 val->mem_unit = PAGE_SIZE;
1277}
1278
1279EXPORT_SYMBOL(si_meminfo);
1280
1281#ifdef CONFIG_NUMA
1282void si_meminfo_node(struct sysinfo *val, int nid)
1283{
1284 pg_data_t *pgdat = NODE_DATA(nid);
1285
1286 val->totalram = pgdat->node_present_pages;
1287 val->freeram = nr_free_pages_pgdat(pgdat);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001288#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1290 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001291#else
1292 val->totalhigh = 0;
1293 val->freehigh = 0;
1294#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 val->mem_unit = PAGE_SIZE;
1296}
1297#endif
1298
1299#define K(x) ((x) << (PAGE_SHIFT-10))
1300
1301/*
1302 * Show free area list (used inside shift_scroll-lock stuff)
1303 * We also calculate the percentage fragmentation. We do this by counting the
1304 * memory on each free list with the exception of the first item on the list.
1305 */
1306void show_free_areas(void)
1307{
Jes Sorensenc7241912006-09-27 01:50:05 -07001308 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 unsigned long active;
1310 unsigned long inactive;
1311 unsigned long free;
1312 struct zone *zone;
1313
1314 for_each_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07001315 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07001317
1318 show_node(zone);
1319 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
Dave Jones6b482c62005-11-10 15:45:56 -05001321 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 struct per_cpu_pageset *pageset;
1323
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001324 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
Jes Sorensenc7241912006-09-27 01:50:05 -07001326 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1327 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1328 cpu, pageset->pcp[0].high,
1329 pageset->pcp[0].batch, pageset->pcp[0].count,
1330 pageset->pcp[1].high, pageset->pcp[1].batch,
1331 pageset->pcp[1].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 }
1333 }
1334
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 get_zone_counts(&active, &inactive, &free);
1336
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1338 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1339 active,
1340 inactive,
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07001341 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07001342 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001343 global_page_state(NR_UNSTABLE_NFS),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 nr_free_pages(),
Christoph Lameter972d1a72006-09-25 23:31:51 -07001345 global_page_state(NR_SLAB_RECLAIMABLE) +
1346 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001347 global_page_state(NR_FILE_MAPPED),
Christoph Lameterdf849a12006-06-30 01:55:38 -07001348 global_page_state(NR_PAGETABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
1350 for_each_zone(zone) {
1351 int i;
1352
Jes Sorensenc7241912006-09-27 01:50:05 -07001353 if (!populated_zone(zone))
1354 continue;
1355
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 show_node(zone);
1357 printk("%s"
1358 " free:%lukB"
1359 " min:%lukB"
1360 " low:%lukB"
1361 " high:%lukB"
1362 " active:%lukB"
1363 " inactive:%lukB"
1364 " present:%lukB"
1365 " pages_scanned:%lu"
1366 " all_unreclaimable? %s"
1367 "\n",
1368 zone->name,
1369 K(zone->free_pages),
1370 K(zone->pages_min),
1371 K(zone->pages_low),
1372 K(zone->pages_high),
1373 K(zone->nr_active),
1374 K(zone->nr_inactive),
1375 K(zone->present_pages),
1376 zone->pages_scanned,
1377 (zone->all_unreclaimable ? "yes" : "no")
1378 );
1379 printk("lowmem_reserve[]:");
1380 for (i = 0; i < MAX_NR_ZONES; i++)
1381 printk(" %lu", zone->lowmem_reserve[i]);
1382 printk("\n");
1383 }
1384
1385 for_each_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001386 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
Jes Sorensenc7241912006-09-27 01:50:05 -07001388 if (!populated_zone(zone))
1389 continue;
1390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 show_node(zone);
1392 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393
1394 spin_lock_irqsave(&zone->lock, flags);
1395 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001396 nr[order] = zone->free_area[order].nr_free;
1397 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 }
1399 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001400 for (order = 0; order < MAX_ORDER; order++)
1401 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 printk("= %lukB\n", K(total));
1403 }
1404
1405 show_swap_cache_info();
1406}
1407
1408/*
1409 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001410 *
1411 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001413static int __meminit build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001414 struct zonelist *zonelist, int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415{
Christoph Lameter1a932052006-01-06 00:11:16 -08001416 struct zone *zone;
1417
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001418 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001419 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001420
1421 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001422 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08001423 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001424 if (populated_zone(zone)) {
Christoph Lameter070f8032006-01-06 00:11:19 -08001425 zonelist->zones[nr_zones++] = zone;
1426 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08001428
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001429 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08001430 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431}
1432
1433#ifdef CONFIG_NUMA
1434#define MAX_NODE_LOAD (num_online_nodes())
Yasunori Goto86356ab2006-06-23 02:03:09 -07001435static int __meminitdata node_load[MAX_NUMNODES];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001437 * 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 -07001438 * @node: node whose fallback list we're appending
1439 * @used_node_mask: nodemask_t of already used nodes
1440 *
1441 * We use a number of factors to determine which is the next node that should
1442 * appear on a given node's fallback list. The node should not have appeared
1443 * already in @node's fallback list, and it should be the next closest node
1444 * according to the distance array (which contains arbitrary distance values
1445 * from each node to each node in the system), and should also prefer nodes
1446 * with no CPUs, since presumably they'll have very little allocation pressure
1447 * on them otherwise.
1448 * It returns -1 if no node is found.
1449 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001450static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001452 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 int min_val = INT_MAX;
1454 int best_node = -1;
1455
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001456 /* Use the local node if we haven't already */
1457 if (!node_isset(node, *used_node_mask)) {
1458 node_set(node, *used_node_mask);
1459 return node;
1460 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001462 for_each_online_node(n) {
1463 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
1465 /* Don't want a node to appear more than once */
1466 if (node_isset(n, *used_node_mask))
1467 continue;
1468
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 /* Use the distance array to find the distance */
1470 val = node_distance(node, n);
1471
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001472 /* Penalize nodes under us ("prefer the next node") */
1473 val += (n < node);
1474
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 /* Give preference to headless and unused nodes */
1476 tmp = node_to_cpumask(n);
1477 if (!cpus_empty(tmp))
1478 val += PENALTY_FOR_NODE_WITH_CPUS;
1479
1480 /* Slight preference for less loaded node */
1481 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1482 val += node_load[n];
1483
1484 if (val < min_val) {
1485 min_val = val;
1486 best_node = n;
1487 }
1488 }
1489
1490 if (best_node >= 0)
1491 node_set(best_node, *used_node_mask);
1492
1493 return best_node;
1494}
1495
Yasunori Goto86356ab2006-06-23 02:03:09 -07001496static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497{
Christoph Lameter19655d32006-09-25 23:31:19 -07001498 int j, node, local_node;
1499 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 int prev_node, load;
1501 struct zonelist *zonelist;
1502 nodemask_t used_mask;
1503
1504 /* initialize zonelists */
Christoph Lameter19655d32006-09-25 23:31:19 -07001505 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 zonelist = pgdat->node_zonelists + i;
1507 zonelist->zones[0] = NULL;
1508 }
1509
1510 /* NUMA-aware ordering of nodes */
1511 local_node = pgdat->node_id;
1512 load = num_online_nodes();
1513 prev_node = local_node;
1514 nodes_clear(used_mask);
1515 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001516 int distance = node_distance(local_node, node);
1517
1518 /*
1519 * If another node is sufficiently far away then it is better
1520 * to reclaim pages in a zone before going off node.
