blob: ea77c999047ec634746d3b6457724bcb7c5eac5c [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
17#include <linux/config.h>
18#include <linux/stddef.h>
19#include <linux/mm.h>
20#include <linux/swap.h>
21#include <linux/interrupt.h>
22#include <linux/pagemap.h>
23#include <linux/bootmem.h>
24#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070025#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/module.h>
27#include <linux/suspend.h>
28#include <linux/pagevec.h>
29#include <linux/blkdev.h>
30#include <linux/slab.h>
31#include <linux/notifier.h>
32#include <linux/topology.h>
33#include <linux/sysctl.h>
34#include <linux/cpu.h>
35#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070036#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/nodemask.h>
38#include <linux/vmalloc.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080039#include <linux/mempolicy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#include <asm/tlbflush.h>
42#include "internal.h"
43
44/*
45 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
46 * initializer cleaner
47 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070048nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
Dean Nelson7223a932005-03-23 19:00:00 -070049EXPORT_SYMBOL(node_online_map);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070050nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
Dean Nelson7223a932005-03-23 19:00:00 -070051EXPORT_SYMBOL(node_possible_map);
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070052unsigned long totalram_pages __read_mostly;
53unsigned long totalhigh_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 */
Andi Kleena2f1b422005-11-05 17:25:53 +010071int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 256, 32 };
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
73EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
75/*
76 * Used by page_zone() to look up the address of the struct zone whose
77 * id is encoded in the upper bits of page->flags
78 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070079struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070080EXPORT_SYMBOL(zone_table);
81
Andi Kleena2f1b422005-11-05 17:25:53 +010082static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem" };
Linus Torvalds1da177e2005-04-16 15:20:36 -070083int min_free_kbytes = 1024;
84
85unsigned long __initdata nr_kernel_pages;
86unsigned long __initdata nr_all_pages;
87
Nick Piggin13e74442006-01-06 00:10:58 -080088#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -070089static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070090{
Dave Hansenbdc8cb92005-10-29 18:16:53 -070091 int ret = 0;
92 unsigned seq;
93 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -070094
Dave Hansenbdc8cb92005-10-29 18:16:53 -070095 do {
96 seq = zone_span_seqbegin(zone);
97 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
98 ret = 1;
99 else if (pfn < zone->zone_start_pfn)
100 ret = 1;
101 } while (zone_span_seqretry(zone, seq));
102
103 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700104}
105
106static int page_is_consistent(struct zone *zone, struct page *page)
107{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108#ifdef CONFIG_HOLES_IN_ZONE
109 if (!pfn_valid(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700110 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111#endif
112 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700113 return 0;
114
115 return 1;
116}
117/*
118 * Temporary debugging check for pages not lying within a given zone.
119 */
120static int bad_range(struct zone *zone, struct page *page)
121{
122 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700124 if (!page_is_consistent(zone, page))
125 return 1;
126
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 return 0;
128}
129
Nick Piggin13e74442006-01-06 00:10:58 -0800130#else
131static inline int bad_range(struct zone *zone, struct page *page)
132{
133 return 0;
134}
135#endif
136
Nick Piggin224abf92006-01-06 00:11:11 -0800137static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138{
Nick Piggin224abf92006-01-06 00:11:11 -0800139 printk(KERN_EMERG "Bad page state in process '%s'\n"
Hugh Dickins7365f3d2006-01-11 12:17:18 -0800140 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
141 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
142 KERN_EMERG "Backtrace:\n",
Nick Piggin224abf92006-01-06 00:11:11 -0800143 current->comm, page, (int)(2*sizeof(unsigned long)),
144 (unsigned long)page->flags, page->mapping,
145 page_mapcount(page), page_count(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 dump_stack();
Hugh Dickins334795e2005-06-21 17:15:08 -0700147 page->flags &= ~(1 << PG_lru |
148 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 1 << PG_active |
151 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700152 1 << PG_reclaim |
153 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 1 << PG_swapcache |
Nick Piggin676165a2006-04-10 11:21:48 +1000155 1 << PG_writeback |
156 1 << PG_buddy );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 set_page_count(page, 0);
158 reset_page_mapcount(page);
159 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700160 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161}
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163/*
164 * Higher-order pages are called "compound pages". They are structured thusly:
165 *
166 * The first PAGE_SIZE page is called the "head page".
167 *
168 * The remaining PAGE_SIZE pages are called "tail pages".
169 *
170 * All pages have PG_compound set. All pages have their ->private pointing at
171 * the head page (even the head page has this).
172 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800173 * The first tail page's ->lru.next holds the address of the compound page's
174 * put_page() function. Its ->lru.prev holds the order of allocation.
175 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800177
178static void free_compound_page(struct page *page)
179{
180 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
181}
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183static void prep_compound_page(struct page *page, unsigned long order)
184{
185 int i;
186 int nr_pages = 1 << order;
187
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800188 page[1].lru.next = (void *)free_compound_page; /* set dtor */
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800189 page[1].lru.prev = (void *)order;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 for (i = 0; i < nr_pages; i++) {
191 struct page *p = page + i;
192
Nick Piggin5e9dace2006-03-22 00:08:01 -0800193 __SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700194 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 }
196}
197
198static void destroy_compound_page(struct page *page, unsigned long order)
199{
200 int i;
201 int nr_pages = 1 << order;
202
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800203 if (unlikely((unsigned long)page[1].lru.prev != order))
Nick Piggin224abf92006-01-06 00:11:11 -0800204 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
206 for (i = 0; i < nr_pages; i++) {
207 struct page *p = page + i;
208
Nick Piggin224abf92006-01-06 00:11:11 -0800209 if (unlikely(!PageCompound(p) |
210 (page_private(p) != (unsigned long)page)))
211 bad_page(page);
Nick Piggin5e9dace2006-03-22 00:08:01 -0800212 __ClearPageCompound(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 }
214}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
Nick Piggin17cf4402006-03-22 00:08:41 -0800216static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
217{
218 int i;
219
220 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
Andrew Morton6626c5d2006-03-22 00:08:42 -0800221 /*
222 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
223 * and __GFP_HIGHMEM from hard or soft interrupt context.
224 */
225 BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800226 for (i = 0; i < (1 << order); i++)
227 clear_highpage(page + i);
228}
229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230/*
231 * function for dealing with page's order in buddy system.
232 * zone->lock is already acquired when we use these.
233 * So, we don't need atomic page->flags operations here.
234 */
Andrew Morton6aa30012006-04-18 22:20:52 -0700235static inline unsigned long page_order(struct page *page)
236{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700237 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238}
239
Andrew Morton6aa30012006-04-18 22:20:52 -0700240static inline void set_page_order(struct page *page, int order)
241{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700242 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000243 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244}
245
246static inline void rmv_page_order(struct page *page)
247{
Nick Piggin676165a2006-04-10 11:21:48 +1000248 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700249 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250}
251
252/*
253 * Locate the struct page for both the matching buddy in our
254 * pair (buddy1) and the combined O(n+1) page they form (page).
255 *
256 * 1) Any buddy B1 will have an order O twin B2 which satisfies
257 * the following equation:
258 * B2 = B1 ^ (1 << O)
259 * For example, if the starting buddy (buddy2) is #8 its order
260 * 1 buddy is #10:
261 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
262 *
263 * 2) Any buddy B will have an order O+1 parent P which
264 * satisfies the following equation:
265 * P = B & ~(1 << O)
266 *
267 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
268 */
269static inline struct page *
270__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
271{
272 unsigned long buddy_idx = page_idx ^ (1 << order);
273
274 return page + (buddy_idx - page_idx);
275}
276
277static inline unsigned long
278__find_combined_index(unsigned long page_idx, unsigned int order)
279{
280 return (page_idx & ~(1 << order));
281}
282
283/*
284 * This function checks whether a page is free && is the buddy
285 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800286 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000287 * (b) the buddy is in the buddy system &&
288 * (c) a page and its buddy have the same order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 *
Nick Piggin676165a2006-04-10 11:21:48 +1000290 * For recording whether a page is in the buddy system, we use PG_buddy.
291 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
292 *
293 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 */
295static inline int page_is_buddy(struct page *page, int order)
296{
Nick Piggin13e74442006-01-06 00:10:58 -0800297#ifdef CONFIG_HOLES_IN_ZONE
298 if (!pfn_valid(page_to_pfn(page)))
299 return 0;
300#endif
301
Nick Piggin676165a2006-04-10 11:21:48 +1000302 if (PageBuddy(page) && page_order(page) == order) {
303 BUG_ON(page_count(page) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700304 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000305 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700306 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307}
308
309/*
310 * Freeing function for a buddy system allocator.
311 *
312 * The concept of a buddy system is to maintain direct-mapped table
313 * (containing bit values) for memory blocks of various "orders".
314 * The bottom level table contains the map for the smallest allocatable
315 * units of memory (here, pages), and each level above it describes
316 * pairs of units from the levels below, hence, "buddies".
317 * At a high level, all that happens here is marking the table entry
318 * at the bottom level available, and propagating the changes upward
319 * as necessary, plus some accounting needed to play nicely with other
320 * parts of the VM system.
321 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000322 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700323 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 * So when we are allocating or freeing one, we can derive the state of the
325 * other. That is, if we allocate a small block, and both were
326 * free, the remainder of the region must be split into blocks.
327 * If a block is freed, and its buddy is also free, then this
328 * triggers coalescing into a block of larger size.
329 *
330 * -- wli
331 */
332
Nick Piggin48db57f2006-01-08 01:00:42 -0800333static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 struct zone *zone, unsigned int order)
335{
336 unsigned long page_idx;
337 int order_size = 1 << order;
338
Nick Piggin224abf92006-01-06 00:11:11 -0800339 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 destroy_compound_page(page, order);
341
342 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
343
344 BUG_ON(page_idx & (order_size - 1));
345 BUG_ON(bad_range(zone, page));
346
347 zone->free_pages += order_size;
348 while (order < MAX_ORDER-1) {
349 unsigned long combined_idx;
350 struct free_area *area;
351 struct page *buddy;
352
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 buddy = __page_find_buddy(page, page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 if (!page_is_buddy(buddy, order))
355 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800356
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 list_del(&buddy->lru);
358 area = zone->free_area + order;
359 area->nr_free--;
360 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800361 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 page = page + (combined_idx - page_idx);
363 page_idx = combined_idx;
364 order++;
365 }
366 set_page_order(page, order);
367 list_add(&page->lru, &zone->free_area[order].free_list);
368 zone->free_area[order].nr_free++;
369}
370
Nick Piggin224abf92006-01-06 00:11:11 -0800371static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372{
Nick Piggin92be2e32006-01-06 00:10:57 -0800373 if (unlikely(page_mapcount(page) |
374 (page->mapping != NULL) |
375 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 (page->flags & (
377 1 << PG_lru |
378 1 << PG_private |
379 1 << PG_locked |
380 1 << PG_active |
381 1 << PG_reclaim |
382 1 << PG_slab |
383 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700384 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000385 1 << PG_reserved |
386 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800387 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700389 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800390 /*
391 * For now, we report if PG_reserved was found set, but do not
392 * clear it, and do not free the page. But we shall soon need
393 * to do more, for when the ZERO_PAGE count wraps negative.
