blob: cace22b3ac25dc0f4af6fe10a7cc0a74c9d32778 [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
Helge Deller15ad7cd2006-12-06 20:40:36 -080086static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter2f1b6242006-09-25 23:31:13 -070087 "DMA",
Christoph Lameterfb0e7942006-09-25 23:31:13 -070088#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -070089 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -070090#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -070091 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -070092#ifdef CONFIG_HIGHMEM
Christoph Lameter2f1b6242006-09-25 23:31:13 -070093 "HighMem"
Christoph Lametere53ef382006-09-25 23:31:14 -070094#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -070095};
96
Linus Torvalds1da177e2005-04-16 15:20:36 -070097int min_free_kbytes = 1024;
98
Yasunori Goto86356ab2006-06-23 02:03:09 -070099unsigned long __meminitdata nr_kernel_pages;
100unsigned long __meminitdata nr_all_pages;
Mel Gorman0e0b8642006-09-27 01:49:56 -0700101static unsigned long __initdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
Mel Gormanc7132162006-09-27 01:49:43 -0700103#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
104 /*
105 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
106 * ranges of memory (RAM) that may be registered with add_active_range().
107 * Ranges passed to add_active_range() will be merged if possible
108 * so the number of times add_active_range() can be called is
109 * related to the number of nodes and the number of holes
110 */
111 #ifdef CONFIG_MAX_ACTIVE_REGIONS
112 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
113 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
114 #else
115 #if MAX_NUMNODES >= 32
116 /* If there can be many nodes, allow up to 50 holes per node */
117 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
118 #else
119 /* By default, allow up to 256 distinct regions */
120 #define MAX_ACTIVE_REGIONS 256
121 #endif
122 #endif
123
124 struct node_active_region __initdata early_node_map[MAX_ACTIVE_REGIONS];
125 int __initdata nr_nodemap_entries;
126 unsigned long __initdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
127 unsigned long __initdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Mel Gormanfb014392006-09-27 01:49:59 -0700128#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
129 unsigned long __initdata node_boundary_start_pfn[MAX_NUMNODES];
130 unsigned long __initdata node_boundary_end_pfn[MAX_NUMNODES];
131#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -0700132#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
133
Nick Piggin13e74442006-01-06 00:10:58 -0800134#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700135static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700137 int ret = 0;
138 unsigned seq;
139 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700140
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700141 do {
142 seq = zone_span_seqbegin(zone);
143 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
144 ret = 1;
145 else if (pfn < zone->zone_start_pfn)
146 ret = 1;
147 } while (zone_span_seqretry(zone, seq));
148
149 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700150}
151
152static int page_is_consistent(struct zone *zone, struct page *page)
153{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154#ifdef CONFIG_HOLES_IN_ZONE
155 if (!pfn_valid(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700156 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157#endif
158 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700159 return 0;
160
161 return 1;
162}
163/*
164 * Temporary debugging check for pages not lying within a given zone.
165 */
166static int bad_range(struct zone *zone, struct page *page)
167{
168 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700170 if (!page_is_consistent(zone, page))
171 return 1;
172
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 return 0;
174}
Nick Piggin13e74442006-01-06 00:10:58 -0800175#else
176static inline int bad_range(struct zone *zone, struct page *page)
177{
178 return 0;
179}
180#endif
181
Nick Piggin224abf92006-01-06 00:11:11 -0800182static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
Nick Piggin224abf92006-01-06 00:11:11 -0800184 printk(KERN_EMERG "Bad page state in process '%s'\n"
Hugh Dickins7365f3d2006-01-11 12:17:18 -0800185 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
186 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
187 KERN_EMERG "Backtrace:\n",
Nick Piggin224abf92006-01-06 00:11:11 -0800188 current->comm, page, (int)(2*sizeof(unsigned long)),
189 (unsigned long)page->flags, page->mapping,
190 page_mapcount(page), page_count(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 dump_stack();
Hugh Dickins334795e2005-06-21 17:15:08 -0700192 page->flags &= ~(1 << PG_lru |
193 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 1 << PG_active |
196 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700197 1 << PG_reclaim |
198 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 1 << PG_swapcache |
Nick Piggin676165a2006-04-10 11:21:48 +1000200 1 << PG_writeback |
201 1 << PG_buddy );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 set_page_count(page, 0);
203 reset_page_mapcount(page);
204 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700205 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206}
207
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208/*
209 * Higher-order pages are called "compound pages". They are structured thusly:
210 *
211 * The first PAGE_SIZE page is called the "head page".
212 *
213 * The remaining PAGE_SIZE pages are called "tail pages".
214 *
215 * All pages have PG_compound set. All pages have their ->private pointing at
216 * the head page (even the head page has this).
217 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800218 * The first tail page's ->lru.next holds the address of the compound page's
219 * put_page() function. Its ->lru.prev holds the order of allocation.
220 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800222
223static void free_compound_page(struct page *page)
224{
225 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
226}
227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228static void prep_compound_page(struct page *page, unsigned long order)
229{
230 int i;
231 int nr_pages = 1 << order;
232
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800233 set_compound_page_dtor(page, free_compound_page);
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800234 page[1].lru.prev = (void *)order;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 for (i = 0; i < nr_pages; i++) {
236 struct page *p = page + i;
237
Nick Piggin5e9dace2006-03-22 00:08:01 -0800238 __SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700239 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 }
241}
242
243static void destroy_compound_page(struct page *page, unsigned long order)
244{
245 int i;
246 int nr_pages = 1 << order;
247
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800248 if (unlikely((unsigned long)page[1].lru.prev != order))
Nick Piggin224abf92006-01-06 00:11:11 -0800249 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
251 for (i = 0; i < nr_pages; i++) {
252 struct page *p = page + i;
253
Nick Piggin224abf92006-01-06 00:11:11 -0800254 if (unlikely(!PageCompound(p) |
255 (page_private(p) != (unsigned long)page)))
256 bad_page(page);
Nick Piggin5e9dace2006-03-22 00:08:01 -0800257 __ClearPageCompound(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 }
259}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
Nick Piggin17cf4402006-03-22 00:08:41 -0800261static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
262{
263 int i;
264
Nick Piggin725d7042006-09-25 23:30:55 -0700265 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
Andrew Morton6626c5d2006-03-22 00:08:42 -0800266 /*
267 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
268 * and __GFP_HIGHMEM from hard or soft interrupt context.
269 */
Nick Piggin725d7042006-09-25 23:30:55 -0700270 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800271 for (i = 0; i < (1 << order); i++)
272 clear_highpage(page + i);
273}
274
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275/*
276 * function for dealing with page's order in buddy system.
277 * zone->lock is already acquired when we use these.
278 * So, we don't need atomic page->flags operations here.
279 */
Andrew Morton6aa30012006-04-18 22:20:52 -0700280static inline unsigned long page_order(struct page *page)
281{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700282 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283}
284
Andrew Morton6aa30012006-04-18 22:20:52 -0700285static inline void set_page_order(struct page *page, int order)
286{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700287 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000288 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289}
290
291static inline void rmv_page_order(struct page *page)
292{
Nick Piggin676165a2006-04-10 11:21:48 +1000293 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700294 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
296
297/*
298 * Locate the struct page for both the matching buddy in our
299 * pair (buddy1) and the combined O(n+1) page they form (page).
300 *
301 * 1) Any buddy B1 will have an order O twin B2 which satisfies
302 * the following equation:
303 * B2 = B1 ^ (1 << O)
304 * For example, if the starting buddy (buddy2) is #8 its order
305 * 1 buddy is #10:
306 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
307 *
308 * 2) Any buddy B will have an order O+1 parent P which
309 * satisfies the following equation:
310 * P = B & ~(1 << O)
311 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200312 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 */
314static inline struct page *
315__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
316{
317 unsigned long buddy_idx = page_idx ^ (1 << order);
318
319 return page + (buddy_idx - page_idx);
320}
321
322static inline unsigned long
323__find_combined_index(unsigned long page_idx, unsigned int order)
324{
325 return (page_idx & ~(1 << order));
326}
327
328/*
329 * This function checks whether a page is free && is the buddy
330 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800331 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000332 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700333 * (c) a page and its buddy have the same order &&
334 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 *
Nick Piggin676165a2006-04-10 11:21:48 +1000336 * For recording whether a page is in the buddy system, we use PG_buddy.
337 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
338 *
339 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700341static inline int page_is_buddy(struct page *page, struct page *buddy,
342 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343{
Nick Piggin13e74442006-01-06 00:10:58 -0800344#ifdef CONFIG_HOLES_IN_ZONE
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700345 if (!pfn_valid(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800346 return 0;
347#endif
348
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700349 if (page_zone_id(page) != page_zone_id(buddy))
350 return 0;
351
352 if (PageBuddy(buddy) && page_order(buddy) == order) {
353 BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700354 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000355 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700356 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357}
358
359/*
360 * Freeing function for a buddy system allocator.
361 *
362 * The concept of a buddy system is to maintain direct-mapped table
363 * (containing bit values) for memory blocks of various "orders".
364 * The bottom level table contains the map for the smallest allocatable
365 * units of memory (here, pages), and each level above it describes
366 * pairs of units from the levels below, hence, "buddies".
367 * At a high level, all that happens here is marking the table entry
368 * at the bottom level available, and propagating the changes upward
369 * as necessary, plus some accounting needed to play nicely with other
370 * parts of the VM system.
371 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000372 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700373 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 * So when we are allocating or freeing one, we can derive the state of the
375 * other. That is, if we allocate a small block, and both were
376 * free, the remainder of the region must be split into blocks.
377 * If a block is freed, and its buddy is also free, then this
378 * triggers coalescing into a block of larger size.
379 *
380 * -- wli
381 */
382
Nick Piggin48db57f2006-01-08 01:00:42 -0800383static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 struct zone *zone, unsigned int order)
385{
386 unsigned long page_idx;
387 int order_size = 1 << order;
388
Nick Piggin224abf92006-01-06 00:11:11 -0800389 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 destroy_compound_page(page, order);
391
392 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
393
Nick Piggin725d7042006-09-25 23:30:55 -0700394 VM_BUG_ON(page_idx & (order_size - 1));
395 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
397 zone->free_pages += order_size;
398 while (order < MAX_ORDER-1) {
399 unsigned long combined_idx;
400 struct free_area *area;
401 struct page *buddy;
402
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700404 if (!page_is_buddy(page, buddy, order))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800406
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 list_del(&buddy->lru);
408 area = zone->free_area + order;
409 area->nr_free--;
410 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800411 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 page = page + (combined_idx - page_idx);
413 page_idx = combined_idx;
414 order++;
415 }
416 set_page_order(page, order);
417 list_add(&page->lru, &zone->free_area[order].free_list);
418 zone->free_area[order].nr_free++;
419}
420
Nick Piggin224abf92006-01-06 00:11:11 -0800421static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422{
Nick Piggin92be2e32006-01-06 00:10:57 -0800423 if (unlikely(page_mapcount(page) |
424 (page->mapping != NULL) |
425 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 (page->flags & (
427 1 << PG_lru |
428 1 << PG_private |
429 1 << PG_locked |
430 1 << PG_active |
431 1 << PG_reclaim |
432 1 << PG_slab |
433 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700434 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000435 1 << PG_reserved |
436 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800437 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700439 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800440 /*
441 * For now, we report if PG_reserved was found set, but do not
442 * clear it, and do not free the page. But we shall soon need
443 * to do more, for when the ZERO_PAGE count wraps negative.
444 */
445 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446}
447
448/*
449 * Frees a list of pages.
450 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700451 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 *
453 * If the zone was previously in an "all pages pinned" state then look to
454 * see if this freeing clears that state.
455 *
456 * And clear the zone's pages_scanned counter, to hold off the "all pages are
457 * pinned" detection logic.