1521 */
1522 if (distance > RECLAIM_DISTANCE)
1523 zone_reclaim_mode = 1;
1524
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 /*
1526 * We don't want to pressure a particular node.
1527 * So adding penalty to the first node in same
1528 * distance group to make it round-robin.
1529 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08001530
1531 if (distance != node_distance(local_node, prev_node))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 node_load[node] += load;
1533 prev_node = node;
1534 load--;
Christoph Lameter19655d32006-09-25 23:31:19 -07001535 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 zonelist = pgdat->node_zonelists + i;
1537 for (j = 0; zonelist->zones[j] != NULL; j++);
1538
Christoph Lameter19655d32006-09-25 23:31:19 -07001539 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 zonelist->zones[j] = NULL;
1541 }
1542 }
1543}
1544
1545#else /* CONFIG_NUMA */
1546
Yasunori Goto86356ab2006-06-23 02:03:09 -07001547static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548{
Christoph Lameter19655d32006-09-25 23:31:19 -07001549 int node, local_node;
1550 enum zone_type i,j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551
1552 local_node = pgdat->node_id;
Christoph Lameter19655d32006-09-25 23:31:19 -07001553 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 struct zonelist *zonelist;
1555
1556 zonelist = pgdat->node_zonelists + i;
1557
Christoph Lameter19655d32006-09-25 23:31:19 -07001558 j = build_zonelists_node(pgdat, zonelist, 0, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 /*
1560 * Now we build the zonelist so that it contains the zones
1561 * of all the other nodes.
1562 * We don't want to pressure a particular node, so when
1563 * building the zones for node N, we make sure that the
1564 * zones coming right after the local ones are those from
1565 * node N+1 (modulo N)
1566 */
1567 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1568 if (!node_online(node))
1569 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07001570 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 }
1572 for (node = 0; node < local_node; node++) {
1573 if (!node_online(node))
1574 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07001575 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 }
1577
1578 zonelist->zones[j] = NULL;
1579 }
1580}
1581
1582#endif /* CONFIG_NUMA */
1583
Yasunori Goto68113782006-06-23 02:03:11 -07001584/* return values int ....just for stop_machine_run() */
1585static int __meminit __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586{
Yasunori Goto68113782006-06-23 02:03:11 -07001587 int nid;
1588 for_each_online_node(nid)
1589 build_zonelists(NODE_DATA(nid));
1590 return 0;
1591}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592
Yasunori Goto68113782006-06-23 02:03:11 -07001593void __meminit build_all_zonelists(void)
1594{
1595 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07001596 __build_all_zonelists(NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07001597 cpuset_init_current_mems_allowed();
1598 } else {
1599 /* we have to stop all cpus to guaranntee there is no user
1600 of zonelist */
1601 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
1602 /* cpuset refresh routine should be here */
1603 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07001604 vm_total_pages = nr_free_pagecache_pages();
1605 printk("Built %i zonelists. Total pages: %ld\n",
1606 num_online_nodes(), vm_total_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607}
1608
1609/*
1610 * Helper functions to size the waitqueue hash table.
1611 * Essentially these want to choose hash table sizes sufficiently
1612 * large so that collisions trying to wait on pages are rare.
1613 * But in fact, the number of active page waitqueues on typical
1614 * systems is ridiculously low, less than 200. So this is even
1615 * conservative, even though it seems large.
1616 *
1617 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1618 * waitqueues, i.e. the size of the waitq table given the number of pages.
1619 */
1620#define PAGES_PER_WAITQUEUE 256
1621
Yasunori Gotocca448f2006-06-23 02:03:10 -07001622#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07001623static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624{
1625 unsigned long size = 1;
1626
1627 pages /= PAGES_PER_WAITQUEUE;
1628
1629 while (size < pages)
1630 size <<= 1;
1631
1632 /*
1633 * Once we have dozens or even hundreds of threads sleeping
1634 * on IO we've got bigger problems than wait queue collision.
1635 * Limit the size of the wait table to a reasonable size.
1636 */
1637 size = min(size, 4096UL);
1638
1639 return max(size, 4UL);
1640}
Yasunori Gotocca448f2006-06-23 02:03:10 -07001641#else
1642/*
1643 * A zone's size might be changed by hot-add, so it is not possible to determine
1644 * a suitable size for its wait_table. So we use the maximum size now.
1645 *
1646 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1647 *
1648 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1649 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1650 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1651 *
1652 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1653 * or more by the traditional way. (See above). It equals:
1654 *
1655 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1656 * ia64(16K page size) : = ( 8G + 4M)byte.
1657 * powerpc (64K page size) : = (32G +16M)byte.
1658 */
1659static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1660{
1661 return 4096UL;
1662}
1663#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
1665/*
1666 * This is an integer logarithm so that shifts can be used later
1667 * to extract the more random high bits from the multiplicative
1668 * hash function before the remainder is taken.
1669 */
1670static inline unsigned long wait_table_bits(unsigned long size)
1671{
1672 return ffz(~size);
1673}
1674
1675#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1676
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677/*
1678 * Initially all pages are reserved - free ones are freed
1679 * up by free_all_bootmem() once the early boot process is
1680 * done. Non-atomic initialization, single-pass.
1681 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001682void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 unsigned long start_pfn)
1684{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001686 unsigned long end_pfn = start_pfn + size;
1687 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001689 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001690 if (!early_pfn_valid(pfn))
1691 continue;
Andy Whitcroft75167952006-10-21 10:24:14 -07001692 if (!early_pfn_in_nid(pfn, nid))
1693 continue;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001694 page = pfn_to_page(pfn);
1695 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08001696 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 reset_page_mapcount(page);
1698 SetPageReserved(page);
1699 INIT_LIST_HEAD(&page->lru);
1700#ifdef WANT_PAGE_VIRTUAL
1701 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1702 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001703 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 }
1706}
1707
1708void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1709 unsigned long size)
1710{
1711 int order;
1712 for (order = 0; order < MAX_ORDER ; order++) {
1713 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1714 zone->free_area[order].nr_free = 0;
1715 }
1716}
1717
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001718#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
Christoph Lameter2f1b6242006-09-25 23:31:13 -07001719void zonetable_add(struct zone *zone, int nid, enum zone_type zid,
1720 unsigned long pfn, unsigned long size)
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001721{
1722 unsigned long snum = pfn_to_section_nr(pfn);
1723 unsigned long end = pfn_to_section_nr(pfn + size);
1724
1725 if (FLAGS_HAS_NODE)
1726 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1727 else
1728 for (; snum <= end; snum++)
1729 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1730}
1731
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732#ifndef __HAVE_ARCH_MEMMAP_INIT
1733#define memmap_init(size, nid, zone, start_pfn) \
1734 memmap_init_zone((size), (nid), (zone), (start_pfn))
1735#endif
1736
Ashok Raj6292d9a2006-02-01 03:04:44 -08001737static int __cpuinit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001738{
1739 int batch;
1740
1741 /*
1742 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001743 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001744 *
1745 * OK, so we don't know how big the cache is. So guess.