394 */
395 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396}
397
398/*
399 * Frees a list of pages.
400 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700401 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 *
403 * If the zone was previously in an "all pages pinned" state then look to
404 * see if this freeing clears that state.
405 *
406 * And clear the zone's pages_scanned counter, to hold off the "all pages are
407 * pinned" detection logic.
408 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800409static void free_pages_bulk(struct zone *zone, int count,
410 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411{
Nick Pigginc54ad302006-01-06 00:10:56 -0800412 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 zone->all_unreclaimable = 0;
414 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800415 while (count--) {
416 struct page *page;
417
418 BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800420 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800422 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800424 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425}
426
Nick Piggin48db57f2006-01-08 01:00:42 -0800427static void free_one_page(struct zone *zone, struct page *page, int order)
428{
429 LIST_HEAD(list);
430 list_add(&page->lru, &list);
431 free_pages_bulk(zone, 1, &list, order);
432}
433
434static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435{
Nick Pigginc54ad302006-01-06 00:10:56 -0800436 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800438 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
440 arch_free_page(page, order);
Ingo Molnarde5097c2006-01-09 15:59:21 -0800441 if (!PageHighMem(page))
442 mutex_debug_check_no_locks_freed(page_address(page),
David Woodhousea4fc7ab2006-01-11 14:41:26 +0000443 PAGE_SIZE<<order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800446 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800447 if (reserved)
448 return;
449
Nick Piggin48db57f2006-01-08 01:00:42 -0800450 kernel_map_pages(page, 1 << order, 0);
Nick Pigginc54ad302006-01-06 00:10:56 -0800451 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800452 __mod_page_state(pgfree, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800453 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800454 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455}
456
David Howellsa226f6c2006-01-06 00:11:08 -0800457/*
458 * permit the bootmem allocator to evade page validation on high-order frees
459 */
460void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
461{
462 if (order == 0) {
463 __ClearPageReserved(page);
464 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800465 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800466 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800467 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800468 int loop;
469
Nick Piggin545b1ea2006-03-22 00:08:07 -0800470 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800471 for (loop = 0; loop < BITS_PER_LONG; loop++) {
472 struct page *p = &page[loop];
473
Nick Piggin545b1ea2006-03-22 00:08:07 -0800474 if (loop + 1 < BITS_PER_LONG)
475 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800476 __ClearPageReserved(p);
477 set_page_count(p, 0);
478 }
479
Nick Piggin7835e982006-03-22 00:08:40 -0800480 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800481 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800482 }
483}
484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
486/*
487 * The order of subdivision here is critical for the IO subsystem.
488 * Please do not alter this order without good reasons and regression
489 * testing. Specifically, as large blocks of memory are subdivided,
490 * the order in which smaller blocks are delivered depends on the order
491 * they're subdivided in this function. This is the primary factor
492 * influencing the order in which pages are delivered to the IO
493 * subsystem according to empirical testing, and this is also justified
494 * by considering the behavior of a buddy system containing a single
495 * large block of memory acted on by a series of small allocations.
496 * This behavior is a critical factor in sglist merging's success.
497 *
498 * -- wli
499 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800500static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 int low, int high, struct free_area *area)
502{
503 unsigned long size = 1 << high;
504
505 while (high > low) {
506 area--;
507 high--;
508 size >>= 1;
509 BUG_ON(bad_range(zone, &page[size]));
510 list_add(&page[size].lru, &area->free_list);
511 area->nr_free++;
512 set_page_order(&page[size], high);
513 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514}
515
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516/*
517 * This page is about to be returned from the page allocator
518 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800519static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520{
Nick Piggin92be2e32006-01-06 00:10:57 -0800521 if (unlikely(page_mapcount(page) |
522 (page->mapping != NULL) |
523 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700524 (page->flags & (
525 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 1 << PG_private |
527 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 1 << PG_active |
529 1 << PG_dirty |
530 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700531 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700533 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000534 1 << PG_reserved |
535 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800536 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
Hugh Dickins689bceb2005-11-21 21:32:20 -0800538 /*
539 * For now, we report if PG_reserved was found set, but do not
540 * clear it, and do not allocate the page: as a safety net.
541 */
542 if (PageReserved(page))
543 return 1;
544
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
546 1 << PG_referenced | 1 << PG_arch_1 |
547 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700548 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800549 set_page_refcounted(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800551
552 if (gfp_flags & __GFP_ZERO)
553 prep_zero_page(page, order, gfp_flags);
554
555 if (order && (gfp_flags & __GFP_COMP))
556 prep_compound_page(page, order);
557
Hugh Dickins689bceb2005-11-21 21:32:20 -0800558 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559}
560
561/*
562 * Do the hard work of removing an element from the buddy allocator.
563 * Call me with the zone->lock already held.
564 */
565static struct page *__rmqueue(struct zone *zone, unsigned int order)
566{
567 struct free_area * area;
568 unsigned int current_order;
569 struct page *page;
570
571 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
572 area = zone->free_area + current_order;
573 if (list_empty(&area->free_list))
574 continue;
575
576 page = list_entry(area->free_list.next, struct page, lru);
577 list_del(&page->lru);
578 rmv_page_order(page);
579 area->nr_free--;
580 zone->free_pages -= 1UL << order;
Nick Piggin085cc7d2006-01-06 00:11:01 -0800581 expand(zone, page, order, current_order, area);
582 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 }
584
585 return NULL;
586}
587
588/*
589 * Obtain a specified number of elements from the buddy allocator, all under
590 * a single hold of the lock, for efficiency. Add them to the supplied list.
591 * Returns the number of new pages which were placed at *list.
592 */
593static int rmqueue_bulk(struct zone *zone, unsigned int order,
594 unsigned long count, struct list_head *list)
595{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
Nick Pigginc54ad302006-01-06 00:10:56 -0800598 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 for (i = 0; i < count; ++i) {
Nick Piggin085cc7d2006-01-06 00:11:01 -0800600 struct page *page = __rmqueue(zone, order);
601 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 list_add_tail(&page->lru, list);
604 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800605 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800606 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607}
608
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700609#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800610/*
611 * Called from the slab reaper to drain pagesets on a particular node that
612 * belong to the currently executing processor.
Christoph Lameter879336c2006-03-22 00:09:08 -0800613 * Note that this function must be called with the thread pinned to
614 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800615 */
616void drain_node_pages(int nodeid)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700617{
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800618 int i, z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700619 unsigned long flags;
620
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800621 for (z = 0; z < MAX_NR_ZONES; z++) {
622 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700623 struct per_cpu_pageset *pset;
624
Nick Piggin23316bc2006-01-08 01:00:41 -0800625 pset = zone_pcp(zone, smp_processor_id());
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700626 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
627 struct per_cpu_pages *pcp;
628
629 pcp = &pset->pcp[i];
Christoph Lameter879336c2006-03-22 00:09:08 -0800630 if (pcp->count) {
631 local_irq_save(flags);
632 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
633 pcp->count = 0;
634 local_irq_restore(flags);
635 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700636 }
637 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700638}
639#endif
640
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
642static void __drain_pages(unsigned int cpu)
643{
Nick Pigginc54ad302006-01-06 00:10:56 -0800644 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 struct zone *zone;
646 int i;
647
648 for_each_zone(zone) {
649 struct per_cpu_pageset *pset;
650
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700651 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
653 struct per_cpu_pages *pcp;
654
655 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800656 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800657 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
658 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800659 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 }
661 }
662}
663#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
664
665#ifdef CONFIG_PM
666
667void mark_free_pages(struct zone *zone)
668{
669 unsigned long zone_pfn, flags;
670 int order;
671 struct list_head *curr;
672
673 if (!zone->spanned_pages)
674 return;
675
676 spin_lock_irqsave(&zone->lock, flags);
677 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
678 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
679
680 for (order = MAX_ORDER - 1; order >= 0; --order)
681 list_for_each(curr, &zone->free_area[order].free_list) {
682 unsigned long start_pfn, i;
683
684 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
685
686 for (i=0; i < (1<<order); i++)
687 SetPageNosaveFree(pfn_to_page(start_pfn+i));
688 }
689 spin_unlock_irqrestore(&zone->lock, flags);
690}
691
692/*
693 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
694 */
695void drain_local_pages(void)
696{
697 unsigned long flags;
698
699 local_irq_save(flags);
700 __drain_pages(smp_processor_id());
701 local_irq_restore(flags);
702}
703#endif /* CONFIG_PM */
704
Nick Piggina74609f2006-01-06 00:11:20 -0800705static void zone_statistics(struct zonelist *zonelist, struct zone *z, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
707#ifdef CONFIG_NUMA
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 pg_data_t *pg = z->zone_pgdat;
709 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
710 struct per_cpu_pageset *p;
711
Nick Piggina74609f2006-01-06 00:11:20 -0800712 p = zone_pcp(z, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 if (pg == orig) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700714 p->numa_hit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 } else {
716 p->numa_miss++;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700717 zone_pcp(zonelist->zones[0], cpu)->numa_foreign++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 }
719 if (pg == NODE_DATA(numa_node_id()))
720 p->local_node++;
721 else
722 p->other_node++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723#endif
724}
725
726/*
727 * Free a 0-order page
728 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729static void fastcall free_hot_cold_page(struct page *page, int cold)
730{
731 struct zone *zone = page_zone(page);
732 struct per_cpu_pages *pcp;
733 unsigned long flags;
734
735 arch_free_page(page, 0);
736
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 if (PageAnon(page))
738 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800739 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800740 return;
741
Hugh Dickins689bceb2005-11-21 21:32:20 -0800742 kernel_map_pages(page, 1, 0);
743
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700744 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800746 __inc_page_state(pgfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 list_add(&page->lru, &pcp->list);
748 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800749 if (pcp->count >= pcp->high) {
750 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
751 pcp->count -= pcp->batch;
752 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 local_irq_restore(flags);
754 put_cpu();
755}
756
757void fastcall free_hot_page(struct page *page)
758{
759 free_hot_cold_page(page, 0);
760}
761
762void fastcall free_cold_page(struct page *page)
763{
764 free_hot_cold_page(page, 1);
765}
766
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800767/*
768 * split_page takes a non-compound higher-order page, and splits it into
769 * n (1<<order) sub-pages: page[0..n]
770 * Each sub-page must be freed individually.
771 *
772 * Note: this is probably too low level an operation for use in drivers.
773 * Please consult with lkml before using this in your driver.
774 */
775void split_page(struct page *page, unsigned int order)
776{
777 int i;
778
779 BUG_ON(PageCompound(page));
780 BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -0800781 for (i = 1; i < (1 << order); i++)
782 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800783}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800784
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785/*
786 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
787 * we cheat by calling it from here, in the order > 0 path. Saves a branch
788 * or two.
789 */
Nick Piggina74609f2006-01-06 00:11:20 -0800790static struct page *buffered_rmqueue(struct zonelist *zonelist,
791 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792{
793 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800794 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800796 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
Hugh Dickins689bceb2005-11-21 21:32:20 -0800798again:
Nick Piggina74609f2006-01-06 00:11:20 -0800799 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800800 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 struct per_cpu_pages *pcp;
802
Nick Piggina74609f2006-01-06 00:11:20 -0800803 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800805 if (!pcp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 pcp->count += rmqueue_bulk(zone, 0,
807 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800808 if (unlikely(!pcp->count))
809 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 }
Nick Piggina74609f2006-01-06 00:11:20 -0800811 page = list_entry(pcp->list.next, struct page, lru);
812 list_del(&page->lru);
813 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800814 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 spin_lock_irqsave(&zone->lock, flags);
816 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800817 spin_unlock(&zone->lock);
818 if (!page)
819 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 }
821
Nick Piggina74609f2006-01-06 00:11:20 -0800822 __mod_page_state_zone(zone, pgalloc, 1 << order);
823 zone_statistics(zonelist, zone, cpu);
824 local_irq_restore(flags);
825 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
Nick Piggina74609f2006-01-06 00:11:20 -0800827 BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -0800828 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -0800829 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800831
832failed:
833 local_irq_restore(flags);
834 put_cpu();
835 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836}
837
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800838#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800839#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
840#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
841#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
842#define ALLOC_HARDER 0x10 /* try to alloc harder */
843#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
844#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800845
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846/*
847 * Return 1 if free pages are above 'mark'. This takes into account the order
848 * of the allocation.
849 */
850int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800851 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852{
853 /* free_pages my go negative - that's OK */
854 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
855 int o;
856
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800857 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800859 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 min -= min / 4;
861
862 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
863 return 0;
864 for (o = 0; o < order; o++) {
865 /* At the next order, this order's pages become unavailable */
866 free_pages -= z->free_area[o].nr_free << o;
867
868 /* Require fewer higher order pages to be free */
869 min >>= 1;
870
871 if (free_pages <= min)
872 return 0;
873 }
874 return 1;
875}
876
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800877/*
878 * get_page_from_freeliest goes through the zonelist trying to allocate
879 * a page.
880 */
881static struct page *
882get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
883 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -0700884{
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800885 struct zone **z = zonelist->zones;
886 struct page *page = NULL;
887 int classzone_idx = zone_idx(*z);
888
889 /*
890 * Go through the zonelist once, looking for a zone with enough free.
891 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
892 */
893 do {
894 if ((alloc_flags & ALLOC_CPUSET) &&
895 !cpuset_zone_allowed(*z, gfp_mask))
896 continue;
897
898 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -0800899 unsigned long mark;
900 if (alloc_flags & ALLOC_WMARK_MIN)
901 mark = (*z)->pages_min;
902 else if (alloc_flags & ALLOC_WMARK_LOW)
903 mark = (*z)->pages_low;
904 else
905 mark = (*z)->pages_high;
906 if (!zone_watermark_ok(*z, order, mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800907 classzone_idx, alloc_flags))
Christoph Lameter9eeff232006-01-18 17:42:31 -0800908 if (!zone_reclaim_mode ||
909 !zone_reclaim(*z, gfp_mask, order))
910 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800911 }
912
Nick Piggina74609f2006-01-06 00:11:20 -0800913 page = buffered_rmqueue(zonelist, *z, order, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800914 if (page) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800915 break;
916 }
917 } while (*(++z) != NULL);
918 return page;
Martin Hicks753ee722005-06-21 17:14:41 -0700919}
920
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921/*
922 * This is the 'heart' of the zoned buddy allocator.
923 */
924struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +0100925__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 struct zonelist *zonelist)
927{
Al Viro260b2362005-10-21 03:22:44 -0400928 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800929 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 struct page *page;
931 struct reclaim_state reclaim_state;
932 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800934 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 int did_some_progress;
936
937 might_sleep_if(wait);
938
Jens Axboe6b1de912005-11-17 21:35:02 +0100939restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800940 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800942 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 /* Should this ever happen?? */
944 return NULL;
945 }
Jens Axboe6b1de912005-11-17 21:35:02 +0100946
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800947 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -0800948 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800949 if (page)
950 goto got_pg;
951
Jens Axboe6b1de912005-11-17 21:35:02 +0100952 do {
Christoph Lameter0b1303f2006-03-24 03:15:59 -0800953 if (cpuset_zone_allowed(*z, gfp_mask))
954 wakeup_kswapd(*z, order);
Jens Axboe6b1de912005-11-17 21:35:02 +0100955 } while (*(++z));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800956
Paul Jackson9bf22292005-09-06 15:18:12 -0700957 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800958 * OK, we're below the kswapd watermark and have kicked background
959 * reclaim. Now things get more complex, so set up alloc_flags according
960 * to how we want to proceed.
961 *
962 * The caller may dip into page reserves a bit more if the caller
963 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -0800964 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
965 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -0700966 */
Nick Piggin31488902005-11-28 13:44:03 -0800967 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800968 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
969 alloc_flags |= ALLOC_HARDER;
970 if (gfp_mask & __GFP_HIGH)
971 alloc_flags |= ALLOC_HIGH;
Paul Jackson47f3a862006-01-06 00:10:32 -0800972 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
974 /*
975 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800976 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 *
978 * This is the last chance, in general, before the goto nopage.
979 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -0700980 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800982 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
983 if (page)
984 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
986 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -0700987
988 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
989 && !in_interrupt()) {
990 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -0800991nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -0700992 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800993 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -0800994 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800995 if (page)
996 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -0800997 if (gfp_mask & __GFP_NOFAIL) {
998 blk_congestion_wait(WRITE, HZ/50);
999 goto nofail_alloc;
1000 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 }
1002 goto nopage;
1003 }
1004
1005 /* Atomic allocations - we can't balance anything */
1006 if (!wait)
1007 goto nopage;
1008
1009rebalance:
1010 cond_resched();
1011
1012 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001013 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 p->flags |= PF_MEMALLOC;
1015 reclaim_state.reclaimed_slab = 0;
1016 p->reclaim_state = &reclaim_state;
1017
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001018 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
1020 p->reclaim_state = NULL;
1021 p->flags &= ~PF_MEMALLOC;
1022
1023 cond_resched();
1024
1025 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001026 page = get_page_from_freelist(gfp_mask, order,
1027 zonelist, alloc_flags);
1028 if (page)
1029 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1031 /*
1032 * Go through the zonelist yet one more time, keep
1033 * very high watermark here, this is only to catch
1034 * a parallel oom killing, we must fail if we're still
1035 * under heavy pressure.
1036 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001037 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001038 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001039 if (page)
1040 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001042 out_of_memory(zonelist, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 goto restart;
1044 }
1045
1046 /*
1047 * Don't let big-order allocations loop unless the caller explicitly
1048 * requests that. Wait for some write requests to complete then retry.
1049 *
1050 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1051 * <= 3, but that may not be true in other implementations.
1052 */
1053 do_retry = 0;
1054 if (!(gfp_mask & __GFP_NORETRY)) {
1055 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1056 do_retry = 1;
1057 if (gfp_mask & __GFP_NOFAIL)
1058 do_retry = 1;
1059 }
1060 if (do_retry) {
1061 blk_congestion_wait(WRITE, HZ/50);
1062 goto rebalance;
1063 }
1064
1065nopage:
1066 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1067 printk(KERN_WARNING "%s: page allocation failure."
1068 " order:%d, mode:0x%x\n",
1069 p->comm, order, gfp_mask);
1070 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001071 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 return page;
1075}
1076
1077EXPORT_SYMBOL(__alloc_pages);
1078
1079/*
1080 * Common helper functions.
1081 */
Al Virodd0fc662005-10-07 07:46:04 +01001082fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083{
1084 struct page * page;
1085 page = alloc_pages(gfp_mask, order);
1086 if (!page)
1087 return 0;
1088 return (unsigned long) page_address(page);
1089}
1090
1091EXPORT_SYMBOL(__get_free_pages);
1092
Al Virodd0fc662005-10-07 07:46:04 +01001093fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094{
1095 struct page * page;
1096
1097 /*
1098 * get_zeroed_page() returns a 32-bit address, which cannot represent
1099 * a highmem page
1100 */
Al Viro260b2362005-10-21 03:22:44 -04001101 BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
1103 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1104 if (page)
1105 return (unsigned long) page_address(page);
1106 return 0;
1107}
1108
1109EXPORT_SYMBOL(get_zeroed_page);
1110
1111void __pagevec_free(struct pagevec *pvec)
1112{
1113 int i = pagevec_count(pvec);
1114
1115 while (--i >= 0)
1116 free_hot_cold_page(pvec->pages[i], pvec->cold);
1117}
1118
1119fastcall void __free_pages(struct page *page, unsigned int order)
1120{
Nick Pigginb5810032005-10-29 18:16:12 -07001121 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 if (order == 0)
1123 free_hot_page(page);
1124 else
1125 __free_pages_ok(page, order);
1126 }
1127}
1128
1129EXPORT_SYMBOL(__free_pages);
1130
1131fastcall void free_pages(unsigned long addr, unsigned int order)
1132{
1133 if (addr != 0) {
1134 BUG_ON(!virt_addr_valid((void *)addr));
1135 __free_pages(virt_to_page((void *)addr), order);
1136 }
1137}
1138
1139EXPORT_SYMBOL(free_pages);
1140
1141/*
1142 * Total amount of free (allocatable) RAM:
1143 */
1144unsigned int nr_free_pages(void)
1145{
1146 unsigned int sum = 0;
1147 struct zone *zone;
1148
1149 for_each_zone(zone)
1150 sum += zone->free_pages;
1151
1152 return sum;
1153}
1154
1155EXPORT_SYMBOL(nr_free_pages);
1156
1157#ifdef CONFIG_NUMA
1158unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1159{
1160 unsigned int i, sum = 0;
1161
1162 for (i = 0; i < MAX_NR_ZONES; i++)
1163 sum += pgdat->node_zones[i].free_pages;
1164
1165 return sum;
1166}
1167#endif
1168
1169static unsigned int nr_free_zone_pages(int offset)
1170{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001171 /* Just pick one node, since fallback list is circular */
1172 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 unsigned int sum = 0;
1174
Martin J. Blighe310fd42005-07-29 22:59:18 -07001175 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1176 struct zone **zonep = zonelist->zones;
1177 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178
Martin J. Blighe310fd42005-07-29 22:59:18 -07001179 for (zone = *zonep++; zone; zone = *zonep++) {
1180 unsigned long size = zone->present_pages;
1181 unsigned long high = zone->pages_high;
1182 if (size > high)
1183 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 }
1185
1186 return sum;
1187}
1188
1189/*
1190 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1191 */
1192unsigned int nr_free_buffer_pages(void)
1193{
Al Viroaf4ca452005-10-21 02:55:38 -04001194 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195}
1196
1197/*
1198 * Amount of free RAM allocatable within all zones
1199 */
1200unsigned int nr_free_pagecache_pages(void)
1201{
Al Viroaf4ca452005-10-21 02:55:38 -04001202 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203}
1204
1205#ifdef CONFIG_HIGHMEM
1206unsigned int nr_free_highpages (void)
1207{
1208 pg_data_t *pgdat;
1209 unsigned int pages = 0;
1210
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08001211 for_each_online_pgdat(pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1213
1214 return pages;
1215}
1216#endif
1217
1218#ifdef CONFIG_NUMA
1219static void show_node(struct zone *zone)
1220{
1221 printk("Node %d ", zone->zone_pgdat->node_id);
1222}
1223#else
1224#define show_node(zone) do { } while (0)
1225#endif
1226
1227/*
1228 * Accumulate the page_state information across all CPUs.