458 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800459static void free_pages_bulk(struct zone *zone, int count,
460 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
Nick Pigginc54ad302006-01-06 00:10:56 -0800462 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 zone->all_unreclaimable = 0;
464 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800465 while (count--) {
466 struct page *page;
467
Nick Piggin725d7042006-09-25 23:30:55 -0700468 VM_BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800470 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800472 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800474 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475}
476
Nick Piggin48db57f2006-01-08 01:00:42 -0800477static void free_one_page(struct zone *zone, struct page *page, int order)
478{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700479 spin_lock(&zone->lock);
480 zone->all_unreclaimable = 0;
481 zone->pages_scanned = 0;
Paul Jackson0798e512006-12-06 20:31:38 -0800482 __free_one_page(page, zone, order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700483 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800484}
485
486static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487{
Nick Pigginc54ad302006-01-06 00:10:56 -0800488 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800490 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800493 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800494 if (reserved)
495 return;
496
Nick Piggin9858db52006-10-11 01:21:30 -0700497 if (!PageHighMem(page))
498 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Nick Piggindafb1362006-10-11 01:21:30 -0700499 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800500 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700501
Nick Pigginc54ad302006-01-06 00:10:56 -0800502 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700503 __count_vm_events(PGFREE, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800504 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800505 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506}
507
David Howellsa226f6c2006-01-06 00:11:08 -0800508/*
509 * permit the bootmem allocator to evade page validation on high-order frees
510 */
511void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
512{
513 if (order == 0) {
514 __ClearPageReserved(page);
515 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800516 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800517 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800518 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800519 int loop;
520
Nick Piggin545b1ea2006-03-22 00:08:07 -0800521 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800522 for (loop = 0; loop < BITS_PER_LONG; loop++) {
523 struct page *p = &page[loop];
524
Nick Piggin545b1ea2006-03-22 00:08:07 -0800525 if (loop + 1 < BITS_PER_LONG)
526 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800527 __ClearPageReserved(p);
528 set_page_count(p, 0);
529 }
530
Nick Piggin7835e982006-03-22 00:08:40 -0800531 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800532 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800533 }
534}
535
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536
537/*
538 * The order of subdivision here is critical for the IO subsystem.
539 * Please do not alter this order without good reasons and regression
540 * testing. Specifically, as large blocks of memory are subdivided,
541 * the order in which smaller blocks are delivered depends on the order
542 * they're subdivided in this function. This is the primary factor
543 * influencing the order in which pages are delivered to the IO
544 * subsystem according to empirical testing, and this is also justified
545 * by considering the behavior of a buddy system containing a single
546 * large block of memory acted on by a series of small allocations.
547 * This behavior is a critical factor in sglist merging's success.
548 *
549 * -- wli
550 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800551static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 int low, int high, struct free_area *area)
553{
554 unsigned long size = 1 << high;
555
556 while (high > low) {
557 area--;
558 high--;
559 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700560 VM_BUG_ON(bad_range(zone, &page[size]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 list_add(&page[size].lru, &area->free_list);
562 area->nr_free++;
563 set_page_order(&page[size], high);
564 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565}
566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567/*
568 * This page is about to be returned from the page allocator
569 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800570static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571{
Nick Piggin92be2e32006-01-06 00:10:57 -0800572 if (unlikely(page_mapcount(page) |
573 (page->mapping != NULL) |
574 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700575 (page->flags & (
576 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 1 << PG_private |
578 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 1 << PG_active |
580 1 << PG_dirty |
581 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700582 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700584 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000585 1 << PG_reserved |
586 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800587 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Hugh Dickins689bceb2005-11-21 21:32:20 -0800589 /*
590 * For now, we report if PG_reserved was found set, but do not
591 * clear it, and do not allocate the page: as a safety net.
592 */
593 if (PageReserved(page))
594 return 1;
595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
597 1 << PG_referenced | 1 << PG_arch_1 |
598 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700599 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800600 set_page_refcounted(page);
Nick Piggincc102502006-12-06 20:32:00 -0800601
602 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800604
605 if (gfp_flags & __GFP_ZERO)
606 prep_zero_page(page, order, gfp_flags);
607
608 if (order && (gfp_flags & __GFP_COMP))
609 prep_compound_page(page, order);
610
Hugh Dickins689bceb2005-11-21 21:32:20 -0800611 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612}
613
614/*
615 * Do the hard work of removing an element from the buddy allocator.
616 * Call me with the zone->lock already held.
617 */
618static struct page *__rmqueue(struct zone *zone, unsigned int order)
619{
620 struct free_area * area;
621 unsigned int current_order;
622 struct page *page;
623
624 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
625 area = zone->free_area + current_order;
626 if (list_empty(&area->free_list))
627 continue;
628
629 page = list_entry(area->free_list.next, struct page, lru);
630 list_del(&page->lru);
631 rmv_page_order(page);
632 area->nr_free--;
633 zone->free_pages -= 1UL << order;
Nick Piggin085cc7d2006-01-06 00:11:01 -0800634 expand(zone, page, order, current_order, area);
635 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 }
637
638 return NULL;
639}
640
641/*
642 * Obtain a specified number of elements from the buddy allocator, all under
643 * a single hold of the lock, for efficiency. Add them to the supplied list.
644 * Returns the number of new pages which were placed at *list.
645 */
646static int rmqueue_bulk(struct zone *zone, unsigned int order,
647 unsigned long count, struct list_head *list)
648{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Nick Pigginc54ad302006-01-06 00:10:56 -0800651 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 for (i = 0; i < count; ++i) {
Nick Piggin085cc7d2006-01-06 00:11:01 -0800653 struct page *page = __rmqueue(zone, order);
654 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 list_add_tail(&page->lru, list);
657 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800658 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800659 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660}
661
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700662#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800663/*
664 * Called from the slab reaper to drain pagesets on a particular node that
Christoph Lameter39bbcb82006-09-25 23:31:49 -0700665 * belongs to the currently executing processor.
Christoph Lameter879336c2006-03-22 00:09:08 -0800666 * Note that this function must be called with the thread pinned to
667 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800668 */
669void drain_node_pages(int nodeid)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700670{
Christoph Lameter2f6726e2006-09-25 23:31:18 -0700671 int i;
672 enum zone_type z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700673 unsigned long flags;
674
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800675 for (z = 0; z < MAX_NR_ZONES; z++) {
676 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700677 struct per_cpu_pageset *pset;
678
Christoph Lameter39bbcb82006-09-25 23:31:49 -0700679 if (!populated_zone(zone))
680 continue;
681
Nick Piggin23316bc2006-01-08 01:00:41 -0800682 pset = zone_pcp(zone, smp_processor_id());
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700683 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
684 struct per_cpu_pages *pcp;
685
686 pcp = &pset->pcp[i];
Christoph Lameter879336c2006-03-22 00:09:08 -0800687 if (pcp->count) {
Christoph Lameterbc4ba392006-12-06 20:33:02 -0800688 int to_drain;
689
Christoph Lameter879336c2006-03-22 00:09:08 -0800690 local_irq_save(flags);
Christoph Lameterbc4ba392006-12-06 20:33:02 -0800691 if (pcp->count >= pcp->batch)
692 to_drain = pcp->batch;
693 else
694 to_drain = pcp->count;
695 free_pages_bulk(zone, to_drain, &pcp->list, 0);
696 pcp->count -= to_drain;
Christoph Lameter879336c2006-03-22 00:09:08 -0800697 local_irq_restore(flags);
698 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700699 }
700 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700701}
702#endif
703
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704static 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}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
726#ifdef CONFIG_PM
727
728void mark_free_pages(struct zone *zone)
729{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700730 unsigned long pfn, max_zone_pfn;
731 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 int order;
733 struct list_head *curr;
734
735 if (!zone->spanned_pages)
736 return;
737
738 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700739
740 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
741 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
742 if (pfn_valid(pfn)) {
743 struct page *page = pfn_to_page(pfn);
744
745 if (!PageNosave(page))
746 ClearPageNosaveFree(page);
747 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
749 for (order = MAX_ORDER - 1; order >= 0; --order)
750 list_for_each(curr, &zone->free_area[order].free_list) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700751 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700753 pfn = page_to_pfn(list_entry(curr, struct page, lru));
754 for (i = 0; i < (1UL << order); i++)
755 SetPageNosaveFree(pfn_to_page(pfn + i));
756 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 spin_unlock_irqrestore(&zone->lock, flags);
759}
760
761/*
762 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
763 */
764void drain_local_pages(void)
765{
766 unsigned long flags;
767
768 local_irq_save(flags);
769 __drain_pages(smp_processor_id());
770 local_irq_restore(flags);
771}
772#endif /* CONFIG_PM */
773
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774/*
775 * Free a 0-order page
776 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777static void fastcall free_hot_cold_page(struct page *page, int cold)
778{
779 struct zone *zone = page_zone(page);
780 struct per_cpu_pages *pcp;
781 unsigned long flags;
782
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 if (PageAnon(page))
784 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800785 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800786 return;
787
Nick Piggin9858db52006-10-11 01:21:30 -0700788 if (!PageHighMem(page))
789 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Nick Piggindafb1362006-10-11 01:21:30 -0700790 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800791 kernel_map_pages(page, 1, 0);
792
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700793 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700795 __count_vm_event(PGFREE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 list_add(&page->lru, &pcp->list);
797 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800798 if (pcp->count >= pcp->high) {
799 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
800 pcp->count -= pcp->batch;
801 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 local_irq_restore(flags);
803 put_cpu();
804}
805
806void fastcall free_hot_page(struct page *page)
807{
808 free_hot_cold_page(page, 0);
809}
810
811void fastcall free_cold_page(struct page *page)
812{
813 free_hot_cold_page(page, 1);
814}
815
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800816/*
817 * split_page takes a non-compound higher-order page, and splits it into
818 * n (1<<order) sub-pages: page[0..n]
819 * Each sub-page must be freed individually.
820 *
821 * Note: this is probably too low level an operation for use in drivers.
822 * Please consult with lkml before using this in your driver.
823 */
824void split_page(struct page *page, unsigned int order)
825{
826 int i;
827
Nick Piggin725d7042006-09-25 23:30:55 -0700828 VM_BUG_ON(PageCompound(page));
829 VM_BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -0800830 for (i = 1; i < (1 << order); i++)
831 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800832}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800833
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834/*
835 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
836 * we cheat by calling it from here, in the order > 0 path. Saves a branch
837 * or two.
838 */
Nick Piggina74609f2006-01-06 00:11:20 -0800839static struct page *buffered_rmqueue(struct zonelist *zonelist,
840 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841{
842 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800843 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800845 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
Hugh Dickins689bceb2005-11-21 21:32:20 -0800847again:
Nick Piggina74609f2006-01-06 00:11:20 -0800848 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800849 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 struct per_cpu_pages *pcp;
851
Nick Piggina74609f2006-01-06 00:11:20 -0800852 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800854 if (!pcp->count) {
nkalmala941c7102006-11-02 22:07:04 -0800855 pcp->count = rmqueue_bulk(zone, 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800857 if (unlikely(!pcp->count))
858 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 }
Nick Piggina74609f2006-01-06 00:11:20 -0800860 page = list_entry(pcp->list.next, struct page, lru);
861 list_del(&page->lru);
862 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800863 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 spin_lock_irqsave(&zone->lock, flags);
865 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800866 spin_unlock(&zone->lock);
867 if (!page)
868 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 }
870
Christoph Lameterf8891e52006-06-30 01:55:45 -0700871 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Christoph Lameterca889e62006-06-30 01:55:44 -0700872 zone_statistics(zonelist, zone);
Nick Piggina74609f2006-01-06 00:11:20 -0800873 local_irq_restore(flags);
874 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
Nick Piggin725d7042006-09-25 23:30:55 -0700876 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -0800877 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -0800878 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800880
881failed:
882 local_irq_restore(flags);
883 put_cpu();
884 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885}
886
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800887#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800888#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
889#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
890#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
891#define ALLOC_HARDER 0x10 /* try to alloc harder */
892#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
893#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800894
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895/*
896 * Return 1 if free pages are above 'mark'. This takes into account the order
897 * of the allocation.
898 */
899int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800900 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901{
902 /* free_pages my go negative - that's OK */
Nick Piggine80ee882006-10-04 02:15:23 -0700903 unsigned long min = mark;
904 long free_pages = z->free_pages - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 int o;
906
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800907 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800909 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 min -= min / 4;
911
912 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
913 return 0;
914 for (o = 0; o < order; o++) {
915 /* At the next order, this order's pages become unavailable */
916 free_pages -= z->free_area[o].nr_free << o;
917
918 /* Require fewer higher order pages to be free */
919 min >>= 1;
920
921 if (free_pages <= min)
922 return 0;
923 }
924 return 1;
925}
926
Paul Jackson9276b1bc2006-12-06 20:31:48 -0800927#ifdef CONFIG_NUMA
928/*
929 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
930 * skip over zones that are not allowed by the cpuset, or that have
931 * been recently (in last second) found to be nearly full. See further
932 * comments in mmzone.h. Reduces cache footprint of zonelist scans
933 * that have to skip over alot of full or unallowed zones.
934 *
935 * If the zonelist cache is present in the passed in zonelist, then
936 * returns a pointer to the allowed node mask (either the current
937 * tasks mems_allowed, or node_online_map.)
938 *
939 * If the zonelist cache is not available for this zonelist, does
940 * nothing and returns NULL.
941 *
942 * If the fullzones BITMAP in the zonelist cache is stale (more than
943 * a second since last zap'd) then we zap it out (clear its bits.)
944 *
945 * We hold off even calling zlc_setup, until after we've checked the
946 * first zone in the zonelist, on the theory that most allocations will
947 * be satisfied from that first zone, so best to examine that zone as
948 * quickly as we can.
949 */
950static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
951{
952 struct zonelist_cache *zlc; /* cached zonelist speedup info */
953 nodemask_t *allowednodes; /* zonelist_cache approximation */
954
955 zlc = zonelist->zlcache_ptr;
956 if (!zlc)
957 return NULL;
958
959 if (jiffies - zlc->last_full_zap > 1 * HZ) {
960 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
961 zlc->last_full_zap = jiffies;
962 }
963
964 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
965 &cpuset_current_mems_allowed :
966 &node_online_map;
967 return allowednodes;
968}
969
970/*
971 * Given 'z' scanning a zonelist, run a couple of quick checks to see
972 * if it is worth looking at further for free memory:
973 * 1) Check that the zone isn't thought to be full (doesn't have its
974 * bit set in the zonelist_cache fullzones BITMAP).
975 * 2) Check that the zones node (obtained from the zonelist_cache
976 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
977 * Return true (non-zero) if zone is worth looking at further, or
978 * else return false (zero) if it is not.
979 *
980 * This check -ignores- the distinction between various watermarks,
981 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
982 * found to be full for any variation of these watermarks, it will
983 * be considered full for up to one second by all requests, unless
984 * we are so low on memory on all allowed nodes that we are forced
985 * into the second scan of the zonelist.
986 *
987 * In the second scan we ignore this zonelist cache and exactly
988 * apply the watermarks to all zones, even it is slower to do so.
989 * We are low on memory in the second scan, and should leave no stone
990 * unturned looking for a free page.
991 */
992static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
993 nodemask_t *allowednodes)
994{
995 struct zonelist_cache *zlc; /* cached zonelist speedup info */
996 int i; /* index of *z in zonelist zones */
997 int n; /* node that zone *z is on */
998
999 zlc = zonelist->zlcache_ptr;
1000 if (!zlc)
1001 return 1;
1002
1003 i = z - zonelist->zones;
1004 n = zlc->z_to_n[i];
1005
1006 /* This zone is worth trying if it is allowed but not full */
1007 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1008}
1009
1010/*
1011 * Given 'z' scanning a zonelist, set the corresponding bit in
1012 * zlc->fullzones, so that subsequent attempts to allocate a page
1013 * from that zone don't waste time re-examining it.
1014 */
1015static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1016{
1017 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1018 int i; /* index of *z in zonelist zones */
1019
1020 zlc = zonelist->zlcache_ptr;
1021 if (!zlc)
1022 return;
1023
1024 i = z - zonelist->zones;
1025
1026 set_bit(i, zlc->fullzones);
1027}
1028
1029#else /* CONFIG_NUMA */
1030
1031static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1032{
1033 return NULL;
1034}
1035
1036static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1037 nodemask_t *allowednodes)
1038{
1039 return 1;
1040}
1041
1042static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1043{
1044}
1045#endif /* CONFIG_NUMA */
1046
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001047/*
Paul Jackson0798e512006-12-06 20:31:38 -08001048 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001049 * a page.
1050 */
1051static struct page *
1052get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1053 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -07001054{
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001055 struct zone **z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001056 struct page *page = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001057 int classzone_idx = zone_idx(zonelist->zones[0]);
Christoph Lameter1192d522006-09-25 23:31:45 -07001058 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001059 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1060 int zlc_active = 0; /* set if using zonelist_cache */
1061 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001062
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001063zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001064 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001065 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001066 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1067 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001068 z = zonelist->zones;
1069
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001070 do {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001071 if (NUMA_BUILD && zlc_active &&
1072 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1073 continue;
Christoph Lameter1192d522006-09-25 23:31:45 -07001074 zone = *z;
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001075 if (unlikely(NUMA_BUILD && (gfp_mask & __GFP_THISNODE) &&
Christoph Lameter1192d522006-09-25 23:31:45 -07001076 zone->zone_pgdat != zonelist->zones[0]->zone_pgdat))
Christoph Lameter9b819d22006-09-25 23:31:40 -07001077 break;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001078 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson0798e512006-12-06 20:31:38 -08001079 !cpuset_zone_allowed(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001080 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001081
1082 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001083 unsigned long mark;
1084 if (alloc_flags & ALLOC_WMARK_MIN)
Christoph Lameter1192d522006-09-25 23:31:45 -07001085 mark = zone->pages_min;
Nick Piggin31488902005-11-28 13:44:03 -08001086 else if (alloc_flags & ALLOC_WMARK_LOW)
Christoph Lameter1192d522006-09-25 23:31:45 -07001087 mark = zone->pages_low;
Nick Piggin31488902005-11-28 13:44:03 -08001088 else
Christoph Lameter1192d522006-09-25 23:31:45 -07001089 mark = zone->pages_high;
Paul Jackson0798e512006-12-06 20:31:38 -08001090 if (!zone_watermark_ok(zone, order, mark,
1091 classzone_idx, alloc_flags)) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001092 if (!zone_reclaim_mode ||
Christoph Lameter1192d522006-09-25 23:31:45 -07001093 !zone_reclaim(zone, gfp_mask, order))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001094 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001095 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001096 }
1097
Christoph Lameter1192d522006-09-25 23:31:45 -07001098 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
Paul Jackson0798e512006-12-06 20:31:38 -08001099 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001100 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001101this_zone_full:
1102 if (NUMA_BUILD)
1103 zlc_mark_zone_full(zonelist, z);
1104try_next_zone:
1105 if (NUMA_BUILD && !did_zlc_setup) {
1106 /* we do zlc_setup after the first zone is tried */
1107 allowednodes = zlc_setup(zonelist, alloc_flags);
1108 zlc_active = 1;
1109 did_zlc_setup = 1;
1110 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001111 } while (*(++z) != NULL);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001112
1113 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1114 /* Disable zlc cache for second zonelist scan */
1115 zlc_active = 0;
1116 goto zonelist_scan;
1117 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001118 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001119}
1120
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121/*
1122 * This is the 'heart' of the zoned buddy allocator.
1123 */
1124struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +01001125__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 struct zonelist *zonelist)
1127{
Al Viro260b2362005-10-21 03:22:44 -04001128 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001129 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 struct page *page;
1131 struct reclaim_state reclaim_state;
1132 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001134 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 int did_some_progress;
1136
1137 might_sleep_if(wait);
1138
Jens Axboe6b1de912005-11-17 21:35:02 +01001139restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001140 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001142 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 /* Should this ever happen?? */
1144 return NULL;
1145 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001146
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001147 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001148 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001149 if (page)
1150 goto got_pg;
1151
Christoph Lameter952f3b52006-12-06 20:33:26 -08001152 /*
1153 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1154 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1155 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1156 * using a larger set of nodes after it has established that the
1157 * allowed per node queues are empty and that nodes are
1158 * over allocated.
1159 */
1160 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1161 goto nopage;
1162
Paul Jackson0798e512006-12-06 20:31:38 -08001163 for (z = zonelist->zones; *z; z++)
Chris Wright43b0bc02006-06-25 05:47:55 -07001164 wakeup_kswapd(*z, order);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001165
Paul Jackson9bf22292005-09-06 15:18:12 -07001166 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001167 * OK, we're below the kswapd watermark and have kicked background
1168 * reclaim. Now things get more complex, so set up alloc_flags according
1169 * to how we want to proceed.
1170 *
1171 * The caller may dip into page reserves a bit more if the caller
1172 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -08001173 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1174 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -07001175 */
Nick Piggin31488902005-11-28 13:44:03 -08001176 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001177 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1178 alloc_flags |= ALLOC_HARDER;
1179 if (gfp_mask & __GFP_HIGH)
1180 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -07001181 if (wait)
1182 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
1184 /*
1185 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001186 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 *
1188 * This is the last chance, in general, before the goto nopage.
1189 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -07001190 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001192 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1193 if (page)
1194 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
1196 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -07001197
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001198rebalance:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001199 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1200 && !in_interrupt()) {
1201 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -08001202nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001203 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001204 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -08001205 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001206 if (page)
1207 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -08001208 if (gfp_mask & __GFP_NOFAIL) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001209 congestion_wait(WRITE, HZ/50);
Kirill Korotaev885036d2005-11-13 16:06:41 -08001210 goto nofail_alloc;
1211 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 }
1213 goto nopage;
1214 }
1215
1216 /* Atomic allocations - we can't balance anything */
1217 if (!wait)
1218 goto nopage;
1219
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 cond_resched();
1221
1222 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001223 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 p->flags |= PF_MEMALLOC;
1225 reclaim_state.reclaimed_slab = 0;
1226 p->reclaim_state = &reclaim_state;
1227
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001228 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
1230 p->reclaim_state = NULL;
1231 p->flags &= ~PF_MEMALLOC;
1232
1233 cond_resched();
1234
1235 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001236 page = get_page_from_freelist(gfp_mask, order,
1237 zonelist, alloc_flags);
1238 if (page)
1239 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1241 /*
1242 * Go through the zonelist yet one more time, keep
1243 * very high watermark here, this is only to catch
1244 * a parallel oom killing, we must fail if we're still
1245 * under heavy pressure.
1246 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001247 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001248 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001249 if (page)
1250 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001252 out_of_memory(zonelist, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 goto restart;
1254 }
1255
1256 /*
1257 * Don't let big-order allocations loop unless the caller explicitly
1258 * requests that. Wait for some write requests to complete then retry.
1259 *
1260 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1261 * <= 3, but that may not be true in other implementations.
1262 */
1263 do_retry = 0;
1264 if (!(gfp_mask & __GFP_NORETRY)) {
1265 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1266 do_retry = 1;
1267 if (gfp_mask & __GFP_NOFAIL)
1268 do_retry = 1;
1269 }
1270 if (do_retry) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001271 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 goto rebalance;
1273 }
1274
1275nopage:
1276 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1277 printk(KERN_WARNING "%s: page allocation failure."
1278 " order:%d, mode:0x%x\n",
1279 p->comm, order, gfp_mask);
1280 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001281 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 return page;
1285}
1286
1287EXPORT_SYMBOL(__alloc_pages);
1288
1289/*
1290 * Common helper functions.
1291 */
Al Virodd0fc662005-10-07 07:46:04 +01001292fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293{
1294 struct page * page;
1295 page = alloc_pages(gfp_mask, order);
1296 if (!page)
1297 return 0;
1298 return (unsigned long) page_address(page);
1299}
1300
1301EXPORT_SYMBOL(__get_free_pages);
1302
Al Virodd0fc662005-10-07 07:46:04 +01001303fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304{
1305 struct page * page;
1306
1307 /*
1308 * get_zeroed_page() returns a 32-bit address, which cannot represent
1309 * a highmem page
1310 */
Nick Piggin725d7042006-09-25 23:30:55 -07001311 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
1313 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1314 if (page)
1315 return (unsigned long) page_address(page);
1316 return 0;
1317}
1318
1319EXPORT_SYMBOL(get_zeroed_page);
1320
1321void __pagevec_free(struct pagevec *pvec)
1322{
1323 int i = pagevec_count(pvec);
1324
1325 while (--i >= 0)
1326 free_hot_cold_page(pvec->pages[i], pvec->cold);
1327}
1328
1329fastcall void __free_pages(struct page *page, unsigned int order)
1330{
Nick Pigginb5810032005-10-29 18:16:12 -07001331 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 if (order == 0)
1333 free_hot_page(page);
1334 else
1335 __free_pages_ok(page, order);
1336 }
1337}
1338
1339EXPORT_SYMBOL(__free_pages);
1340
1341fastcall void free_pages(unsigned long addr, unsigned int order)
1342{
1343 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001344 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 __free_pages(virt_to_page((void *)addr), order);
1346 }
1347}
1348
1349EXPORT_SYMBOL(free_pages);
1350
1351/*
1352 * Total amount of free (allocatable) RAM:
1353 */
1354unsigned int nr_free_pages(void)
1355{
1356 unsigned int sum = 0;
1357 struct zone *zone;
1358
1359 for_each_zone(zone)
1360 sum += zone->free_pages;
1361
1362 return sum;
1363}
1364
1365EXPORT_SYMBOL(nr_free_pages);
1366
1367#ifdef CONFIG_NUMA
1368unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1369{
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001370 unsigned int sum = 0;
1371 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
1373 for (i = 0; i < MAX_NR_ZONES; i++)
1374 sum += pgdat->node_zones[i].free_pages;
1375
1376 return sum;
1377}
1378#endif
1379
1380static unsigned int nr_free_zone_pages(int offset)
1381{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001382 /* Just pick one node, since fallback list is circular */
1383 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 unsigned int sum = 0;
1385
Martin J. Blighe310fd42005-07-29 22:59:18 -07001386 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1387 struct zone **zonep = zonelist->zones;
1388 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
Martin J. Blighe310fd42005-07-29 22:59:18 -07001390 for (zone = *zonep++; zone; zone = *zonep++) {
1391 unsigned long size = zone->present_pages;
1392 unsigned long high = zone->pages_high;
1393 if (size > high)
1394 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 }
1396
1397 return sum;
1398}
1399
1400/*
1401 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1402 */
1403unsigned int nr_free_buffer_pages(void)
1404{
Al Viroaf4ca452005-10-21 02:55:38 -04001405 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406}
1407
1408/*
1409 * Amount of free RAM allocatable within all zones
1410 */
1411unsigned int nr_free_pagecache_pages(void)
1412{
Al Viroaf4ca452005-10-21 02:55:38 -04001413 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001415
1416static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001418 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08001419 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422void si_meminfo(struct sysinfo *val)
1423{
1424 val->totalram = totalram_pages;
1425 val->sharedram = 0;
1426 val->freeram = nr_free_pages();
1427 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 val->totalhigh = totalhigh_pages;
1429 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 val->mem_unit = PAGE_SIZE;
1431}
1432
1433EXPORT_SYMBOL(si_meminfo);
1434
1435#ifdef CONFIG_NUMA
1436void si_meminfo_node(struct sysinfo *val, int nid)
1437{
1438 pg_data_t *pgdat = NODE_DATA(nid);
1439
1440 val->totalram = pgdat->node_present_pages;
1441 val->freeram = nr_free_pages_pgdat(pgdat);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001442#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1444 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001445#else
1446 val->totalhigh = 0;
1447 val->freehigh = 0;
1448#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 val->mem_unit = PAGE_SIZE;
1450}
1451#endif
1452
1453#define K(x) ((x) << (PAGE_SHIFT-10))
1454
1455/*
1456 * Show free area list (used inside shift_scroll-lock stuff)
1457 * We also calculate the percentage fragmentation. We do this by counting the
1458 * memory on each free list with the exception of the first item on the list.
1459 */
1460void show_free_areas(void)
1461{
Jes Sorensenc7241912006-09-27 01:50:05 -07001462 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 unsigned long active;
1464 unsigned long inactive;
1465 unsigned long free;
1466 struct zone *zone;
1467
1468 for_each_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07001469 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07001471
1472 show_node(zone);
1473 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474
Dave Jones6b482c62005-11-10 15:45:56 -05001475 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 struct per_cpu_pageset *pageset;
1477
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001478 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479
Jes Sorensenc7241912006-09-27 01:50:05 -07001480 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1481 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1482 cpu, pageset->pcp[0].high,
1483 pageset->pcp[0].batch, pageset->pcp[0].count,
1484 pageset->pcp[1].high, pageset->pcp[1].batch,
1485 pageset->pcp[1].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 }
1487 }
1488
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 get_zone_counts(&active, &inactive, &free);
1490
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1492 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1493 active,
1494 inactive,
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07001495 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07001496 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001497 global_page_state(NR_UNSTABLE_NFS),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 nr_free_pages(),
Christoph Lameter972d1a72006-09-25 23:31:51 -07001499 global_page_state(NR_SLAB_RECLAIMABLE) +
1500 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001501 global_page_state(NR_FILE_MAPPED),
Christoph Lameterdf849a12006-06-30 01:55:38 -07001502 global_page_state(NR_PAGETABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503
1504 for_each_zone(zone) {
1505 int i;
1506
Jes Sorensenc7241912006-09-27 01:50:05 -07001507 if (!populated_zone(zone))
1508 continue;
1509
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 show_node(zone);
1511 printk("%s"
1512 " free:%lukB"
1513 " min:%lukB"
1514 " low:%lukB"
1515 " high:%lukB"
1516 " active:%lukB"
1517 " inactive:%lukB"
1518 " present:%lukB"
1519 " pages_scanned:%lu"
1520 " all_unreclaimable? %s"
1521 "\n",
1522 zone->name,
1523 K(zone->free_pages),
1524 K(zone->pages_min),
1525 K(zone->pages_low),
1526 K(zone->pages_high),
1527 K(zone->nr_active),
1528 K(zone->nr_inactive),
1529 K(zone->present_pages),
1530 zone->pages_scanned,
1531 (zone->all_unreclaimable ? "yes" : "no")
1532 );
1533 printk("lowmem_reserve[]:");
1534 for (i = 0; i < MAX_NR_ZONES; i++)
1535 printk(" %lu", zone->lowmem_reserve[i]);
1536 printk("\n");
1537 }
1538
1539 for_each_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001540 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541
Jes Sorensenc7241912006-09-27 01:50:05 -07001542 if (!populated_zone(zone))
1543 continue;
1544
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 show_node(zone);
1546 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
1548 spin_lock_irqsave(&zone->lock, flags);
1549 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001550 nr[order] = zone->free_area[order].nr_free;
1551 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 }
1553 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001554 for (order = 0; order < MAX_ORDER; order++)
1555 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 printk("= %lukB\n", K(total));
1557 }
1558
1559 show_swap_cache_info();
1560}
1561
1562/*
1563 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001564 *
1565 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001567static int __meminit build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001568 struct zonelist *zonelist, int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569{
Christoph Lameter1a932052006-01-06 00:11:16 -08001570 struct zone *zone;
1571
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001572 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001573 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001574
1575 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001576 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08001577 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001578 if (populated_zone(zone)) {
Christoph Lameter070f8032006-01-06 00:11:19 -08001579 zonelist->zones[nr_zones++] = zone;
1580 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08001582
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001583 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08001584 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585}
1586
1587#ifdef CONFIG_NUMA
1588#define MAX_NODE_LOAD (num_online_nodes())
Yasunori Goto86356ab2006-06-23 02:03:09 -07001589static int __meminitdata node_load[MAX_NUMNODES];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001591 * 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 -07001592 * @node: node whose fallback list we're appending
1593 * @used_node_mask: nodemask_t of already used nodes
1594 *
1595 * We use a number of factors to determine which is the next node that should
1596 * appear on a given node's fallback list. The node should not have appeared
1597 * already in @node's fallback list, and it should be the next closest node
1598 * according to the distance array (which contains arbitrary distance values
1599 * from each node to each node in the system), and should also prefer nodes
1600 * with no CPUs, since presumably they'll have very little allocation pressure
1601 * on them otherwise.
1602 * It returns -1 if no node is found.
1603 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001604static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001606 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 int min_val = INT_MAX;
1608 int best_node = -1;
1609
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001610 /* Use the local node if we haven't already */
1611 if (!node_isset(node, *used_node_mask)) {
1612 node_set(node, *used_node_mask);
1613 return node;
1614 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001616 for_each_online_node(n) {
1617 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
1619 /* Don't want a node to appear more than once */
1620 if (node_isset(n, *used_node_mask))
1621 continue;
1622
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 /* Use the distance array to find the distance */
1624 val = node_distance(node, n);
1625
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001626 /* Penalize nodes under us ("prefer the next node") */
1627 val += (n < node);
1628
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 /* Give preference to headless and unused nodes */
1630 tmp = node_to_cpumask(n);
1631 if (!cpus_empty(tmp))
1632 val += PENALTY_FOR_NODE_WITH_CPUS;
1633
1634 /* Slight preference for less loaded node */
1635 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1636 val += node_load[n];
1637
1638 if (val < min_val) {
1639 min_val = val;
1640 best_node = n;
1641 }
1642 }
1643
1644 if (best_node >= 0)
1645 node_set(best_node, *used_node_mask);
1646
1647 return best_node;
1648}
1649
Yasunori Goto86356ab2006-06-23 02:03:09 -07001650static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651{
Christoph Lameter19655d32006-09-25 23:31:19 -07001652 int j, node, local_node;
1653 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 int prev_node, load;
1655 struct zonelist *zonelist;
1656 nodemask_t used_mask;
1657
1658 /* initialize zonelists */
Christoph Lameter19655d32006-09-25 23:31:19 -07001659 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 zonelist = pgdat->node_zonelists + i;
1661 zonelist->zones[0] = NULL;
1662 }
1663
1664 /* NUMA-aware ordering of nodes */
1665 local_node = pgdat->node_id;
1666 load = num_online_nodes();
1667 prev_node = local_node;
1668 nodes_clear(used_mask);
1669 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001670 int distance = node_distance(local_node, node);
1671
1672 /*
1673 * If another node is sufficiently far away then it is better
1674 * to reclaim pages in a zone before going off node.
1675 */
1676 if (distance > RECLAIM_DISTANCE)
1677 zone_reclaim_mode = 1;
1678
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 /*
1680 * We don't want to pressure a particular node.
1681 * So adding penalty to the first node in same
1682 * distance group to make it round-robin.
1683 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08001684
1685 if (distance != node_distance(local_node, prev_node))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 node_load[node] += load;
1687 prev_node = node;
1688 load--;
Christoph Lameter19655d32006-09-25 23:31:19 -07001689 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 zonelist = pgdat->node_zonelists + i;
1691 for (j = 0; zonelist->zones[j] != NULL; j++);
1692
Christoph Lameter19655d32006-09-25 23:31:19 -07001693 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 zonelist->zones[j] = NULL;
1695 }
1696 }
1697}
1698
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001699/* Construct the zonelist performance cache - see further mmzone.h */
1700static void __meminit build_zonelist_cache(pg_data_t *pgdat)
1701{
1702 int i;
1703
1704 for (i = 0; i < MAX_NR_ZONES; i++) {
1705 struct zonelist *zonelist;
1706 struct zonelist_cache *zlc;
1707 struct zone **z;
1708
1709 zonelist = pgdat->node_zonelists + i;
1710 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
1711 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1712 for (z = zonelist->zones; *z; z++)
1713 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
1714 }
1715}
1716
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717#else /* CONFIG_NUMA */
1718
Yasunori Goto86356ab2006-06-23 02:03:09 -07001719static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720{
Christoph Lameter19655d32006-09-25 23:31:19 -07001721 int node, local_node;
1722 enum zone_type i,j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
1724 local_node = pgdat->node_id;
Christoph Lameter19655d32006-09-25 23:31:19 -07001725 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 struct zonelist *zonelist;
1727
1728 zonelist = pgdat->node_zonelists + i;
1729
Christoph Lameter19655d32006-09-25 23:31:19 -07001730 j = build_zonelists_node(pgdat, zonelist, 0, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 /*
1732 * Now we build the zonelist so that it contains the zones
1733 * of all the other nodes.
1734 * We don't want to pressure a particular node, so when
1735 * building the zones for node N, we make sure that the
1736 * zones coming right after the local ones are those from
1737 * node N+1 (modulo N)
1738 */
1739 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1740 if (!node_online(node))
1741 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07001742 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 }
1744 for (node = 0; node < local_node; node++) {
1745 if (!node_online(node))
1746 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07001747 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 }
1749
1750 zonelist->zones[j] = NULL;
1751 }
1752}
1753
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001754/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
1755static void __meminit build_zonelist_cache(pg_data_t *pgdat)
1756{
1757 int i;
1758
1759 for (i = 0; i < MAX_NR_ZONES; i++)
1760 pgdat->node_zonelists[i].zlcache_ptr = NULL;
1761}
1762
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763#endif /* CONFIG_NUMA */
1764
Yasunori Goto68113782006-06-23 02:03:11 -07001765/* return values int ....just for stop_machine_run() */
1766static int __meminit __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767{
Yasunori Goto68113782006-06-23 02:03:11 -07001768 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001769
1770 for_each_online_node(nid) {
Yasunori Goto68113782006-06-23 02:03:11 -07001771 build_zonelists(NODE_DATA(nid));
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001772 build_zonelist_cache(NODE_DATA(nid));
1773 }
Yasunori Goto68113782006-06-23 02:03:11 -07001774 return 0;
1775}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776
Yasunori Goto68113782006-06-23 02:03:11 -07001777void __meminit build_all_zonelists(void)
1778{
1779 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07001780 __build_all_zonelists(NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07001781 cpuset_init_current_mems_allowed();
1782 } else {
1783 /* we have to stop all cpus to guaranntee there is no user
1784 of zonelist */
1785 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
1786 /* cpuset refresh routine should be here */
1787 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07001788 vm_total_pages = nr_free_pagecache_pages();
1789 printk("Built %i zonelists. Total pages: %ld\n",
1790 num_online_nodes(), vm_total_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791}
1792
1793/*
1794 * Helper functions to size the waitqueue hash table.
1795 * Essentially these want to choose hash table sizes sufficiently
1796 * large so that collisions trying to wait on pages are rare.
1797 * But in fact, the number of active page waitqueues on typical
1798 * systems is ridiculously low, less than 200. So this is even
1799 * conservative, even though it seems large.
1800 *
1801 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1802 * waitqueues, i.e. the size of the waitq table given the number of pages.
1803 */
1804#define PAGES_PER_WAITQUEUE 256
1805
Yasunori Gotocca448f2006-06-23 02:03:10 -07001806#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07001807static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808{
1809 unsigned long size = 1;
1810
1811 pages /= PAGES_PER_WAITQUEUE;
1812
1813 while (size < pages)
1814 size <<= 1;
1815
1816 /*
1817 * Once we have dozens or even hundreds of threads sleeping
1818 * on IO we've got bigger problems than wait queue collision.
1819 * Limit the size of the wait table to a reasonable size.
1820 */
1821 size = min(size, 4096UL);
1822
1823 return max(size, 4UL);
1824}
Yasunori Gotocca448f2006-06-23 02:03:10 -07001825#else
1826/*
1827 * A zone's size might be changed by hot-add, so it is not possible to determine
1828 * a suitable size for its wait_table. So we use the maximum size now.
1829 *
1830 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1831 *
1832 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1833 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1834 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1835 *
1836 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1837 * or more by the traditional way. (See above). It equals:
1838 *
1839 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1840 * ia64(16K page size) : = ( 8G + 4M)byte.
1841 * powerpc (64K page size) : = (32G +16M)byte.
1842 */
1843static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1844{
1845 return 4096UL;
1846}
1847#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848
1849/*
1850 * This is an integer logarithm so that shifts can be used later
1851 * to extract the more random high bits from the multiplicative
1852 * hash function before the remainder is taken.
1853 */
1854static inline unsigned long wait_table_bits(unsigned long size)
1855{
1856 return ffz(~size);
1857}
1858
1859#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1860
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861/*
1862 * Initially all pages are reserved - free ones are freed
1863 * up by free_all_bootmem() once the early boot process is
1864 * done. Non-atomic initialization, single-pass.
1865 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001866void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 unsigned long start_pfn)
1868{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001870 unsigned long end_pfn = start_pfn + size;
1871 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001873 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001874 if (!early_pfn_valid(pfn))
1875 continue;
Andy Whitcroft75167952006-10-21 10:24:14 -07001876 if (!early_pfn_in_nid(pfn, nid))
1877 continue;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001878 page = pfn_to_page(pfn);
1879 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08001880 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 reset_page_mapcount(page);
1882 SetPageReserved(page);
1883 INIT_LIST_HEAD(&page->lru);
1884#ifdef WANT_PAGE_VIRTUAL
1885 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1886 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001887 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 }
1890}
1891
1892void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1893 unsigned long size)
1894{
1895 int order;
1896 for (order = 0; order < MAX_ORDER ; order++) {
1897 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1898 zone->free_area[order].nr_free = 0;
1899 }
1900}
1901
1902#ifndef __HAVE_ARCH_MEMMAP_INIT
1903#define memmap_init(size, nid, zone, start_pfn) \
1904 memmap_init_zone((size), (nid), (zone), (start_pfn))
1905#endif
1906
Ashok Raj6292d9a2006-02-01 03:04:44 -08001907static int __cpuinit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001908{
1909 int batch;
1910
1911 /*
1912 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001913 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001914 *
1915 * OK, so we don't know how big the cache is. So guess.
1916 */
1917 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001918 if (batch * PAGE_SIZE > 512 * 1024)
1919 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001920 batch /= 4; /* We effectively *= 4 below */
1921 if (batch < 1)
1922 batch = 1;
1923
1924 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001925 * Clamp the batch to a 2^n - 1 value. Having a power
1926 * of 2 value was found to be more likely to have
1927 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001928 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001929 * For example if 2 tasks are alternately allocating
1930 * batches of pages, one task can end up with a lot
1931 * of pages of one half of the possible page colors
1932 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001933 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001934 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001935
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001936 return batch;
1937}
1938
Christoph Lameter2caaad42005-06-21 17:15:00 -07001939inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1940{
1941 struct per_cpu_pages *pcp;
1942
Magnus Damm1c6fe942005-10-26 01:58:59 -07001943 memset(p, 0, sizeof(*p));
1944
Christoph Lameter2caaad42005-06-21 17:15:00 -07001945 pcp = &p->pcp[0]; /* hot */
1946 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001947 pcp->high = 6 * batch;
1948 pcp->batch = max(1UL, 1 * batch);
1949 INIT_LIST_HEAD(&pcp->list);
1950
1951 pcp = &p->pcp[1]; /* cold*/
1952 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001953 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001954 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001955 INIT_LIST_HEAD(&pcp->list);
1956}
1957
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001958/*
1959 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1960 * to the value high for the pageset p.
1961 */
1962
1963static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1964 unsigned long high)
1965{
1966 struct per_cpu_pages *pcp;
1967
1968 pcp = &p->pcp[0]; /* hot list */
1969 pcp->high = high;
1970 pcp->batch = max(1UL, high/4);
1971 if ((high/4) > (PAGE_SHIFT * 8))
1972 pcp->batch = PAGE_SHIFT * 8;
1973}
1974
1975
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001976#ifdef CONFIG_NUMA
1977/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001978 * Boot pageset table. One per cpu which is going to be used for all
1979 * zones and all nodes. The parameters will be set in such a way
1980 * that an item put on a list will immediately be handed over to
1981 * the buddy list. This is safe since pageset manipulation is done
1982 * with interrupts disabled.
1983 *
1984 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001985 *
1986 * The boot_pagesets must be kept even after bootup is complete for
1987 * unused processors and/or zones. They do play a role for bootstrapping
1988 * hotplugged processors.
1989 *
1990 * zoneinfo_show() and maybe other functions do
1991 * not check if the processor is online before following the pageset pointer.
1992 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001993 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001994static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001995
1996/*
1997 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001998 * per cpu pageset array in struct zone.
1999 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08002000static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002001{
2002 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002003
2004 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002005
Christoph Lameter66a55032006-09-27 01:50:09 -07002006 if (!populated_zone(zone))
2007 continue;
2008
Nick Piggin23316bc2006-01-08 01:00:41 -08002009 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002010 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08002011 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002012 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002013
Nick Piggin23316bc2006-01-08 01:00:41 -08002014 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002015
2016 if (percpu_pagelist_fraction)
2017 setup_pagelist_highmark(zone_pcp(zone, cpu),
2018 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002019 }
2020
2021 return 0;
2022bad:
2023 for_each_zone(dzone) {
2024 if (dzone == zone)
2025 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08002026 kfree(zone_pcp(dzone, cpu));
2027 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002028 }
2029 return -ENOMEM;
2030}
2031
2032static inline void free_zone_pagesets(int cpu)
2033{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002034 struct zone *zone;
2035
2036 for_each_zone(zone) {
2037 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2038
David Rientjesf3ef9ea2006-09-25 16:24:57 -07002039 /* Free per_cpu_pageset if it is slab allocated */
2040 if (pset != &boot_pageset[cpu])
2041 kfree(pset);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002042 zone_pcp(zone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002043 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002044}
2045
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07002046static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002047 unsigned long action,
2048 void *hcpu)
2049{
2050 int cpu = (long)hcpu;
2051 int ret = NOTIFY_OK;
2052
2053 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08002054 case CPU_UP_PREPARE:
2055 if (process_zones(cpu))
2056 ret = NOTIFY_BAD;
2057 break;
2058 case CPU_UP_CANCELED:
2059 case CPU_DEAD:
2060 free_zone_pagesets(cpu);
2061 break;
2062 default:
2063 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002064 }
2065 return ret;
2066}
2067
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07002068static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002069 { &pageset_cpuup_callback, NULL, 0 };
2070
Al Viro78d99552005-12-15 09:18:25 +00002071void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002072{
2073 int err;
2074
2075 /* Initialize per_cpu_pageset for cpu 0.
2076 * A cpuup callback will do this for every cpu
2077 * as it comes online
2078 */
2079 err = process_zones(smp_processor_id());
2080 BUG_ON(err);
2081 register_cpu_notifier(&pageset_notifier);
2082}
2083
2084#endif
2085
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002086static __meminit
Yasunori Gotocca448f2006-06-23 02:03:10 -07002087int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07002088{
2089 int i;
2090 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002091 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07002092
2093 /*
2094 * The per-page waitqueue mechanism uses hashed waitqueues
2095 * per zone.
2096 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07002097 zone->wait_table_hash_nr_entries =
2098 wait_table_hash_nr_entries(zone_size_pages);
2099 zone->wait_table_bits =
2100 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002101 alloc_size = zone->wait_table_hash_nr_entries
2102 * sizeof(wait_queue_head_t);
2103
2104 if (system_state == SYSTEM_BOOTING) {
2105 zone->wait_table = (wait_queue_head_t *)
2106 alloc_bootmem_node(pgdat, alloc_size);
2107 } else {
2108 /*
2109 * This case means that a zone whose size was 0 gets new memory
2110 * via memory hot-add.
2111 * But it may be the case that a new node was hot-added. In
2112 * this case vmalloc() will not be able to use this new node's
2113 * memory - this wait_table must be initialized to use this new
2114 * node itself as well.
2115 * To use this new node's memory, further consideration will be
2116 * necessary.
2117 */
2118 zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size);
2119 }
2120 if (!zone->wait_table)
2121 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07002122
Yasunori Goto02b694d2006-06-23 02:03:08 -07002123 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07002124 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002125
2126 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002127}
2128
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002129static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07002130{
2131 int cpu;
2132 unsigned long batch = zone_batchsize(zone);
2133
2134 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2135#ifdef CONFIG_NUMA
2136 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002137 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002138 setup_pageset(&boot_pageset[cpu],0);
2139#else
2140 setup_pageset(zone_pcp(zone,cpu), batch);
2141#endif
2142 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08002143 if (zone->present_pages)
2144 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2145 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07002146}
2147
Yasunori Goto718127c2006-06-23 02:03:10 -07002148__meminit int init_currently_empty_zone(struct zone *zone,
2149 unsigned long zone_start_pfn,
2150 unsigned long size)
Dave Hansened8ece22005-10-29 18:16:50 -07002151{
2152 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002153 int ret;
2154 ret = zone_wait_table_init(zone, size);
2155 if (ret)
2156 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07002157 pgdat->nr_zones = zone_idx(zone) + 1;
2158
Dave Hansened8ece22005-10-29 18:16:50 -07002159 zone->zone_start_pfn = zone_start_pfn;
2160
2161 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2162
2163 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
Yasunori Goto718127c2006-06-23 02:03:10 -07002164
2165 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002166}
2167
Mel Gormanc7132162006-09-27 01:49:43 -07002168#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2169/*
2170 * Basic iterator support. Return the first range of PFNs for a node
2171 * Note: nid == MAX_NUMNODES returns first region regardless of node
2172 */
2173static int __init first_active_region_index_in_nid(int nid)
2174{
2175 int i;
2176
2177 for (i = 0; i < nr_nodemap_entries; i++)
2178 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2179 return i;
2180
2181 return -1;
2182}
2183
2184/*
2185 * Basic iterator support. Return the next active range of PFNs for a node
2186 * Note: nid == MAX_NUMNODES returns next region regardles of node
2187 */
2188static int __init next_active_region_index_in_nid(int index, int nid)
2189{
2190 for (index = index + 1; index < nr_nodemap_entries; index++)
2191 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2192 return index;
2193
2194 return -1;
2195}
2196
2197#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2198/*
2199 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2200 * Architectures may implement their own version but if add_active_range()
2201 * was used and there are no special requirements, this is a convenient
2202 * alternative
2203 */
2204int __init early_pfn_to_nid(unsigned long pfn)
2205{
2206 int i;
2207
2208 for (i = 0; i < nr_nodemap_entries; i++) {
2209 unsigned long start_pfn = early_node_map[i].start_pfn;
2210 unsigned long end_pfn = early_node_map[i].end_pfn;
2211
2212 if (start_pfn <= pfn && pfn < end_pfn)
2213 return early_node_map[i].nid;
2214 }
2215
2216 return 0;
2217}
2218#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2219
2220/* Basic iterator support to walk early_node_map[] */
2221#define for_each_active_range_index_in_nid(i, nid) \
2222 for (i = first_active_region_index_in_nid(nid); i != -1; \
2223 i = next_active_region_index_in_nid(i, nid))
2224
2225/**
2226 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002227 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2228 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07002229 *
2230 * If an architecture guarantees that all ranges registered with
2231 * add_active_ranges() contain no holes and may be freed, this
2232 * this function may be used instead of calling free_bootmem() manually.
2233 */
2234void __init free_bootmem_with_active_regions(int nid,
2235 unsigned long max_low_pfn)
2236{
2237 int i;
2238
2239 for_each_active_range_index_in_nid(i, nid) {
2240 unsigned long size_pages = 0;
2241 unsigned long end_pfn = early_node_map[i].end_pfn;
2242
2243 if (early_node_map[i].start_pfn >= max_low_pfn)
2244 continue;
2245
2246 if (end_pfn > max_low_pfn)
2247 end_pfn = max_low_pfn;
2248
2249 size_pages = end_pfn - early_node_map[i].start_pfn;
2250 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2251 PFN_PHYS(early_node_map[i].start_pfn),
2252 size_pages << PAGE_SHIFT);
2253 }
2254}
2255
2256/**
2257 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002258 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07002259 *
2260 * If an architecture guarantees that all ranges registered with
2261 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002262 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07002263 */
2264void __init sparse_memory_present_with_active_regions(int nid)
2265{
2266 int i;
2267
2268 for_each_active_range_index_in_nid(i, nid)
2269 memory_present(early_node_map[i].nid,
2270 early_node_map[i].start_pfn,
2271 early_node_map[i].end_pfn);
2272}
2273
2274/**
Mel Gormanfb014392006-09-27 01:49:59 -07002275 * push_node_boundaries - Push node boundaries to at least the requested boundary
2276 * @nid: The nid of the node to push the boundary for
2277 * @start_pfn: The start pfn of the node
2278 * @end_pfn: The end pfn of the node
2279 *
2280 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2281 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2282 * be hotplugged even though no physical memory exists. This function allows
2283 * an arch to push out the node boundaries so mem_map is allocated that can
2284 * be used later.
2285 */
2286#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2287void __init push_node_boundaries(unsigned int nid,
2288 unsigned long start_pfn, unsigned long end_pfn)
2289{
2290 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2291 nid, start_pfn, end_pfn);
2292
2293 /* Initialise the boundary for this node if necessary */
2294 if (node_boundary_end_pfn[nid] == 0)
2295 node_boundary_start_pfn[nid] = -1UL;
2296
2297 /* Update the boundaries */
2298 if (node_boundary_start_pfn[nid] > start_pfn)
2299 node_boundary_start_pfn[nid] = start_pfn;
2300 if (node_boundary_end_pfn[nid] < end_pfn)
2301 node_boundary_end_pfn[nid] = end_pfn;
2302}
2303
2304/* If necessary, push the node boundary out for reserve hotadd */
2305static void __init account_node_boundary(unsigned int nid,
2306 unsigned long *start_pfn, unsigned long *end_pfn)
2307{
2308 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2309 nid, *start_pfn, *end_pfn);
2310
2311 /* Return if boundary information has not been provided */
2312 if (node_boundary_end_pfn[nid] == 0)
2313 return;
2314
2315 /* Check the boundaries and update if necessary */
2316 if (node_boundary_start_pfn[nid] < *start_pfn)
2317 *start_pfn = node_boundary_start_pfn[nid];
2318 if (node_boundary_end_pfn[nid] > *end_pfn)
2319 *end_pfn = node_boundary_end_pfn[nid];
2320}
2321#else
2322void __init push_node_boundaries(unsigned int nid,
2323 unsigned long start_pfn, unsigned long end_pfn) {}
2324
2325static void __init account_node_boundary(unsigned int nid,
2326 unsigned long *start_pfn, unsigned long *end_pfn) {}
2327#endif
2328
2329
2330/**
Mel Gormanc7132162006-09-27 01:49:43 -07002331 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002332 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2333 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2334 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07002335 *
2336 * It returns the start and end page frame of a node based on information
2337 * provided by an arch calling add_active_range(). If called for a node
2338 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002339 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07002340 */
2341void __init get_pfn_range_for_nid(unsigned int nid,
2342 unsigned long *start_pfn, unsigned long *end_pfn)
2343{
2344 int i;
2345 *start_pfn = -1UL;
2346 *end_pfn = 0;
2347
2348 for_each_active_range_index_in_nid(i, nid) {
2349 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
2350 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
2351 }
2352
2353 if (*start_pfn == -1UL) {
2354 printk(KERN_WARNING "Node %u active with no memory\n", nid);
2355 *start_pfn = 0;
2356 }
Mel Gormanfb014392006-09-27 01:49:59 -07002357
2358 /* Push the node boundaries out if requested */
2359 account_node_boundary(nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07002360}
2361
2362/*
2363 * Return the number of pages a zone spans in a node, including holes
2364 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2365 */
2366unsigned long __init zone_spanned_pages_in_node(int nid,
2367 unsigned long zone_type,
2368 unsigned long *ignored)
2369{
2370 unsigned long node_start_pfn, node_end_pfn;
2371 unsigned long zone_start_pfn, zone_end_pfn;
2372
2373 /* Get the start and end of the node and zone */
2374 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2375 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
2376 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
2377
2378 /* Check that this node has pages within the zone's required range */
2379 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
2380 return 0;
2381
2382 /* Move the zone boundaries inside the node if necessary */
2383 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
2384 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
2385
2386 /* Return the spanned pages */
2387 return zone_end_pfn - zone_start_pfn;
2388}
2389
2390/*
2391 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002392 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07002393 */
2394unsigned long __init __absent_pages_in_range(int nid,
2395 unsigned long range_start_pfn,
2396 unsigned long range_end_pfn)
2397{
2398 int i = 0;
2399 unsigned long prev_end_pfn = 0, hole_pages = 0;
2400 unsigned long start_pfn;
2401
2402 /* Find the end_pfn of the first active range of pfns in the node */
2403 i = first_active_region_index_in_nid(nid);
2404 if (i == -1)
2405 return 0;
2406
Mel Gorman9c7cd682006-09-27 01:49:58 -07002407 /* Account for ranges before physical memory on this node */
2408 if (early_node_map[i].start_pfn > range_start_pfn)
2409 hole_pages = early_node_map[i].start_pfn - range_start_pfn;
2410
Mel Gormanc7132162006-09-27 01:49:43 -07002411 prev_end_pfn = early_node_map[i].start_pfn;
2412
2413 /* Find all holes for the zone within the node */
2414 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
2415
2416 /* No need to continue if prev_end_pfn is outside the zone */
2417 if (prev_end_pfn >= range_end_pfn)
2418 break;
2419
2420 /* Make sure the end of the zone is not within the hole */
2421 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
2422 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
2423
2424 /* Update the hole size cound and move on */
2425 if (start_pfn > range_start_pfn) {
2426 BUG_ON(prev_end_pfn > start_pfn);
2427 hole_pages += start_pfn - prev_end_pfn;
2428 }
2429 prev_end_pfn = early_node_map[i].end_pfn;
2430 }
2431
Mel Gorman9c7cd682006-09-27 01:49:58 -07002432 /* Account for ranges past physical memory on this node */
2433 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07002434 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07002435 max(range_start_pfn, prev_end_pfn);
2436
Mel Gormanc7132162006-09-27 01:49:43 -07002437 return hole_pages;
2438}
2439
2440/**
2441 * absent_pages_in_range - Return number of page frames in holes within a range
2442 * @start_pfn: The start PFN to start searching for holes
2443 * @end_pfn: The end PFN to stop searching for holes
2444 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002445 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07002446 */
2447unsigned long __init absent_pages_in_range(unsigned long start_pfn,
2448 unsigned long end_pfn)
2449{
2450 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
2451}
2452
2453/* Return the number of page frames in holes in a zone on a node */
2454unsigned long __init zone_absent_pages_in_node(int nid,
2455 unsigned long zone_type,
2456 unsigned long *ignored)
2457{
Mel Gorman9c7cd682006-09-27 01:49:58 -07002458 unsigned long node_start_pfn, node_end_pfn;
2459 unsigned long zone_start_pfn, zone_end_pfn;
2460
2461 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2462 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
2463 node_start_pfn);
2464 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
2465 node_end_pfn);
2466
2467 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07002468}
Mel Gorman0e0b8642006-09-27 01:49:56 -07002469
Mel Gormanc7132162006-09-27 01:49:43 -07002470#else
2471static inline unsigned long zone_spanned_pages_in_node(int nid,
2472 unsigned long zone_type,
2473 unsigned long *zones_size)
2474{
2475 return zones_size[zone_type];
2476}
2477
2478static inline unsigned long zone_absent_pages_in_node(int nid,
2479 unsigned long zone_type,
2480 unsigned long *zholes_size)
2481{
2482 if (!zholes_size)
2483 return 0;
2484
2485 return zholes_size[zone_type];
2486}
Mel Gorman0e0b8642006-09-27 01:49:56 -07002487
Mel Gormanc7132162006-09-27 01:49:43 -07002488#endif
2489
2490static void __init calculate_node_totalpages(struct pglist_data *pgdat,
2491 unsigned long *zones_size, unsigned long *zholes_size)
2492{
2493 unsigned long realtotalpages, totalpages = 0;
2494 enum zone_type i;
2495
2496 for (i = 0; i < MAX_NR_ZONES; i++)
2497 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
2498 zones_size);
2499 pgdat->node_spanned_pages = totalpages;
2500
2501 realtotalpages = totalpages;
2502 for (i = 0; i < MAX_NR_ZONES; i++)
2503 realtotalpages -=
2504 zone_absent_pages_in_node(pgdat->node_id, i,
2505 zholes_size);
2506 pgdat->node_present_pages = realtotalpages;
2507 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
2508 realtotalpages);
2509}
2510
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511/*
2512 * Set up the zone data structures:
2513 * - mark all pages reserved
2514 * - mark all memory queues empty
2515 * - clear the memory bitmaps
2516 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07002517static void __meminit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 unsigned long *zones_size, unsigned long *zholes_size)
2519{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07002520 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07002521 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07002523 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524
Dave Hansen208d54e2005-10-29 18:16:52 -07002525 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 pgdat->nr_zones = 0;
2527 init_waitqueue_head(&pgdat->kswapd_wait);
2528 pgdat->kswapd_max_order = 0;
2529
2530 for (j = 0; j < MAX_NR_ZONES; j++) {
2531 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07002532 unsigned long size, realsize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533
Mel Gormanc7132162006-09-27 01:49:43 -07002534 size = zone_spanned_pages_in_node(nid, j, zones_size);
2535 realsize = size - zone_absent_pages_in_node(nid, j,
2536 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537
Mel Gorman0e0b8642006-09-27 01:49:56 -07002538 /*
2539 * Adjust realsize so that it accounts for how much memory
2540 * is used by this zone for memmap. This affects the watermark
2541 * and per-cpu initialisations
2542 */
2543 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
2544 if (realsize >= memmap_pages) {
2545 realsize -= memmap_pages;
2546 printk(KERN_DEBUG
2547 " %s zone: %lu pages used for memmap\n",
2548 zone_names[j], memmap_pages);
2549 } else
2550 printk(KERN_WARNING
2551 " %s zone: %lu pages exceeds realsize %lu\n",
2552 zone_names[j], memmap_pages, realsize);
2553
2554 /* Account for reserved DMA pages */
2555 if (j == ZONE_DMA && realsize > dma_reserve) {
2556 realsize -= dma_reserve;
2557 printk(KERN_DEBUG " DMA zone: %lu pages reserved\n",
2558 dma_reserve);
2559 }
2560
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002561 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 nr_kernel_pages += realsize;
2563 nr_all_pages += realsize;
2564
2565 zone->spanned_pages = size;
2566 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07002567#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07002568 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07002569 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07002570 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07002571 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07002572#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 zone->name = zone_names[j];
2574 spin_lock_init(&zone->lock);
2575 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002576 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 zone->zone_pgdat = pgdat;
2578 zone->free_pages = 0;
2579
Martin Bligh3bb1a852006-10-28 10:38:24 -07002580 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581
Dave Hansened8ece22005-10-29 18:16:50 -07002582 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 INIT_LIST_HEAD(&zone->active_list);
2584 INIT_LIST_HEAD(&zone->inactive_list);
2585 zone->nr_scan_active = 0;
2586 zone->nr_scan_inactive = 0;
2587 zone->nr_active = 0;
2588 zone->nr_inactive = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07002589 zap_zone_vm_stats(zone);
Martin Hicks53e9a612005-09-03 15:54:51 -07002590 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 if (!size)
2592 continue;
2593
Yasunori Goto718127c2006-06-23 02:03:10 -07002594 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
2595 BUG_ON(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 }
2598}
2599
2600static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2601{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 /* Skip empty nodes */
2603 if (!pgdat->node_spanned_pages)
2604 return;
2605
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002606#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 /* ia64 gets its own node_mem_map, before this, without bootmem */
2608 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07002609 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002610 struct page *map;
2611
Bob Piccoe984bb42006-05-20 15:00:31 -07002612 /*
2613 * The zone's endpoints aren't required to be MAX_ORDER
2614 * aligned but the node_mem_map endpoints must be in order
2615 * for the buddy allocator to function correctly.
2616 */
2617 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
2618 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
2619 end = ALIGN(end, MAX_ORDER_NR_PAGES);
2620 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002621 map = alloc_remap(pgdat->node_id, size);
2622 if (!map)
2623 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07002624 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002626#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 /*
2628 * With no DISCONTIG, the global mem_map is just set as node 0's
2629 */
Mel Gormanc7132162006-09-27 01:49:43 -07002630 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07002632#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2633 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
2634 mem_map -= pgdat->node_start_pfn;
2635#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2636 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002638#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639}
2640
Yasunori Goto86356ab2006-06-23 02:03:09 -07002641void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 unsigned long *zones_size, unsigned long node_start_pfn,
2643 unsigned long *zholes_size)
2644{
2645 pgdat->node_id = nid;
2646 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07002647 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648
2649 alloc_node_mem_map(pgdat);
2650
2651 free_area_init_core(pgdat, zones_size, zholes_size);
2652}
2653
Mel Gormanc7132162006-09-27 01:49:43 -07002654#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2655/**
2656 * add_active_range - Register a range of PFNs backed by physical memory
2657 * @nid: The node ID the range resides on
2658 * @start_pfn: The start PFN of the available physical memory
2659 * @end_pfn: The end PFN of the available physical memory
2660 *
2661 * These ranges are stored in an early_node_map[] and later used by
2662 * free_area_init_nodes() to calculate zone sizes and holes. If the
2663 * range spans a memory hole, it is up to the architecture to ensure
2664 * the memory is not freed by the bootmem allocator. If possible
2665 * the range being registered will be merged with existing ranges.
2666 */
2667void __init add_active_range(unsigned int nid, unsigned long start_pfn,
2668 unsigned long end_pfn)
2669{
2670 int i;
2671
2672 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
2673 "%d entries of %d used\n",
2674 nid, start_pfn, end_pfn,
2675 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
2676
2677 /* Merge with existing active regions if possible */
2678 for (i = 0; i < nr_nodemap_entries; i++) {
2679 if (early_node_map[i].nid != nid)
2680 continue;
2681
2682 /* Skip if an existing region covers this new one */
2683 if (start_pfn >= early_node_map[i].start_pfn &&
2684 end_pfn <= early_node_map[i].end_pfn)
2685 return;
2686
2687 /* Merge forward if suitable */
2688 if (start_pfn <= early_node_map[i].end_pfn &&
2689 end_pfn > early_node_map[i].end_pfn) {
2690 early_node_map[i].end_pfn = end_pfn;
2691 return;
2692 }
2693
2694 /* Merge backward if suitable */
2695 if (start_pfn < early_node_map[i].end_pfn &&
2696 end_pfn >= early_node_map[i].start_pfn) {
2697 early_node_map[i].start_pfn = start_pfn;
2698 return;
2699 }
2700 }
2701
2702 /* Check that early_node_map is large enough */
2703 if (i >= MAX_ACTIVE_REGIONS) {
2704 printk(KERN_CRIT "More than %d memory regions, truncating\n",
2705 MAX_ACTIVE_REGIONS);
2706 return;
2707 }
2708
2709 early_node_map[i].nid = nid;
2710 early_node_map[i].start_pfn = start_pfn;
2711 early_node_map[i].end_pfn = end_pfn;
2712 nr_nodemap_entries = i + 1;
2713}
2714
2715/**
2716 * shrink_active_range - Shrink an existing registered range of PFNs
2717 * @nid: The node id the range is on that should be shrunk
2718 * @old_end_pfn: The old end PFN of the range
2719 * @new_end_pfn: The new PFN of the range
2720 *
2721 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
2722 * The map is kept at the end physical page range that has already been
2723 * registered with add_active_range(). This function allows an arch to shrink
2724 * an existing registered range.
2725 */
2726void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
2727 unsigned long new_end_pfn)
2728{
2729 int i;
2730
2731 /* Find the old active region end and shrink */
2732 for_each_active_range_index_in_nid(i, nid)
2733 if (early_node_map[i].end_pfn == old_end_pfn) {
2734 early_node_map[i].end_pfn = new_end_pfn;
2735 break;
2736 }
2737}
2738
2739/**
2740 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002741 *
Mel Gormanc7132162006-09-27 01:49:43 -07002742 * During discovery, it may be found that a table like SRAT is invalid
2743 * and an alternative discovery method must be used. This function removes
2744 * all currently registered regions.
2745 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002746void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07002747{
2748 memset(early_node_map, 0, sizeof(early_node_map));
2749 nr_nodemap_entries = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07002750#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2751 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
2752 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
2753#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -07002754}
2755
2756/* Compare two active node_active_regions */
2757static int __init cmp_node_active_region(const void *a, const void *b)
2758{
2759 struct node_active_region *arange = (struct node_active_region *)a;
2760 struct node_active_region *brange = (struct node_active_region *)b;
2761
2762 /* Done this way to avoid overflows */
2763 if (arange->start_pfn > brange->start_pfn)
2764 return 1;
2765 if (arange->start_pfn < brange->start_pfn)
2766 return -1;
2767
2768 return 0;
2769}
2770
2771/* sort the node_map by start_pfn */
2772static void __init sort_node_map(void)
2773{
2774 sort(early_node_map, (size_t)nr_nodemap_entries,
2775 sizeof(struct node_active_region),
2776 cmp_node_active_region, NULL);
2777}
2778
2779/* Find the lowest pfn for a node. This depends on a sorted early_node_map */
2780unsigned long __init find_min_pfn_for_node(unsigned long nid)
2781{
2782 int i;
2783
Mel Gorman1abbfb42006-11-23 12:01:41 +00002784 /* Regions in the early_node_map can be in any order */
2785 sort_node_map();
2786
Mel Gormanc7132162006-09-27 01:49:43 -07002787 /* Assuming a sorted map, the first range found has the starting pfn */
2788 for_each_active_range_index_in_nid(i, nid)
2789 return early_node_map[i].start_pfn;
2790
2791 printk(KERN_WARNING "Could not find start_pfn for node %lu\n", nid);
2792 return 0;
2793}
2794
2795/**
2796 * find_min_pfn_with_active_regions - Find the minimum PFN registered
2797 *
2798 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002799 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07002800 */
2801unsigned long __init find_min_pfn_with_active_regions(void)
2802{
2803 return find_min_pfn_for_node(MAX_NUMNODES);
2804}
2805
2806/**
2807 * find_max_pfn_with_active_regions - Find the maximum PFN registered
2808 *
2809 * It returns the maximum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002810 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07002811 */
2812unsigned long __init find_max_pfn_with_active_regions(void)
2813{
2814 int i;
2815 unsigned long max_pfn = 0;
2816
2817 for (i = 0; i < nr_nodemap_entries; i++)
2818 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
2819
2820 return max_pfn;
2821}
2822
2823/**
2824 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002825 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07002826 *
2827 * This will call free_area_init_node() for each active node in the system.
2828 * Using the page ranges provided by add_active_range(), the size of each
2829 * zone in each node and their holes is calculated. If the maximum PFN
2830 * between two adjacent zones match, it is assumed that the zone is empty.
2831 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
2832 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
2833 * starts where the previous one ended. For example, ZONE_DMA32 starts
2834 * at arch_max_dma_pfn.
2835 */
2836void __init free_area_init_nodes(unsigned long *max_zone_pfn)
2837{
2838 unsigned long nid;
2839 enum zone_type i;
2840
2841 /* Record where the zone boundaries are */
2842 memset(arch_zone_lowest_possible_pfn, 0,
2843 sizeof(arch_zone_lowest_possible_pfn));
2844 memset(arch_zone_highest_possible_pfn, 0,
2845 sizeof(arch_zone_highest_possible_pfn));
2846 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
2847 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
2848 for (i = 1; i < MAX_NR_ZONES; i++) {
2849 arch_zone_lowest_possible_pfn[i] =
2850 arch_zone_highest_possible_pfn[i-1];
2851 arch_zone_highest_possible_pfn[i] =
2852 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
2853 }
2854
Mel Gormanc7132162006-09-27 01:49:43 -07002855 /* Print out the zone ranges */
2856 printk("Zone PFN ranges:\n");
2857 for (i = 0; i < MAX_NR_ZONES; i++)
2858 printk(" %-8s %8lu -> %8lu\n",
2859 zone_names[i],
2860 arch_zone_lowest_possible_pfn[i],
2861 arch_zone_highest_possible_pfn[i]);
2862
2863 /* Print out the early_node_map[] */
2864 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
2865 for (i = 0; i < nr_nodemap_entries; i++)
2866 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
2867 early_node_map[i].start_pfn,
2868 early_node_map[i].end_pfn);
2869
2870 /* Initialise every node */
2871 for_each_online_node(nid) {
2872 pg_data_t *pgdat = NODE_DATA(nid);
2873 free_area_init_node(nid, pgdat, NULL,
2874 find_min_pfn_for_node(nid), NULL);
2875 }
2876}
2877#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2878
Mel Gorman0e0b8642006-09-27 01:49:56 -07002879/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002880 * set_dma_reserve - set the specified number of pages reserved in the first zone
2881 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07002882 *
2883 * The per-cpu batchsize and zone watermarks are determined by present_pages.
2884 * In the DMA zone, a significant percentage may be consumed by kernel image
2885 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002886 * function may optionally be used to account for unfreeable pages in the
2887 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
2888 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07002889 */
2890void __init set_dma_reserve(unsigned long new_dma_reserve)
2891{
2892 dma_reserve = new_dma_reserve;
2893}
2894
Dave Hansen93b75042005-06-23 00:07:47 -07002895#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896static bootmem_data_t contig_bootmem_data;
2897struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2898
2899EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002900#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901
2902void __init free_area_init(unsigned long *zones_size)
2903{
Dave Hansen93b75042005-06-23 00:07:47 -07002904 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2906}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908static int page_alloc_cpu_notify(struct notifier_block *self,
2909 unsigned long action, void *hcpu)
2910{
2911 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912
2913 if (action == CPU_DEAD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914 local_irq_disable();
2915 __drain_pages(cpu);
Christoph Lameterf8891e52006-06-30 01:55:45 -07002916 vm_events_fold_cpu(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 local_irq_enable();
Christoph Lameter2244b952006-06-30 01:55:33 -07002918 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919 }
2920 return NOTIFY_OK;
2921}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922
2923void __init page_alloc_init(void)
2924{
2925 hotcpu_notifier(page_alloc_cpu_notify, 0);
2926}
2927
2928/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002929 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2930 * or min_free_kbytes changes.
2931 */
2932static void calculate_totalreserve_pages(void)
2933{
2934 struct pglist_data *pgdat;
2935 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002936 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002937
2938 for_each_online_pgdat(pgdat) {
2939 for (i = 0; i < MAX_NR_ZONES; i++) {
2940 struct zone *zone = pgdat->node_zones + i;
2941 unsigned long max = 0;
2942
2943 /* Find valid and maximum lowmem_reserve in the zone */
2944 for (j = i; j < MAX_NR_ZONES; j++) {
2945 if (zone->lowmem_reserve[j] > max)
2946 max = zone->lowmem_reserve[j];
2947 }
2948
2949 /* we treat pages_high as reserved pages. */
2950 max += zone->pages_high;
2951
2952 if (max > zone->present_pages)
2953 max = zone->present_pages;
2954 reserve_pages += max;
2955 }
2956 }
2957 totalreserve_pages = reserve_pages;
2958}
2959
2960/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 * setup_per_zone_lowmem_reserve - called whenever
2962 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2963 * has a correct pages reserved value, so an adequate number of
2964 * pages are left in the zone after a successful __alloc_pages().
2965 */
2966static void setup_per_zone_lowmem_reserve(void)
2967{
2968 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002969 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08002971 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 for (j = 0; j < MAX_NR_ZONES; j++) {
2973 struct zone *zone = pgdat->node_zones + j;
2974 unsigned long present_pages = zone->present_pages;
2975
2976 zone->lowmem_reserve[j] = 0;
2977
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002978 idx = j;
2979 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 struct zone *lower_zone;
2981
Christoph Lameter2f6726e2006-09-25 23:31:18 -07002982 idx--;
2983
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2985 sysctl_lowmem_reserve_ratio[idx] = 1;
2986
2987 lower_zone = pgdat->node_zones + idx;
2988 lower_zone->lowmem_reserve[j] = present_pages /
2989 sysctl_lowmem_reserve_ratio[idx];
2990 present_pages += lower_zone->present_pages;
2991 }
2992 }
2993 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002994
2995 /* update totalreserve_pages */
2996 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997}
2998
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002999/**
3000 * setup_per_zone_pages_min - called when min_free_kbytes changes.
3001 *
3002 * Ensures that the pages_{min,low,high} values for each zone are set correctly
3003 * with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 */
Dave Hansen3947be12005-10-29 18:16:54 -07003005void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006{
3007 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
3008 unsigned long lowmem_pages = 0;
3009 struct zone *zone;
3010 unsigned long flags;
3011
3012 /* Calculate total number of !ZONE_HIGHMEM pages */
3013 for_each_zone(zone) {
3014 if (!is_highmem(zone))
3015 lowmem_pages += zone->present_pages;
3016 }
3017
3018 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07003019 u64 tmp;
3020
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07003022 tmp = (u64)pages_min * zone->present_pages;
3023 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 if (is_highmem(zone)) {
3025 /*
Nick Piggin669ed172005-11-13 16:06:45 -08003026 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
3027 * need highmem pages, so cap pages_min to a small
3028 * value here.
3029 *
3030 * The (pages_high-pages_low) and (pages_low-pages_min)
3031 * deltas controls asynch page reclaim, and so should
3032 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 */
3034 int min_pages;
3035
3036 min_pages = zone->present_pages / 1024;
3037 if (min_pages < SWAP_CLUSTER_MAX)
3038 min_pages = SWAP_CLUSTER_MAX;
3039 if (min_pages > 128)
3040 min_pages = 128;
3041 zone->pages_min = min_pages;
3042 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08003043 /*
3044 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 * proportionate to the zone's size.
3046 */
Nick Piggin669ed172005-11-13 16:06:45 -08003047 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048 }
3049
Andrew Mortonac924c62006-05-15 09:43:59 -07003050 zone->pages_low = zone->pages_min + (tmp >> 2);
3051 zone->pages_high = zone->pages_min + (tmp >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052 spin_unlock_irqrestore(&zone->lru_lock, flags);
3053 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003054
3055 /* update totalreserve_pages */
3056 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057}
3058
3059/*
3060 * Initialise min_free_kbytes.
3061 *
3062 * For small machines we want it small (128k min). For large machines
3063 * we want it large (64MB max). But it is not linear, because network
3064 * bandwidth does not increase linearly with machine size. We use
3065 *
3066 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
3067 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
3068 *
3069 * which yields
3070 *
3071 * 16MB: 512k
3072 * 32MB: 724k
3073 * 64MB: 1024k
3074 * 128MB: 1448k
3075 * 256MB: 2048k
3076 * 512MB: 2896k
3077 * 1024MB: 4096k
3078 * 2048MB: 5792k
3079 * 4096MB: 8192k
3080 * 8192MB: 11584k
3081 * 16384MB: 16384k
3082 */
3083static int __init init_per_zone_pages_min(void)
3084{
3085 unsigned long lowmem_kbytes;
3086
3087 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
3088
3089 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
3090 if (min_free_kbytes < 128)
3091 min_free_kbytes = 128;
3092 if (min_free_kbytes > 65536)
3093 min_free_kbytes = 65536;
3094 setup_per_zone_pages_min();
3095 setup_per_zone_lowmem_reserve();
3096 return 0;
3097}
3098module_init(init_per_zone_pages_min)
3099
3100/*
3101 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
3102 * that we can call two helper functions whenever min_free_kbytes
3103 * changes.
3104 */
3105int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
3106 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3107{
3108 proc_dointvec(table, write, file, buffer, length, ppos);
3109 setup_per_zone_pages_min();
3110 return 0;
3111}
3112
Christoph Lameter96146342006-07-03 00:24:13 -07003113#ifdef CONFIG_NUMA
3114int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
3115 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3116{
3117 struct zone *zone;
3118 int rc;
3119
3120 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3121 if (rc)
3122 return rc;
3123
3124 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07003125 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07003126 sysctl_min_unmapped_ratio) / 100;
3127 return 0;
3128}
Christoph Lameter0ff38492006-09-25 23:31:52 -07003129
3130int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
3131 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3132{
3133 struct zone *zone;
3134 int rc;
3135
3136 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3137 if (rc)
3138 return rc;
3139
3140 for_each_zone(zone)
3141 zone->min_slab_pages = (zone->present_pages *
3142 sysctl_min_slab_ratio) / 100;
3143 return 0;
3144}
Christoph Lameter96146342006-07-03 00:24:13 -07003145#endif
3146
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147/*
3148 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
3149 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
3150 * whenever sysctl_lowmem_reserve_ratio changes.
3151 *
3152 * The reserve ratio obviously has absolutely no relation with the
3153 * pages_min watermarks. The lowmem reserve ratio can only make sense
3154 * if in function of the boot time zone sizes.
3155 */
3156int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
3157 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3158{
3159 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3160 setup_per_zone_lowmem_reserve();
3161 return 0;
3162}
3163
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003164/*
3165 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
3166 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
3167 * can have before it gets flushed back to buddy allocator.
3168 */
3169
3170int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
3171 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3172{
3173 struct zone *zone;
3174 unsigned int cpu;
3175 int ret;
3176
3177 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3178 if (!write || (ret == -EINVAL))
3179 return ret;
3180 for_each_zone(zone) {
3181 for_each_online_cpu(cpu) {
3182 unsigned long high;
3183 high = zone->present_pages / percpu_pagelist_fraction;
3184 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
3185 }
3186 }
3187 return 0;
3188}
3189
David S. Millerf034b5d2006-08-24 03:08:07 -07003190int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191
3192#ifdef CONFIG_NUMA
3193static int __init set_hashdist(char *str)
3194{
3195 if (!str)
3196 return 0;
3197 hashdist = simple_strtoul(str, &str, 0);
3198 return 1;
3199}
3200__setup("hashdist=", set_hashdist);
3201#endif
3202
3203/*
3204 * allocate a large system hash table from bootmem
3205 * - it is assumed that the hash table must contain an exact power-of-2
3206 * quantity of entries
3207 * - limit is the number of hash buckets, not the total allocation size
3208 */
3209void *__init alloc_large_system_hash(const char *tablename,
3210 unsigned long bucketsize,
3211 unsigned long numentries,
3212 int scale,
3213 int flags,
3214 unsigned int *_hash_shift,
3215 unsigned int *_hash_mask,
3216 unsigned long limit)
3217{
3218 unsigned long long max = limit;
3219 unsigned long log2qty, size;
3220 void *table = NULL;
3221
3222 /* allow the kernel cmdline to have a say */
3223 if (!numentries) {
3224 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08003225 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
3227 numentries >>= 20 - PAGE_SHIFT;
3228 numentries <<= 20 - PAGE_SHIFT;
3229
3230 /* limit to 1 bucket per 2^scale bytes of low memory */
3231 if (scale > PAGE_SHIFT)
3232 numentries >>= (scale - PAGE_SHIFT);
3233 else
3234 numentries <<= (PAGE_SHIFT - scale);
3235 }
John Hawkes6e692ed2006-03-25 03:08:02 -08003236 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237
3238 /* limit allocation size to 1/16 total memory by default */
3239 if (max == 0) {
3240 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
3241 do_div(max, bucketsize);
3242 }
3243
3244 if (numentries > max)
3245 numentries = max;
3246
3247 log2qty = long_log2(numentries);
3248
3249 do {
3250 size = bucketsize << log2qty;
3251 if (flags & HASH_EARLY)
3252 table = alloc_bootmem(size);
3253 else if (hashdist)
3254 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
3255 else {
3256 unsigned long order;
3257 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
3258 ;
3259 table = (void*) __get_free_pages(GFP_ATOMIC, order);
3260 }
3261 } while (!table && size > PAGE_SIZE && --log2qty);
3262
3263 if (!table)
3264 panic("Failed to allocate %s hash table\n", tablename);
3265
3266 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
3267 tablename,
3268 (1U << log2qty),
3269 long_log2(size) - PAGE_SHIFT,
3270 size);
3271
3272 if (_hash_shift)
3273 *_hash_shift = log2qty;
3274 if (_hash_mask)
3275 *_hash_mask = (1 << log2qty) - 1;
3276
3277 return table;
3278}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003279
3280#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003281struct page *pfn_to_page(unsigned long pfn)
3282{
Andy Whitcroft67de6482006-06-23 02:03:12 -07003283 return __pfn_to_page(pfn);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003284}
3285unsigned long page_to_pfn(struct page *page)
3286{
Andy Whitcroft67de6482006-06-23 02:03:12 -07003287 return __page_to_pfn(page);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003288}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003289EXPORT_SYMBOL(pfn_to_page);
3290EXPORT_SYMBOL(page_to_pfn);
3291#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
Andrew Morton6220ec72006-10-19 23:29:05 -07003292
3293#if MAX_NUMNODES > 1
3294/*
3295 * Find the highest possible node id.
3296 */
3297int highest_possible_node_id(void)
3298{
3299 unsigned int node;
3300 unsigned int highest = 0;
3301
3302 for_each_node_mask(node, node_possible_map)
3303 highest = node;
3304 return highest;
3305}
3306EXPORT_SYMBOL(highest_possible_node_id);
3307#endif