1746 */
1747 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001748 if (batch * PAGE_SIZE > 512 * 1024)
1749 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001750 batch /= 4; /* We effectively *= 4 below */
1751 if (batch < 1)
1752 batch = 1;
1753
1754 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001755 * Clamp the batch to a 2^n - 1 value. Having a power
1756 * of 2 value was found to be more likely to have
1757 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001758 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001759 * For example if 2 tasks are alternately allocating
1760 * batches of pages, one task can end up with a lot
1761 * of pages of one half of the possible page colors
1762 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001763 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001764 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001765
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001766 return batch;
1767}
1768
Christoph Lameter2caaad42005-06-21 17:15:00 -07001769inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1770{
1771 struct per_cpu_pages *pcp;
1772
Magnus Damm1c6fe942005-10-26 01:58:59 -07001773 memset(p, 0, sizeof(*p));
1774
Christoph Lameter2caaad42005-06-21 17:15:00 -07001775 pcp = &p->pcp[0]; /* hot */
1776 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001777 pcp->high = 6 * batch;
1778 pcp->batch = max(1UL, 1 * batch);
1779 INIT_LIST_HEAD(&pcp->list);
1780
1781 pcp = &p->pcp[1]; /* cold*/
1782 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001783 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001784 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001785 INIT_LIST_HEAD(&pcp->list);
1786}
1787
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001788/*
1789 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1790 * to the value high for the pageset p.
1791 */
1792
1793static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1794 unsigned long high)
1795{
1796 struct per_cpu_pages *pcp;
1797
1798 pcp = &p->pcp[0]; /* hot list */
1799 pcp->high = high;
1800 pcp->batch = max(1UL, high/4);
1801 if ((high/4) > (PAGE_SHIFT * 8))
1802 pcp->batch = PAGE_SHIFT * 8;
1803}
1804
1805
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001806#ifdef CONFIG_NUMA
1807/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001808 * Boot pageset table. One per cpu which is going to be used for all
1809 * zones and all nodes. The parameters will be set in such a way
1810 * that an item put on a list will immediately be handed over to
1811 * the buddy list. This is safe since pageset manipulation is done
1812 * with interrupts disabled.
1813 *
1814 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001815 *
1816 * The boot_pagesets must be kept even after bootup is complete for
1817 * unused processors and/or zones. They do play a role for bootstrapping
1818 * hotplugged processors.
1819 *
1820 * zoneinfo_show() and maybe other functions do
1821 * not check if the processor is online before following the pageset pointer.
1822 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001823 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001824static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001825
1826/*
1827 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001828 * per cpu pageset array in struct zone.
1829 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08001830static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001831{
1832 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001833
1834 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001835
Christoph Lameter66a55032006-09-27 01:50:09 -07001836 if (!populated_zone(zone))
1837 continue;
1838
Nick Piggin23316bc2006-01-08 01:00:41 -08001839 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001840 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08001841 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001842 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001843
Nick Piggin23316bc2006-01-08 01:00:41 -08001844 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001845
1846 if (percpu_pagelist_fraction)
1847 setup_pagelist_highmark(zone_pcp(zone, cpu),
1848 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001849 }
1850
1851 return 0;
1852bad:
1853 for_each_zone(dzone) {
1854 if (dzone == zone)
1855 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08001856 kfree(zone_pcp(dzone, cpu));
1857 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001858 }
1859 return -ENOMEM;
1860}
1861
1862static inline void free_zone_pagesets(int cpu)
1863{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001864 struct zone *zone;
1865
1866 for_each_zone(zone) {
1867 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1868
David Rientjesf3ef9ea2006-09-25 16:24:57 -07001869 /* Free per_cpu_pageset if it is slab allocated */
1870 if (pset != &boot_pageset[cpu])
1871 kfree(pset);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001872 zone_pcp(zone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001873 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001874}
1875
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07001876static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001877 unsigned long action,
1878 void *hcpu)
1879{
1880 int cpu = (long)hcpu;
1881 int ret = NOTIFY_OK;
1882
1883 switch (action) {
1884 case CPU_UP_PREPARE:
1885 if (process_zones(cpu))
1886 ret = NOTIFY_BAD;
1887 break;
Andi Kleenb0d41692005-11-05 17:25:53 +01001888 case CPU_UP_CANCELED:
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001889 case CPU_DEAD:
1890 free_zone_pagesets(cpu);
1891 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001892 default:
1893 break;
1894 }
1895 return ret;
1896}
1897
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07001898static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001899 { &pageset_cpuup_callback, NULL, 0 };
1900
Al Viro78d99552005-12-15 09:18:25 +00001901void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001902{
1903 int err;
1904
1905 /* Initialize per_cpu_pageset for cpu 0.
1906 * A cpuup callback will do this for every cpu
1907 * as it comes online
1908 */
1909 err = process_zones(smp_processor_id());
1910 BUG_ON(err);
1911 register_cpu_notifier(&pageset_notifier);
1912}
1913
1914#endif
1915
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001916static __meminit
Yasunori Gotocca448f2006-06-23 02:03:10 -07001917int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07001918{
1919 int i;
1920 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07001921 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07001922
1923 /*
1924 * The per-page waitqueue mechanism uses hashed waitqueues
1925 * per zone.
1926 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07001927 zone->wait_table_hash_nr_entries =
1928 wait_table_hash_nr_entries(zone_size_pages);
1929 zone->wait_table_bits =
1930 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07001931 alloc_size = zone->wait_table_hash_nr_entries
1932 * sizeof(wait_queue_head_t);
1933
1934 if (system_state == SYSTEM_BOOTING) {
1935 zone->wait_table = (wait_queue_head_t *)
1936 alloc_bootmem_node(pgdat, alloc_size);
1937 } else {
1938 /*
1939 * This case means that a zone whose size was 0 gets new memory
1940 * via memory hot-add.
1941 * But it may be the case that a new node was hot-added. In
1942 * this case vmalloc() will not be able to use this new node's
1943 * memory - this wait_table must be initialized to use this new
1944 * node itself as well.
1945 * To use this new node's memory, further consideration will be
1946 * necessary.
1947 */
1948 zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size);
1949 }
1950 if (!zone->wait_table)
1951 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07001952
Yasunori Goto02b694d2006-06-23 02:03:08 -07001953 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07001954 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07001955
1956 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07001957}
1958
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001959static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07001960{
1961 int cpu;
1962 unsigned long batch = zone_batchsize(zone);
1963
1964 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1965#ifdef CONFIG_NUMA
1966 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08001967 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07001968 setup_pageset(&boot_pageset[cpu],0);
1969#else
1970 setup_pageset(zone_pcp(zone,cpu), batch);
1971#endif
1972 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08001973 if (zone->present_pages)
1974 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
1975 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07001976}
1977
Yasunori Goto718127c2006-06-23 02:03:10 -07001978__meminit int init_currently_empty_zone(struct zone *zone,
1979 unsigned long zone_start_pfn,
1980 unsigned long size)
Dave Hansened8ece22005-10-29 18:16:50 -07001981{
1982 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07001983 int ret;
1984 ret = zone_wait_table_init(zone, size);
1985 if (ret)
1986 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07001987 pgdat->nr_zones = zone_idx(zone) + 1;
1988
Dave Hansened8ece22005-10-29 18:16:50 -07001989 zone->zone_start_pfn = zone_start_pfn;
1990
1991 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
1992
1993 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
Yasunori Goto718127c2006-06-23 02:03:10 -07001994
1995 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07001996}
1997
Mel Gormanc7132162006-09-27 01:49:43 -07001998#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
1999/*
2000 * Basic iterator support. Return the first range of PFNs for a node
2001 * Note: nid == MAX_NUMNODES returns first region regardless of node
2002 */
2003static int __init first_active_region_index_in_nid(int nid)
2004{
2005 int i;
2006
2007 for (i = 0; i < nr_nodemap_entries; i++)
2008 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2009 return i;
2010
2011 return -1;
2012}
2013
2014/*
2015 * Basic iterator support. Return the next active range of PFNs for a node
2016 * Note: nid == MAX_NUMNODES returns next region regardles of node
2017 */
2018static int __init next_active_region_index_in_nid(int index, int nid)
2019{
2020 for (index = index + 1; index < nr_nodemap_entries; index++)
2021 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2022 return index;
2023
2024 return -1;
2025}
2026
2027#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2028/*
2029 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2030 * Architectures may implement their own version but if add_active_range()
2031 * was used and there are no special requirements, this is a convenient
2032 * alternative
2033 */
2034int __init early_pfn_to_nid(unsigned long pfn)
2035{
2036 int i;
2037
2038 for (i = 0; i < nr_nodemap_entries; i++) {
2039 unsigned long start_pfn = early_node_map[i].start_pfn;
2040 unsigned long end_pfn = early_node_map[i].end_pfn;
2041
2042 if (start_pfn <= pfn && pfn < end_pfn)
2043 return early_node_map[i].nid;
2044 }
2045
2046 return 0;
2047}
2048#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2049
2050/* Basic iterator support to walk early_node_map[] */
2051#define for_each_active_range_index_in_nid(i, nid) \
2052 for (i = first_active_region_index_in_nid(nid); i != -1; \
2053 i = next_active_region_index_in_nid(i, nid))
2054
2055/**
2056 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002057 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2058 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07002059 *
2060 * If an architecture guarantees that all ranges registered with
2061 * add_active_ranges() contain no holes and may be freed, this
2062 * this function may be used instead of calling free_bootmem() manually.
2063 */
2064void __init free_bootmem_with_active_regions(int nid,
2065 unsigned long max_low_pfn)
2066{
2067 int i;
2068
2069 for_each_active_range_index_in_nid(i, nid) {
2070 unsigned long size_pages = 0;
2071 unsigned long end_pfn = early_node_map[i].end_pfn;
2072
2073 if (early_node_map[i].start_pfn >= max_low_pfn)
2074 continue;
2075
2076 if (end_pfn > max_low_pfn)
2077 end_pfn = max_low_pfn;
2078
2079 size_pages = end_pfn - early_node_map[i].start_pfn;
2080 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2081 PFN_PHYS(early_node_map[i].start_pfn),
2082 size_pages << PAGE_SHIFT);
2083 }
2084}
2085
2086/**
2087 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002088 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07002089 *
2090 * If an architecture guarantees that all ranges registered with
2091 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002092 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07002093 */
2094void __init sparse_memory_present_with_active_regions(int nid)
2095{
2096 int i;
2097
2098 for_each_active_range_index_in_nid(i, nid)
2099 memory_present(early_node_map[i].nid,
2100 early_node_map[i].start_pfn,
2101 early_node_map[i].end_pfn);
2102}
2103
2104/**
Mel Gormanfb014392006-09-27 01:49:59 -07002105 * push_node_boundaries - Push node boundaries to at least the requested boundary
2106 * @nid: The nid of the node to push the boundary for
2107 * @start_pfn: The start pfn of the node
2108 * @end_pfn: The end pfn of the node
2109 *
2110 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2111 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2112 * be hotplugged even though no physical memory exists. This function allows
2113 * an arch to push out the node boundaries so mem_map is allocated that can
2114 * be used later.
2115 */
2116#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2117void __init push_node_boundaries(unsigned int nid,
2118 unsigned long start_pfn, unsigned long end_pfn)
2119{
2120 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2121 nid, start_pfn, end_pfn);
2122
2123 /* Initialise the boundary for this node if necessary */
2124 if (node_boundary_end_pfn[nid] == 0)
2125 node_boundary_start_pfn[nid] = -1UL;
2126
2127 /* Update the boundaries */
2128 if (node_boundary_start_pfn[nid] > start_pfn)
2129 node_boundary_start_pfn[nid] = start_pfn;
2130 if (node_boundary_end_pfn[nid] < end_pfn)
2131 node_boundary_end_pfn[nid] = end_pfn;
2132}
2133
2134/* If necessary, push the node boundary out for reserve hotadd */
2135static void __init account_node_boundary(unsigned int nid,
2136 unsigned long *start_pfn, unsigned long *end_pfn)
2137{
2138 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2139 nid, *start_pfn, *end_pfn);
2140
2141 /* Return if boundary information has not been provided */
2142 if (node_boundary_end_pfn[nid] == 0)
2143 return;
2144
2145 /* Check the boundaries and update if necessary */
2146 if (node_boundary_start_pfn[nid] < *start_pfn)
2147 *start_pfn = node_boundary_start_pfn[nid];
2148 if (node_boundary_end_pfn[nid] > *end_pfn)
2149 *end_pfn = node_boundary_end_pfn[nid];
2150}
2151#else
2152void __init push_node_boundaries(unsigned int nid,
2153 unsigned long start_pfn, unsigned long end_pfn) {}
2154
2155static void __init account_node_boundary(unsigned int nid,
2156 unsigned long *start_pfn, unsigned long *end_pfn) {}
2157#endif
2158
2159
2160/**
Mel Gormanc7132162006-09-27 01:49:43 -07002161 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002162 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2163 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2164 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07002165 *
2166 * It returns the start and end page frame of a node based on information
2167 * provided by an arch calling add_active_range(). If called for a node
2168 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002169 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07002170 */
2171void __init get_pfn_range_for_nid(unsigned int nid,
2172 unsigned long *start_pfn, unsigned long *end_pfn)
2173{
2174 int i;
2175 *start_pfn = -1UL;
2176 *end_pfn = 0;
2177
2178 for_each_active_range_index_in_nid(i, nid) {
2179 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
2180 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
2181 }
2182
2183 if (*start_pfn == -1UL) {
2184 printk(KERN_WARNING "Node %u active with no memory\n", nid);
2185 *start_pfn = 0;
2186 }
Mel Gormanfb014392006-09-27 01:49:59 -07002187
2188 /* Push the node boundaries out if requested */
2189 account_node_boundary(nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07002190}
2191
2192/*
2193 * Return the number of pages a zone spans in a node, including holes
2194 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2195 */
2196unsigned long __init zone_spanned_pages_in_node(int nid,
2197 unsigned long zone_type,
2198 unsigned long *ignored)
2199{
2200 unsigned long node_start_pfn, node_end_pfn;
2201 unsigned long zone_start_pfn, zone_end_pfn;
2202
2203 /* Get the start and end of the node and zone */
2204 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2205 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
2206 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
2207
2208 /* Check that this node has pages within the zone's required range */
2209 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
2210 return 0;
2211
2212 /* Move the zone boundaries inside the node if necessary */
2213 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
2214 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
2215
2216 /* Return the spanned pages */
2217 return zone_end_pfn - zone_start_pfn;
2218}
2219
2220/*
2221 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002222 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07002223 */
2224unsigned long __init __absent_pages_in_range(int nid,
2225 unsigned long range_start_pfn,
2226 unsigned long range_end_pfn)
2227{
2228 int i = 0;
2229 unsigned long prev_end_pfn = 0, hole_pages = 0;
2230 unsigned long start_pfn;
2231
2232 /* Find the end_pfn of the first active range of pfns in the node */
2233 i = first_active_region_index_in_nid(nid);
2234 if (i == -1)
2235 return 0;
2236
Mel Gorman9c7cd682006-09-27 01:49:58 -07002237 /* Account for ranges before physical memory on this node */
2238 if (early_node_map[i].start_pfn > range_start_pfn)
2239 hole_pages = early_node_map[i].start_pfn - range_start_pfn;
2240
Mel Gormanc7132162006-09-27 01:49:43 -07002241 prev_end_pfn = early_node_map[i].start_pfn;
2242
2243 /* Find all holes for the zone within the node */
2244 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
2245
2246 /* No need to continue if prev_end_pfn is outside the zone */
2247 if (prev_end_pfn >= range_end_pfn)
2248 break;
2249
2250 /* Make sure the end of the zone is not within the hole */
2251 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
2252 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
2253
2254 /* Update the hole size cound and move on */
2255 if (start_pfn > range_start_pfn) {
2256 BUG_ON(prev_end_pfn > start_pfn);
2257 hole_pages += start_pfn - prev_end_pfn;
2258 }
2259 prev_end_pfn = early_node_map[i].end_pfn;
2260 }
2261
Mel Gorman9c7cd682006-09-27 01:49:58 -07002262 /* Account for ranges past physical memory on this node */
2263 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07002264 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07002265 max(range_start_pfn, prev_end_pfn);
2266
Mel Gormanc7132162006-09-27 01:49:43 -07002267 return hole_pages;
2268}
2269
2270/**
2271 * absent_pages_in_range - Return number of page frames in holes within a range
2272 * @start_pfn: The start PFN to start searching for holes
2273 * @end_pfn: The end PFN to stop searching for holes
2274 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002275 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07002276 */
2277unsigned long __init absent_pages_in_range(unsigned long start_pfn,
2278 unsigned long end_pfn)
2279{
2280 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
2281}
2282
2283/* Return the number of page frames in holes in a zone on a node */
2284unsigned long __init zone_absent_pages_in_node(int nid,
2285 unsigned long zone_type,
2286 unsigned long *ignored)
2287{
Mel Gorman9c7cd682006-09-27 01:49:58 -07002288 unsigned long node_start_pfn, node_end_pfn;
2289 unsigned long zone_start_pfn, zone_end_pfn;
2290
2291 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2292 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
2293 node_start_pfn);
2294 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
2295 node_end_pfn);
2296
2297 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07002298}
Mel Gorman0e0b8642006-09-27 01:49:56 -07002299
Mel Gormanc7132162006-09-27 01:49:43 -07002300#else
2301static inline unsigned long zone_spanned_pages_in_node(int nid,
2302 unsigned long zone_type,
2303 unsigned long *zones_size)
2304{
2305 return zones_size[zone_type];
2306}
2307
2308static inline unsigned long zone_absent_pages_in_node(int nid,
2309 unsigned long zone_type,
2310 unsigned long *zholes_size)
2311{
2312 if (!zholes_size)
2313 return 0;
2314
2315 return zholes_size[zone_type];
2316}
Mel Gorman0e0b8642006-09-27 01:49:56 -07002317
Mel Gormanc7132162006-09-27 01:49:43 -07002318#endif
2319
2320static void __init calculate_node_totalpages(struct pglist_data *pgdat,
2321 unsigned long *zones_size, unsigned long *zholes_size)
2322{
2323 unsigned long realtotalpages, totalpages = 0;
2324 enum zone_type i;
2325
2326 for (i = 0; i < MAX_NR_ZONES; i++)
2327 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
2328 zones_size);
2329 pgdat->node_spanned_pages = totalpages;
2330
2331 realtotalpages = totalpages;
2332 for (i = 0; i < MAX_NR_ZONES; i++)
2333 realtotalpages -=
2334 zone_absent_pages_in_node(pgdat->node_id, i,
2335 zholes_size);
2336 pgdat->node_present_pages = realtotalpages;
2337 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
2338 realtotalpages);
2339}
2340
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341/*
2342 * Set up the zone data structures:
2343 * - mark all pages reserved
2344 * - mark all memory queues empty
2345 * - clear the memory bitmaps
2346 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07002347static void __meminit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 unsigned long *zones_size, unsigned long *zholes_size)
2349{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07002350 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07002351 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07002353 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354
Dave Hansen208d54e2005-10-29 18:16:52 -07002355 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 pgdat->nr_zones = 0;
2357 init_waitqueue_head(&pgdat->kswapd_wait);
2358 pgdat->kswapd_max_order = 0;
2359
2360 for (j = 0; j < MAX_NR_ZONES; j++) {
2361 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07002362 unsigned long size, realsize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363
Mel Gormanc7132162006-09-27 01:49:43 -07002364 size = zone_spanned_pages_in_node(nid, j, zones_size);
2365 realsize = size - zone_absent_pages_in_node(nid, j,
2366 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367
Mel Gorman0e0b8642006-09-27 01:49:56 -07002368 /*
2369 * Adjust realsize so that it accounts for how much memory
2370 * is used by this zone for memmap. This affects the watermark
2371 * and per-cpu initialisations
2372 */
2373 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
2374 if (realsize >= memmap_pages) {
2375 realsize -= memmap_pages;
2376 printk(KERN_DEBUG
2377 " %s zone: %lu pages used for memmap\n",
2378 zone_names[j], memmap_pages);
2379 } else
2380 printk(KERN_WARNING
2381 " %s zone: %lu pages exceeds realsize %lu\n",
2382 zone_names[j], memmap_pages, realsize);
2383
2384 /* Account for reserved DMA pages */
2385 if (j == ZONE_DMA && realsize > dma_reserve) {
2386 realsize -= dma_reserve;
2387 printk(KERN_DEBUG " DMA zone: %lu pages reserved\n",
2388 dma_reserve);
2389 }
2390
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002391 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 nr_kernel_pages += realsize;
2393 nr_all_pages += realsize;
2394
2395 zone->spanned_pages = size;
2396 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07002397#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07002398 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07002399 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07002400 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07002401 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07002402#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 zone->name = zone_names[j];
2404 spin_lock_init(&zone->lock);
2405 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002406 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 zone->zone_pgdat = pgdat;
2408 zone->free_pages = 0;
2409
Martin Bligh3bb1a8522006-10-28 10:38:24 -07002410 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411
Dave Hansened8ece22005-10-29 18:16:50 -07002412 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 INIT_LIST_HEAD(&zone->active_list);
2414 INIT_LIST_HEAD(&zone->inactive_list);
2415 zone->nr_scan_active = 0;
2416 zone->nr_scan_inactive = 0;
2417 zone->nr_active = 0;
2418 zone->nr_inactive = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07002419 zap_zone_vm_stats(zone);
Martin Hicks53e9a612005-09-03 15:54:51 -07002420 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 if (!size)
2422 continue;
2423
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002424 zonetable_add(zone, nid, j, zone_start_pfn, size);
Yasunori Goto718127c2006-06-23 02:03:10 -07002425 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
2426 BUG_ON(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 }
2429}
2430
2431static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2432{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 /* Skip empty nodes */
2434 if (!pgdat->node_spanned_pages)
2435 return;
2436
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002437#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 /* ia64 gets its own node_mem_map, before this, without bootmem */
2439 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07002440 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002441 struct page *map;
2442
Bob Piccoe984bb42006-05-20 15:00:31 -07002443 /*
2444 * The zone's endpoints aren't required to be MAX_ORDER
2445 * aligned but the node_mem_map endpoints must be in order
2446 * for the buddy allocator to function correctly.
2447 */
2448 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
2449 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
2450 end = ALIGN(end, MAX_ORDER_NR_PAGES);
2451 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002452 map = alloc_remap(pgdat->node_id, size);
2453 if (!map)
2454 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07002455 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002457#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 /*
2459 * With no DISCONTIG, the global mem_map is just set as node 0's
2460 */
Mel Gormanc7132162006-09-27 01:49:43 -07002461 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07002463#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2464 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
2465 mem_map -= pgdat->node_start_pfn;
2466#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2467 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002469#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470}
2471
Yasunori Goto86356ab2006-06-23 02:03:09 -07002472void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 unsigned long *zones_size, unsigned long node_start_pfn,
2474 unsigned long *zholes_size)
2475{
2476 pgdat->node_id = nid;
2477 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07002478 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479
2480 alloc_node_mem_map(pgdat);
2481
2482 free_area_init_core(pgdat, zones_size, zholes_size);
2483}
2484
Mel Gormanc7132162006-09-27 01:49:43 -07002485#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2486/**
2487 * add_active_range - Register a range of PFNs backed by physical memory
2488 * @nid: The node ID the range resides on
2489 * @start_pfn: The start PFN of the available physical memory
2490 * @end_pfn: The end PFN of the available physical memory
2491 *
2492 * These ranges are stored in an early_node_map[] and later used by
2493 * free_area_init_nodes() to calculate zone sizes and holes. If the
2494 * range spans a memory hole, it is up to the architecture to ensure
2495 * the memory is not freed by the bootmem allocator. If possible
2496 * the range being registered will be merged with existing ranges.
2497 */
2498void __init add_active_range(unsigned int nid, unsigned long start_pfn,
2499 unsigned long end_pfn)
2500{
2501 int i;
2502
2503 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
2504 "%d entries of %d used\n",
2505 nid, start_pfn, end_pfn,
2506 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
2507
2508 /* Merge with existing active regions if possible */
2509 for (i = 0; i < nr_nodemap_entries; i++) {
2510 if (early_node_map[i].nid != nid)
2511 continue;
2512
2513 /* Skip if an existing region covers this new one */
2514 if (start_pfn >= early_node_map[i].start_pfn &&
2515 end_pfn <= early_node_map[i].end_pfn)
2516 return;
2517
2518 /* Merge forward if suitable */
2519 if (start_pfn <= early_node_map[i].end_pfn &&
2520 end_pfn > early_node_map[i].end_pfn) {
2521 early_node_map[i].end_pfn = end_pfn;
2522 return;
2523 }
2524
2525 /* Merge backward if suitable */
2526 if (start_pfn < early_node_map[i].end_pfn &&
2527 end_pfn >= early_node_map[i].start_pfn) {
2528 early_node_map[i].start_pfn = start_pfn;
2529 return;
2530 }
2531 }
2532
2533 /* Check that early_node_map is large enough */
2534 if (i >= MAX_ACTIVE_REGIONS) {
2535 printk(KERN_CRIT "More than %d memory regions, truncating\n",
2536 MAX_ACTIVE_REGIONS);
2537 return;
2538 }
2539
2540 early_node_map[i].nid = nid;
2541 early_node_map[i].start_pfn = start_pfn;
2542 early_node_map[i].end_pfn = end_pfn;
2543 nr_nodemap_entries = i + 1;
2544}
2545
2546/**
2547 * shrink_active_range - Shrink an existing registered range of PFNs
2548 * @nid: The node id the range is on that should be shrunk
2549 * @old_end_pfn: The old end PFN of the range
2550 * @new_end_pfn: The new PFN of the range
2551 *
2552 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
2553 * The map is kept at the end physical page range that has already been
2554 * registered with add_active_range(). This function allows an arch to shrink
2555 * an existing registered range.
2556 */
2557void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
2558 unsigned long new_end_pfn)
2559{
2560 int i;
2561
2562 /* Find the old active region end and shrink */
2563 for_each_active_range_index_in_nid(i, nid)
2564 if (early_node_map[i].end_pfn == old_end_pfn) {
2565 early_node_map[i].end_pfn = new_end_pfn;
2566 break;
2567 }
2568}
2569
2570/**
2571 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002572 *
Mel Gormanc7132162006-09-27 01:49:43 -07002573 * During discovery, it may be found that a table like SRAT is invalid
2574 * and an alternative discovery method must be used. This function removes
2575 * all currently registered regions.
2576 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002577void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07002578{
2579 memset(early_node_map, 0, sizeof(early_node_map));
2580 nr_nodemap_entries = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07002581#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2582 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
2583 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
2584#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -07002585}
2586
2587/* Compare two active node_active_regions */
2588static int __init cmp_node_active_region(const void *a, const void *b)
2589{
2590 struct node_active_region *arange = (struct node_active_region *)a;
2591 struct node_active_region *brange = (struct node_active_region *)b;
2592
2593 /* Done this way to avoid overflows */
2594 if (arange->start_pfn > brange->start_pfn)
2595 return 1;
2596 if (arange->start_pfn < brange->start_pfn)
2597 return -1;
2598
2599 return 0;
2600}
2601
2602/* sort the node_map by start_pfn */
2603static void __init sort_node_map(void)
2604{
2605 sort(early_node_map, (size_t)nr_nodemap_entries,
2606 sizeof(struct node_active_region),
2607 cmp_node_active_region, NULL);
2608}
2609
2610/* Find the lowest pfn for a node. This depends on a sorted early_node_map */
2611unsigned long __init find_min_pfn_for_node(unsigned long nid)
2612{
2613 int i;
2614
2615 /* Assuming a sorted map, the first range found has the starting pfn */
2616 for_each_active_range_index_in_nid(i, nid)
2617 return early_node_map[i].start_pfn;
2618
2619 printk(KERN_WARNING "Could not find start_pfn for node %lu\n", nid);
2620 return 0;
2621}
2622
2623/**
2624 * find_min_pfn_with_active_regions - Find the minimum PFN registered
2625 *
2626 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002627 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07002628 */
2629unsigned long __init find_min_pfn_with_active_regions(void)
2630{
2631 return find_min_pfn_for_node(MAX_NUMNODES);
2632}
2633
2634/**
2635 * find_max_pfn_with_active_regions - Find the maximum PFN registered
2636 *
2637 * It returns the maximum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002638 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07002639 */
2640unsigned long __init find_max_pfn_with_active_regions(void)
2641{
2642 int i;
2643 unsigned long max_pfn = 0;
2644
2645 for (i = 0; i < nr_nodemap_entries; i++)
2646 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
2647
2648 return max_pfn;
2649}
2650
2651/**
2652 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002653 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07002654 *
2655 * This will call free_area_init_node() for each active node in the system.
2656 * Using the page ranges provided by add_active_range(), the size of each
2657 * zone in each node and their holes is calculated. If the maximum PFN
2658 * between two adjacent zones match, it is assumed that the zone is empty.
2659 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
2660 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
2661 * starts where the previous one ended. For example, ZONE_DMA32 starts
2662 * at arch_max_dma_pfn.
2663 */
2664void __init free_area_init_nodes(unsigned long *max_zone_pfn)
2665{
2666 unsigned long nid;
2667 enum zone_type i;
2668
2669 /* Record where the zone boundaries are */
2670 memset(arch_zone_lowest_possible_pfn, 0,
2671 sizeof(arch_zone_lowest_possible_pfn));
2672 memset(arch_zone_highest_possible_pfn, 0,
2673 sizeof(arch_zone_highest_possible_pfn));
2674 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
2675 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
2676 for (i = 1; i < MAX_NR_ZONES; i++) {
2677 arch_zone_lowest_possible_pfn[i] =
2678 arch_zone_highest_possible_pfn[i-1];
2679 arch_zone_highest_possible_pfn[i] =
2680 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
2681 }
2682
2683 /* Regions in the early_node_map can be in any order */
2684 sort_node_map();
2685
2686 /* Print out the zone ranges */
2687 printk("Zone PFN ranges:\n");
2688 for (i = 0; i < MAX_NR_ZONES; i++)
2689 printk(" %-8s %8lu -> %8lu\n",
2690 zone_names[i],
2691 arch_zone_lowest_possible_pfn[i],
2692 arch_zone_highest_possible_pfn[i]);
2693
2694 /* Print out the early_node_map[] */
2695 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
2696 for (i = 0; i < nr_nodemap_entries; i++)
2697 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
2698 early_node_map[i].start_pfn,
2699 early_node_map[i].end_pfn);
2700
2701 /* Initialise every node */
2702 for_each_online_node(nid) {
2703 pg_data_t *pgdat = NODE_DATA(nid);
2704 free_area_init_node(nid, pgdat, NULL,
2705 find_min_pfn_for_node(nid), NULL);
2706 }
2707}
2708#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2709
Mel Gorman0e0b8642006-09-27 01:49:56 -07002710/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002711 * set_dma_reserve - set the specified number of pages reserved in the first zone
2712 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07002713 *
2714 * The per-cpu batchsize and zone watermarks are determined by present_pages.
2715 * In the DMA zone, a significant percentage may be consumed by kernel image
2716 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002717 * function may optionally be used to account for unfreeable pages in the
2718 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
2719 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07002720 */
2721void __init set_dma_reserve(unsigned long new_dma_reserve)
2722{
2723 dma_reserve = new_dma_reserve;
2724}
2725
Dave Hansen93b75042005-06-23 00:07:47 -07002726#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727static bootmem_data_t contig_bootmem_data;
2728struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2729
2730EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002731#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732
2733void __init free_area_init(unsigned long *zones_size)
2734{
Dave Hansen93b75042005-06-23 00:07:47 -07002735 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2737}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739#ifdef CONFIG_HOTPLUG_CPU
2740static int page_alloc_cpu_notify(struct notifier_block *self,
2741 unsigned long action, void *hcpu)
2742{
2743 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744
2745 if (action == CPU_DEAD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 local_irq_disable();
2747 __drain_pages(cpu);
Christoph Lameterf8891e52006-06-30 01:55:45 -07002748 vm_events_fold_cpu(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749 local_irq_enable();
Christoph Lameter2244b952006-06-30 01:55:33 -07002750 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 }
2752 return NOTIFY_OK;
2753}
2754#endif /* CONFIG_HOTPLUG_CPU */
2755
2756void __init page_alloc_init(void)
2757{
2758 hotcpu_notifier(page_alloc_cpu_notify, 0);
2759}
2760
2761/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002762 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2763 * or min_free_kbytes changes.
2764 */
2765static void calculate_totalreserve_pages(void)
2766{
2767 struct pglist_data *pgdat;
2768 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002769 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002770
2771 for_each_online_pgdat(pgdat) {
2772 for (i = 0; i < MAX_NR_ZONES; i++) {
2773 struct zone *zone = pgdat->node_zones + i;
2774 unsigned long max = 0;
2775
2776 /* Find valid and maximum lowmem_reserve in the zone */
2777 for (j = i; j < MAX_NR_ZONES; j++) {
2778 if (zone->lowmem_reserve[j] > max)
2779 max = zone->lowmem_reserve[j];
2780 }
2781
2782 /* we treat pages_high as reserved pages. */
2783 max += zone->pages_high;
2784
2785 if (max > zone->present_pages)
2786 max = zone->present_pages;
2787 reserve_pages += max;
2788 }
2789 }
2790 totalreserve_pages = reserve_pages;
2791}
2792
2793/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 * setup_per_zone_lowmem_reserve - called whenever
2795 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2796 * has a correct pages reserved value, so an adequate number of
2797 * pages are left in the zone after a successful __alloc_pages().
2798 */
2799static void setup_per_zone_lowmem_reserve(void)
2800{
2801 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002802 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08002804 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 for (j = 0; j < MAX_NR_ZONES; j++) {
2806 struct zone *zone = pgdat->node_zones + j;
2807 unsigned long present_pages = zone->present_pages;
2808
2809 zone->lowmem_reserve[j] = 0;
2810
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002811 idx = j;
2812 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813 struct zone *lower_zone;
2814
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002815 idx--;
2816
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2818 sysctl_lowmem_reserve_ratio[idx] = 1;
2819
2820 lower_zone = pgdat->node_zones + idx;
2821 lower_zone->lowmem_reserve[j] = present_pages /
2822 sysctl_lowmem_reserve_ratio[idx];
2823 present_pages += lower_zone->present_pages;
2824 }
2825 }
2826 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002827
2828 /* update totalreserve_pages */
2829 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830}
2831
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002832/**
2833 * setup_per_zone_pages_min - called when min_free_kbytes changes.
2834 *
2835 * Ensures that the pages_{min,low,high} values for each zone are set correctly
2836 * with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 */
Dave Hansen3947be12005-10-29 18:16:54 -07002838void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839{
2840 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2841 unsigned long lowmem_pages = 0;
2842 struct zone *zone;
2843 unsigned long flags;
2844
2845 /* Calculate total number of !ZONE_HIGHMEM pages */
2846 for_each_zone(zone) {
2847 if (!is_highmem(zone))
2848 lowmem_pages += zone->present_pages;
2849 }
2850
2851 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07002852 u64 tmp;
2853
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07002855 tmp = (u64)pages_min * zone->present_pages;
2856 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 if (is_highmem(zone)) {
2858 /*
Nick Piggin669ed172005-11-13 16:06:45 -08002859 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2860 * need highmem pages, so cap pages_min to a small
2861 * value here.
2862 *
2863 * The (pages_high-pages_low) and (pages_low-pages_min)
2864 * deltas controls asynch page reclaim, and so should
2865 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 */
2867 int min_pages;
2868
2869 min_pages = zone->present_pages / 1024;
2870 if (min_pages < SWAP_CLUSTER_MAX)
2871 min_pages = SWAP_CLUSTER_MAX;
2872 if (min_pages > 128)
2873 min_pages = 128;
2874 zone->pages_min = min_pages;
2875 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08002876 /*
2877 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 * proportionate to the zone's size.
2879 */
Nick Piggin669ed172005-11-13 16:06:45 -08002880 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 }
2882
Andrew Mortonac924c62006-05-15 09:43:59 -07002883 zone->pages_low = zone->pages_min + (tmp >> 2);
2884 zone->pages_high = zone->pages_min + (tmp >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 spin_unlock_irqrestore(&zone->lru_lock, flags);
2886 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002887
2888 /* update totalreserve_pages */
2889 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890}
2891
2892/*
2893 * Initialise min_free_kbytes.
2894 *
2895 * For small machines we want it small (128k min). For large machines
2896 * we want it large (64MB max). But it is not linear, because network
2897 * bandwidth does not increase linearly with machine size. We use
2898 *
2899 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2900 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2901 *
2902 * which yields
2903 *
2904 * 16MB: 512k
2905 * 32MB: 724k
2906 * 64MB: 1024k
2907 * 128MB: 1448k
2908 * 256MB: 2048k
2909 * 512MB: 2896k
2910 * 1024MB: 4096k
2911 * 2048MB: 5792k
2912 * 4096MB: 8192k
2913 * 8192MB: 11584k
2914 * 16384MB: 16384k
2915 */
2916static int __init init_per_zone_pages_min(void)
2917{
2918 unsigned long lowmem_kbytes;
2919
2920 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2921
2922 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2923 if (min_free_kbytes < 128)
2924 min_free_kbytes = 128;
2925 if (min_free_kbytes > 65536)
2926 min_free_kbytes = 65536;
2927 setup_per_zone_pages_min();
2928 setup_per_zone_lowmem_reserve();
2929 return 0;
2930}
2931module_init(init_per_zone_pages_min)
2932
2933/*
2934 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2935 * that we can call two helper functions whenever min_free_kbytes
2936 * changes.
2937 */
2938int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2939 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2940{
2941 proc_dointvec(table, write, file, buffer, length, ppos);
2942 setup_per_zone_pages_min();
2943 return 0;
2944}
2945
Christoph Lameter96146342006-07-03 00:24:13 -07002946#ifdef CONFIG_NUMA
2947int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
2948 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2949{
2950 struct zone *zone;
2951 int rc;
2952
2953 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2954 if (rc)
2955 return rc;
2956
2957 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07002958 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07002959 sysctl_min_unmapped_ratio) / 100;
2960 return 0;
2961}
Christoph Lameter0ff38492006-09-25 23:31:52 -07002962
2963int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
2964 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2965{
2966 struct zone *zone;
2967 int rc;
2968
2969 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2970 if (rc)
2971 return rc;
2972
2973 for_each_zone(zone)
2974 zone->min_slab_pages = (zone->present_pages *
2975 sysctl_min_slab_ratio) / 100;
2976 return 0;
2977}
Christoph Lameter96146342006-07-03 00:24:13 -07002978#endif
2979
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980/*
2981 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2982 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2983 * whenever sysctl_lowmem_reserve_ratio changes.
2984 *
2985 * The reserve ratio obviously has absolutely no relation with the
2986 * pages_min watermarks. The lowmem reserve ratio can only make sense
2987 * if in function of the boot time zone sizes.
2988 */
2989int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2990 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2991{
2992 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2993 setup_per_zone_lowmem_reserve();
2994 return 0;
2995}
2996
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002997/*
2998 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2999 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
3000 * can have before it gets flushed back to buddy allocator.
3001 */
3002
3003int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
3004 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3005{
3006 struct zone *zone;
3007 unsigned int cpu;
3008 int ret;
3009
3010 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3011 if (!write || (ret == -EINVAL))
3012 return ret;
3013 for_each_zone(zone) {
3014 for_each_online_cpu(cpu) {
3015 unsigned long high;
3016 high = zone->present_pages / percpu_pagelist_fraction;
3017 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
3018 }
3019 }
3020 return 0;
3021}
3022
David S. Millerf034b5d2006-08-24 03:08:07 -07003023int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024
3025#ifdef CONFIG_NUMA
3026static int __init set_hashdist(char *str)
3027{
3028 if (!str)
3029 return 0;
3030 hashdist = simple_strtoul(str, &str, 0);
3031 return 1;
3032}
3033__setup("hashdist=", set_hashdist);
3034#endif
3035
3036/*
3037 * allocate a large system hash table from bootmem
3038 * - it is assumed that the hash table must contain an exact power-of-2
3039 * quantity of entries
3040 * - limit is the number of hash buckets, not the total allocation size
3041 */
3042void *__init alloc_large_system_hash(const char *tablename,
3043 unsigned long bucketsize,
3044 unsigned long numentries,
3045 int scale,
3046 int flags,
3047 unsigned int *_hash_shift,
3048 unsigned int *_hash_mask,
3049 unsigned long limit)
3050{
3051 unsigned long long max = limit;
3052 unsigned long log2qty, size;
3053 void *table = NULL;
3054
3055 /* allow the kernel cmdline to have a say */
3056 if (!numentries) {
3057 /* round applicable memory size up to nearest megabyte */
3058 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
3059 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
3060 numentries >>= 20 - PAGE_SHIFT;
3061 numentries <<= 20 - PAGE_SHIFT;
3062
3063 /* limit to 1 bucket per 2^scale bytes of low memory */
3064 if (scale > PAGE_SHIFT)
3065 numentries >>= (scale - PAGE_SHIFT);
3066 else
3067 numentries <<= (PAGE_SHIFT - scale);
3068 }
John Hawkes6e692ed2006-03-25 03:08:02 -08003069 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070
3071 /* limit allocation size to 1/16 total memory by default */
3072 if (max == 0) {
3073 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
3074 do_div(max, bucketsize);
3075 }
3076
3077 if (numentries > max)
3078 numentries = max;
3079
3080 log2qty = long_log2(numentries);
3081
3082 do {
3083 size = bucketsize << log2qty;
3084 if (flags & HASH_EARLY)
3085 table = alloc_bootmem(size);
3086 else if (hashdist)
3087 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
3088 else {
3089 unsigned long order;
3090 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
3091 ;
3092 table = (void*) __get_free_pages(GFP_ATOMIC, order);
3093 }
3094 } while (!table && size > PAGE_SIZE && --log2qty);
3095
3096 if (!table)
3097 panic("Failed to allocate %s hash table\n", tablename);
3098
3099 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
3100 tablename,
3101 (1U << log2qty),
3102 long_log2(size) - PAGE_SHIFT,
3103 size);
3104
3105 if (_hash_shift)
3106 *_hash_shift = log2qty;
3107 if (_hash_mask)
3108 *_hash_mask = (1 << log2qty) - 1;
3109
3110 return table;
3111}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003112
3113#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003114struct page *pfn_to_page(unsigned long pfn)
3115{
Andy Whitcroft67de6482006-06-23 02:03:12 -07003116 return __pfn_to_page(pfn);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003117}
3118unsigned long page_to_pfn(struct page *page)
3119{
Andy Whitcroft67de6482006-06-23 02:03:12 -07003120 return __page_to_pfn(page);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003121}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003122EXPORT_SYMBOL(pfn_to_page);
3123EXPORT_SYMBOL(page_to_pfn);
3124#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
Andrew Morton6220ec72006-10-19 23:29:05 -07003125
3126#if MAX_NUMNODES > 1
3127/*
3128 * Find the highest possible node id.
3129 */
3130int highest_possible_node_id(void)
3131{
3132 unsigned int node;
3133 unsigned int highest = 0;
3134
3135 for_each_node_mask(node, node_possible_map)
3136 highest = node;
3137 return highest;
3138}
3139EXPORT_SYMBOL(highest_possible_node_id);
3140#endif