1229 * The result is unavoidably approximate - it can change
1230 * during and after execution of this function.
1231 */
1232static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1233
1234atomic_t nr_pagecache = ATOMIC_INIT(0);
1235EXPORT_SYMBOL(nr_pagecache);
1236#ifdef CONFIG_SMP
1237DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1238#endif
1239
Nick Piggina86b1f52006-01-06 00:11:00 -08001240static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241{
Andrew Mortonb40607f2006-03-22 00:07:39 -08001242 unsigned cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001244 memset(ret, 0, nr * sizeof(unsigned long));
Andrew Morton84c20082006-01-08 01:00:28 -08001245 cpus_and(*cpumask, *cpumask, cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
Andrew Mortonb40607f2006-03-22 00:07:39 -08001247 for_each_cpu_mask(cpu, *cpumask) {
1248 unsigned long *in;
1249 unsigned long *out;
1250 unsigned off;
1251 unsigned next_cpu;
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001252
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 in = (unsigned long *)&per_cpu(page_states, cpu);
1254
Andrew Mortonb40607f2006-03-22 00:07:39 -08001255 next_cpu = next_cpu(cpu, *cpumask);
1256 if (likely(next_cpu < NR_CPUS))
1257 prefetch(&per_cpu(page_states, next_cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
1259 out = (unsigned long *)ret;
1260 for (off = 0; off < nr; off++)
1261 *out++ += *in++;
1262 }
1263}
1264
Martin Hicksc07e02d2005-09-03 15:55:11 -07001265void get_page_state_node(struct page_state *ret, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266{
1267 int nr;
Martin Hicksc07e02d2005-09-03 15:55:11 -07001268 cpumask_t mask = node_to_cpumask(node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269
1270 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1271 nr /= sizeof(unsigned long);
1272
Martin Hicksc07e02d2005-09-03 15:55:11 -07001273 __get_page_state(ret, nr+1, &mask);
1274}
1275
1276void get_page_state(struct page_state *ret)
1277{
1278 int nr;
1279 cpumask_t mask = CPU_MASK_ALL;
1280
1281 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1282 nr /= sizeof(unsigned long);
1283
1284 __get_page_state(ret, nr + 1, &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285}
1286
1287void get_full_page_state(struct page_state *ret)
1288{
Martin Hicksc07e02d2005-09-03 15:55:11 -07001289 cpumask_t mask = CPU_MASK_ALL;
1290
1291 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292}
1293
Nick Piggina74609f2006-01-06 00:11:20 -08001294unsigned long read_page_state_offset(unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295{
1296 unsigned long ret = 0;
1297 int cpu;
1298
Andrew Morton84c20082006-01-08 01:00:28 -08001299 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 unsigned long in;
1301
1302 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1303 ret += *((unsigned long *)in);
1304 }
1305 return ret;
1306}
1307
Nick Piggina74609f2006-01-06 00:11:20 -08001308void __mod_page_state_offset(unsigned long offset, unsigned long delta)
1309{
1310 void *ptr;
1311
1312 ptr = &__get_cpu_var(page_states);
1313 *(unsigned long *)(ptr + offset) += delta;
1314}
1315EXPORT_SYMBOL(__mod_page_state_offset);
1316
1317void mod_page_state_offset(unsigned long offset, unsigned long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318{
1319 unsigned long flags;
Nick Piggina74609f2006-01-06 00:11:20 -08001320 void *ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
1322 local_irq_save(flags);
1323 ptr = &__get_cpu_var(page_states);
Nick Piggina74609f2006-01-06 00:11:20 -08001324 *(unsigned long *)(ptr + offset) += delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 local_irq_restore(flags);
1326}
Nick Piggina74609f2006-01-06 00:11:20 -08001327EXPORT_SYMBOL(mod_page_state_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
1329void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1330 unsigned long *free, struct pglist_data *pgdat)
1331{
1332 struct zone *zones = pgdat->node_zones;
1333 int i;
1334
1335 *active = 0;
1336 *inactive = 0;
1337 *free = 0;
1338 for (i = 0; i < MAX_NR_ZONES; i++) {
1339 *active += zones[i].nr_active;
1340 *inactive += zones[i].nr_inactive;
1341 *free += zones[i].free_pages;
1342 }
1343}
1344
1345void get_zone_counts(unsigned long *active,
1346 unsigned long *inactive, unsigned long *free)
1347{
1348 struct pglist_data *pgdat;
1349
1350 *active = 0;
1351 *inactive = 0;
1352 *free = 0;
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08001353 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 unsigned long l, m, n;
1355 __get_zone_counts(&l, &m, &n, pgdat);
1356 *active += l;
1357 *inactive += m;
1358 *free += n;
1359 }
1360}
1361
1362void si_meminfo(struct sysinfo *val)
1363{
1364 val->totalram = totalram_pages;
1365 val->sharedram = 0;
1366 val->freeram = nr_free_pages();
1367 val->bufferram = nr_blockdev_pages();
1368#ifdef CONFIG_HIGHMEM
1369 val->totalhigh = totalhigh_pages;
1370 val->freehigh = nr_free_highpages();
1371#else
1372 val->totalhigh = 0;
1373 val->freehigh = 0;
1374#endif
1375 val->mem_unit = PAGE_SIZE;
1376}
1377
1378EXPORT_SYMBOL(si_meminfo);
1379
1380#ifdef CONFIG_NUMA
1381void si_meminfo_node(struct sysinfo *val, int nid)
1382{
1383 pg_data_t *pgdat = NODE_DATA(nid);
1384
1385 val->totalram = pgdat->node_present_pages;
1386 val->freeram = nr_free_pages_pgdat(pgdat);
1387 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1388 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1389 val->mem_unit = PAGE_SIZE;
1390}
1391#endif
1392
1393#define K(x) ((x) << (PAGE_SHIFT-10))
1394
1395/*
1396 * Show free area list (used inside shift_scroll-lock stuff)
1397 * We also calculate the percentage fragmentation. We do this by counting the
1398 * memory on each free list with the exception of the first item on the list.
1399 */
1400void show_free_areas(void)
1401{
1402 struct page_state ps;
1403 int cpu, temperature;
1404 unsigned long active;
1405 unsigned long inactive;
1406 unsigned long free;
1407 struct zone *zone;
1408
1409 for_each_zone(zone) {
1410 show_node(zone);
1411 printk("%s per-cpu:", zone->name);
1412
Con Kolivasf3fe6512006-01-06 00:11:15 -08001413 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 printk(" empty\n");
1415 continue;
1416 } else
1417 printk("\n");
1418
Dave Jones6b482c62005-11-10 15:45:56 -05001419 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 struct per_cpu_pageset *pageset;
1421
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001422 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
1424 for (temperature = 0; temperature < 2; temperature++)
Nick Piggin2d92c5c2006-01-06 00:10:59 -08001425 printk("cpu %d %s: high %d, batch %d used:%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 cpu,
1427 temperature ? "cold" : "hot",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 pageset->pcp[temperature].high,
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001429 pageset->pcp[temperature].batch,
1430 pageset->pcp[temperature].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 }
1432 }
1433
1434 get_page_state(&ps);
1435 get_zone_counts(&active, &inactive, &free);
1436
Denis Vlasenkoc0d62212005-06-21 17:15:14 -07001437 printk("Free pages: %11ukB (%ukB HighMem)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 K(nr_free_pages()),
1439 K(nr_free_highpages()));
1440
1441 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1442 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1443 active,
1444 inactive,
1445 ps.nr_dirty,
1446 ps.nr_writeback,
1447 ps.nr_unstable,
1448 nr_free_pages(),
1449 ps.nr_slab,
1450 ps.nr_mapped,
1451 ps.nr_page_table_pages);
1452
1453 for_each_zone(zone) {
1454 int i;
1455
1456 show_node(zone);
1457 printk("%s"
1458 " free:%lukB"
1459 " min:%lukB"
1460 " low:%lukB"
1461 " high:%lukB"
1462 " active:%lukB"
1463 " inactive:%lukB"
1464 " present:%lukB"
1465 " pages_scanned:%lu"
1466 " all_unreclaimable? %s"
1467 "\n",
1468 zone->name,
1469 K(zone->free_pages),
1470 K(zone->pages_min),
1471 K(zone->pages_low),
1472 K(zone->pages_high),
1473 K(zone->nr_active),
1474 K(zone->nr_inactive),
1475 K(zone->present_pages),
1476 zone->pages_scanned,
1477 (zone->all_unreclaimable ? "yes" : "no")
1478 );
1479 printk("lowmem_reserve[]:");
1480 for (i = 0; i < MAX_NR_ZONES; i++)
1481 printk(" %lu", zone->lowmem_reserve[i]);
1482 printk("\n");
1483 }
1484
1485 for_each_zone(zone) {
1486 unsigned long nr, flags, order, total = 0;
1487
1488 show_node(zone);
1489 printk("%s: ", zone->name);
Con Kolivasf3fe6512006-01-06 00:11:15 -08001490 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 printk("empty\n");
1492 continue;
1493 }
1494
1495 spin_lock_irqsave(&zone->lock, flags);
1496 for (order = 0; order < MAX_ORDER; order++) {
1497 nr = zone->free_area[order].nr_free;
1498 total += nr << order;
1499 printk("%lu*%lukB ", nr, K(1UL) << order);
1500 }
1501 spin_unlock_irqrestore(&zone->lock, flags);
1502 printk("= %lukB\n", K(total));
1503 }
1504
1505 show_swap_cache_info();
1506}
1507
1508/*
1509 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001510 *
1511 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 */
Christoph Lameter1a932052006-01-06 00:11:16 -08001513static int __init build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter070f8032006-01-06 00:11:19 -08001514 struct zonelist *zonelist, int nr_zones, int zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515{
Christoph Lameter1a932052006-01-06 00:11:16 -08001516 struct zone *zone;
1517
Christoph Lameter070f8032006-01-06 00:11:19 -08001518 BUG_ON(zone_type > ZONE_HIGHMEM);
Christoph Lameter02a68a52006-01-06 00:11:18 -08001519
1520 do {
Christoph Lameter070f8032006-01-06 00:11:19 -08001521 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001522 if (populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523#ifndef CONFIG_HIGHMEM
Christoph Lameter070f8032006-01-06 00:11:19 -08001524 BUG_ON(zone_type > ZONE_NORMAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525#endif
Christoph Lameter070f8032006-01-06 00:11:19 -08001526 zonelist->zones[nr_zones++] = zone;
1527 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 }
Christoph Lameter070f8032006-01-06 00:11:19 -08001529 zone_type--;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001530
Christoph Lameter070f8032006-01-06 00:11:19 -08001531 } while (zone_type >= 0);
1532 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533}
1534
Al Viro260b2362005-10-21 03:22:44 -04001535static inline int highest_zone(int zone_bits)
1536{
1537 int res = ZONE_NORMAL;
1538 if (zone_bits & (__force int)__GFP_HIGHMEM)
1539 res = ZONE_HIGHMEM;
Andi Kleena2f1b422005-11-05 17:25:53 +01001540 if (zone_bits & (__force int)__GFP_DMA32)
1541 res = ZONE_DMA32;
Al Viro260b2362005-10-21 03:22:44 -04001542 if (zone_bits & (__force int)__GFP_DMA)
1543 res = ZONE_DMA;
1544 return res;
1545}
1546
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547#ifdef CONFIG_NUMA
1548#define MAX_NODE_LOAD (num_online_nodes())
1549static int __initdata node_load[MAX_NUMNODES];
1550/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001551 * 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 -07001552 * @node: node whose fallback list we're appending
1553 * @used_node_mask: nodemask_t of already used nodes
1554 *
1555 * We use a number of factors to determine which is the next node that should
1556 * appear on a given node's fallback list. The node should not have appeared
1557 * already in @node's fallback list, and it should be the next closest node
1558 * according to the distance array (which contains arbitrary distance values
1559 * from each node to each node in the system), and should also prefer nodes
1560 * with no CPUs, since presumably they'll have very little allocation pressure
1561 * on them otherwise.
1562 * It returns -1 if no node is found.
1563 */
1564static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1565{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001566 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 int min_val = INT_MAX;
1568 int best_node = -1;
1569
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001570 /* Use the local node if we haven't already */
1571 if (!node_isset(node, *used_node_mask)) {
1572 node_set(node, *used_node_mask);
1573 return node;
1574 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001576 for_each_online_node(n) {
1577 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578
1579 /* Don't want a node to appear more than once */
1580 if (node_isset(n, *used_node_mask))
1581 continue;
1582
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 /* Use the distance array to find the distance */
1584 val = node_distance(node, n);
1585
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001586 /* Penalize nodes under us ("prefer the next node") */
1587 val += (n < node);
1588
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 /* Give preference to headless and unused nodes */
1590 tmp = node_to_cpumask(n);
1591 if (!cpus_empty(tmp))
1592 val += PENALTY_FOR_NODE_WITH_CPUS;
1593
1594 /* Slight preference for less loaded node */
1595 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1596 val += node_load[n];
1597
1598 if (val < min_val) {
1599 min_val = val;
1600 best_node = n;
1601 }
1602 }
1603
1604 if (best_node >= 0)
1605 node_set(best_node, *used_node_mask);
1606
1607 return best_node;
1608}
1609
1610static void __init build_zonelists(pg_data_t *pgdat)
1611{
1612 int i, j, k, node, local_node;
1613 int prev_node, load;
1614 struct zonelist *zonelist;
1615 nodemask_t used_mask;
1616
1617 /* initialize zonelists */
1618 for (i = 0; i < GFP_ZONETYPES; i++) {
1619 zonelist = pgdat->node_zonelists + i;
1620 zonelist->zones[0] = NULL;
1621 }
1622
1623 /* NUMA-aware ordering of nodes */
1624 local_node = pgdat->node_id;
1625 load = num_online_nodes();
1626 prev_node = local_node;
1627 nodes_clear(used_mask);
1628 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001629 int distance = node_distance(local_node, node);
1630
1631 /*
1632 * If another node is sufficiently far away then it is better
1633 * to reclaim pages in a zone before going off node.
1634 */
1635 if (distance > RECLAIM_DISTANCE)
1636 zone_reclaim_mode = 1;
1637
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 /*
1639 * We don't want to pressure a particular node.
1640 * So adding penalty to the first node in same
1641 * distance group to make it round-robin.
1642 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08001643
1644 if (distance != node_distance(local_node, prev_node))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 node_load[node] += load;
1646 prev_node = node;
1647 load--;
1648 for (i = 0; i < GFP_ZONETYPES; i++) {
1649 zonelist = pgdat->node_zonelists + i;
1650 for (j = 0; zonelist->zones[j] != NULL; j++);
1651
Al Viro260b2362005-10-21 03:22:44 -04001652 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
1654 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1655 zonelist->zones[j] = NULL;
1656 }
1657 }
1658}
1659
1660#else /* CONFIG_NUMA */
1661
1662static void __init build_zonelists(pg_data_t *pgdat)
1663{
1664 int i, j, k, node, local_node;
1665
1666 local_node = pgdat->node_id;
1667 for (i = 0; i < GFP_ZONETYPES; i++) {
1668 struct zonelist *zonelist;
1669
1670 zonelist = pgdat->node_zonelists + i;
1671
1672 j = 0;
Al Viro260b2362005-10-21 03:22:44 -04001673 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 j = build_zonelists_node(pgdat, zonelist, j, k);
1675 /*
1676 * Now we build the zonelist so that it contains the zones
1677 * of all the other nodes.
1678 * We don't want to pressure a particular node, so when
1679 * building the zones for node N, we make sure that the
1680 * zones coming right after the local ones are those from
1681 * node N+1 (modulo N)
1682 */
1683 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1684 if (!node_online(node))
1685 continue;
1686 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1687 }
1688 for (node = 0; node < local_node; node++) {
1689 if (!node_online(node))
1690 continue;
1691 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1692 }
1693
1694 zonelist->zones[j] = NULL;
1695 }
1696}
1697
1698#endif /* CONFIG_NUMA */
1699
1700void __init build_all_zonelists(void)
1701{
1702 int i;
1703
1704 for_each_online_node(i)
1705 build_zonelists(NODE_DATA(i));
1706 printk("Built %i zonelists\n", num_online_nodes());
1707 cpuset_init_current_mems_allowed();
1708}
1709
1710/*
1711 * Helper functions to size the waitqueue hash table.
1712 * Essentially these want to choose hash table sizes sufficiently
1713 * large so that collisions trying to wait on pages are rare.
1714 * But in fact, the number of active page waitqueues on typical
1715 * systems is ridiculously low, less than 200. So this is even
1716 * conservative, even though it seems large.
1717 *
1718 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1719 * waitqueues, i.e. the size of the waitq table given the number of pages.
1720 */
1721#define PAGES_PER_WAITQUEUE 256
1722
1723static inline unsigned long wait_table_size(unsigned long pages)
1724{
1725 unsigned long size = 1;
1726
1727 pages /= PAGES_PER_WAITQUEUE;
1728
1729 while (size < pages)
1730 size <<= 1;
1731
1732 /*
1733 * Once we have dozens or even hundreds of threads sleeping
1734 * on IO we've got bigger problems than wait queue collision.
1735 * Limit the size of the wait table to a reasonable size.
1736 */
1737 size = min(size, 4096UL);
1738
1739 return max(size, 4UL);
1740}
1741
1742/*
1743 * This is an integer logarithm so that shifts can be used later
1744 * to extract the more random high bits from the multiplicative
1745 * hash function before the remainder is taken.
1746 */
1747static inline unsigned long wait_table_bits(unsigned long size)
1748{
1749 return ffz(~size);
1750}
1751
1752#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1753
1754static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1755 unsigned long *zones_size, unsigned long *zholes_size)
1756{
1757 unsigned long realtotalpages, totalpages = 0;
1758 int i;
1759
1760 for (i = 0; i < MAX_NR_ZONES; i++)
1761 totalpages += zones_size[i];
1762 pgdat->node_spanned_pages = totalpages;
1763
1764 realtotalpages = totalpages;
1765 if (zholes_size)
1766 for (i = 0; i < MAX_NR_ZONES; i++)
1767 realtotalpages -= zholes_size[i];
1768 pgdat->node_present_pages = realtotalpages;
1769 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1770}
1771
1772
1773/*
1774 * Initially all pages are reserved - free ones are freed
1775 * up by free_all_bootmem() once the early boot process is
1776 * done. Non-atomic initialization, single-pass.
1777 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001778void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 unsigned long start_pfn)
1780{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001782 unsigned long end_pfn = start_pfn + size;
1783 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001785 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001786 if (!early_pfn_valid(pfn))
1787 continue;
1788 page = pfn_to_page(pfn);
1789 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08001790 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 reset_page_mapcount(page);
1792 SetPageReserved(page);
1793 INIT_LIST_HEAD(&page->lru);
1794#ifdef WANT_PAGE_VIRTUAL
1795 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1796 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001797 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 }
1800}
1801
1802void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1803 unsigned long size)
1804{
1805 int order;
1806 for (order = 0; order < MAX_ORDER ; order++) {
1807 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1808 zone->free_area[order].nr_free = 0;
1809 }
1810}
1811
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001812#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1813void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1814 unsigned long size)
1815{
1816 unsigned long snum = pfn_to_section_nr(pfn);
1817 unsigned long end = pfn_to_section_nr(pfn + size);
1818
1819 if (FLAGS_HAS_NODE)
1820 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1821 else
1822 for (; snum <= end; snum++)
1823 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1824}
1825
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826#ifndef __HAVE_ARCH_MEMMAP_INIT
1827#define memmap_init(size, nid, zone, start_pfn) \
1828 memmap_init_zone((size), (nid), (zone), (start_pfn))
1829#endif
1830
Ashok Raj6292d9a2006-02-01 03:04:44 -08001831static int __cpuinit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001832{
1833 int batch;
1834
1835 /*
1836 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001837 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001838 *
1839 * OK, so we don't know how big the cache is. So guess.
1840 */
1841 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001842 if (batch * PAGE_SIZE > 512 * 1024)
1843 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001844 batch /= 4; /* We effectively *= 4 below */
1845 if (batch < 1)
1846 batch = 1;
1847
1848 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001849 * Clamp the batch to a 2^n - 1 value. Having a power
1850 * of 2 value was found to be more likely to have
1851 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001852 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001853 * For example if 2 tasks are alternately allocating
1854 * batches of pages, one task can end up with a lot
1855 * of pages of one half of the possible page colors
1856 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001857 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001858 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001859
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001860 return batch;
1861}
1862
Christoph Lameter2caaad42005-06-21 17:15:00 -07001863inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1864{
1865 struct per_cpu_pages *pcp;
1866
Magnus Damm1c6fe942005-10-26 01:58:59 -07001867 memset(p, 0, sizeof(*p));
1868
Christoph Lameter2caaad42005-06-21 17:15:00 -07001869 pcp = &p->pcp[0]; /* hot */
1870 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001871 pcp->high = 6 * batch;
1872 pcp->batch = max(1UL, 1 * batch);
1873 INIT_LIST_HEAD(&pcp->list);
1874
1875 pcp = &p->pcp[1]; /* cold*/
1876 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001877 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001878 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001879 INIT_LIST_HEAD(&pcp->list);
1880}
1881
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001882/*
1883 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1884 * to the value high for the pageset p.
1885 */
1886
1887static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1888 unsigned long high)
1889{
1890 struct per_cpu_pages *pcp;
1891
1892 pcp = &p->pcp[0]; /* hot list */
1893 pcp->high = high;
1894 pcp->batch = max(1UL, high/4);
1895 if ((high/4) > (PAGE_SHIFT * 8))
1896 pcp->batch = PAGE_SHIFT * 8;
1897}
1898
1899
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001900#ifdef CONFIG_NUMA
1901/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001902 * Boot pageset table. One per cpu which is going to be used for all
1903 * zones and all nodes. The parameters will be set in such a way
1904 * that an item put on a list will immediately be handed over to
1905 * the buddy list. This is safe since pageset manipulation is done
1906 * with interrupts disabled.
1907 *
1908 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001909 *
1910 * The boot_pagesets must be kept even after bootup is complete for
1911 * unused processors and/or zones. They do play a role for bootstrapping
1912 * hotplugged processors.
1913 *
1914 * zoneinfo_show() and maybe other functions do
1915 * not check if the processor is online before following the pageset pointer.
1916 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001917 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001918static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001919
1920/*
1921 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001922 * per cpu pageset array in struct zone.
1923 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08001924static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001925{
1926 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001927
1928 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001929
Nick Piggin23316bc2006-01-08 01:00:41 -08001930 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001931 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08001932 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001933 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001934
Nick Piggin23316bc2006-01-08 01:00:41 -08001935 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001936
1937 if (percpu_pagelist_fraction)
1938 setup_pagelist_highmark(zone_pcp(zone, cpu),
1939 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001940 }
1941
1942 return 0;
1943bad:
1944 for_each_zone(dzone) {
1945 if (dzone == zone)
1946 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08001947 kfree(zone_pcp(dzone, cpu));
1948 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001949 }
1950 return -ENOMEM;
1951}
1952
1953static inline void free_zone_pagesets(int cpu)
1954{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001955 struct zone *zone;
1956
1957 for_each_zone(zone) {
1958 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1959
1960 zone_pcp(zone, cpu) = NULL;
1961 kfree(pset);
1962 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001963}
1964
Chandra Seetharaman83d722f2006-04-24 19:35:21 -07001965static int pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001966 unsigned long action,
1967 void *hcpu)
1968{
1969 int cpu = (long)hcpu;
1970 int ret = NOTIFY_OK;
1971
1972 switch (action) {
1973 case CPU_UP_PREPARE:
1974 if (process_zones(cpu))
1975 ret = NOTIFY_BAD;
1976 break;
Andi Kleenb0d41692005-11-05 17:25:53 +01001977 case CPU_UP_CANCELED:
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001978 case CPU_DEAD:
1979 free_zone_pagesets(cpu);
1980 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001981 default:
1982 break;
1983 }
1984 return ret;
1985}
1986
1987static struct notifier_block pageset_notifier =
1988 { &pageset_cpuup_callback, NULL, 0 };
1989
Al Viro78d99552005-12-15 09:18:25 +00001990void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001991{
1992 int err;
1993
1994 /* Initialize per_cpu_pageset for cpu 0.
1995 * A cpuup callback will do this for every cpu
1996 * as it comes online
1997 */
1998 err = process_zones(smp_processor_id());
1999 BUG_ON(err);
2000 register_cpu_notifier(&pageset_notifier);
2001}
2002
2003#endif
2004
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002005static __meminit
Dave Hansened8ece22005-10-29 18:16:50 -07002006void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2007{
2008 int i;
2009 struct pglist_data *pgdat = zone->zone_pgdat;
2010
2011 /*
2012 * The per-page waitqueue mechanism uses hashed waitqueues
2013 * per zone.
2014 */
2015 zone->wait_table_size = wait_table_size(zone_size_pages);
2016 zone->wait_table_bits = wait_table_bits(zone->wait_table_size);
2017 zone->wait_table = (wait_queue_head_t *)
2018 alloc_bootmem_node(pgdat, zone->wait_table_size
2019 * sizeof(wait_queue_head_t));
2020
2021 for(i = 0; i < zone->wait_table_size; ++i)
2022 init_waitqueue_head(zone->wait_table + i);
2023}
2024
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002025static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07002026{
2027 int cpu;
2028 unsigned long batch = zone_batchsize(zone);
2029
2030 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2031#ifdef CONFIG_NUMA
2032 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002033 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002034 setup_pageset(&boot_pageset[cpu],0);
2035#else
2036 setup_pageset(zone_pcp(zone,cpu), batch);
2037#endif
2038 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08002039 if (zone->present_pages)
2040 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2041 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07002042}
2043
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002044static __meminit void init_currently_empty_zone(struct zone *zone,
Dave Hansened8ece22005-10-29 18:16:50 -07002045 unsigned long zone_start_pfn, unsigned long size)
2046{
2047 struct pglist_data *pgdat = zone->zone_pgdat;
2048
2049 zone_wait_table_init(zone, size);
2050 pgdat->nr_zones = zone_idx(zone) + 1;
2051
Dave Hansened8ece22005-10-29 18:16:50 -07002052 zone->zone_start_pfn = zone_start_pfn;
2053
2054 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2055
2056 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2057}
2058
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059/*
2060 * Set up the zone data structures:
2061 * - mark all pages reserved
2062 * - mark all memory queues empty
2063 * - clear the memory bitmaps
2064 */
2065static void __init free_area_init_core(struct pglist_data *pgdat,
2066 unsigned long *zones_size, unsigned long *zholes_size)
2067{
Dave Hansened8ece22005-10-29 18:16:50 -07002068 unsigned long j;
2069 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 unsigned long zone_start_pfn = pgdat->node_start_pfn;
2071
Dave Hansen208d54e2005-10-29 18:16:52 -07002072 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 pgdat->nr_zones = 0;
2074 init_waitqueue_head(&pgdat->kswapd_wait);
2075 pgdat->kswapd_max_order = 0;
2076
2077 for (j = 0; j < MAX_NR_ZONES; j++) {
2078 struct zone *zone = pgdat->node_zones + j;
2079 unsigned long size, realsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 realsize = size = zones_size[j];
2082 if (zholes_size)
2083 realsize -= zholes_size[j];
2084
Andi Kleena2f1b422005-11-05 17:25:53 +01002085 if (j < ZONE_HIGHMEM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 nr_kernel_pages += realsize;
2087 nr_all_pages += realsize;
2088
2089 zone->spanned_pages = size;
2090 zone->present_pages = realsize;
2091 zone->name = zone_names[j];
2092 spin_lock_init(&zone->lock);
2093 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002094 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 zone->zone_pgdat = pgdat;
2096 zone->free_pages = 0;
2097
2098 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2099
Dave Hansened8ece22005-10-29 18:16:50 -07002100 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 INIT_LIST_HEAD(&zone->active_list);
2102 INIT_LIST_HEAD(&zone->inactive_list);
2103 zone->nr_scan_active = 0;
2104 zone->nr_scan_inactive = 0;
2105 zone->nr_active = 0;
2106 zone->nr_inactive = 0;
Martin Hicks53e9a612005-09-03 15:54:51 -07002107 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 if (!size)
2109 continue;
2110
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002111 zonetable_add(zone, nid, j, zone_start_pfn, size);
Dave Hansened8ece22005-10-29 18:16:50 -07002112 init_currently_empty_zone(zone, zone_start_pfn, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 }
2115}
2116
2117static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2118{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 /* Skip empty nodes */
2120 if (!pgdat->node_spanned_pages)
2121 return;
2122
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002123#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 /* ia64 gets its own node_mem_map, before this, without bootmem */
2125 if (!pgdat->node_mem_map) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002126 unsigned long size;
2127 struct page *map;
2128
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002130 map = alloc_remap(pgdat->node_id, size);
2131 if (!map)
2132 map = alloc_bootmem_node(pgdat, size);
2133 pgdat->node_mem_map = map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002135#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 /*
2137 * With no DISCONTIG, the global mem_map is just set as node 0's
2138 */
2139 if (pgdat == NODE_DATA(0))
2140 mem_map = NODE_DATA(0)->node_mem_map;
2141#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002142#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143}
2144
2145void __init free_area_init_node(int nid, struct pglist_data *pgdat,
2146 unsigned long *zones_size, unsigned long node_start_pfn,
2147 unsigned long *zholes_size)
2148{
2149 pgdat->node_id = nid;
2150 pgdat->node_start_pfn = node_start_pfn;
2151 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2152
2153 alloc_node_mem_map(pgdat);
2154
2155 free_area_init_core(pgdat, zones_size, zholes_size);
2156}
2157
Dave Hansen93b75042005-06-23 00:07:47 -07002158#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159static bootmem_data_t contig_bootmem_data;
2160struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2161
2162EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002163#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164
2165void __init free_area_init(unsigned long *zones_size)
2166{
Dave Hansen93b75042005-06-23 00:07:47 -07002167 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2169}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170
2171#ifdef CONFIG_PROC_FS
2172
2173#include <linux/seq_file.h>
2174
2175static void *frag_start(struct seq_file *m, loff_t *pos)
2176{
2177 pg_data_t *pgdat;
2178 loff_t node = *pos;
KAMEZAWA Hiroyukiae0f15f2006-03-27 01:16:01 -08002179 for (pgdat = first_online_pgdat();
2180 pgdat && node;
2181 pgdat = next_online_pgdat(pgdat))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 --node;
2183
2184 return pgdat;
2185}
2186
2187static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
2188{
2189 pg_data_t *pgdat = (pg_data_t *)arg;
2190
2191 (*pos)++;
KAMEZAWA Hiroyukiae0f15f2006-03-27 01:16:01 -08002192 return next_online_pgdat(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193}
2194
2195static void frag_stop(struct seq_file *m, void *arg)
2196{
2197}
2198
2199/*
2200 * This walks the free areas for each zone.
2201 */
2202static int frag_show(struct seq_file *m, void *arg)
2203{
2204 pg_data_t *pgdat = (pg_data_t *)arg;
2205 struct zone *zone;
2206 struct zone *node_zones = pgdat->node_zones;
2207 unsigned long flags;
2208 int order;
2209
2210 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
Con Kolivasf3fe6512006-01-06 00:11:15 -08002211 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 continue;
2213
2214 spin_lock_irqsave(&zone->lock, flags);
2215 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2216 for (order = 0; order < MAX_ORDER; ++order)
2217 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2218 spin_unlock_irqrestore(&zone->lock, flags);
2219 seq_putc(m, '\n');
2220 }
2221 return 0;
2222}
2223
2224struct seq_operations fragmentation_op = {
2225 .start = frag_start,
2226 .next = frag_next,
2227 .stop = frag_stop,
2228 .show = frag_show,
2229};
2230
Nikita Danilov295ab932005-06-21 17:14:38 -07002231/*
2232 * Output information about zones in @pgdat.
2233 */
2234static int zoneinfo_show(struct seq_file *m, void *arg)
2235{
2236 pg_data_t *pgdat = arg;
2237 struct zone *zone;
2238 struct zone *node_zones = pgdat->node_zones;
2239 unsigned long flags;
2240
2241 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2242 int i;
2243
Con Kolivasf3fe6512006-01-06 00:11:15 -08002244 if (!populated_zone(zone))
Nikita Danilov295ab932005-06-21 17:14:38 -07002245 continue;
2246
2247 spin_lock_irqsave(&zone->lock, flags);
2248 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2249 seq_printf(m,
2250 "\n pages free %lu"
2251 "\n min %lu"
2252 "\n low %lu"
2253 "\n high %lu"
2254 "\n active %lu"
2255 "\n inactive %lu"
2256 "\n scanned %lu (a: %lu i: %lu)"
2257 "\n spanned %lu"
2258 "\n present %lu",
2259 zone->free_pages,
2260 zone->pages_min,
2261 zone->pages_low,
2262 zone->pages_high,
2263 zone->nr_active,
2264 zone->nr_inactive,
2265 zone->pages_scanned,
2266 zone->nr_scan_active, zone->nr_scan_inactive,
2267 zone->spanned_pages,
2268 zone->present_pages);
2269 seq_printf(m,
2270 "\n protection: (%lu",
2271 zone->lowmem_reserve[0]);
2272 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2273 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2274 seq_printf(m,
2275 ")"
2276 "\n pagesets");
Nick Piggin23316bc2006-01-08 01:00:41 -08002277 for_each_online_cpu(i) {
Nikita Danilov295ab932005-06-21 17:14:38 -07002278 struct per_cpu_pageset *pageset;
2279 int j;
2280
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002281 pageset = zone_pcp(zone, i);
Nikita Danilov295ab932005-06-21 17:14:38 -07002282 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2283 if (pageset->pcp[j].count)
2284 break;
2285 }
2286 if (j == ARRAY_SIZE(pageset->pcp))
2287 continue;
2288 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2289 seq_printf(m,
2290 "\n cpu: %i pcp: %i"
2291 "\n count: %i"
Nikita Danilov295ab932005-06-21 17:14:38 -07002292 "\n high: %i"
2293 "\n batch: %i",
2294 i, j,
2295 pageset->pcp[j].count,
Nikita Danilov295ab932005-06-21 17:14:38 -07002296 pageset->pcp[j].high,
2297 pageset->pcp[j].batch);
2298 }
2299#ifdef CONFIG_NUMA
2300 seq_printf(m,
2301 "\n numa_hit: %lu"
2302 "\n numa_miss: %lu"
2303 "\n numa_foreign: %lu"
2304 "\n interleave_hit: %lu"
2305 "\n local_node: %lu"
2306 "\n other_node: %lu",
2307 pageset->numa_hit,
2308 pageset->numa_miss,
2309 pageset->numa_foreign,
2310 pageset->interleave_hit,
2311 pageset->local_node,
2312 pageset->other_node);
2313#endif
2314 }
2315 seq_printf(m,
2316 "\n all_unreclaimable: %u"
2317 "\n prev_priority: %i"
2318 "\n temp_priority: %i"
2319 "\n start_pfn: %lu",
2320 zone->all_unreclaimable,
2321 zone->prev_priority,
2322 zone->temp_priority,
2323 zone->zone_start_pfn);
2324 spin_unlock_irqrestore(&zone->lock, flags);
2325 seq_putc(m, '\n');
2326 }
2327 return 0;
2328}
2329
2330struct seq_operations zoneinfo_op = {
2331 .start = frag_start, /* iterate over all zones. The same as in
2332 * fragmentation. */
2333 .next = frag_next,
2334 .stop = frag_stop,
2335 .show = zoneinfo_show,
2336};
2337
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338static char *vmstat_text[] = {
2339 "nr_dirty",
2340 "nr_writeback",
2341 "nr_unstable",
2342 "nr_page_table_pages",
2343 "nr_mapped",
2344 "nr_slab",
2345
2346 "pgpgin",
2347 "pgpgout",
2348 "pswpin",
2349 "pswpout",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350
Nick Piggin9328b8f2006-01-06 00:11:10 -08002351 "pgalloc_high",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 "pgalloc_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002353 "pgalloc_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 "pgalloc_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002355
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 "pgfree",
2357 "pgactivate",
2358 "pgdeactivate",
2359
2360 "pgfault",
2361 "pgmajfault",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002362
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 "pgrefill_high",
2364 "pgrefill_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002365 "pgrefill_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 "pgrefill_dma",
2367
2368 "pgsteal_high",
2369 "pgsteal_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002370 "pgsteal_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 "pgsteal_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002372
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 "pgscan_kswapd_high",
2374 "pgscan_kswapd_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002375 "pgscan_kswapd_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 "pgscan_kswapd_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002377
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 "pgscan_direct_high",
2379 "pgscan_direct_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002380 "pgscan_direct_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 "pgscan_direct_dma",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382
Nick Piggin9328b8f2006-01-06 00:11:10 -08002383 "pginodesteal",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 "slabs_scanned",
2385 "kswapd_steal",
2386 "kswapd_inodesteal",
2387 "pageoutrun",
2388 "allocstall",
2389
2390 "pgrotated",
KAMEZAWA Hiroyukiedfbe2b2005-05-01 08:58:37 -07002391 "nr_bounce",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392};
2393
2394static void *vmstat_start(struct seq_file *m, loff_t *pos)
2395{
2396 struct page_state *ps;
2397
2398 if (*pos >= ARRAY_SIZE(vmstat_text))
2399 return NULL;
2400
2401 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2402 m->private = ps;
2403 if (!ps)
2404 return ERR_PTR(-ENOMEM);
2405 get_full_page_state(ps);
2406 ps->pgpgin /= 2; /* sectors -> kbytes */
2407 ps->pgpgout /= 2;
2408 return (unsigned long *)ps + *pos;
2409}
2410
2411static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2412{
2413 (*pos)++;
2414 if (*pos >= ARRAY_SIZE(vmstat_text))
2415 return NULL;
2416 return (unsigned long *)m->private + *pos;
2417}
2418
2419static int vmstat_show(struct seq_file *m, void *arg)
2420{
2421 unsigned long *l = arg;
2422 unsigned long off = l - (unsigned long *)m->private;
2423
2424 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2425 return 0;
2426}
2427
2428static void vmstat_stop(struct seq_file *m, void *arg)
2429{
2430 kfree(m->private);
2431 m->private = NULL;
2432}
2433
2434struct seq_operations vmstat_op = {
2435 .start = vmstat_start,
2436 .next = vmstat_next,
2437 .stop = vmstat_stop,
2438 .show = vmstat_show,
2439};
2440
2441#endif /* CONFIG_PROC_FS */
2442
2443#ifdef CONFIG_HOTPLUG_CPU
2444static int page_alloc_cpu_notify(struct notifier_block *self,
2445 unsigned long action, void *hcpu)
2446{
2447 int cpu = (unsigned long)hcpu;
2448 long *count;
2449 unsigned long *src, *dest;
2450
2451 if (action == CPU_DEAD) {
2452 int i;
2453
2454 /* Drain local pagecache count. */
2455 count = &per_cpu(nr_pagecache_local, cpu);
2456 atomic_add(*count, &nr_pagecache);
2457 *count = 0;
2458 local_irq_disable();
2459 __drain_pages(cpu);
2460
2461 /* Add dead cpu's page_states to our own. */
2462 dest = (unsigned long *)&__get_cpu_var(page_states);
2463 src = (unsigned long *)&per_cpu(page_states, cpu);
2464
2465 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2466 i++) {
2467 dest[i] += src[i];
2468 src[i] = 0;
2469 }
2470
2471 local_irq_enable();
2472 }
2473 return NOTIFY_OK;
2474}
2475#endif /* CONFIG_HOTPLUG_CPU */
2476
2477void __init page_alloc_init(void)
2478{
2479 hotcpu_notifier(page_alloc_cpu_notify, 0);
2480}
2481
2482/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002483 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2484 * or min_free_kbytes changes.
2485 */
2486static void calculate_totalreserve_pages(void)
2487{
2488 struct pglist_data *pgdat;
2489 unsigned long reserve_pages = 0;
2490 int i, j;
2491
2492 for_each_online_pgdat(pgdat) {
2493 for (i = 0; i < MAX_NR_ZONES; i++) {
2494 struct zone *zone = pgdat->node_zones + i;
2495 unsigned long max = 0;
2496
2497 /* Find valid and maximum lowmem_reserve in the zone */
2498 for (j = i; j < MAX_NR_ZONES; j++) {
2499 if (zone->lowmem_reserve[j] > max)
2500 max = zone->lowmem_reserve[j];
2501 }
2502
2503 /* we treat pages_high as reserved pages. */
2504 max += zone->pages_high;
2505
2506 if (max > zone->present_pages)
2507 max = zone->present_pages;
2508 reserve_pages += max;
2509 }
2510 }
2511 totalreserve_pages = reserve_pages;
2512}
2513
2514/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 * setup_per_zone_lowmem_reserve - called whenever
2516 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2517 * has a correct pages reserved value, so an adequate number of
2518 * pages are left in the zone after a successful __alloc_pages().
2519 */
2520static void setup_per_zone_lowmem_reserve(void)
2521{
2522 struct pglist_data *pgdat;
2523 int j, idx;
2524
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08002525 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 for (j = 0; j < MAX_NR_ZONES; j++) {
2527 struct zone *zone = pgdat->node_zones + j;
2528 unsigned long present_pages = zone->present_pages;
2529
2530 zone->lowmem_reserve[j] = 0;
2531
2532 for (idx = j-1; idx >= 0; idx--) {
2533 struct zone *lower_zone;
2534
2535 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2536 sysctl_lowmem_reserve_ratio[idx] = 1;
2537
2538 lower_zone = pgdat->node_zones + idx;
2539 lower_zone->lowmem_reserve[j] = present_pages /
2540 sysctl_lowmem_reserve_ratio[idx];
2541 present_pages += lower_zone->present_pages;
2542 }
2543 }
2544 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002545
2546 /* update totalreserve_pages */
2547 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548}
2549
2550/*
2551 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2552 * that the pages_{min,low,high} values for each zone are set correctly
2553 * with respect to min_free_kbytes.
2554 */
Dave Hansen3947be12005-10-29 18:16:54 -07002555void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556{
2557 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2558 unsigned long lowmem_pages = 0;
2559 struct zone *zone;
2560 unsigned long flags;
2561
2562 /* Calculate total number of !ZONE_HIGHMEM pages */
2563 for_each_zone(zone) {
2564 if (!is_highmem(zone))
2565 lowmem_pages += zone->present_pages;
2566 }
2567
2568 for_each_zone(zone) {
Nick Piggin669ed172005-11-13 16:06:45 -08002569 unsigned long tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 spin_lock_irqsave(&zone->lru_lock, flags);
Nick Piggin669ed172005-11-13 16:06:45 -08002571 tmp = (pages_min * zone->present_pages) / lowmem_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 if (is_highmem(zone)) {
2573 /*
Nick Piggin669ed172005-11-13 16:06:45 -08002574 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2575 * need highmem pages, so cap pages_min to a small
2576 * value here.
2577 *
2578 * The (pages_high-pages_low) and (pages_low-pages_min)
2579 * deltas controls asynch page reclaim, and so should
2580 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 */
2582 int min_pages;
2583
2584 min_pages = zone->present_pages / 1024;
2585 if (min_pages < SWAP_CLUSTER_MAX)
2586 min_pages = SWAP_CLUSTER_MAX;
2587 if (min_pages > 128)
2588 min_pages = 128;
2589 zone->pages_min = min_pages;
2590 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08002591 /*
2592 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 * proportionate to the zone's size.
2594 */
Nick Piggin669ed172005-11-13 16:06:45 -08002595 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 }
2597
Nick Piggin669ed172005-11-13 16:06:45 -08002598 zone->pages_low = zone->pages_min + tmp / 4;
2599 zone->pages_high = zone->pages_min + tmp / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 spin_unlock_irqrestore(&zone->lru_lock, flags);
2601 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002602
2603 /* update totalreserve_pages */
2604 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605}
2606
2607/*
2608 * Initialise min_free_kbytes.
2609 *
2610 * For small machines we want it small (128k min). For large machines
2611 * we want it large (64MB max). But it is not linear, because network
2612 * bandwidth does not increase linearly with machine size. We use
2613 *
2614 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2615 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2616 *
2617 * which yields
2618 *
2619 * 16MB: 512k
2620 * 32MB: 724k
2621 * 64MB: 1024k
2622 * 128MB: 1448k
2623 * 256MB: 2048k
2624 * 512MB: 2896k
2625 * 1024MB: 4096k
2626 * 2048MB: 5792k
2627 * 4096MB: 8192k
2628 * 8192MB: 11584k
2629 * 16384MB: 16384k
2630 */
2631static int __init init_per_zone_pages_min(void)
2632{
2633 unsigned long lowmem_kbytes;
2634
2635 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2636
2637 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2638 if (min_free_kbytes < 128)
2639 min_free_kbytes = 128;
2640 if (min_free_kbytes > 65536)
2641 min_free_kbytes = 65536;
2642 setup_per_zone_pages_min();
2643 setup_per_zone_lowmem_reserve();
2644 return 0;
2645}
2646module_init(init_per_zone_pages_min)
2647
2648/*
2649 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2650 * that we can call two helper functions whenever min_free_kbytes
2651 * changes.
2652 */
2653int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2654 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2655{
2656 proc_dointvec(table, write, file, buffer, length, ppos);
2657 setup_per_zone_pages_min();
2658 return 0;
2659}
2660
2661/*
2662 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2663 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2664 * whenever sysctl_lowmem_reserve_ratio changes.
2665 *
2666 * The reserve ratio obviously has absolutely no relation with the
2667 * pages_min watermarks. The lowmem reserve ratio can only make sense
2668 * if in function of the boot time zone sizes.
2669 */
2670int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2671 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2672{
2673 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2674 setup_per_zone_lowmem_reserve();
2675 return 0;
2676}
2677
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002678/*
2679 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2680 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2681 * can have before it gets flushed back to buddy allocator.
2682 */
2683
2684int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2685 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2686{
2687 struct zone *zone;
2688 unsigned int cpu;
2689 int ret;
2690
2691 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2692 if (!write || (ret == -EINVAL))
2693 return ret;
2694 for_each_zone(zone) {
2695 for_each_online_cpu(cpu) {
2696 unsigned long high;
2697 high = zone->present_pages / percpu_pagelist_fraction;
2698 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2699 }
2700 }
2701 return 0;
2702}
2703
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704__initdata int hashdist = HASHDIST_DEFAULT;
2705
2706#ifdef CONFIG_NUMA
2707static int __init set_hashdist(char *str)
2708{
2709 if (!str)
2710 return 0;
2711 hashdist = simple_strtoul(str, &str, 0);
2712 return 1;
2713}
2714__setup("hashdist=", set_hashdist);
2715#endif
2716
2717/*
2718 * allocate a large system hash table from bootmem
2719 * - it is assumed that the hash table must contain an exact power-of-2
2720 * quantity of entries
2721 * - limit is the number of hash buckets, not the total allocation size
2722 */
2723void *__init alloc_large_system_hash(const char *tablename,
2724 unsigned long bucketsize,
2725 unsigned long numentries,
2726 int scale,
2727 int flags,
2728 unsigned int *_hash_shift,
2729 unsigned int *_hash_mask,
2730 unsigned long limit)
2731{
2732 unsigned long long max = limit;
2733 unsigned long log2qty, size;
2734 void *table = NULL;
2735
2736 /* allow the kernel cmdline to have a say */
2737 if (!numentries) {
2738 /* round applicable memory size up to nearest megabyte */
2739 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2740 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2741 numentries >>= 20 - PAGE_SHIFT;
2742 numentries <<= 20 - PAGE_SHIFT;
2743
2744 /* limit to 1 bucket per 2^scale bytes of low memory */
2745 if (scale > PAGE_SHIFT)
2746 numentries >>= (scale - PAGE_SHIFT);
2747 else
2748 numentries <<= (PAGE_SHIFT - scale);
2749 }
John Hawkes6e692ed2006-03-25 03:08:02 -08002750 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751
2752 /* limit allocation size to 1/16 total memory by default */
2753 if (max == 0) {
2754 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2755 do_div(max, bucketsize);
2756 }
2757
2758 if (numentries > max)
2759 numentries = max;
2760
2761 log2qty = long_log2(numentries);
2762
2763 do {
2764 size = bucketsize << log2qty;
2765 if (flags & HASH_EARLY)
2766 table = alloc_bootmem(size);
2767 else if (hashdist)
2768 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2769 else {
2770 unsigned long order;
2771 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2772 ;
2773 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2774 }
2775 } while (!table && size > PAGE_SIZE && --log2qty);
2776
2777 if (!table)
2778 panic("Failed to allocate %s hash table\n", tablename);
2779
2780 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2781 tablename,
2782 (1U << log2qty),
2783 long_log2(size) - PAGE_SHIFT,
2784 size);
2785
2786 if (_hash_shift)
2787 *_hash_shift = log2qty;
2788 if (_hash_mask)
2789 *_hash_mask = (1 << log2qty) - 1;
2790
2791 return table;
2792}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002793
2794#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
2795/*
2796 * pfn <-> page translation. out-of-line version.
2797 * (see asm-generic/memory_model.h)
2798 */
2799#if defined(CONFIG_FLATMEM)
2800struct page *pfn_to_page(unsigned long pfn)
2801{
2802 return mem_map + (pfn - ARCH_PFN_OFFSET);
2803}
2804unsigned long page_to_pfn(struct page *page)
2805{
2806 return (page - mem_map) + ARCH_PFN_OFFSET;
2807}
2808#elif defined(CONFIG_DISCONTIGMEM)
2809struct page *pfn_to_page(unsigned long pfn)
2810{
2811 int nid = arch_pfn_to_nid(pfn);
2812 return NODE_DATA(nid)->node_mem_map + arch_local_page_offset(pfn,nid);
2813}
2814unsigned long page_to_pfn(struct page *page)
2815{
KAMEZAWA Hiroyukia0140c12006-03-27 01:15:55 -08002816 struct pglist_data *pgdat = NODE_DATA(page_to_nid(page));
2817 return (page - pgdat->node_mem_map) + pgdat->node_start_pfn;
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002818}
2819#elif defined(CONFIG_SPARSEMEM)
2820struct page *pfn_to_page(unsigned long pfn)
2821{
2822 return __section_mem_map_addr(__pfn_to_section(pfn)) + pfn;
2823}
2824
2825unsigned long page_to_pfn(struct page *page)
2826{
2827 long section_id = page_to_section(page);
2828 return page - __section_mem_map_addr(__nr_to_section(section_id));
2829}
2830#endif /* CONFIG_FLATMEM/DISCONTIGMME/SPARSEMEM */
2831EXPORT_SYMBOL(pfn_to_page);
2832EXPORT_SYMBOL(page_to_pfn);
2833#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */