blob: 5b5cbb5e181604b75502f72ec5757ac861b285a9 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070042#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include "internal.h"
44
45/*
46 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
47 * initializer cleaner
48 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070049nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
Dean Nelson7223a932005-03-23 19:00:00 -070050EXPORT_SYMBOL(node_online_map);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070051nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
Dean Nelson7223a932005-03-23 19:00:00 -070052EXPORT_SYMBOL(node_possible_map);
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070053unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070054unsigned long totalreserve_pages __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055long nr_swap_pages;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080056int percpu_pagelist_fraction;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080058static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080059
Linus Torvalds1da177e2005-04-16 15:20:36 -070060/*
61 * results with 256, 32 in the lowmem_reserve sysctl:
62 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
63 * 1G machine -> (16M dma, 784M normal, 224M high)
64 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
65 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
66 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010067 *
68 * TBD: should special case ZONE_DMA32 machines here - in those we normally
69 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -070071int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
72 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -070073#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -070074 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -070075#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -070076 32
77};
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
79EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
81/*
82 * Used by page_zone() to look up the address of the struct zone whose
83 * id is encoded in the upper bits of page->flags
84 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070085struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070086EXPORT_SYMBOL(zone_table);
87
Christoph Lameter2f1b6242006-09-25 23:31:13 -070088static char *zone_names[MAX_NR_ZONES] = {
89 "DMA",
Christoph Lameterfb0e7942006-09-25 23:31:13 -070090#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -070091 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -070092#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -070093 "Normal",
94 "HighMem"
95};
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Nick Piggin13e74442006-01-06 00:10:58 -0800102#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700103static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700105 int ret = 0;
106 unsigned seq;
107 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700108
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700109 do {
110 seq = zone_span_seqbegin(zone);
111 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
112 ret = 1;
113 else if (pfn < zone->zone_start_pfn)
114 ret = 1;
115 } while (zone_span_seqretry(zone, seq));
116
117 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700118}
119
120static int page_is_consistent(struct zone *zone, struct page *page)
121{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122#ifdef CONFIG_HOLES_IN_ZONE
123 if (!pfn_valid(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700124 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125#endif
126 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700127 return 0;
128
129 return 1;
130}
131/*
132 * Temporary debugging check for pages not lying within a given zone.
133 */
134static int bad_range(struct zone *zone, struct page *page)
135{
136 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700138 if (!page_is_consistent(zone, page))
139 return 1;
140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 return 0;
142}
Nick Piggin13e74442006-01-06 00:10:58 -0800143#else
144static inline int bad_range(struct zone *zone, struct page *page)
145{
146 return 0;
147}
148#endif
149
Nick Piggin224abf92006-01-06 00:11:11 -0800150static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151{
Nick Piggin224abf92006-01-06 00:11:11 -0800152 printk(KERN_EMERG "Bad page state in process '%s'\n"
Hugh Dickins7365f3d2006-01-11 12:17:18 -0800153 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
154 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
155 KERN_EMERG "Backtrace:\n",
Nick Piggin224abf92006-01-06 00:11:11 -0800156 current->comm, page, (int)(2*sizeof(unsigned long)),
157 (unsigned long)page->flags, page->mapping,
158 page_mapcount(page), page_count(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 dump_stack();
Hugh Dickins334795e2005-06-21 17:15:08 -0700160 page->flags &= ~(1 << PG_lru |
161 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 1 << PG_active |
164 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700165 1 << PG_reclaim |
166 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 1 << PG_swapcache |
Nick Piggin676165a2006-04-10 11:21:48 +1000168 1 << PG_writeback |
169 1 << PG_buddy );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 set_page_count(page, 0);
171 reset_page_mapcount(page);
172 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700173 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174}
175
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176/*
177 * Higher-order pages are called "compound pages". They are structured thusly:
178 *
179 * The first PAGE_SIZE page is called the "head page".
180 *
181 * The remaining PAGE_SIZE pages are called "tail pages".
182 *
183 * All pages have PG_compound set. All pages have their ->private pointing at
184 * the head page (even the head page has this).
185 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800186 * The first tail page's ->lru.next holds the address of the compound page's
187 * put_page() function. Its ->lru.prev holds the order of allocation.
188 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800190
191static void free_compound_page(struct page *page)
192{
193 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
194}
195
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196static void prep_compound_page(struct page *page, unsigned long order)
197{
198 int i;
199 int nr_pages = 1 << order;
200
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800201 page[1].lru.next = (void *)free_compound_page; /* set dtor */
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800202 page[1].lru.prev = (void *)order;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 for (i = 0; i < nr_pages; i++) {
204 struct page *p = page + i;
205
Nick Piggin5e9dace2006-03-22 00:08:01 -0800206 __SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700207 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 }
209}
210
211static void destroy_compound_page(struct page *page, unsigned long order)
212{
213 int i;
214 int nr_pages = 1 << order;
215
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800216 if (unlikely((unsigned long)page[1].lru.prev != order))
Nick Piggin224abf92006-01-06 00:11:11 -0800217 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
219 for (i = 0; i < nr_pages; i++) {
220 struct page *p = page + i;
221
Nick Piggin224abf92006-01-06 00:11:11 -0800222 if (unlikely(!PageCompound(p) |
223 (page_private(p) != (unsigned long)page)))
224 bad_page(page);
Nick Piggin5e9dace2006-03-22 00:08:01 -0800225 __ClearPageCompound(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 }
227}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
Nick Piggin17cf4402006-03-22 00:08:41 -0800229static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
230{
231 int i;
232
Nick Piggin725d7042006-09-25 23:30:55 -0700233 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
Andrew Morton6626c5d2006-03-22 00:08:42 -0800234 /*
235 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
236 * and __GFP_HIGHMEM from hard or soft interrupt context.
237 */
Nick Piggin725d7042006-09-25 23:30:55 -0700238 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800239 for (i = 0; i < (1 << order); i++)
240 clear_highpage(page + i);
241}
242
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243/*
244 * function for dealing with page's order in buddy system.
245 * zone->lock is already acquired when we use these.
246 * So, we don't need atomic page->flags operations here.
247 */
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700248static inline unsigned long page_order(struct page *page)
249{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700250 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251}
252
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700253static inline void set_page_order(struct page *page, int order)
254{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700255 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000256 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257}
258
259static inline void rmv_page_order(struct page *page)
260{
Nick Piggin676165a2006-04-10 11:21:48 +1000261 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700262 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263}
264
265/*
266 * Locate the struct page for both the matching buddy in our
267 * pair (buddy1) and the combined O(n+1) page they form (page).
268 *
269 * 1) Any buddy B1 will have an order O twin B2 which satisfies
270 * the following equation:
271 * B2 = B1 ^ (1 << O)
272 * For example, if the starting buddy (buddy2) is #8 its order
273 * 1 buddy is #10:
274 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
275 *
276 * 2) Any buddy B will have an order O+1 parent P which
277 * satisfies the following equation:
278 * P = B & ~(1 << O)
279 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200280 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 */
282static inline struct page *
283__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
284{
285 unsigned long buddy_idx = page_idx ^ (1 << order);
286
287 return page + (buddy_idx - page_idx);
288}
289
290static inline unsigned long
291__find_combined_index(unsigned long page_idx, unsigned int order)
292{
293 return (page_idx & ~(1 << order));
294}
295
296/*
297 * This function checks whether a page is free && is the buddy
298 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800299 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000300 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700301 * (c) a page and its buddy have the same order &&
302 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 *
Nick Piggin676165a2006-04-10 11:21:48 +1000304 * For recording whether a page is in the buddy system, we use PG_buddy.
305 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
306 *
307 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700309static inline int page_is_buddy(struct page *page, struct page *buddy,
310 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311{
Nick Piggin13e74442006-01-06 00:10:58 -0800312#ifdef CONFIG_HOLES_IN_ZONE
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700313 if (!pfn_valid(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800314 return 0;
315#endif
316
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700317 if (page_zone_id(page) != page_zone_id(buddy))
318 return 0;
319
320 if (PageBuddy(buddy) && page_order(buddy) == order) {
321 BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700322 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000323 }
Andrew Morton6aa3001b22006-04-18 22:20:52 -0700324 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325}
326
327/*
328 * Freeing function for a buddy system allocator.
329 *
330 * The concept of a buddy system is to maintain direct-mapped table
331 * (containing bit values) for memory blocks of various "orders".
332 * The bottom level table contains the map for the smallest allocatable
333 * units of memory (here, pages), and each level above it describes
334 * pairs of units from the levels below, hence, "buddies".
335 * At a high level, all that happens here is marking the table entry
336 * at the bottom level available, and propagating the changes upward
337 * as necessary, plus some accounting needed to play nicely with other
338 * parts of the VM system.
339 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000340 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700341 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 * So when we are allocating or freeing one, we can derive the state of the
343 * other. That is, if we allocate a small block, and both were
344 * free, the remainder of the region must be split into blocks.
345 * If a block is freed, and its buddy is also free, then this
346 * triggers coalescing into a block of larger size.
347 *
348 * -- wli
349 */
350
Nick Piggin48db57f2006-01-08 01:00:42 -0800351static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 struct zone *zone, unsigned int order)
353{
354 unsigned long page_idx;
355 int order_size = 1 << order;
356
Nick Piggin224abf92006-01-06 00:11:11 -0800357 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 destroy_compound_page(page, order);
359
360 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
361
Nick Piggin725d7042006-09-25 23:30:55 -0700362 VM_BUG_ON(page_idx & (order_size - 1));
363 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
365 zone->free_pages += order_size;
366 while (order < MAX_ORDER-1) {
367 unsigned long combined_idx;
368 struct free_area *area;
369 struct page *buddy;
370
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700372 if (!page_is_buddy(page, buddy, order))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800374
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 list_del(&buddy->lru);
376 area = zone->free_area + order;
377 area->nr_free--;
378 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800379 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 page = page + (combined_idx - page_idx);
381 page_idx = combined_idx;
382 order++;
383 }
384 set_page_order(page, order);
385 list_add(&page->lru, &zone->free_area[order].free_list);
386 zone->free_area[order].nr_free++;
387}
388
Nick Piggin224abf92006-01-06 00:11:11 -0800389static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390{
Nick Piggin92be2e332006-01-06 00:10:57 -0800391 if (unlikely(page_mapcount(page) |
392 (page->mapping != NULL) |
393 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 (page->flags & (
395 1 << PG_lru |
396 1 << PG_private |
397 1 << PG_locked |
398 1 << PG_active |
399 1 << PG_reclaim |
400 1 << PG_slab |
401 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700402 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000403 1 << PG_reserved |
404 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800405 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700407 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800408 /*
409 * For now, we report if PG_reserved was found set, but do not
410 * clear it, and do not free the page. But we shall soon need
411 * to do more, for when the ZERO_PAGE count wraps negative.
412 */
413 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414}
415
416/*
417 * Frees a list of pages.
418 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700419 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 *
421 * If the zone was previously in an "all pages pinned" state then look to
422 * see if this freeing clears that state.
423 *
424 * And clear the zone's pages_scanned counter, to hold off the "all pages are
425 * pinned" detection logic.
426 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800427static void free_pages_bulk(struct zone *zone, int count,
428 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429{
Nick Pigginc54ad302006-01-06 00:10:56 -0800430 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 zone->all_unreclaimable = 0;
432 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800433 while (count--) {
434 struct page *page;
435
Nick Piggin725d7042006-09-25 23:30:55 -0700436 VM_BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800438 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800440 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800442 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443}
444
Nick Piggin48db57f2006-01-08 01:00:42 -0800445static void free_one_page(struct zone *zone, struct page *page, int order)
446{
447 LIST_HEAD(list);
448 list_add(&page->lru, &list);
449 free_pages_bulk(zone, 1, &list, order);
450}
451
452static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
Nick Pigginc54ad302006-01-06 00:10:56 -0800454 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800456 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
458 arch_free_page(page, order);
Ingo Molnarde5097c2006-01-09 15:59:21 -0800459 if (!PageHighMem(page))
Ingo Molnarf9b84042006-06-27 02:54:49 -0700460 debug_check_no_locks_freed(page_address(page),
461 PAGE_SIZE<<order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800464 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800465 if (reserved)
466 return;
467
Nick Piggin48db57f2006-01-08 01:00:42 -0800468 kernel_map_pages(page, 1 << order, 0);
Nick Pigginc54ad302006-01-06 00:10:56 -0800469 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700470 __count_vm_events(PGFREE, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800471 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800472 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473}
474
David Howellsa226f6c2006-01-06 00:11:08 -0800475/*
476 * permit the bootmem allocator to evade page validation on high-order frees
477 */
478void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
479{
480 if (order == 0) {
481 __ClearPageReserved(page);
482 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800483 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800484 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800485 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800486 int loop;
487
Nick Piggin545b1ea2006-03-22 00:08:07 -0800488 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800489 for (loop = 0; loop < BITS_PER_LONG; loop++) {
490 struct page *p = &page[loop];
491
Nick Piggin545b1ea2006-03-22 00:08:07 -0800492 if (loop + 1 < BITS_PER_LONG)
493 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800494 __ClearPageReserved(p);
495 set_page_count(p, 0);
496 }
497
Nick Piggin7835e982006-03-22 00:08:40 -0800498 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800499 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800500 }
501}
502
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
504/*
505 * The order of subdivision here is critical for the IO subsystem.
506 * Please do not alter this order without good reasons and regression
507 * testing. Specifically, as large blocks of memory are subdivided,
508 * the order in which smaller blocks are delivered depends on the order
509 * they're subdivided in this function. This is the primary factor
510 * influencing the order in which pages are delivered to the IO
511 * subsystem according to empirical testing, and this is also justified
512 * by considering the behavior of a buddy system containing a single
513 * large block of memory acted on by a series of small allocations.
514 * This behavior is a critical factor in sglist merging's success.
515 *
516 * -- wli
517 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800518static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 int low, int high, struct free_area *area)
520{
521 unsigned long size = 1 << high;
522
523 while (high > low) {
524 area--;
525 high--;
526 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700527 VM_BUG_ON(bad_range(zone, &page[size]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 list_add(&page[size].lru, &area->free_list);
529 area->nr_free++;
530 set_page_order(&page[size], high);
531 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532}
533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534/*
535 * This page is about to be returned from the page allocator
536 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800537static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538{
Nick Piggin92be2e332006-01-06 00:10:57 -0800539 if (unlikely(page_mapcount(page) |
540 (page->mapping != NULL) |
541 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700542 (page->flags & (
543 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 1 << PG_private |
545 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 1 << PG_active |
547 1 << PG_dirty |
548 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700549 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700551 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000552 1 << PG_reserved |
553 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800554 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Hugh Dickins689bceb2005-11-21 21:32:20 -0800556 /*
557 * For now, we report if PG_reserved was found set, but do not
558 * clear it, and do not allocate the page: as a safety net.
559 */
560 if (PageReserved(page))
561 return 1;
562
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
564 1 << PG_referenced | 1 << PG_arch_1 |
565 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700566 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800567 set_page_refcounted(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800569
570 if (gfp_flags & __GFP_ZERO)
571 prep_zero_page(page, order, gfp_flags);
572
573 if (order && (gfp_flags & __GFP_COMP))
574 prep_compound_page(page, order);
575
Hugh Dickins689bceb2005-11-21 21:32:20 -0800576 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577}
578
579/*
580 * Do the hard work of removing an element from the buddy allocator.
581 * Call me with the zone->lock already held.
582 */
583static struct page *__rmqueue(struct zone *zone, unsigned int order)
584{
585 struct free_area * area;
586 unsigned int current_order;
587 struct page *page;
588
589 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
590 area = zone->free_area + current_order;
591 if (list_empty(&area->free_list))
592 continue;
593
594 page = list_entry(area->free_list.next, struct page, lru);
595 list_del(&page->lru);
596 rmv_page_order(page);
597 area->nr_free--;
598 zone->free_pages -= 1UL << order;
Nick Piggin085cc7d2006-01-06 00:11:01 -0800599 expand(zone, page, order, current_order, area);
600 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 }
602
603 return NULL;
604}
605
606/*
607 * Obtain a specified number of elements from the buddy allocator, all under
608 * a single hold of the lock, for efficiency. Add them to the supplied list.
609 * Returns the number of new pages which were placed at *list.
610 */
611static int rmqueue_bulk(struct zone *zone, unsigned int order,
612 unsigned long count, struct list_head *list)
613{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
Nick Pigginc54ad302006-01-06 00:10:56 -0800616 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 for (i = 0; i < count; ++i) {
Nick Piggin085cc7d2006-01-06 00:11:01 -0800618 struct page *page = __rmqueue(zone, order);
619 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 list_add_tail(&page->lru, list);
622 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800623 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800624 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625}
626
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700627#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800628/*
629 * Called from the slab reaper to drain pagesets on a particular node that
630 * belong to the currently executing processor.
Christoph Lameter879336c2006-03-22 00:09:08 -0800631 * Note that this function must be called with the thread pinned to
632 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800633 */
634void drain_node_pages(int nodeid)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700635{
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800636 int i, z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700637 unsigned long flags;
638
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800639 for (z = 0; z < MAX_NR_ZONES; z++) {
640 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700641 struct per_cpu_pageset *pset;
642
Nick Piggin23316bc2006-01-08 01:00:41 -0800643 pset = zone_pcp(zone, smp_processor_id());
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700644 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
645 struct per_cpu_pages *pcp;
646
647 pcp = &pset->pcp[i];
Christoph Lameter879336c2006-03-22 00:09:08 -0800648 if (pcp->count) {
649 local_irq_save(flags);
650 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
651 pcp->count = 0;
652 local_irq_restore(flags);
653 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700654 }
655 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700656}
657#endif
658
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
660static void __drain_pages(unsigned int cpu)
661{
Nick Pigginc54ad302006-01-06 00:10:56 -0800662 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 struct zone *zone;
664 int i;
665
666 for_each_zone(zone) {
667 struct per_cpu_pageset *pset;
668
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700669 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
671 struct per_cpu_pages *pcp;
672
673 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800674 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800675 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
676 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800677 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 }
679 }
680}
681#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
682
683#ifdef CONFIG_PM
684
685void mark_free_pages(struct zone *zone)
686{
687 unsigned long zone_pfn, flags;
688 int order;
689 struct list_head *curr;
690
691 if (!zone->spanned_pages)
692 return;
693
694 spin_lock_irqsave(&zone->lock, flags);
695 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
696 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
697
698 for (order = MAX_ORDER - 1; order >= 0; --order)
699 list_for_each(curr, &zone->free_area[order].free_list) {
700 unsigned long start_pfn, i;
701
702 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
703
704 for (i=0; i < (1<<order); i++)
705 SetPageNosaveFree(pfn_to_page(start_pfn+i));
706 }
707 spin_unlock_irqrestore(&zone->lock, flags);
708}
709
710/*
711 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
712 */
713void drain_local_pages(void)
714{
715 unsigned long flags;
716
717 local_irq_save(flags);
718 __drain_pages(smp_processor_id());
719 local_irq_restore(flags);
720}
721#endif /* CONFIG_PM */
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723/*
724 * Free a 0-order page
725 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726static void fastcall free_hot_cold_page(struct page *page, int cold)
727{
728 struct zone *zone = page_zone(page);
729 struct per_cpu_pages *pcp;
730 unsigned long flags;
731
732 arch_free_page(page, 0);
733
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 if (PageAnon(page))
735 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800736 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800737 return;
738
Hugh Dickins689bceb2005-11-21 21:32:20 -0800739 kernel_map_pages(page, 1, 0);
740
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700741 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700743 __count_vm_event(PGFREE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 list_add(&page->lru, &pcp->list);
745 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800746 if (pcp->count >= pcp->high) {
747 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
748 pcp->count -= pcp->batch;
749 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 local_irq_restore(flags);
751 put_cpu();
752}
753
754void fastcall free_hot_page(struct page *page)
755{
756 free_hot_cold_page(page, 0);
757}
758
759void fastcall free_cold_page(struct page *page)
760{
761 free_hot_cold_page(page, 1);
762}
763
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800764/*
765 * split_page takes a non-compound higher-order page, and splits it into
766 * n (1<<order) sub-pages: page[0..n]
767 * Each sub-page must be freed individually.
768 *
769 * Note: this is probably too low level an operation for use in drivers.
770 * Please consult with lkml before using this in your driver.
771 */
772void split_page(struct page *page, unsigned int order)
773{
774 int i;
775
Nick Piggin725d7042006-09-25 23:30:55 -0700776 VM_BUG_ON(PageCompound(page));
777 VM_BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -0800778 for (i = 1; i < (1 << order); i++)
779 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800780}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782/*
783 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
784 * we cheat by calling it from here, in the order > 0 path. Saves a branch
785 * or two.
786 */
Nick Piggina74609f2006-01-06 00:11:20 -0800787static struct page *buffered_rmqueue(struct zonelist *zonelist,
788 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789{
790 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800791 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800793 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794
Hugh Dickins689bceb2005-11-21 21:32:20 -0800795again:
Nick Piggina74609f2006-01-06 00:11:20 -0800796 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800797 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 struct per_cpu_pages *pcp;
799
Nick Piggina74609f2006-01-06 00:11:20 -0800800 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800802 if (!pcp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 pcp->count += rmqueue_bulk(zone, 0,
804 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800805 if (unlikely(!pcp->count))
806 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 }
Nick Piggina74609f2006-01-06 00:11:20 -0800808 page = list_entry(pcp->list.next, struct page, lru);
809 list_del(&page->lru);
810 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800811 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 spin_lock_irqsave(&zone->lock, flags);
813 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800814 spin_unlock(&zone->lock);
815 if (!page)
816 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 }
818
Christoph Lameterf8891e52006-06-30 01:55:45 -0700819 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Christoph Lameterca889e62006-06-30 01:55:44 -0700820 zone_statistics(zonelist, zone);
Nick Piggina74609f2006-01-06 00:11:20 -0800821 local_irq_restore(flags);
822 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823
Nick Piggin725d7042006-09-25 23:30:55 -0700824 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -0800825 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -0800826 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800828
829failed:
830 local_irq_restore(flags);
831 put_cpu();
832 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833}
834
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800835#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800836#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
837#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
838#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
839#define ALLOC_HARDER 0x10 /* try to alloc harder */
840#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
841#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800842
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843/*
844 * Return 1 if free pages are above 'mark'. This takes into account the order
845 * of the allocation.
846 */
847int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800848 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849{
850 /* free_pages my go negative - that's OK */
851 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
852 int o;
853
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800854 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800856 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 min -= min / 4;
858
859 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
860 return 0;
861 for (o = 0; o < order; o++) {
862 /* At the next order, this order's pages become unavailable */
863 free_pages -= z->free_area[o].nr_free << o;
864
865 /* Require fewer higher order pages to be free */
866 min >>= 1;
867
868 if (free_pages <= min)
869 return 0;
870 }
871 return 1;
872}
873
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800874/*
875 * get_page_from_freeliest goes through the zonelist trying to allocate
876 * a page.
877 */
878static struct page *
879get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
880 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -0700881{
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800882 struct zone **z = zonelist->zones;
883 struct page *page = NULL;
884 int classzone_idx = zone_idx(*z);
885
886 /*
887 * Go through the zonelist once, looking for a zone with enough free.
888 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
889 */
890 do {
891 if ((alloc_flags & ALLOC_CPUSET) &&
892 !cpuset_zone_allowed(*z, gfp_mask))
893 continue;
894
895 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -0800896 unsigned long mark;
897 if (alloc_flags & ALLOC_WMARK_MIN)
898 mark = (*z)->pages_min;
899 else if (alloc_flags & ALLOC_WMARK_LOW)
900 mark = (*z)->pages_low;
901 else
902 mark = (*z)->pages_high;
903 if (!zone_watermark_ok(*z, order, mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800904 classzone_idx, alloc_flags))
Christoph Lameter9eeff232006-01-18 17:42:31 -0800905 if (!zone_reclaim_mode ||
906 !zone_reclaim(*z, gfp_mask, order))
907 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800908 }
909
Nick Piggina74609f2006-01-06 00:11:20 -0800910 page = buffered_rmqueue(zonelist, *z, order, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800911 if (page) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800912 break;
913 }
914 } while (*(++z) != NULL);
915 return page;
Martin Hicks753ee722005-06-21 17:14:41 -0700916}
917
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918/*
919 * This is the 'heart' of the zoned buddy allocator.
920 */
921struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +0100922__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 struct zonelist *zonelist)
924{
Al Viro260b2362005-10-21 03:22:44 -0400925 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800926 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 struct page *page;
928 struct reclaim_state reclaim_state;
929 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800931 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 int did_some_progress;
933
934 might_sleep_if(wait);
935
Jens Axboe6b1de912005-11-17 21:35:02 +0100936restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800937 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800939 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 /* Should this ever happen?? */
941 return NULL;
942 }
Jens Axboe6b1de912005-11-17 21:35:02 +0100943
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800944 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -0800945 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800946 if (page)
947 goto got_pg;
948
Jens Axboe6b1de912005-11-17 21:35:02 +0100949 do {
Chris Wright43b0bc02006-06-25 05:47:55 -0700950 wakeup_kswapd(*z, order);
Jens Axboe6b1de912005-11-17 21:35:02 +0100951 } while (*(++z));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800952
Paul Jackson9bf22292005-09-06 15:18:12 -0700953 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800954 * OK, we're below the kswapd watermark and have kicked background
955 * reclaim. Now things get more complex, so set up alloc_flags according
956 * to how we want to proceed.
957 *
958 * The caller may dip into page reserves a bit more if the caller
959 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -0800960 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
961 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -0700962 */
Nick Piggin31488902005-11-28 13:44:03 -0800963 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800964 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
965 alloc_flags |= ALLOC_HARDER;
966 if (gfp_mask & __GFP_HIGH)
967 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -0700968 if (wait)
969 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
971 /*
972 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800973 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 *
975 * This is the last chance, in general, before the goto nopage.
976 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -0700977 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800979 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
980 if (page)
981 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
983 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -0700984
985 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
986 && !in_interrupt()) {
987 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -0800988nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -0700989 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800990 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -0800991 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800992 if (page)
993 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -0800994 if (gfp_mask & __GFP_NOFAIL) {
995 blk_congestion_wait(WRITE, HZ/50);
996 goto nofail_alloc;
997 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 }
999 goto nopage;
1000 }
1001
1002 /* Atomic allocations - we can't balance anything */
1003 if (!wait)
1004 goto nopage;
1005
1006rebalance:
1007 cond_resched();
1008
1009 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001010 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 p->flags |= PF_MEMALLOC;
1012 reclaim_state.reclaimed_slab = 0;
1013 p->reclaim_state = &reclaim_state;
1014
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001015 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
1017 p->reclaim_state = NULL;
1018 p->flags &= ~PF_MEMALLOC;
1019
1020 cond_resched();
1021
1022 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001023 page = get_page_from_freelist(gfp_mask, order,
1024 zonelist, alloc_flags);
1025 if (page)
1026 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1028 /*
1029 * Go through the zonelist yet one more time, keep
1030 * very high watermark here, this is only to catch
1031 * a parallel oom killing, we must fail if we're still
1032 * under heavy pressure.
1033 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001034 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001035 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001036 if (page)
1037 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001039 out_of_memory(zonelist, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 goto restart;
1041 }
1042
1043 /*
1044 * Don't let big-order allocations loop unless the caller explicitly
1045 * requests that. Wait for some write requests to complete then retry.
1046 *
1047 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1048 * <= 3, but that may not be true in other implementations.
1049 */
1050 do_retry = 0;
1051 if (!(gfp_mask & __GFP_NORETRY)) {
1052 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1053 do_retry = 1;
1054 if (gfp_mask & __GFP_NOFAIL)
1055 do_retry = 1;
1056 }
1057 if (do_retry) {
1058 blk_congestion_wait(WRITE, HZ/50);
1059 goto rebalance;
1060 }
1061
1062nopage:
1063 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1064 printk(KERN_WARNING "%s: page allocation failure."
1065 " order:%d, mode:0x%x\n",
1066 p->comm, order, gfp_mask);
1067 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001068 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 return page;
1072}
1073
1074EXPORT_SYMBOL(__alloc_pages);
1075
1076/*
1077 * Common helper functions.
1078 */
Al Virodd0fc662005-10-07 07:46:04 +01001079fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080{
1081 struct page * page;
1082 page = alloc_pages(gfp_mask, order);
1083 if (!page)
1084 return 0;
1085 return (unsigned long) page_address(page);
1086}
1087
1088EXPORT_SYMBOL(__get_free_pages);
1089
Al Virodd0fc662005-10-07 07:46:04 +01001090fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091{
1092 struct page * page;
1093
1094 /*
1095 * get_zeroed_page() returns a 32-bit address, which cannot represent
1096 * a highmem page
1097 */
Nick Piggin725d7042006-09-25 23:30:55 -07001098 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
1100 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1101 if (page)
1102 return (unsigned long) page_address(page);
1103 return 0;
1104}
1105
1106EXPORT_SYMBOL(get_zeroed_page);
1107
1108void __pagevec_free(struct pagevec *pvec)
1109{
1110 int i = pagevec_count(pvec);
1111
1112 while (--i >= 0)
1113 free_hot_cold_page(pvec->pages[i], pvec->cold);
1114}
1115
1116fastcall void __free_pages(struct page *page, unsigned int order)
1117{
Nick Pigginb5810032005-10-29 18:16:12 -07001118 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 if (order == 0)
1120 free_hot_page(page);
1121 else
1122 __free_pages_ok(page, order);
1123 }
1124}
1125
1126EXPORT_SYMBOL(__free_pages);
1127
1128fastcall void free_pages(unsigned long addr, unsigned int order)
1129{
1130 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001131 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 __free_pages(virt_to_page((void *)addr), order);
1133 }
1134}
1135
1136EXPORT_SYMBOL(free_pages);
1137
1138/*
1139 * Total amount of free (allocatable) RAM:
1140 */
1141unsigned int nr_free_pages(void)
1142{
1143 unsigned int sum = 0;
1144 struct zone *zone;
1145
1146 for_each_zone(zone)
1147 sum += zone->free_pages;
1148
1149 return sum;
1150}
1151
1152EXPORT_SYMBOL(nr_free_pages);
1153
1154#ifdef CONFIG_NUMA
1155unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1156{
1157 unsigned int i, sum = 0;
1158
1159 for (i = 0; i < MAX_NR_ZONES; i++)
1160 sum += pgdat->node_zones[i].free_pages;
1161
1162 return sum;
1163}
1164#endif
1165
1166static unsigned int nr_free_zone_pages(int offset)
1167{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001168 /* Just pick one node, since fallback list is circular */
1169 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 unsigned int sum = 0;
1171
Martin J. Blighe310fd42005-07-29 22:59:18 -07001172 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1173 struct zone **zonep = zonelist->zones;
1174 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175
Martin J. Blighe310fd42005-07-29 22:59:18 -07001176 for (zone = *zonep++; zone; zone = *zonep++) {
1177 unsigned long size = zone->present_pages;
1178 unsigned long high = zone->pages_high;
1179 if (size > high)
1180 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 }
1182
1183 return sum;
1184}
1185
1186/*
1187 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1188 */
1189unsigned int nr_free_buffer_pages(void)
1190{
Al Viroaf4ca452005-10-21 02:55:38 -04001191 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192}
1193
1194/*
1195 * Amount of free RAM allocatable within all zones
1196 */
1197unsigned int nr_free_pagecache_pages(void)
1198{
Al Viroaf4ca452005-10-21 02:55:38 -04001199 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201#ifdef CONFIG_NUMA
1202static void show_node(struct zone *zone)
1203{
1204 printk("Node %d ", zone->zone_pgdat->node_id);
1205}
1206#else
1207#define show_node(zone) do { } while (0)
1208#endif
1209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210void si_meminfo(struct sysinfo *val)
1211{
1212 val->totalram = totalram_pages;
1213 val->sharedram = 0;
1214 val->freeram = nr_free_pages();
1215 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 val->totalhigh = totalhigh_pages;
1217 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 val->mem_unit = PAGE_SIZE;
1219}
1220
1221EXPORT_SYMBOL(si_meminfo);
1222
1223#ifdef CONFIG_NUMA
1224void si_meminfo_node(struct sysinfo *val, int nid)
1225{
1226 pg_data_t *pgdat = NODE_DATA(nid);
1227
1228 val->totalram = pgdat->node_present_pages;
1229 val->freeram = nr_free_pages_pgdat(pgdat);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001230#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1232 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001233#else
1234 val->totalhigh = 0;
1235 val->freehigh = 0;
1236#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 val->mem_unit = PAGE_SIZE;
1238}
1239#endif
1240
1241#define K(x) ((x) << (PAGE_SHIFT-10))
1242
1243/*
1244 * Show free area list (used inside shift_scroll-lock stuff)
1245 * We also calculate the percentage fragmentation. We do this by counting the
1246 * memory on each free list with the exception of the first item on the list.
1247 */
1248void show_free_areas(void)
1249{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 int cpu, temperature;
1251 unsigned long active;
1252 unsigned long inactive;
1253 unsigned long free;
1254 struct zone *zone;
1255
1256 for_each_zone(zone) {
1257 show_node(zone);
1258 printk("%s per-cpu:", zone->name);
1259
Con Kolivasf3fe6512006-01-06 00:11:15 -08001260 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 printk(" empty\n");
1262 continue;
1263 } else
1264 printk("\n");
1265
Dave Jones6b482c62005-11-10 15:45:56 -05001266 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 struct per_cpu_pageset *pageset;
1268
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001269 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
1271 for (temperature = 0; temperature < 2; temperature++)
Nick Piggin2d92c5c2006-01-06 00:10:59 -08001272 printk("cpu %d %s: high %d, batch %d used:%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 cpu,
1274 temperature ? "cold" : "hot",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 pageset->pcp[temperature].high,
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001276 pageset->pcp[temperature].batch,
1277 pageset->pcp[temperature].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 }
1279 }
1280
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 get_zone_counts(&active, &inactive, &free);
1282
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1284 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1285 active,
1286 inactive,
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07001287 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07001288 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001289 global_page_state(NR_UNSTABLE_NFS),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 nr_free_pages(),
Christoph Lameter9a865ff2006-06-30 01:55:38 -07001291 global_page_state(NR_SLAB),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001292 global_page_state(NR_FILE_MAPPED),
Christoph Lameterdf849a12006-06-30 01:55:38 -07001293 global_page_state(NR_PAGETABLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
1295 for_each_zone(zone) {
1296 int i;
1297
1298 show_node(zone);
1299 printk("%s"
1300 " free:%lukB"
1301 " min:%lukB"
1302 " low:%lukB"
1303 " high:%lukB"
1304 " active:%lukB"
1305 " inactive:%lukB"
1306 " present:%lukB"
1307 " pages_scanned:%lu"
1308 " all_unreclaimable? %s"
1309 "\n",
1310 zone->name,
1311 K(zone->free_pages),
1312 K(zone->pages_min),
1313 K(zone->pages_low),
1314 K(zone->pages_high),
1315 K(zone->nr_active),
1316 K(zone->nr_inactive),
1317 K(zone->present_pages),
1318 zone->pages_scanned,
1319 (zone->all_unreclaimable ? "yes" : "no")
1320 );
1321 printk("lowmem_reserve[]:");
1322 for (i = 0; i < MAX_NR_ZONES; i++)
1323 printk(" %lu", zone->lowmem_reserve[i]);
1324 printk("\n");
1325 }
1326
1327 for_each_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001328 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
1330 show_node(zone);
1331 printk("%s: ", zone->name);
Con Kolivasf3fe6512006-01-06 00:11:15 -08001332 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 printk("empty\n");
1334 continue;
1335 }
1336
1337 spin_lock_irqsave(&zone->lock, flags);
1338 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001339 nr[order] = zone->free_area[order].nr_free;
1340 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 }
1342 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001343 for (order = 0; order < MAX_ORDER; order++)
1344 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 printk("= %lukB\n", K(total));
1346 }
1347
1348 show_swap_cache_info();
1349}
1350
1351/*
1352 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001353 *
1354 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001356static int __meminit build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter070f8032006-01-06 00:11:19 -08001357 struct zonelist *zonelist, int nr_zones, int zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358{
Christoph Lameter1a932052006-01-06 00:11:16 -08001359 struct zone *zone;
1360
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001361 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter02a68a52006-01-06 00:11:18 -08001362
1363 do {
Christoph Lameter070f8032006-01-06 00:11:19 -08001364 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001365 if (populated_zone(zone)) {
Christoph Lameter070f8032006-01-06 00:11:19 -08001366 zonelist->zones[nr_zones++] = zone;
1367 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 }
Christoph Lameter070f8032006-01-06 00:11:19 -08001369 zone_type--;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001370
Christoph Lameter070f8032006-01-06 00:11:19 -08001371 } while (zone_type >= 0);
1372 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373}
1374
Al Viro260b2362005-10-21 03:22:44 -04001375static inline int highest_zone(int zone_bits)
1376{
1377 int res = ZONE_NORMAL;
1378 if (zone_bits & (__force int)__GFP_HIGHMEM)
1379 res = ZONE_HIGHMEM;
Christoph Lameterfb0e7942006-09-25 23:31:13 -07001380#ifdef CONFIG_ZONE_DMA32
Andi Kleena2f1b422005-11-05 17:25:53 +01001381 if (zone_bits & (__force int)__GFP_DMA32)
1382 res = ZONE_DMA32;
Christoph Lameterfb0e7942006-09-25 23:31:13 -07001383#endif
Al Viro260b2362005-10-21 03:22:44 -04001384 if (zone_bits & (__force int)__GFP_DMA)
1385 res = ZONE_DMA;
1386 return res;
1387}
1388
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389#ifdef CONFIG_NUMA
1390#define MAX_NODE_LOAD (num_online_nodes())
Yasunori Goto86356ab2006-06-23 02:03:09 -07001391static int __meminitdata node_load[MAX_NUMNODES];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001393 * 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 -07001394 * @node: node whose fallback list we're appending
1395 * @used_node_mask: nodemask_t of already used nodes
1396 *
1397 * We use a number of factors to determine which is the next node that should
1398 * appear on a given node's fallback list. The node should not have appeared
1399 * already in @node's fallback list, and it should be the next closest node
1400 * according to the distance array (which contains arbitrary distance values
1401 * from each node to each node in the system), and should also prefer nodes
1402 * with no CPUs, since presumably they'll have very little allocation pressure
1403 * on them otherwise.
1404 * It returns -1 if no node is found.
1405 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001406static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001408 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 int min_val = INT_MAX;
1410 int best_node = -1;
1411
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001412 /* Use the local node if we haven't already */
1413 if (!node_isset(node, *used_node_mask)) {
1414 node_set(node, *used_node_mask);
1415 return node;
1416 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001418 for_each_online_node(n) {
1419 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
1421 /* Don't want a node to appear more than once */
1422 if (node_isset(n, *used_node_mask))
1423 continue;
1424
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 /* Use the distance array to find the distance */
1426 val = node_distance(node, n);
1427
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001428 /* Penalize nodes under us ("prefer the next node") */
1429 val += (n < node);
1430
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 /* Give preference to headless and unused nodes */
1432 tmp = node_to_cpumask(n);
1433 if (!cpus_empty(tmp))
1434 val += PENALTY_FOR_NODE_WITH_CPUS;
1435
1436 /* Slight preference for less loaded node */
1437 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1438 val += node_load[n];
1439
1440 if (val < min_val) {
1441 min_val = val;
1442 best_node = n;
1443 }
1444 }
1445
1446 if (best_node >= 0)
1447 node_set(best_node, *used_node_mask);
1448
1449 return best_node;
1450}
1451
Yasunori Goto86356ab2006-06-23 02:03:09 -07001452static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453{
1454 int i, j, k, node, local_node;
1455 int prev_node, load;
1456 struct zonelist *zonelist;
1457 nodemask_t used_mask;
1458
1459 /* initialize zonelists */
1460 for (i = 0; i < GFP_ZONETYPES; i++) {
1461 zonelist = pgdat->node_zonelists + i;
1462 zonelist->zones[0] = NULL;
1463 }
1464
1465 /* NUMA-aware ordering of nodes */
1466 local_node = pgdat->node_id;
1467 load = num_online_nodes();
1468 prev_node = local_node;
1469 nodes_clear(used_mask);
1470 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001471 int distance = node_distance(local_node, node);
1472
1473 /*
1474 * If another node is sufficiently far away then it is better
1475 * to reclaim pages in a zone before going off node.
1476 */
1477 if (distance > RECLAIM_DISTANCE)
1478 zone_reclaim_mode = 1;
1479
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 /*
1481 * We don't want to pressure a particular node.
1482 * So adding penalty to the first node in same
1483 * distance group to make it round-robin.
1484 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08001485
1486 if (distance != node_distance(local_node, prev_node))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 node_load[node] += load;
1488 prev_node = node;
1489 load--;
1490 for (i = 0; i < GFP_ZONETYPES; i++) {
1491 zonelist = pgdat->node_zonelists + i;
1492 for (j = 0; zonelist->zones[j] != NULL; j++);
1493
Al Viro260b2362005-10-21 03:22:44 -04001494 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495
1496 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1497 zonelist->zones[j] = NULL;
1498 }
1499 }
1500}
1501
1502#else /* CONFIG_NUMA */
1503
Yasunori Goto86356ab2006-06-23 02:03:09 -07001504static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07001506 int i, node, local_node;
1507 enum zone_type k;
1508 enum zone_type j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
1510 local_node = pgdat->node_id;
1511 for (i = 0; i < GFP_ZONETYPES; i++) {
1512 struct zonelist *zonelist;
1513
1514 zonelist = pgdat->node_zonelists + i;
1515
1516 j = 0;
Al Viro260b2362005-10-21 03:22:44 -04001517 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 j = build_zonelists_node(pgdat, zonelist, j, k);
1519 /*
1520 * Now we build the zonelist so that it contains the zones
1521 * of all the other nodes.
1522 * We don't want to pressure a particular node, so when
1523 * building the zones for node N, we make sure that the
1524 * zones coming right after the local ones are those from
1525 * node N+1 (modulo N)
1526 */
1527 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1528 if (!node_online(node))
1529 continue;
1530 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1531 }
1532 for (node = 0; node < local_node; node++) {
1533 if (!node_online(node))
1534 continue;
1535 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1536 }
1537
1538 zonelist->zones[j] = NULL;
1539 }
1540}
1541
1542#endif /* CONFIG_NUMA */
1543
Yasunori Goto68113782006-06-23 02:03:11 -07001544/* return values int ....just for stop_machine_run() */
1545static int __meminit __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546{
Yasunori Goto68113782006-06-23 02:03:11 -07001547 int nid;
1548 for_each_online_node(nid)
1549 build_zonelists(NODE_DATA(nid));
1550 return 0;
1551}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
Yasunori Goto68113782006-06-23 02:03:11 -07001553void __meminit build_all_zonelists(void)
1554{
1555 if (system_state == SYSTEM_BOOTING) {
1556 __build_all_zonelists(0);
1557 cpuset_init_current_mems_allowed();
1558 } else {
1559 /* we have to stop all cpus to guaranntee there is no user
1560 of zonelist */
1561 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
1562 /* cpuset refresh routine should be here */
1563 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07001564 vm_total_pages = nr_free_pagecache_pages();
1565 printk("Built %i zonelists. Total pages: %ld\n",
1566 num_online_nodes(), vm_total_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567}
1568
1569/*
1570 * Helper functions to size the waitqueue hash table.
1571 * Essentially these want to choose hash table sizes sufficiently
1572 * large so that collisions trying to wait on pages are rare.
1573 * But in fact, the number of active page waitqueues on typical
1574 * systems is ridiculously low, less than 200. So this is even
1575 * conservative, even though it seems large.
1576 *
1577 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1578 * waitqueues, i.e. the size of the waitq table given the number of pages.
1579 */
1580#define PAGES_PER_WAITQUEUE 256
1581
Yasunori Gotocca448f2006-06-23 02:03:10 -07001582#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07001583static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584{
1585 unsigned long size = 1;
1586
1587 pages /= PAGES_PER_WAITQUEUE;
1588
1589 while (size < pages)
1590 size <<= 1;
1591
1592 /*
1593 * Once we have dozens or even hundreds of threads sleeping
1594 * on IO we've got bigger problems than wait queue collision.
1595 * Limit the size of the wait table to a reasonable size.
1596 */
1597 size = min(size, 4096UL);
1598
1599 return max(size, 4UL);
1600}
Yasunori Gotocca448f2006-06-23 02:03:10 -07001601#else
1602/*
1603 * A zone's size might be changed by hot-add, so it is not possible to determine
1604 * a suitable size for its wait_table. So we use the maximum size now.
1605 *
1606 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1607 *
1608 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1609 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1610 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1611 *
1612 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1613 * or more by the traditional way. (See above). It equals:
1614 *
1615 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1616 * ia64(16K page size) : = ( 8G + 4M)byte.
1617 * powerpc (64K page size) : = (32G +16M)byte.
1618 */
1619static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1620{
1621 return 4096UL;
1622}
1623#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
1625/*
1626 * This is an integer logarithm so that shifts can be used later
1627 * to extract the more random high bits from the multiplicative
1628 * hash function before the remainder is taken.
1629 */
1630static inline unsigned long wait_table_bits(unsigned long size)
1631{
1632 return ffz(~size);
1633}
1634
1635#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1636
1637static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1638 unsigned long *zones_size, unsigned long *zholes_size)
1639{
1640 unsigned long realtotalpages, totalpages = 0;
1641 int i;
1642
1643 for (i = 0; i < MAX_NR_ZONES; i++)
1644 totalpages += zones_size[i];
1645 pgdat->node_spanned_pages = totalpages;
1646
1647 realtotalpages = totalpages;
1648 if (zholes_size)
1649 for (i = 0; i < MAX_NR_ZONES; i++)
1650 realtotalpages -= zholes_size[i];
1651 pgdat->node_present_pages = realtotalpages;
1652 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1653}
1654
1655
1656/*
1657 * Initially all pages are reserved - free ones are freed
1658 * up by free_all_bootmem() once the early boot process is
1659 * done. Non-atomic initialization, single-pass.
1660 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001661void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 unsigned long start_pfn)
1663{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001665 unsigned long end_pfn = start_pfn + size;
1666 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001668 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001669 if (!early_pfn_valid(pfn))
1670 continue;
1671 page = pfn_to_page(pfn);
1672 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08001673 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 reset_page_mapcount(page);
1675 SetPageReserved(page);
1676 INIT_LIST_HEAD(&page->lru);
1677#ifdef WANT_PAGE_VIRTUAL
1678 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1679 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001680 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 }
1683}
1684
1685void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1686 unsigned long size)
1687{
1688 int order;
1689 for (order = 0; order < MAX_ORDER ; order++) {
1690 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1691 zone->free_area[order].nr_free = 0;
1692 }
1693}
1694
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001695#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
Christoph Lameter2f1b6242006-09-25 23:31:13 -07001696void zonetable_add(struct zone *zone, int nid, enum zone_type zid,
1697 unsigned long pfn, unsigned long size)
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001698{
1699 unsigned long snum = pfn_to_section_nr(pfn);
1700 unsigned long end = pfn_to_section_nr(pfn + size);
1701
1702 if (FLAGS_HAS_NODE)
1703 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1704 else
1705 for (; snum <= end; snum++)
1706 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1707}
1708
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709#ifndef __HAVE_ARCH_MEMMAP_INIT
1710#define memmap_init(size, nid, zone, start_pfn) \
1711 memmap_init_zone((size), (nid), (zone), (start_pfn))
1712#endif
1713
Ashok Raj6292d9a2006-02-01 03:04:44 -08001714static int __cpuinit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001715{
1716 int batch;
1717
1718 /*
1719 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001720 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001721 *
1722 * OK, so we don't know how big the cache is. So guess.
1723 */
1724 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001725 if (batch * PAGE_SIZE > 512 * 1024)
1726 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001727 batch /= 4; /* We effectively *= 4 below */
1728 if (batch < 1)
1729 batch = 1;
1730
1731 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001732 * Clamp the batch to a 2^n - 1 value. Having a power
1733 * of 2 value was found to be more likely to have
1734 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001735 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001736 * For example if 2 tasks are alternately allocating
1737 * batches of pages, one task can end up with a lot
1738 * of pages of one half of the possible page colors
1739 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001740 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001741 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001742
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001743 return batch;
1744}
1745
Christoph Lameter2caaad42005-06-21 17:15:00 -07001746inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1747{
1748 struct per_cpu_pages *pcp;
1749
Magnus Damm1c6fe942005-10-26 01:58:59 -07001750 memset(p, 0, sizeof(*p));
1751
Christoph Lameter2caaad42005-06-21 17:15:00 -07001752 pcp = &p->pcp[0]; /* hot */
1753 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001754 pcp->high = 6 * batch;
1755 pcp->batch = max(1UL, 1 * batch);
1756 INIT_LIST_HEAD(&pcp->list);
1757
1758 pcp = &p->pcp[1]; /* cold*/
1759 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001760 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001761 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001762 INIT_LIST_HEAD(&pcp->list);
1763}
1764
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001765/*
1766 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1767 * to the value high for the pageset p.
1768 */
1769
1770static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1771 unsigned long high)
1772{
1773 struct per_cpu_pages *pcp;
1774
1775 pcp = &p->pcp[0]; /* hot list */
1776 pcp->high = high;
1777 pcp->batch = max(1UL, high/4);
1778 if ((high/4) > (PAGE_SHIFT * 8))
1779 pcp->batch = PAGE_SHIFT * 8;
1780}
1781
1782
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001783#ifdef CONFIG_NUMA
1784/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001785 * Boot pageset table. One per cpu which is going to be used for all
1786 * zones and all nodes. The parameters will be set in such a way
1787 * that an item put on a list will immediately be handed over to
1788 * the buddy list. This is safe since pageset manipulation is done
1789 * with interrupts disabled.
1790 *
1791 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001792 *
1793 * The boot_pagesets must be kept even after bootup is complete for
1794 * unused processors and/or zones. They do play a role for bootstrapping
1795 * hotplugged processors.
1796 *
1797 * zoneinfo_show() and maybe other functions do
1798 * not check if the processor is online before following the pageset pointer.
1799 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001800 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001801static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001802
1803/*
1804 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001805 * per cpu pageset array in struct zone.
1806 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08001807static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001808{
1809 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001810
1811 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001812
Nick Piggin23316bc2006-01-08 01:00:41 -08001813 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001814 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08001815 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001816 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001817
Nick Piggin23316bc2006-01-08 01:00:41 -08001818 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001819
1820 if (percpu_pagelist_fraction)
1821 setup_pagelist_highmark(zone_pcp(zone, cpu),
1822 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001823 }
1824
1825 return 0;
1826bad:
1827 for_each_zone(dzone) {
1828 if (dzone == zone)
1829 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08001830 kfree(zone_pcp(dzone, cpu));
1831 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001832 }
1833 return -ENOMEM;
1834}
1835
1836static inline void free_zone_pagesets(int cpu)
1837{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001838 struct zone *zone;
1839
1840 for_each_zone(zone) {
1841 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1842
David Rientjesf3ef9ea2006-09-25 16:24:57 -07001843 /* Free per_cpu_pageset if it is slab allocated */
1844 if (pset != &boot_pageset[cpu])
1845 kfree(pset);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001846 zone_pcp(zone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001847 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001848}
1849
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07001850static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001851 unsigned long action,
1852 void *hcpu)
1853{
1854 int cpu = (long)hcpu;
1855 int ret = NOTIFY_OK;
1856
1857 switch (action) {
1858 case CPU_UP_PREPARE:
1859 if (process_zones(cpu))
1860 ret = NOTIFY_BAD;
1861 break;
Andi Kleenb0d41692005-11-05 17:25:53 +01001862 case CPU_UP_CANCELED:
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001863 case CPU_DEAD:
1864 free_zone_pagesets(cpu);
1865 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001866 default:
1867 break;
1868 }
1869 return ret;
1870}
1871
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07001872static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001873 { &pageset_cpuup_callback, NULL, 0 };
1874
Al Viro78d99552005-12-15 09:18:25 +00001875void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001876{
1877 int err;
1878
1879 /* Initialize per_cpu_pageset for cpu 0.
1880 * A cpuup callback will do this for every cpu
1881 * as it comes online
1882 */
1883 err = process_zones(smp_processor_id());
1884 BUG_ON(err);
1885 register_cpu_notifier(&pageset_notifier);
1886}
1887
1888#endif
1889
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001890static __meminit
Yasunori Gotocca448f2006-06-23 02:03:10 -07001891int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07001892{
1893 int i;
1894 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07001895 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07001896
1897 /*
1898 * The per-page waitqueue mechanism uses hashed waitqueues
1899 * per zone.
1900 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07001901 zone->wait_table_hash_nr_entries =
1902 wait_table_hash_nr_entries(zone_size_pages);
1903 zone->wait_table_bits =
1904 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07001905 alloc_size = zone->wait_table_hash_nr_entries
1906 * sizeof(wait_queue_head_t);
1907
1908 if (system_state == SYSTEM_BOOTING) {
1909 zone->wait_table = (wait_queue_head_t *)
1910 alloc_bootmem_node(pgdat, alloc_size);
1911 } else {
1912 /*
1913 * This case means that a zone whose size was 0 gets new memory
1914 * via memory hot-add.
1915 * But it may be the case that a new node was hot-added. In
1916 * this case vmalloc() will not be able to use this new node's
1917 * memory - this wait_table must be initialized to use this new
1918 * node itself as well.
1919 * To use this new node's memory, further consideration will be
1920 * necessary.
1921 */
1922 zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size);
1923 }
1924 if (!zone->wait_table)
1925 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07001926
Yasunori Goto02b694d2006-06-23 02:03:08 -07001927 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07001928 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07001929
1930 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07001931}
1932
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001933static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07001934{
1935 int cpu;
1936 unsigned long batch = zone_batchsize(zone);
1937
1938 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1939#ifdef CONFIG_NUMA
1940 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08001941 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07001942 setup_pageset(&boot_pageset[cpu],0);
1943#else
1944 setup_pageset(zone_pcp(zone,cpu), batch);
1945#endif
1946 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08001947 if (zone->present_pages)
1948 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
1949 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07001950}
1951
Yasunori Goto718127c2006-06-23 02:03:10 -07001952__meminit int init_currently_empty_zone(struct zone *zone,
1953 unsigned long zone_start_pfn,
1954 unsigned long size)
Dave Hansened8ece22005-10-29 18:16:50 -07001955{
1956 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07001957 int ret;
1958 ret = zone_wait_table_init(zone, size);
1959 if (ret)
1960 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07001961 pgdat->nr_zones = zone_idx(zone) + 1;
1962
Dave Hansened8ece22005-10-29 18:16:50 -07001963 zone->zone_start_pfn = zone_start_pfn;
1964
1965 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
1966
1967 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
Yasunori Goto718127c2006-06-23 02:03:10 -07001968
1969 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07001970}
1971
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972/*
1973 * Set up the zone data structures:
1974 * - mark all pages reserved
1975 * - mark all memory queues empty
1976 * - clear the memory bitmaps
1977 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001978static void __meminit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 unsigned long *zones_size, unsigned long *zholes_size)
1980{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07001981 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07001982 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07001984 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985
Dave Hansen208d54e2005-10-29 18:16:52 -07001986 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 pgdat->nr_zones = 0;
1988 init_waitqueue_head(&pgdat->kswapd_wait);
1989 pgdat->kswapd_max_order = 0;
1990
1991 for (j = 0; j < MAX_NR_ZONES; j++) {
1992 struct zone *zone = pgdat->node_zones + j;
1993 unsigned long size, realsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 realsize = size = zones_size[j];
1996 if (zholes_size)
1997 realsize -= zholes_size[j];
1998
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001999 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 nr_kernel_pages += realsize;
2001 nr_all_pages += realsize;
2002
2003 zone->spanned_pages = size;
2004 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07002005#ifdef CONFIG_NUMA
2006 zone->min_unmapped_ratio = (realsize*sysctl_min_unmapped_ratio)
2007 / 100;
2008#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 zone->name = zone_names[j];
2010 spin_lock_init(&zone->lock);
2011 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002012 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 zone->zone_pgdat = pgdat;
2014 zone->free_pages = 0;
2015
2016 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2017
Dave Hansened8ece22005-10-29 18:16:50 -07002018 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 INIT_LIST_HEAD(&zone->active_list);
2020 INIT_LIST_HEAD(&zone->inactive_list);
2021 zone->nr_scan_active = 0;
2022 zone->nr_scan_inactive = 0;
2023 zone->nr_active = 0;
2024 zone->nr_inactive = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07002025 zap_zone_vm_stats(zone);
Martin Hicks53e9a612005-09-03 15:54:51 -07002026 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 if (!size)
2028 continue;
2029
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002030 zonetable_add(zone, nid, j, zone_start_pfn, size);
Yasunori Goto718127c2006-06-23 02:03:10 -07002031 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
2032 BUG_ON(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 }
2035}
2036
2037static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2038{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 /* Skip empty nodes */
2040 if (!pgdat->node_spanned_pages)
2041 return;
2042
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002043#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 /* ia64 gets its own node_mem_map, before this, without bootmem */
2045 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07002046 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002047 struct page *map;
2048
Bob Piccoe984bb42006-05-20 15:00:31 -07002049 /*
2050 * The zone's endpoints aren't required to be MAX_ORDER
2051 * aligned but the node_mem_map endpoints must be in order
2052 * for the buddy allocator to function correctly.
2053 */
2054 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
2055 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
2056 end = ALIGN(end, MAX_ORDER_NR_PAGES);
2057 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002058 map = alloc_remap(pgdat->node_id, size);
2059 if (!map)
2060 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07002061 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002063#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 /*
2065 * With no DISCONTIG, the global mem_map is just set as node 0's
2066 */
2067 if (pgdat == NODE_DATA(0))
2068 mem_map = NODE_DATA(0)->node_mem_map;
2069#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002070#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071}
2072
Yasunori Goto86356ab2006-06-23 02:03:09 -07002073void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 unsigned long *zones_size, unsigned long node_start_pfn,
2075 unsigned long *zholes_size)
2076{
2077 pgdat->node_id = nid;
2078 pgdat->node_start_pfn = node_start_pfn;
2079 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2080
2081 alloc_node_mem_map(pgdat);
2082
2083 free_area_init_core(pgdat, zones_size, zholes_size);
2084}
2085
Dave Hansen93b75042005-06-23 00:07:47 -07002086#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087static bootmem_data_t contig_bootmem_data;
2088struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2089
2090EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002091#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092
2093void __init free_area_init(unsigned long *zones_size)
2094{
Dave Hansen93b75042005-06-23 00:07:47 -07002095 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2097}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099#ifdef CONFIG_HOTPLUG_CPU
2100static int page_alloc_cpu_notify(struct notifier_block *self,
2101 unsigned long action, void *hcpu)
2102{
2103 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
2105 if (action == CPU_DEAD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 local_irq_disable();
2107 __drain_pages(cpu);
Christoph Lameterf8891e52006-06-30 01:55:45 -07002108 vm_events_fold_cpu(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 local_irq_enable();
Christoph Lameter2244b952006-06-30 01:55:33 -07002110 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 }
2112 return NOTIFY_OK;
2113}
2114#endif /* CONFIG_HOTPLUG_CPU */
2115
2116void __init page_alloc_init(void)
2117{
2118 hotcpu_notifier(page_alloc_cpu_notify, 0);
2119}
2120
2121/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002122 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2123 * or min_free_kbytes changes.
2124 */
2125static void calculate_totalreserve_pages(void)
2126{
2127 struct pglist_data *pgdat;
2128 unsigned long reserve_pages = 0;
2129 int i, j;
2130
2131 for_each_online_pgdat(pgdat) {
2132 for (i = 0; i < MAX_NR_ZONES; i++) {
2133 struct zone *zone = pgdat->node_zones + i;
2134 unsigned long max = 0;
2135
2136 /* Find valid and maximum lowmem_reserve in the zone */
2137 for (j = i; j < MAX_NR_ZONES; j++) {
2138 if (zone->lowmem_reserve[j] > max)
2139 max = zone->lowmem_reserve[j];
2140 }
2141
2142 /* we treat pages_high as reserved pages. */
2143 max += zone->pages_high;
2144
2145 if (max > zone->present_pages)
2146 max = zone->present_pages;
2147 reserve_pages += max;
2148 }
2149 }
2150 totalreserve_pages = reserve_pages;
2151}
2152
2153/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 * setup_per_zone_lowmem_reserve - called whenever
2155 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2156 * has a correct pages reserved value, so an adequate number of
2157 * pages are left in the zone after a successful __alloc_pages().
2158 */
2159static void setup_per_zone_lowmem_reserve(void)
2160{
2161 struct pglist_data *pgdat;
2162 int j, idx;
2163
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08002164 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 for (j = 0; j < MAX_NR_ZONES; j++) {
2166 struct zone *zone = pgdat->node_zones + j;
2167 unsigned long present_pages = zone->present_pages;
2168
2169 zone->lowmem_reserve[j] = 0;
2170
2171 for (idx = j-1; idx >= 0; idx--) {
2172 struct zone *lower_zone;
2173
2174 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2175 sysctl_lowmem_reserve_ratio[idx] = 1;
2176
2177 lower_zone = pgdat->node_zones + idx;
2178 lower_zone->lowmem_reserve[j] = present_pages /
2179 sysctl_lowmem_reserve_ratio[idx];
2180 present_pages += lower_zone->present_pages;
2181 }
2182 }
2183 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002184
2185 /* update totalreserve_pages */
2186 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187}
2188
2189/*
2190 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2191 * that the pages_{min,low,high} values for each zone are set correctly
2192 * with respect to min_free_kbytes.
2193 */
Dave Hansen3947be12005-10-29 18:16:54 -07002194void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195{
2196 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2197 unsigned long lowmem_pages = 0;
2198 struct zone *zone;
2199 unsigned long flags;
2200
2201 /* Calculate total number of !ZONE_HIGHMEM pages */
2202 for_each_zone(zone) {
2203 if (!is_highmem(zone))
2204 lowmem_pages += zone->present_pages;
2205 }
2206
2207 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07002208 u64 tmp;
2209
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07002211 tmp = (u64)pages_min * zone->present_pages;
2212 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 if (is_highmem(zone)) {
2214 /*
Nick Piggin669ed172005-11-13 16:06:45 -08002215 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2216 * need highmem pages, so cap pages_min to a small
2217 * value here.
2218 *
2219 * The (pages_high-pages_low) and (pages_low-pages_min)
2220 * deltas controls asynch page reclaim, and so should
2221 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 */
2223 int min_pages;
2224
2225 min_pages = zone->present_pages / 1024;
2226 if (min_pages < SWAP_CLUSTER_MAX)
2227 min_pages = SWAP_CLUSTER_MAX;
2228 if (min_pages > 128)
2229 min_pages = 128;
2230 zone->pages_min = min_pages;
2231 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08002232 /*
2233 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 * proportionate to the zone's size.
2235 */
Nick Piggin669ed172005-11-13 16:06:45 -08002236 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 }
2238
Andrew Mortonac924c62006-05-15 09:43:59 -07002239 zone->pages_low = zone->pages_min + (tmp >> 2);
2240 zone->pages_high = zone->pages_min + (tmp >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 spin_unlock_irqrestore(&zone->lru_lock, flags);
2242 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002243
2244 /* update totalreserve_pages */
2245 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246}
2247
2248/*
2249 * Initialise min_free_kbytes.
2250 *
2251 * For small machines we want it small (128k min). For large machines
2252 * we want it large (64MB max). But it is not linear, because network
2253 * bandwidth does not increase linearly with machine size. We use
2254 *
2255 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2256 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2257 *
2258 * which yields
2259 *
2260 * 16MB: 512k
2261 * 32MB: 724k
2262 * 64MB: 1024k
2263 * 128MB: 1448k
2264 * 256MB: 2048k
2265 * 512MB: 2896k
2266 * 1024MB: 4096k
2267 * 2048MB: 5792k
2268 * 4096MB: 8192k
2269 * 8192MB: 11584k
2270 * 16384MB: 16384k
2271 */
2272static int __init init_per_zone_pages_min(void)
2273{
2274 unsigned long lowmem_kbytes;
2275
2276 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2277
2278 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2279 if (min_free_kbytes < 128)
2280 min_free_kbytes = 128;
2281 if (min_free_kbytes > 65536)
2282 min_free_kbytes = 65536;
2283 setup_per_zone_pages_min();
2284 setup_per_zone_lowmem_reserve();
2285 return 0;
2286}
2287module_init(init_per_zone_pages_min)
2288
2289/*
2290 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2291 * that we can call two helper functions whenever min_free_kbytes
2292 * changes.
2293 */
2294int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2295 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2296{
2297 proc_dointvec(table, write, file, buffer, length, ppos);
2298 setup_per_zone_pages_min();
2299 return 0;
2300}
2301
Christoph Lameter96146342006-07-03 00:24:13 -07002302#ifdef CONFIG_NUMA
2303int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
2304 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2305{
2306 struct zone *zone;
2307 int rc;
2308
2309 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2310 if (rc)
2311 return rc;
2312
2313 for_each_zone(zone)
2314 zone->min_unmapped_ratio = (zone->present_pages *
2315 sysctl_min_unmapped_ratio) / 100;
2316 return 0;
2317}
2318#endif
2319
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320/*
2321 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2322 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2323 * whenever sysctl_lowmem_reserve_ratio changes.
2324 *
2325 * The reserve ratio obviously has absolutely no relation with the
2326 * pages_min watermarks. The lowmem reserve ratio can only make sense
2327 * if in function of the boot time zone sizes.
2328 */
2329int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2330 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2331{
2332 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2333 setup_per_zone_lowmem_reserve();
2334 return 0;
2335}
2336
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002337/*
2338 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2339 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2340 * can have before it gets flushed back to buddy allocator.
2341 */
2342
2343int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2344 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2345{
2346 struct zone *zone;
2347 unsigned int cpu;
2348 int ret;
2349
2350 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2351 if (!write || (ret == -EINVAL))
2352 return ret;
2353 for_each_zone(zone) {
2354 for_each_online_cpu(cpu) {
2355 unsigned long high;
2356 high = zone->present_pages / percpu_pagelist_fraction;
2357 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2358 }
2359 }
2360 return 0;
2361}
2362
David S. Millerf034b5d2006-08-24 03:08:07 -07002363int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364
2365#ifdef CONFIG_NUMA
2366static int __init set_hashdist(char *str)
2367{
2368 if (!str)
2369 return 0;
2370 hashdist = simple_strtoul(str, &str, 0);
2371 return 1;
2372}
2373__setup("hashdist=", set_hashdist);
2374#endif
2375
2376/*
2377 * allocate a large system hash table from bootmem
2378 * - it is assumed that the hash table must contain an exact power-of-2
2379 * quantity of entries
2380 * - limit is the number of hash buckets, not the total allocation size
2381 */
2382void *__init alloc_large_system_hash(const char *tablename,
2383 unsigned long bucketsize,
2384 unsigned long numentries,
2385 int scale,
2386 int flags,
2387 unsigned int *_hash_shift,
2388 unsigned int *_hash_mask,
2389 unsigned long limit)
2390{
2391 unsigned long long max = limit;
2392 unsigned long log2qty, size;
2393 void *table = NULL;
2394
2395 /* allow the kernel cmdline to have a say */
2396 if (!numentries) {
2397 /* round applicable memory size up to nearest megabyte */
2398 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2399 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2400 numentries >>= 20 - PAGE_SHIFT;
2401 numentries <<= 20 - PAGE_SHIFT;
2402
2403 /* limit to 1 bucket per 2^scale bytes of low memory */
2404 if (scale > PAGE_SHIFT)
2405 numentries >>= (scale - PAGE_SHIFT);
2406 else
2407 numentries <<= (PAGE_SHIFT - scale);
2408 }
John Hawkes6e692ed2006-03-25 03:08:02 -08002409 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410
2411 /* limit allocation size to 1/16 total memory by default */
2412 if (max == 0) {
2413 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2414 do_div(max, bucketsize);
2415 }
2416
2417 if (numentries > max)
2418 numentries = max;
2419
2420 log2qty = long_log2(numentries);
2421
2422 do {
2423 size = bucketsize << log2qty;
2424 if (flags & HASH_EARLY)
2425 table = alloc_bootmem(size);
2426 else if (hashdist)
2427 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2428 else {
2429 unsigned long order;
2430 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2431 ;
2432 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2433 }
2434 } while (!table && size > PAGE_SIZE && --log2qty);
2435
2436 if (!table)
2437 panic("Failed to allocate %s hash table\n", tablename);
2438
2439 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2440 tablename,
2441 (1U << log2qty),
2442 long_log2(size) - PAGE_SHIFT,
2443 size);
2444
2445 if (_hash_shift)
2446 *_hash_shift = log2qty;
2447 if (_hash_mask)
2448 *_hash_mask = (1 << log2qty) - 1;
2449
2450 return table;
2451}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002452
2453#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002454struct page *pfn_to_page(unsigned long pfn)
2455{
Andy Whitcroft67de6482006-06-23 02:03:12 -07002456 return __pfn_to_page(pfn);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002457}
2458unsigned long page_to_pfn(struct page *page)
2459{
Andy Whitcroft67de6482006-06-23 02:03:12 -07002460 return __page_to_pfn(page);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002461}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002462EXPORT_SYMBOL(pfn_to_page);
2463EXPORT_SYMBOL(page_to_pfn);
2464#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */