blob: 253a450c400df06898de5d864ff2c8863c560043 [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>
Andrew Mortonac924c62006-05-15 09:43:59 -070042#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include "internal.h"
44
45/*
46 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
47 * initializer cleaner
48 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070049nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
Dean Nelson7223a932005-03-23 19:00:00 -070050EXPORT_SYMBOL(node_online_map);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070051nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
Dean Nelson7223a932005-03-23 19:00:00 -070052EXPORT_SYMBOL(node_possible_map);
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070053unsigned long totalram_pages __read_mostly;
54unsigned long totalhigh_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070055unsigned long totalreserve_pages __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056long nr_swap_pages;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080057int percpu_pagelist_fraction;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080059static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080060
Linus Torvalds1da177e2005-04-16 15:20:36 -070061/*
62 * results with 256, 32 in the lowmem_reserve sysctl:
63 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
64 * 1G machine -> (16M dma, 784M normal, 224M high)
65 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
66 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
67 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010068 *
69 * TBD: should special case ZONE_DMA32 machines here - in those we normally
70 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 */
Andi Kleena2f1b422005-11-05 17:25:53 +010072int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 256, 32 };
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
74EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76/*
77 * Used by page_zone() to look up the address of the struct zone whose
78 * id is encoded in the upper bits of page->flags
79 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070080struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081EXPORT_SYMBOL(zone_table);
82
Andi Kleena2f1b422005-11-05 17:25:53 +010083static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem" };
Linus Torvalds1da177e2005-04-16 15:20:36 -070084int min_free_kbytes = 1024;
85
86unsigned long __initdata nr_kernel_pages;
87unsigned long __initdata nr_all_pages;
88
Nick Piggin13e74442006-01-06 00:10:58 -080089#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -070090static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091{
Dave Hansenbdc8cb92005-10-29 18:16:53 -070092 int ret = 0;
93 unsigned seq;
94 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -070095
Dave Hansenbdc8cb92005-10-29 18:16:53 -070096 do {
97 seq = zone_span_seqbegin(zone);
98 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
99 ret = 1;
100 else if (pfn < zone->zone_start_pfn)
101 ret = 1;
102 } while (zone_span_seqretry(zone, seq));
103
104 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700105}
106
107static int page_is_consistent(struct zone *zone, struct page *page)
108{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109#ifdef CONFIG_HOLES_IN_ZONE
110 if (!pfn_valid(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700111 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112#endif
113 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700114 return 0;
115
116 return 1;
117}
118/*
119 * Temporary debugging check for pages not lying within a given zone.
120 */
121static int bad_range(struct zone *zone, struct page *page)
122{
123 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700125 if (!page_is_consistent(zone, page))
126 return 1;
127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 return 0;
129}
130
Nick Piggin13e74442006-01-06 00:10:58 -0800131#else
132static inline int bad_range(struct zone *zone, struct page *page)
133{
134 return 0;
135}
136#endif
137
Nick Piggin224abf92006-01-06 00:11:11 -0800138static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139{
Nick Piggin224abf92006-01-06 00:11:11 -0800140 printk(KERN_EMERG "Bad page state in process '%s'\n"
Hugh Dickins7365f3d2006-01-11 12:17:18 -0800141 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
142 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
143 KERN_EMERG "Backtrace:\n",
Nick Piggin224abf92006-01-06 00:11:11 -0800144 current->comm, page, (int)(2*sizeof(unsigned long)),
145 (unsigned long)page->flags, page->mapping,
146 page_mapcount(page), page_count(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 dump_stack();
Hugh Dickins334795e2005-06-21 17:15:08 -0700148 page->flags &= ~(1 << PG_lru |
149 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 1 << PG_active |
152 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700153 1 << PG_reclaim |
154 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 1 << PG_swapcache |
Nick Piggin676165a2006-04-10 11:21:48 +1000156 1 << PG_writeback |
157 1 << PG_buddy );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 set_page_count(page, 0);
159 reset_page_mapcount(page);
160 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700161 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162}
163
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164/*
165 * Higher-order pages are called "compound pages". They are structured thusly:
166 *
167 * The first PAGE_SIZE page is called the "head page".
168 *
169 * The remaining PAGE_SIZE pages are called "tail pages".
170 *
171 * All pages have PG_compound set. All pages have their ->private pointing at
172 * the head page (even the head page has this).
173 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800174 * The first tail page's ->lru.next holds the address of the compound page's
175 * put_page() function. Its ->lru.prev holds the order of allocation.
176 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800178
179static void free_compound_page(struct page *page)
180{
181 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
182}
183
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184static void prep_compound_page(struct page *page, unsigned long order)
185{
186 int i;
187 int nr_pages = 1 << order;
188
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800189 page[1].lru.next = (void *)free_compound_page; /* set dtor */
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800190 page[1].lru.prev = (void *)order;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 for (i = 0; i < nr_pages; i++) {
192 struct page *p = page + i;
193
Nick Piggin5e9dace2006-03-22 00:08:01 -0800194 __SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700195 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 }
197}
198
199static void destroy_compound_page(struct page *page, unsigned long order)
200{
201 int i;
202 int nr_pages = 1 << order;
203
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800204 if (unlikely((unsigned long)page[1].lru.prev != order))
Nick Piggin224abf92006-01-06 00:11:11 -0800205 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
207 for (i = 0; i < nr_pages; i++) {
208 struct page *p = page + i;
209
Nick Piggin224abf92006-01-06 00:11:11 -0800210 if (unlikely(!PageCompound(p) |
211 (page_private(p) != (unsigned long)page)))
212 bad_page(page);
Nick Piggin5e9dace2006-03-22 00:08:01 -0800213 __ClearPageCompound(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 }
215}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
Nick Piggin17cf4402006-03-22 00:08:41 -0800217static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
218{
219 int i;
220
221 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
Andrew Morton6626c5d2006-03-22 00:08:42 -0800222 /*
223 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
224 * and __GFP_HIGHMEM from hard or soft interrupt context.
225 */
226 BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800227 for (i = 0; i < (1 << order); i++)
228 clear_highpage(page + i);
229}
230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231/*
232 * function for dealing with page's order in buddy system.
233 * zone->lock is already acquired when we use these.
234 * So, we don't need atomic page->flags operations here.
235 */
Andrew Morton6aa30012006-04-18 22:20:52 -0700236static inline unsigned long page_order(struct page *page)
237{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700238 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239}
240
Andrew Morton6aa30012006-04-18 22:20:52 -0700241static inline void set_page_order(struct page *page, int order)
242{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700243 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000244 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245}
246
247static inline void rmv_page_order(struct page *page)
248{
Nick Piggin676165a2006-04-10 11:21:48 +1000249 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700250 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251}
252
253/*
254 * Locate the struct page for both the matching buddy in our
255 * pair (buddy1) and the combined O(n+1) page they form (page).
256 *
257 * 1) Any buddy B1 will have an order O twin B2 which satisfies
258 * the following equation:
259 * B2 = B1 ^ (1 << O)
260 * For example, if the starting buddy (buddy2) is #8 its order
261 * 1 buddy is #10:
262 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
263 *
264 * 2) Any buddy B will have an order O+1 parent P which
265 * satisfies the following equation:
266 * P = B & ~(1 << O)
267 *
268 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
269 */
270static inline struct page *
271__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
272{
273 unsigned long buddy_idx = page_idx ^ (1 << order);
274
275 return page + (buddy_idx - page_idx);
276}
277
278static inline unsigned long
279__find_combined_index(unsigned long page_idx, unsigned int order)
280{
281 return (page_idx & ~(1 << order));
282}
283
284/*
285 * This function checks whether a page is free && is the buddy
286 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800287 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000288 * (b) the buddy is in the buddy system &&
289 * (c) a page and its buddy have the same order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 *
Nick Piggin676165a2006-04-10 11:21:48 +1000291 * For recording whether a page is in the buddy system, we use PG_buddy.
292 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
293 *
294 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 */
296static inline int page_is_buddy(struct page *page, int order)
297{
Nick Piggin13e74442006-01-06 00:10:58 -0800298#ifdef CONFIG_HOLES_IN_ZONE
299 if (!pfn_valid(page_to_pfn(page)))
300 return 0;
301#endif
302
Nick Piggin676165a2006-04-10 11:21:48 +1000303 if (PageBuddy(page) && page_order(page) == order) {
304 BUG_ON(page_count(page) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700305 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000306 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700307 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308}
309
310/*
311 * Freeing function for a buddy system allocator.
312 *
313 * The concept of a buddy system is to maintain direct-mapped table
314 * (containing bit values) for memory blocks of various "orders".
315 * The bottom level table contains the map for the smallest allocatable
316 * units of memory (here, pages), and each level above it describes
317 * pairs of units from the levels below, hence, "buddies".
318 * At a high level, all that happens here is marking the table entry
319 * at the bottom level available, and propagating the changes upward
320 * as necessary, plus some accounting needed to play nicely with other
321 * parts of the VM system.
322 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000323 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700324 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 * So when we are allocating or freeing one, we can derive the state of the
326 * other. That is, if we allocate a small block, and both were
327 * free, the remainder of the region must be split into blocks.
328 * If a block is freed, and its buddy is also free, then this
329 * triggers coalescing into a block of larger size.
330 *
331 * -- wli
332 */
333
Nick Piggin48db57f2006-01-08 01:00:42 -0800334static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 struct zone *zone, unsigned int order)
336{
337 unsigned long page_idx;
338 int order_size = 1 << order;
339
Nick Piggin224abf92006-01-06 00:11:11 -0800340 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 destroy_compound_page(page, order);
342
343 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
344
345 BUG_ON(page_idx & (order_size - 1));
346 BUG_ON(bad_range(zone, page));
347
348 zone->free_pages += order_size;
349 while (order < MAX_ORDER-1) {
350 unsigned long combined_idx;
351 struct free_area *area;
352 struct page *buddy;
353
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 buddy = __page_find_buddy(page, page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 if (!page_is_buddy(buddy, order))
356 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800357
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 list_del(&buddy->lru);
359 area = zone->free_area + order;
360 area->nr_free--;
361 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800362 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 page = page + (combined_idx - page_idx);
364 page_idx = combined_idx;
365 order++;
366 }
367 set_page_order(page, order);
368 list_add(&page->lru, &zone->free_area[order].free_list);
369 zone->free_area[order].nr_free++;
370}
371
Nick Piggin224abf92006-01-06 00:11:11 -0800372static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373{
Nick Piggin92be2e32006-01-06 00:10:57 -0800374 if (unlikely(page_mapcount(page) |
375 (page->mapping != NULL) |
376 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 (page->flags & (
378 1 << PG_lru |
379 1 << PG_private |
380 1 << PG_locked |
381 1 << PG_active |
382 1 << PG_reclaim |
383 1 << PG_slab |
384 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700385 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000386 1 << PG_reserved |
387 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800388 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700390 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800391 /*
392 * For now, we report if PG_reserved was found set, but do not
393 * clear it, and do not free the page. But we shall soon need
394 * to do more, for when the ZERO_PAGE count wraps negative.
395 */
396 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397}
398
399/*
400 * Frees a list of pages.
401 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700402 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 *
404 * If the zone was previously in an "all pages pinned" state then look to
405 * see if this freeing clears that state.
406 *
407 * And clear the zone's pages_scanned counter, to hold off the "all pages are
408 * pinned" detection logic.
409 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800410static void free_pages_bulk(struct zone *zone, int count,
411 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412{
Nick Pigginc54ad302006-01-06 00:10:56 -0800413 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 zone->all_unreclaimable = 0;
415 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800416 while (count--) {
417 struct page *page;
418
419 BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800421 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800423 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800425 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426}
427
Nick Piggin48db57f2006-01-08 01:00:42 -0800428static void free_one_page(struct zone *zone, struct page *page, int order)
429{
430 LIST_HEAD(list);
431 list_add(&page->lru, &list);
432 free_pages_bulk(zone, 1, &list, order);
433}
434
435static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436{
Nick Pigginc54ad302006-01-06 00:10:56 -0800437 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800439 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
441 arch_free_page(page, order);
Ingo Molnarde5097c2006-01-09 15:59:21 -0800442 if (!PageHighMem(page))
443 mutex_debug_check_no_locks_freed(page_address(page),
David Woodhousea4fc7ab2006-01-11 14:41:26 +0000444 PAGE_SIZE<<order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800447 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800448 if (reserved)
449 return;
450
Nick Piggin48db57f2006-01-08 01:00:42 -0800451 kernel_map_pages(page, 1 << order, 0);
Nick Pigginc54ad302006-01-06 00:10:56 -0800452 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800453 __mod_page_state(pgfree, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800454 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800455 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456}
457
David Howellsa226f6c2006-01-06 00:11:08 -0800458/*
459 * permit the bootmem allocator to evade page validation on high-order frees
460 */
461void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
462{
463 if (order == 0) {
464 __ClearPageReserved(page);
465 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800466 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800467 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800468 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800469 int loop;
470
Nick Piggin545b1ea2006-03-22 00:08:07 -0800471 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800472 for (loop = 0; loop < BITS_PER_LONG; loop++) {
473 struct page *p = &page[loop];
474
Nick Piggin545b1ea2006-03-22 00:08:07 -0800475 if (loop + 1 < BITS_PER_LONG)
476 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800477 __ClearPageReserved(p);
478 set_page_count(p, 0);
479 }
480
Nick Piggin7835e982006-03-22 00:08:40 -0800481 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800482 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800483 }
484}
485
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
487/*
488 * The order of subdivision here is critical for the IO subsystem.
489 * Please do not alter this order without good reasons and regression
490 * testing. Specifically, as large blocks of memory are subdivided,
491 * the order in which smaller blocks are delivered depends on the order
492 * they're subdivided in this function. This is the primary factor
493 * influencing the order in which pages are delivered to the IO
494 * subsystem according to empirical testing, and this is also justified
495 * by considering the behavior of a buddy system containing a single
496 * large block of memory acted on by a series of small allocations.
497 * This behavior is a critical factor in sglist merging's success.
498 *
499 * -- wli
500 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800501static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 int low, int high, struct free_area *area)
503{
504 unsigned long size = 1 << high;
505
506 while (high > low) {
507 area--;
508 high--;
509 size >>= 1;
510 BUG_ON(bad_range(zone, &page[size]));
511 list_add(&page[size].lru, &area->free_list);
512 area->nr_free++;
513 set_page_order(&page[size], high);
514 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515}
516
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517/*
518 * This page is about to be returned from the page allocator
519 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800520static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521{
Nick Piggin92be2e32006-01-06 00:10:57 -0800522 if (unlikely(page_mapcount(page) |
523 (page->mapping != NULL) |
524 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700525 (page->flags & (
526 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 1 << PG_private |
528 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 1 << PG_active |
530 1 << PG_dirty |
531 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700532 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700534 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000535 1 << PG_reserved |
536 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800537 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538
Hugh Dickins689bceb2005-11-21 21:32:20 -0800539 /*
540 * For now, we report if PG_reserved was found set, but do not
541 * clear it, and do not allocate the page: as a safety net.
542 */
543 if (PageReserved(page))
544 return 1;
545
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
547 1 << PG_referenced | 1 << PG_arch_1 |
548 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700549 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800550 set_page_refcounted(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800552
553 if (gfp_flags & __GFP_ZERO)
554 prep_zero_page(page, order, gfp_flags);
555
556 if (order && (gfp_flags & __GFP_COMP))
557 prep_compound_page(page, order);
558
Hugh Dickins689bceb2005-11-21 21:32:20 -0800559 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560}
561
562/*
563 * Do the hard work of removing an element from the buddy allocator.
564 * Call me with the zone->lock already held.
565 */
566static struct page *__rmqueue(struct zone *zone, unsigned int order)
567{
568 struct free_area * area;
569 unsigned int current_order;
570 struct page *page;
571
572 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
573 area = zone->free_area + current_order;
574 if (list_empty(&area->free_list))
575 continue;
576
577 page = list_entry(area->free_list.next, struct page, lru);
578 list_del(&page->lru);
579 rmv_page_order(page);
580 area->nr_free--;
581 zone->free_pages -= 1UL << order;
Nick Piggin085cc7d2006-01-06 00:11:01 -0800582 expand(zone, page, order, current_order, area);
583 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 }
585
586 return NULL;
587}
588
589/*
590 * Obtain a specified number of elements from the buddy allocator, all under
591 * a single hold of the lock, for efficiency. Add them to the supplied list.
592 * Returns the number of new pages which were placed at *list.
593 */
594static int rmqueue_bulk(struct zone *zone, unsigned int order,
595 unsigned long count, struct list_head *list)
596{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Nick Pigginc54ad302006-01-06 00:10:56 -0800599 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 for (i = 0; i < count; ++i) {
Nick Piggin085cc7d2006-01-06 00:11:01 -0800601 struct page *page = __rmqueue(zone, order);
602 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 list_add_tail(&page->lru, list);
605 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800606 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800607 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608}
609
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700610#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800611/*
612 * Called from the slab reaper to drain pagesets on a particular node that
613 * belong to the currently executing processor.
Christoph Lameter879336c2006-03-22 00:09:08 -0800614 * Note that this function must be called with the thread pinned to
615 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800616 */
617void drain_node_pages(int nodeid)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700618{
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800619 int i, z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700620 unsigned long flags;
621
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800622 for (z = 0; z < MAX_NR_ZONES; z++) {
623 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700624 struct per_cpu_pageset *pset;
625
Nick Piggin23316bc2006-01-08 01:00:41 -0800626 pset = zone_pcp(zone, smp_processor_id());
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700627 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
628 struct per_cpu_pages *pcp;
629
630 pcp = &pset->pcp[i];
Christoph Lameter879336c2006-03-22 00:09:08 -0800631 if (pcp->count) {
632 local_irq_save(flags);
633 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
634 pcp->count = 0;
635 local_irq_restore(flags);
636 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700637 }
638 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700639}
640#endif
641
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
643static void __drain_pages(unsigned int cpu)
644{
Nick Pigginc54ad302006-01-06 00:10:56 -0800645 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 struct zone *zone;
647 int i;
648
649 for_each_zone(zone) {
650 struct per_cpu_pageset *pset;
651
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700652 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
654 struct per_cpu_pages *pcp;
655
656 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800657 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800658 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
659 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800660 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 }
662 }
663}
664#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
665
666#ifdef CONFIG_PM
667
668void mark_free_pages(struct zone *zone)
669{
670 unsigned long zone_pfn, flags;
671 int order;
672 struct list_head *curr;
673
674 if (!zone->spanned_pages)
675 return;
676
677 spin_lock_irqsave(&zone->lock, flags);
678 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
679 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
680
681 for (order = MAX_ORDER - 1; order >= 0; --order)
682 list_for_each(curr, &zone->free_area[order].free_list) {
683 unsigned long start_pfn, i;
684
685 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
686
687 for (i=0; i < (1<<order); i++)
688 SetPageNosaveFree(pfn_to_page(start_pfn+i));
689 }
690 spin_unlock_irqrestore(&zone->lock, flags);
691}
692
693/*
694 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
695 */
696void drain_local_pages(void)
697{
698 unsigned long flags;
699
700 local_irq_save(flags);
701 __drain_pages(smp_processor_id());
702 local_irq_restore(flags);
703}
704#endif /* CONFIG_PM */
705
Nick Piggina74609f2006-01-06 00:11:20 -0800706static void zone_statistics(struct zonelist *zonelist, struct zone *z, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707{
708#ifdef CONFIG_NUMA
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 pg_data_t *pg = z->zone_pgdat;
710 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
711 struct per_cpu_pageset *p;
712
Nick Piggina74609f2006-01-06 00:11:20 -0800713 p = zone_pcp(z, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 if (pg == orig) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700715 p->numa_hit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 } else {
717 p->numa_miss++;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700718 zone_pcp(zonelist->zones[0], cpu)->numa_foreign++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 }
720 if (pg == NODE_DATA(numa_node_id()))
721 p->local_node++;
722 else
723 p->other_node++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724#endif
725}
726
727/*
728 * Free a 0-order page
729 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730static void fastcall free_hot_cold_page(struct page *page, int cold)
731{
732 struct zone *zone = page_zone(page);
733 struct per_cpu_pages *pcp;
734 unsigned long flags;
735
736 arch_free_page(page, 0);
737
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 if (PageAnon(page))
739 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800740 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800741 return;
742
Hugh Dickins689bceb2005-11-21 21:32:20 -0800743 kernel_map_pages(page, 1, 0);
744
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700745 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800747 __inc_page_state(pgfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 list_add(&page->lru, &pcp->list);
749 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800750 if (pcp->count >= pcp->high) {
751 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
752 pcp->count -= pcp->batch;
753 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 local_irq_restore(flags);
755 put_cpu();
756}
757
758void fastcall free_hot_page(struct page *page)
759{
760 free_hot_cold_page(page, 0);
761}
762
763void fastcall free_cold_page(struct page *page)
764{
765 free_hot_cold_page(page, 1);
766}
767
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800768/*
769 * split_page takes a non-compound higher-order page, and splits it into
770 * n (1<<order) sub-pages: page[0..n]
771 * Each sub-page must be freed individually.
772 *
773 * Note: this is probably too low level an operation for use in drivers.
774 * Please consult with lkml before using this in your driver.
775 */
776void split_page(struct page *page, unsigned int order)
777{
778 int i;
779
780 BUG_ON(PageCompound(page));
781 BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -0800782 for (i = 1; i < (1 << order); i++)
783 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800784}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800785
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786/*
787 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
788 * we cheat by calling it from here, in the order > 0 path. Saves a branch
789 * or two.
790 */
Nick Piggina74609f2006-01-06 00:11:20 -0800791static struct page *buffered_rmqueue(struct zonelist *zonelist,
792 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793{
794 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800795 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800797 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
Hugh Dickins689bceb2005-11-21 21:32:20 -0800799again:
Nick Piggina74609f2006-01-06 00:11:20 -0800800 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800801 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 struct per_cpu_pages *pcp;
803
Nick Piggina74609f2006-01-06 00:11:20 -0800804 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800806 if (!pcp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 pcp->count += rmqueue_bulk(zone, 0,
808 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800809 if (unlikely(!pcp->count))
810 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 }
Nick Piggina74609f2006-01-06 00:11:20 -0800812 page = list_entry(pcp->list.next, struct page, lru);
813 list_del(&page->lru);
814 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800815 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 spin_lock_irqsave(&zone->lock, flags);
817 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800818 spin_unlock(&zone->lock);
819 if (!page)
820 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 }
822
Nick Piggina74609f2006-01-06 00:11:20 -0800823 __mod_page_state_zone(zone, pgalloc, 1 << order);
824 zone_statistics(zonelist, zone, cpu);
825 local_irq_restore(flags);
826 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
Nick Piggina74609f2006-01-06 00:11:20 -0800828 BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -0800829 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -0800830 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800832
833failed:
834 local_irq_restore(flags);
835 put_cpu();
836 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837}
838
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800839#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800840#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
841#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
842#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
843#define ALLOC_HARDER 0x10 /* try to alloc harder */
844#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
845#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800846
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847/*
848 * Return 1 if free pages are above 'mark'. This takes into account the order
849 * of the allocation.
850 */
851int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800852 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853{
854 /* free_pages my go negative - that's OK */
855 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
856 int o;
857
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800858 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800860 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 min -= min / 4;
862
863 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
864 return 0;
865 for (o = 0; o < order; o++) {
866 /* At the next order, this order's pages become unavailable */
867 free_pages -= z->free_area[o].nr_free << o;
868
869 /* Require fewer higher order pages to be free */
870 min >>= 1;
871
872 if (free_pages <= min)
873 return 0;
874 }
875 return 1;
876}
877
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800878/*
879 * get_page_from_freeliest goes through the zonelist trying to allocate
880 * a page.
881 */
882static struct page *
883get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
884 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -0700885{
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800886 struct zone **z = zonelist->zones;
887 struct page *page = NULL;
888 int classzone_idx = zone_idx(*z);
889
890 /*
891 * Go through the zonelist once, looking for a zone with enough free.
892 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
893 */
894 do {
895 if ((alloc_flags & ALLOC_CPUSET) &&
896 !cpuset_zone_allowed(*z, gfp_mask))
897 continue;
898
899 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -0800900 unsigned long mark;
901 if (alloc_flags & ALLOC_WMARK_MIN)
902 mark = (*z)->pages_min;
903 else if (alloc_flags & ALLOC_WMARK_LOW)
904 mark = (*z)->pages_low;
905 else
906 mark = (*z)->pages_high;
907 if (!zone_watermark_ok(*z, order, mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800908 classzone_idx, alloc_flags))
Christoph Lameter9eeff232006-01-18 17:42:31 -0800909 if (!zone_reclaim_mode ||
910 !zone_reclaim(*z, gfp_mask, order))
911 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800912 }
913
Nick Piggina74609f2006-01-06 00:11:20 -0800914 page = buffered_rmqueue(zonelist, *z, order, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800915 if (page) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800916 break;
917 }
918 } while (*(++z) != NULL);
919 return page;
Martin Hicks753ee722005-06-21 17:14:41 -0700920}
921
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922/*
923 * This is the 'heart' of the zoned buddy allocator.
924 */
925struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +0100926__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 struct zonelist *zonelist)
928{
Al Viro260b2362005-10-21 03:22:44 -0400929 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800930 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 struct page *page;
932 struct reclaim_state reclaim_state;
933 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800935 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 int did_some_progress;
937
938 might_sleep_if(wait);
939
Jens Axboe6b1de912005-11-17 21:35:02 +0100940restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800941 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800943 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 /* Should this ever happen?? */
945 return NULL;
946 }
Jens Axboe6b1de912005-11-17 21:35:02 +0100947
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800948 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -0800949 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800950 if (page)
951 goto got_pg;
952
Jens Axboe6b1de912005-11-17 21:35:02 +0100953 do {
Paul Jacksonbdd804f2006-05-20 15:00:09 -0700954 if (cpuset_zone_allowed(*z, gfp_mask|__GFP_HARDWALL))
Christoph Lameter0b1303f2006-03-24 03:15:59 -0800955 wakeup_kswapd(*z, order);
Jens Axboe6b1de912005-11-17 21:35:02 +0100956 } while (*(++z));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800957
Paul Jackson9bf22292005-09-06 15:18:12 -0700958 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800959 * OK, we're below the kswapd watermark and have kicked background
960 * reclaim. Now things get more complex, so set up alloc_flags according
961 * to how we want to proceed.
962 *
963 * The caller may dip into page reserves a bit more if the caller
964 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -0800965 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
966 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -0700967 */
Nick Piggin31488902005-11-28 13:44:03 -0800968 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800969 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
970 alloc_flags |= ALLOC_HARDER;
971 if (gfp_mask & __GFP_HIGH)
972 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -0700973 if (wait)
974 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
976 /*
977 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800978 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 *
980 * This is the last chance, in general, before the goto nopage.
981 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -0700982 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800984 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
985 if (page)
986 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
988 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -0700989
990 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
991 && !in_interrupt()) {
992 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -0800993nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -0700994 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800995 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -0800996 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800997 if (page)
998 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -0800999 if (gfp_mask & __GFP_NOFAIL) {
1000 blk_congestion_wait(WRITE, HZ/50);
1001 goto nofail_alloc;
1002 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 }
1004 goto nopage;
1005 }
1006
1007 /* Atomic allocations - we can't balance anything */
1008 if (!wait)
1009 goto nopage;
1010
1011rebalance:
1012 cond_resched();
1013
1014 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001015 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 p->flags |= PF_MEMALLOC;
1017 reclaim_state.reclaimed_slab = 0;
1018 p->reclaim_state = &reclaim_state;
1019
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001020 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
1022 p->reclaim_state = NULL;
1023 p->flags &= ~PF_MEMALLOC;
1024
1025 cond_resched();
1026
1027 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001028 page = get_page_from_freelist(gfp_mask, order,
1029 zonelist, alloc_flags);
1030 if (page)
1031 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1033 /*
1034 * Go through the zonelist yet one more time, keep
1035 * very high watermark here, this is only to catch
1036 * a parallel oom killing, we must fail if we're still
1037 * under heavy pressure.
1038 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001039 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001040 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001041 if (page)
1042 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001044 out_of_memory(zonelist, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 goto restart;
1046 }
1047
1048 /*
1049 * Don't let big-order allocations loop unless the caller explicitly
1050 * requests that. Wait for some write requests to complete then retry.
1051 *
1052 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1053 * <= 3, but that may not be true in other implementations.
1054 */
1055 do_retry = 0;
1056 if (!(gfp_mask & __GFP_NORETRY)) {
1057 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1058 do_retry = 1;
1059 if (gfp_mask & __GFP_NOFAIL)
1060 do_retry = 1;
1061 }
1062 if (do_retry) {
1063 blk_congestion_wait(WRITE, HZ/50);
1064 goto rebalance;
1065 }
1066
1067nopage:
1068 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1069 printk(KERN_WARNING "%s: page allocation failure."
1070 " order:%d, mode:0x%x\n",
1071 p->comm, order, gfp_mask);
1072 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001073 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 return page;
1077}
1078
1079EXPORT_SYMBOL(__alloc_pages);
1080
1081/*
1082 * Common helper functions.
1083 */
Al Virodd0fc662005-10-07 07:46:04 +01001084fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085{
1086 struct page * page;
1087 page = alloc_pages(gfp_mask, order);
1088 if (!page)
1089 return 0;
1090 return (unsigned long) page_address(page);
1091}
1092
1093EXPORT_SYMBOL(__get_free_pages);
1094
Al Virodd0fc662005-10-07 07:46:04 +01001095fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096{
1097 struct page * page;
1098
1099 /*
1100 * get_zeroed_page() returns a 32-bit address, which cannot represent
1101 * a highmem page
1102 */
Al Viro260b2362005-10-21 03:22:44 -04001103 BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104
1105 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1106 if (page)
1107 return (unsigned long) page_address(page);
1108 return 0;
1109}
1110
1111EXPORT_SYMBOL(get_zeroed_page);
1112
1113void __pagevec_free(struct pagevec *pvec)
1114{
1115 int i = pagevec_count(pvec);
1116
1117 while (--i >= 0)
1118 free_hot_cold_page(pvec->pages[i], pvec->cold);
1119}
1120
1121fastcall void __free_pages(struct page *page, unsigned int order)
1122{
Nick Pigginb5810032005-10-29 18:16:12 -07001123 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 if (order == 0)
1125 free_hot_page(page);
1126 else
1127 __free_pages_ok(page, order);
1128 }
1129}
1130
1131EXPORT_SYMBOL(__free_pages);
1132
1133fastcall void free_pages(unsigned long addr, unsigned int order)
1134{
1135 if (addr != 0) {
1136 BUG_ON(!virt_addr_valid((void *)addr));
1137 __free_pages(virt_to_page((void *)addr), order);
1138 }
1139}
1140
1141EXPORT_SYMBOL(free_pages);
1142
1143/*
1144 * Total amount of free (allocatable) RAM:
1145 */
1146unsigned int nr_free_pages(void)
1147{
1148 unsigned int sum = 0;
1149 struct zone *zone;
1150
1151 for_each_zone(zone)
1152 sum += zone->free_pages;
1153
1154 return sum;
1155}
1156
1157EXPORT_SYMBOL(nr_free_pages);
1158
1159#ifdef CONFIG_NUMA
1160unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1161{
1162 unsigned int i, sum = 0;
1163
1164 for (i = 0; i < MAX_NR_ZONES; i++)
1165 sum += pgdat->node_zones[i].free_pages;
1166
1167 return sum;
1168}
1169#endif
1170
1171static unsigned int nr_free_zone_pages(int offset)
1172{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001173 /* Just pick one node, since fallback list is circular */
1174 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 unsigned int sum = 0;
1176
Martin J. Blighe310fd42005-07-29 22:59:18 -07001177 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1178 struct zone **zonep = zonelist->zones;
1179 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
Martin J. Blighe310fd42005-07-29 22:59:18 -07001181 for (zone = *zonep++; zone; zone = *zonep++) {
1182 unsigned long size = zone->present_pages;
1183 unsigned long high = zone->pages_high;
1184 if (size > high)
1185 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 }
1187
1188 return sum;
1189}
1190
1191/*
1192 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1193 */
1194unsigned int nr_free_buffer_pages(void)
1195{
Al Viroaf4ca452005-10-21 02:55:38 -04001196 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197}
1198
1199/*
1200 * Amount of free RAM allocatable within all zones
1201 */
1202unsigned int nr_free_pagecache_pages(void)
1203{
Al Viroaf4ca452005-10-21 02:55:38 -04001204 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205}
1206
1207#ifdef CONFIG_HIGHMEM
1208unsigned int nr_free_highpages (void)
1209{
1210 pg_data_t *pgdat;
1211 unsigned int pages = 0;
1212
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08001213 for_each_online_pgdat(pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1215
1216 return pages;
1217}
1218#endif
1219
1220#ifdef CONFIG_NUMA
1221static void show_node(struct zone *zone)
1222{
1223 printk("Node %d ", zone->zone_pgdat->node_id);
1224}
1225#else
1226#define show_node(zone) do { } while (0)
1227#endif
1228
1229/*
1230 * Accumulate the page_state information across all CPUs.
1231 * The result is unavoidably approximate - it can change
1232 * during and after execution of this function.
1233 */
1234static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1235
1236atomic_t nr_pagecache = ATOMIC_INIT(0);
1237EXPORT_SYMBOL(nr_pagecache);
1238#ifdef CONFIG_SMP
1239DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1240#endif
1241
Nick Piggina86b1f52006-01-06 00:11:00 -08001242static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243{
Andrew Mortonb40607f2006-03-22 00:07:39 -08001244 unsigned cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001246 memset(ret, 0, nr * sizeof(unsigned long));
Andrew Morton84c20082006-01-08 01:00:28 -08001247 cpus_and(*cpumask, *cpumask, cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
Andrew Mortonb40607f2006-03-22 00:07:39 -08001249 for_each_cpu_mask(cpu, *cpumask) {
1250 unsigned long *in;
1251 unsigned long *out;
1252 unsigned off;
1253 unsigned next_cpu;
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001254
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 in = (unsigned long *)&per_cpu(page_states, cpu);
1256
Andrew Mortonb40607f2006-03-22 00:07:39 -08001257 next_cpu = next_cpu(cpu, *cpumask);
1258 if (likely(next_cpu < NR_CPUS))
1259 prefetch(&per_cpu(page_states, next_cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
1261 out = (unsigned long *)ret;
1262 for (off = 0; off < nr; off++)
1263 *out++ += *in++;
1264 }
1265}
1266
Martin Hicksc07e02d2005-09-03 15:55:11 -07001267void get_page_state_node(struct page_state *ret, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268{
1269 int nr;
Martin Hicksc07e02d2005-09-03 15:55:11 -07001270 cpumask_t mask = node_to_cpumask(node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
1272 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1273 nr /= sizeof(unsigned long);
1274
Martin Hicksc07e02d2005-09-03 15:55:11 -07001275 __get_page_state(ret, nr+1, &mask);
1276}
1277
1278void get_page_state(struct page_state *ret)
1279{
1280 int nr;
1281 cpumask_t mask = CPU_MASK_ALL;
1282
1283 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1284 nr /= sizeof(unsigned long);
1285
1286 __get_page_state(ret, nr + 1, &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287}
1288
1289void get_full_page_state(struct page_state *ret)
1290{
Martin Hicksc07e02d2005-09-03 15:55:11 -07001291 cpumask_t mask = CPU_MASK_ALL;
1292
1293 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294}
1295
Nick Piggina74609f2006-01-06 00:11:20 -08001296unsigned long read_page_state_offset(unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297{
1298 unsigned long ret = 0;
1299 int cpu;
1300
Andrew Morton84c20082006-01-08 01:00:28 -08001301 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 unsigned long in;
1303
1304 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1305 ret += *((unsigned long *)in);
1306 }
1307 return ret;
1308}
1309
Nick Piggina74609f2006-01-06 00:11:20 -08001310void __mod_page_state_offset(unsigned long offset, unsigned long delta)
1311{
1312 void *ptr;
1313
1314 ptr = &__get_cpu_var(page_states);
1315 *(unsigned long *)(ptr + offset) += delta;
1316}
1317EXPORT_SYMBOL(__mod_page_state_offset);
1318
1319void mod_page_state_offset(unsigned long offset, unsigned long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320{
1321 unsigned long flags;
Nick Piggina74609f2006-01-06 00:11:20 -08001322 void *ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
1324 local_irq_save(flags);
1325 ptr = &__get_cpu_var(page_states);
Nick Piggina74609f2006-01-06 00:11:20 -08001326 *(unsigned long *)(ptr + offset) += delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 local_irq_restore(flags);
1328}
Nick Piggina74609f2006-01-06 00:11:20 -08001329EXPORT_SYMBOL(mod_page_state_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
1331void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1332 unsigned long *free, struct pglist_data *pgdat)
1333{
1334 struct zone *zones = pgdat->node_zones;
1335 int i;
1336
1337 *active = 0;
1338 *inactive = 0;
1339 *free = 0;
1340 for (i = 0; i < MAX_NR_ZONES; i++) {
1341 *active += zones[i].nr_active;
1342 *inactive += zones[i].nr_inactive;
1343 *free += zones[i].free_pages;
1344 }
1345}
1346
1347void get_zone_counts(unsigned long *active,
1348 unsigned long *inactive, unsigned long *free)
1349{
1350 struct pglist_data *pgdat;
1351
1352 *active = 0;
1353 *inactive = 0;
1354 *free = 0;
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08001355 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 unsigned long l, m, n;
1357 __get_zone_counts(&l, &m, &n, pgdat);
1358 *active += l;
1359 *inactive += m;
1360 *free += n;
1361 }
1362}
1363
1364void si_meminfo(struct sysinfo *val)
1365{
1366 val->totalram = totalram_pages;
1367 val->sharedram = 0;
1368 val->freeram = nr_free_pages();
1369 val->bufferram = nr_blockdev_pages();
1370#ifdef CONFIG_HIGHMEM
1371 val->totalhigh = totalhigh_pages;
1372 val->freehigh = nr_free_highpages();
1373#else
1374 val->totalhigh = 0;
1375 val->freehigh = 0;
1376#endif
1377 val->mem_unit = PAGE_SIZE;
1378}
1379
1380EXPORT_SYMBOL(si_meminfo);
1381
1382#ifdef CONFIG_NUMA
1383void si_meminfo_node(struct sysinfo *val, int nid)
1384{
1385 pg_data_t *pgdat = NODE_DATA(nid);
1386
1387 val->totalram = pgdat->node_present_pages;
1388 val->freeram = nr_free_pages_pgdat(pgdat);
1389 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1390 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1391 val->mem_unit = PAGE_SIZE;
1392}
1393#endif
1394
1395#define K(x) ((x) << (PAGE_SHIFT-10))
1396
1397/*
1398 * Show free area list (used inside shift_scroll-lock stuff)
1399 * We also calculate the percentage fragmentation. We do this by counting the
1400 * memory on each free list with the exception of the first item on the list.
1401 */
1402void show_free_areas(void)
1403{
1404 struct page_state ps;
1405 int cpu, temperature;
1406 unsigned long active;
1407 unsigned long inactive;
1408 unsigned long free;
1409 struct zone *zone;
1410
1411 for_each_zone(zone) {
1412 show_node(zone);
1413 printk("%s per-cpu:", zone->name);
1414
Con Kolivasf3fe6512006-01-06 00:11:15 -08001415 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 printk(" empty\n");
1417 continue;
1418 } else
1419 printk("\n");
1420
Dave Jones6b482c62005-11-10 15:45:56 -05001421 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 struct per_cpu_pageset *pageset;
1423
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001424 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425
1426 for (temperature = 0; temperature < 2; temperature++)
Nick Piggin2d92c5c2006-01-06 00:10:59 -08001427 printk("cpu %d %s: high %d, batch %d used:%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 cpu,
1429 temperature ? "cold" : "hot",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 pageset->pcp[temperature].high,
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001431 pageset->pcp[temperature].batch,
1432 pageset->pcp[temperature].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 }
1434 }
1435
1436 get_page_state(&ps);
1437 get_zone_counts(&active, &inactive, &free);
1438
Denis Vlasenkoc0d62212005-06-21 17:15:14 -07001439 printk("Free pages: %11ukB (%ukB HighMem)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 K(nr_free_pages()),
1441 K(nr_free_highpages()));
1442
1443 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1444 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1445 active,
1446 inactive,
1447 ps.nr_dirty,
1448 ps.nr_writeback,
1449 ps.nr_unstable,
1450 nr_free_pages(),
1451 ps.nr_slab,
1452 ps.nr_mapped,
1453 ps.nr_page_table_pages);
1454
1455 for_each_zone(zone) {
1456 int i;
1457
1458 show_node(zone);
1459 printk("%s"
1460 " free:%lukB"
1461 " min:%lukB"
1462 " low:%lukB"
1463 " high:%lukB"
1464 " active:%lukB"
1465 " inactive:%lukB"
1466 " present:%lukB"
1467 " pages_scanned:%lu"
1468 " all_unreclaimable? %s"
1469 "\n",
1470 zone->name,
1471 K(zone->free_pages),
1472 K(zone->pages_min),
1473 K(zone->pages_low),
1474 K(zone->pages_high),
1475 K(zone->nr_active),
1476 K(zone->nr_inactive),
1477 K(zone->present_pages),
1478 zone->pages_scanned,
1479 (zone->all_unreclaimable ? "yes" : "no")
1480 );
1481 printk("lowmem_reserve[]:");
1482 for (i = 0; i < MAX_NR_ZONES; i++)
1483 printk(" %lu", zone->lowmem_reserve[i]);
1484 printk("\n");
1485 }
1486
1487 for_each_zone(zone) {
1488 unsigned long nr, flags, order, total = 0;
1489
1490 show_node(zone);
1491 printk("%s: ", zone->name);
Con Kolivasf3fe6512006-01-06 00:11:15 -08001492 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 printk("empty\n");
1494 continue;
1495 }
1496
1497 spin_lock_irqsave(&zone->lock, flags);
1498 for (order = 0; order < MAX_ORDER; order++) {
1499 nr = zone->free_area[order].nr_free;
1500 total += nr << order;
1501 printk("%lu*%lukB ", nr, K(1UL) << order);
1502 }
1503 spin_unlock_irqrestore(&zone->lock, flags);
1504 printk("= %lukB\n", K(total));
1505 }
1506
1507 show_swap_cache_info();
1508}
1509
1510/*
1511 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001512 *
1513 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 */
Christoph Lameter1a932052006-01-06 00:11:16 -08001515static int __init build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter070f8032006-01-06 00:11:19 -08001516 struct zonelist *zonelist, int nr_zones, int zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517{
Christoph Lameter1a932052006-01-06 00:11:16 -08001518 struct zone *zone;
1519
Christoph Lameter070f8032006-01-06 00:11:19 -08001520 BUG_ON(zone_type > ZONE_HIGHMEM);
Christoph Lameter02a68a52006-01-06 00:11:18 -08001521
1522 do {
Christoph Lameter070f8032006-01-06 00:11:19 -08001523 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001524 if (populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525#ifndef CONFIG_HIGHMEM
Christoph Lameter070f8032006-01-06 00:11:19 -08001526 BUG_ON(zone_type > ZONE_NORMAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527#endif
Christoph Lameter070f8032006-01-06 00:11:19 -08001528 zonelist->zones[nr_zones++] = zone;
1529 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 }
Christoph Lameter070f8032006-01-06 00:11:19 -08001531 zone_type--;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001532
Christoph Lameter070f8032006-01-06 00:11:19 -08001533 } while (zone_type >= 0);
1534 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535}
1536
Al Viro260b2362005-10-21 03:22:44 -04001537static inline int highest_zone(int zone_bits)
1538{
1539 int res = ZONE_NORMAL;
1540 if (zone_bits & (__force int)__GFP_HIGHMEM)
1541 res = ZONE_HIGHMEM;
Andi Kleena2f1b422005-11-05 17:25:53 +01001542 if (zone_bits & (__force int)__GFP_DMA32)
1543 res = ZONE_DMA32;
Al Viro260b2362005-10-21 03:22:44 -04001544 if (zone_bits & (__force int)__GFP_DMA)
1545 res = ZONE_DMA;
1546 return res;
1547}
1548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549#ifdef CONFIG_NUMA
1550#define MAX_NODE_LOAD (num_online_nodes())
1551static int __initdata node_load[MAX_NUMNODES];
1552/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001553 * 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 -07001554 * @node: node whose fallback list we're appending
1555 * @used_node_mask: nodemask_t of already used nodes
1556 *
1557 * We use a number of factors to determine which is the next node that should
1558 * appear on a given node's fallback list. The node should not have appeared
1559 * already in @node's fallback list, and it should be the next closest node
1560 * according to the distance array (which contains arbitrary distance values
1561 * from each node to each node in the system), and should also prefer nodes
1562 * with no CPUs, since presumably they'll have very little allocation pressure
1563 * on them otherwise.
1564 * It returns -1 if no node is found.
1565 */
1566static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1567{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001568 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 int min_val = INT_MAX;
1570 int best_node = -1;
1571
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001572 /* Use the local node if we haven't already */
1573 if (!node_isset(node, *used_node_mask)) {
1574 node_set(node, *used_node_mask);
1575 return node;
1576 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001578 for_each_online_node(n) {
1579 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580
1581 /* Don't want a node to appear more than once */
1582 if (node_isset(n, *used_node_mask))
1583 continue;
1584
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 /* Use the distance array to find the distance */
1586 val = node_distance(node, n);
1587
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001588 /* Penalize nodes under us ("prefer the next node") */
1589 val += (n < node);
1590
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 /* Give preference to headless and unused nodes */
1592 tmp = node_to_cpumask(n);
1593 if (!cpus_empty(tmp))
1594 val += PENALTY_FOR_NODE_WITH_CPUS;
1595
1596 /* Slight preference for less loaded node */
1597 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1598 val += node_load[n];
1599
1600 if (val < min_val) {
1601 min_val = val;
1602 best_node = n;
1603 }
1604 }
1605
1606 if (best_node >= 0)
1607 node_set(best_node, *used_node_mask);
1608
1609 return best_node;
1610}
1611
1612static void __init build_zonelists(pg_data_t *pgdat)
1613{
1614 int i, j, k, node, local_node;
1615 int prev_node, load;
1616 struct zonelist *zonelist;
1617 nodemask_t used_mask;
1618
1619 /* initialize zonelists */
1620 for (i = 0; i < GFP_ZONETYPES; i++) {
1621 zonelist = pgdat->node_zonelists + i;
1622 zonelist->zones[0] = NULL;
1623 }
1624
1625 /* NUMA-aware ordering of nodes */
1626 local_node = pgdat->node_id;
1627 load = num_online_nodes();
1628 prev_node = local_node;
1629 nodes_clear(used_mask);
1630 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001631 int distance = node_distance(local_node, node);
1632
1633 /*
1634 * If another node is sufficiently far away then it is better
1635 * to reclaim pages in a zone before going off node.
1636 */
1637 if (distance > RECLAIM_DISTANCE)
1638 zone_reclaim_mode = 1;
1639
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 /*
1641 * We don't want to pressure a particular node.
1642 * So adding penalty to the first node in same
1643 * distance group to make it round-robin.
1644 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08001645
1646 if (distance != node_distance(local_node, prev_node))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 node_load[node] += load;
1648 prev_node = node;
1649 load--;
1650 for (i = 0; i < GFP_ZONETYPES; i++) {
1651 zonelist = pgdat->node_zonelists + i;
1652 for (j = 0; zonelist->zones[j] != NULL; j++);
1653
Al Viro260b2362005-10-21 03:22:44 -04001654 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
1656 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1657 zonelist->zones[j] = NULL;
1658 }
1659 }
1660}
1661
1662#else /* CONFIG_NUMA */
1663
1664static void __init build_zonelists(pg_data_t *pgdat)
1665{
1666 int i, j, k, node, local_node;
1667
1668 local_node = pgdat->node_id;
1669 for (i = 0; i < GFP_ZONETYPES; i++) {
1670 struct zonelist *zonelist;
1671
1672 zonelist = pgdat->node_zonelists + i;
1673
1674 j = 0;
Al Viro260b2362005-10-21 03:22:44 -04001675 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 j = build_zonelists_node(pgdat, zonelist, j, k);
1677 /*
1678 * Now we build the zonelist so that it contains the zones
1679 * of all the other nodes.
1680 * We don't want to pressure a particular node, so when
1681 * building the zones for node N, we make sure that the
1682 * zones coming right after the local ones are those from
1683 * node N+1 (modulo N)
1684 */
1685 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1686 if (!node_online(node))
1687 continue;
1688 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1689 }
1690 for (node = 0; node < local_node; node++) {
1691 if (!node_online(node))
1692 continue;
1693 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1694 }
1695
1696 zonelist->zones[j] = NULL;
1697 }
1698}
1699
1700#endif /* CONFIG_NUMA */
1701
1702void __init build_all_zonelists(void)
1703{
1704 int i;
1705
1706 for_each_online_node(i)
1707 build_zonelists(NODE_DATA(i));
1708 printk("Built %i zonelists\n", num_online_nodes());
1709 cpuset_init_current_mems_allowed();
1710}
1711
1712/*
1713 * Helper functions to size the waitqueue hash table.
1714 * Essentially these want to choose hash table sizes sufficiently
1715 * large so that collisions trying to wait on pages are rare.
1716 * But in fact, the number of active page waitqueues on typical
1717 * systems is ridiculously low, less than 200. So this is even
1718 * conservative, even though it seems large.
1719 *
1720 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1721 * waitqueues, i.e. the size of the waitq table given the number of pages.
1722 */
1723#define PAGES_PER_WAITQUEUE 256
1724
1725static inline unsigned long wait_table_size(unsigned long pages)
1726{
1727 unsigned long size = 1;
1728
1729 pages /= PAGES_PER_WAITQUEUE;
1730
1731 while (size < pages)
1732 size <<= 1;
1733
1734 /*
1735 * Once we have dozens or even hundreds of threads sleeping
1736 * on IO we've got bigger problems than wait queue collision.
1737 * Limit the size of the wait table to a reasonable size.
1738 */
1739 size = min(size, 4096UL);
1740
1741 return max(size, 4UL);
1742}
1743
1744/*
1745 * This is an integer logarithm so that shifts can be used later
1746 * to extract the more random high bits from the multiplicative
1747 * hash function before the remainder is taken.
1748 */
1749static inline unsigned long wait_table_bits(unsigned long size)
1750{
1751 return ffz(~size);
1752}
1753
1754#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1755
1756static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1757 unsigned long *zones_size, unsigned long *zholes_size)
1758{
1759 unsigned long realtotalpages, totalpages = 0;
1760 int i;
1761
1762 for (i = 0; i < MAX_NR_ZONES; i++)
1763 totalpages += zones_size[i];
1764 pgdat->node_spanned_pages = totalpages;
1765
1766 realtotalpages = totalpages;
1767 if (zholes_size)
1768 for (i = 0; i < MAX_NR_ZONES; i++)
1769 realtotalpages -= zholes_size[i];
1770 pgdat->node_present_pages = realtotalpages;
1771 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1772}
1773
1774
1775/*
1776 * Initially all pages are reserved - free ones are freed
1777 * up by free_all_bootmem() once the early boot process is
1778 * done. Non-atomic initialization, single-pass.
1779 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001780void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 unsigned long start_pfn)
1782{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001784 unsigned long end_pfn = start_pfn + size;
1785 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001787 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001788 if (!early_pfn_valid(pfn))
1789 continue;
1790 page = pfn_to_page(pfn);
1791 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08001792 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 reset_page_mapcount(page);
1794 SetPageReserved(page);
1795 INIT_LIST_HEAD(&page->lru);
1796#ifdef WANT_PAGE_VIRTUAL
1797 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1798 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001799 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 }
1802}
1803
1804void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1805 unsigned long size)
1806{
1807 int order;
1808 for (order = 0; order < MAX_ORDER ; order++) {
1809 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1810 zone->free_area[order].nr_free = 0;
1811 }
1812}
1813
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001814#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1815void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1816 unsigned long size)
1817{
1818 unsigned long snum = pfn_to_section_nr(pfn);
1819 unsigned long end = pfn_to_section_nr(pfn + size);
1820
1821 if (FLAGS_HAS_NODE)
1822 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1823 else
1824 for (; snum <= end; snum++)
1825 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1826}
1827
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828#ifndef __HAVE_ARCH_MEMMAP_INIT
1829#define memmap_init(size, nid, zone, start_pfn) \
1830 memmap_init_zone((size), (nid), (zone), (start_pfn))
1831#endif
1832
Ashok Raj6292d9a2006-02-01 03:04:44 -08001833static int __cpuinit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001834{
1835 int batch;
1836
1837 /*
1838 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001839 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001840 *
1841 * OK, so we don't know how big the cache is. So guess.
1842 */
1843 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001844 if (batch * PAGE_SIZE > 512 * 1024)
1845 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001846 batch /= 4; /* We effectively *= 4 below */
1847 if (batch < 1)
1848 batch = 1;
1849
1850 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001851 * Clamp the batch to a 2^n - 1 value. Having a power
1852 * of 2 value was found to be more likely to have
1853 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001854 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001855 * For example if 2 tasks are alternately allocating
1856 * batches of pages, one task can end up with a lot
1857 * of pages of one half of the possible page colors
1858 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001859 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001860 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001861
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001862 return batch;
1863}
1864
Christoph Lameter2caaad42005-06-21 17:15:00 -07001865inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1866{
1867 struct per_cpu_pages *pcp;
1868
Magnus Damm1c6fe942005-10-26 01:58:59 -07001869 memset(p, 0, sizeof(*p));
1870
Christoph Lameter2caaad42005-06-21 17:15:00 -07001871 pcp = &p->pcp[0]; /* hot */
1872 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001873 pcp->high = 6 * batch;
1874 pcp->batch = max(1UL, 1 * batch);
1875 INIT_LIST_HEAD(&pcp->list);
1876
1877 pcp = &p->pcp[1]; /* cold*/
1878 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001879 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001880 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001881 INIT_LIST_HEAD(&pcp->list);
1882}
1883
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001884/*
1885 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1886 * to the value high for the pageset p.
1887 */
1888
1889static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1890 unsigned long high)
1891{
1892 struct per_cpu_pages *pcp;
1893
1894 pcp = &p->pcp[0]; /* hot list */
1895 pcp->high = high;
1896 pcp->batch = max(1UL, high/4);
1897 if ((high/4) > (PAGE_SHIFT * 8))
1898 pcp->batch = PAGE_SHIFT * 8;
1899}
1900
1901
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001902#ifdef CONFIG_NUMA
1903/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001904 * Boot pageset table. One per cpu which is going to be used for all
1905 * zones and all nodes. The parameters will be set in such a way
1906 * that an item put on a list will immediately be handed over to
1907 * the buddy list. This is safe since pageset manipulation is done
1908 * with interrupts disabled.
1909 *
1910 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001911 *
1912 * The boot_pagesets must be kept even after bootup is complete for
1913 * unused processors and/or zones. They do play a role for bootstrapping
1914 * hotplugged processors.
1915 *
1916 * zoneinfo_show() and maybe other functions do
1917 * not check if the processor is online before following the pageset pointer.
1918 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001919 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001920static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001921
1922/*
1923 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001924 * per cpu pageset array in struct zone.
1925 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08001926static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001927{
1928 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001929
1930 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001931
Nick Piggin23316bc2006-01-08 01:00:41 -08001932 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001933 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08001934 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001935 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001936
Nick Piggin23316bc2006-01-08 01:00:41 -08001937 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001938
1939 if (percpu_pagelist_fraction)
1940 setup_pagelist_highmark(zone_pcp(zone, cpu),
1941 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001942 }
1943
1944 return 0;
1945bad:
1946 for_each_zone(dzone) {
1947 if (dzone == zone)
1948 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08001949 kfree(zone_pcp(dzone, cpu));
1950 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001951 }
1952 return -ENOMEM;
1953}
1954
1955static inline void free_zone_pagesets(int cpu)
1956{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001957 struct zone *zone;
1958
1959 for_each_zone(zone) {
1960 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1961
1962 zone_pcp(zone, cpu) = NULL;
1963 kfree(pset);
1964 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001965}
1966
Chandra Seetharaman83d722f2006-04-24 19:35:21 -07001967static int pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001968 unsigned long action,
1969 void *hcpu)
1970{
1971 int cpu = (long)hcpu;
1972 int ret = NOTIFY_OK;
1973
1974 switch (action) {
1975 case CPU_UP_PREPARE:
1976 if (process_zones(cpu))
1977 ret = NOTIFY_BAD;
1978 break;
Andi Kleenb0d41692005-11-05 17:25:53 +01001979 case CPU_UP_CANCELED:
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001980 case CPU_DEAD:
1981 free_zone_pagesets(cpu);
1982 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001983 default:
1984 break;
1985 }
1986 return ret;
1987}
1988
1989static struct notifier_block pageset_notifier =
1990 { &pageset_cpuup_callback, NULL, 0 };
1991
Al Viro78d99552005-12-15 09:18:25 +00001992void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001993{
1994 int err;
1995
1996 /* Initialize per_cpu_pageset for cpu 0.
1997 * A cpuup callback will do this for every cpu
1998 * as it comes online
1999 */
2000 err = process_zones(smp_processor_id());
2001 BUG_ON(err);
2002 register_cpu_notifier(&pageset_notifier);
2003}
2004
2005#endif
2006
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002007static __meminit
Dave Hansened8ece22005-10-29 18:16:50 -07002008void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2009{
2010 int i;
2011 struct pglist_data *pgdat = zone->zone_pgdat;
2012
2013 /*
2014 * The per-page waitqueue mechanism uses hashed waitqueues
2015 * per zone.
2016 */
2017 zone->wait_table_size = wait_table_size(zone_size_pages);
2018 zone->wait_table_bits = wait_table_bits(zone->wait_table_size);
2019 zone->wait_table = (wait_queue_head_t *)
2020 alloc_bootmem_node(pgdat, zone->wait_table_size
2021 * sizeof(wait_queue_head_t));
2022
2023 for(i = 0; i < zone->wait_table_size; ++i)
2024 init_waitqueue_head(zone->wait_table + i);
2025}
2026
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002027static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07002028{
2029 int cpu;
2030 unsigned long batch = zone_batchsize(zone);
2031
2032 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2033#ifdef CONFIG_NUMA
2034 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002035 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002036 setup_pageset(&boot_pageset[cpu],0);
2037#else
2038 setup_pageset(zone_pcp(zone,cpu), batch);
2039#endif
2040 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08002041 if (zone->present_pages)
2042 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2043 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07002044}
2045
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002046static __meminit void init_currently_empty_zone(struct zone *zone,
Dave Hansened8ece22005-10-29 18:16:50 -07002047 unsigned long zone_start_pfn, unsigned long size)
2048{
2049 struct pglist_data *pgdat = zone->zone_pgdat;
2050
2051 zone_wait_table_init(zone, size);
2052 pgdat->nr_zones = zone_idx(zone) + 1;
2053
Dave Hansened8ece22005-10-29 18:16:50 -07002054 zone->zone_start_pfn = zone_start_pfn;
2055
2056 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2057
2058 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2059}
2060
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061/*
2062 * Set up the zone data structures:
2063 * - mark all pages reserved
2064 * - mark all memory queues empty
2065 * - clear the memory bitmaps
2066 */
2067static void __init free_area_init_core(struct pglist_data *pgdat,
2068 unsigned long *zones_size, unsigned long *zholes_size)
2069{
Dave Hansened8ece22005-10-29 18:16:50 -07002070 unsigned long j;
2071 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 unsigned long zone_start_pfn = pgdat->node_start_pfn;
2073
Dave Hansen208d54e2005-10-29 18:16:52 -07002074 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 pgdat->nr_zones = 0;
2076 init_waitqueue_head(&pgdat->kswapd_wait);
2077 pgdat->kswapd_max_order = 0;
2078
2079 for (j = 0; j < MAX_NR_ZONES; j++) {
2080 struct zone *zone = pgdat->node_zones + j;
2081 unsigned long size, realsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 realsize = size = zones_size[j];
2084 if (zholes_size)
2085 realsize -= zholes_size[j];
2086
Andi Kleena2f1b422005-11-05 17:25:53 +01002087 if (j < ZONE_HIGHMEM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 nr_kernel_pages += realsize;
2089 nr_all_pages += realsize;
2090
2091 zone->spanned_pages = size;
2092 zone->present_pages = realsize;
2093 zone->name = zone_names[j];
2094 spin_lock_init(&zone->lock);
2095 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002096 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 zone->zone_pgdat = pgdat;
2098 zone->free_pages = 0;
2099
2100 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2101
Dave Hansened8ece22005-10-29 18:16:50 -07002102 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 INIT_LIST_HEAD(&zone->active_list);
2104 INIT_LIST_HEAD(&zone->inactive_list);
2105 zone->nr_scan_active = 0;
2106 zone->nr_scan_inactive = 0;
2107 zone->nr_active = 0;
2108 zone->nr_inactive = 0;
Martin Hicks53e9a612005-09-03 15:54:51 -07002109 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 if (!size)
2111 continue;
2112
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002113 zonetable_add(zone, nid, j, zone_start_pfn, size);
Dave Hansened8ece22005-10-29 18:16:50 -07002114 init_currently_empty_zone(zone, zone_start_pfn, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 }
2117}
2118
2119static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2120{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 /* Skip empty nodes */
2122 if (!pgdat->node_spanned_pages)
2123 return;
2124
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002125#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 /* ia64 gets its own node_mem_map, before this, without bootmem */
2127 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07002128 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002129 struct page *map;
2130
Bob Piccoe984bb42006-05-20 15:00:31 -07002131 /*
2132 * The zone's endpoints aren't required to be MAX_ORDER
2133 * aligned but the node_mem_map endpoints must be in order
2134 * for the buddy allocator to function correctly.
2135 */
2136 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
2137 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
2138 end = ALIGN(end, MAX_ORDER_NR_PAGES);
2139 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002140 map = alloc_remap(pgdat->node_id, size);
2141 if (!map)
2142 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07002143 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002145#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 /*
2147 * With no DISCONTIG, the global mem_map is just set as node 0's
2148 */
2149 if (pgdat == NODE_DATA(0))
2150 mem_map = NODE_DATA(0)->node_mem_map;
2151#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002152#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153}
2154
2155void __init free_area_init_node(int nid, struct pglist_data *pgdat,
2156 unsigned long *zones_size, unsigned long node_start_pfn,
2157 unsigned long *zholes_size)
2158{
2159 pgdat->node_id = nid;
2160 pgdat->node_start_pfn = node_start_pfn;
2161 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2162
2163 alloc_node_mem_map(pgdat);
2164
2165 free_area_init_core(pgdat, zones_size, zholes_size);
2166}
2167
Dave Hansen93b75042005-06-23 00:07:47 -07002168#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169static bootmem_data_t contig_bootmem_data;
2170struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2171
2172EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002173#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174
2175void __init free_area_init(unsigned long *zones_size)
2176{
Dave Hansen93b75042005-06-23 00:07:47 -07002177 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2179}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180
2181#ifdef CONFIG_PROC_FS
2182
2183#include <linux/seq_file.h>
2184
2185static void *frag_start(struct seq_file *m, loff_t *pos)
2186{
2187 pg_data_t *pgdat;
2188 loff_t node = *pos;
KAMEZAWA Hiroyukiae0f15f2006-03-27 01:16:01 -08002189 for (pgdat = first_online_pgdat();
2190 pgdat && node;
2191 pgdat = next_online_pgdat(pgdat))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 --node;
2193
2194 return pgdat;
2195}
2196
2197static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
2198{
2199 pg_data_t *pgdat = (pg_data_t *)arg;
2200
2201 (*pos)++;
KAMEZAWA Hiroyukiae0f15f2006-03-27 01:16:01 -08002202 return next_online_pgdat(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203}
2204
2205static void frag_stop(struct seq_file *m, void *arg)
2206{
2207}
2208
2209/*
2210 * This walks the free areas for each zone.
2211 */
2212static int frag_show(struct seq_file *m, void *arg)
2213{
2214 pg_data_t *pgdat = (pg_data_t *)arg;
2215 struct zone *zone;
2216 struct zone *node_zones = pgdat->node_zones;
2217 unsigned long flags;
2218 int order;
2219
2220 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
Con Kolivasf3fe6512006-01-06 00:11:15 -08002221 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 continue;
2223
2224 spin_lock_irqsave(&zone->lock, flags);
2225 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2226 for (order = 0; order < MAX_ORDER; ++order)
2227 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2228 spin_unlock_irqrestore(&zone->lock, flags);
2229 seq_putc(m, '\n');
2230 }
2231 return 0;
2232}
2233
2234struct seq_operations fragmentation_op = {
2235 .start = frag_start,
2236 .next = frag_next,
2237 .stop = frag_stop,
2238 .show = frag_show,
2239};
2240
Nikita Danilov295ab932005-06-21 17:14:38 -07002241/*
2242 * Output information about zones in @pgdat.
2243 */
2244static int zoneinfo_show(struct seq_file *m, void *arg)
2245{
2246 pg_data_t *pgdat = arg;
2247 struct zone *zone;
2248 struct zone *node_zones = pgdat->node_zones;
2249 unsigned long flags;
2250
2251 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2252 int i;
2253
Con Kolivasf3fe6512006-01-06 00:11:15 -08002254 if (!populated_zone(zone))
Nikita Danilov295ab932005-06-21 17:14:38 -07002255 continue;
2256
2257 spin_lock_irqsave(&zone->lock, flags);
2258 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2259 seq_printf(m,
2260 "\n pages free %lu"
2261 "\n min %lu"
2262 "\n low %lu"
2263 "\n high %lu"
2264 "\n active %lu"
2265 "\n inactive %lu"
2266 "\n scanned %lu (a: %lu i: %lu)"
2267 "\n spanned %lu"
2268 "\n present %lu",
2269 zone->free_pages,
2270 zone->pages_min,
2271 zone->pages_low,
2272 zone->pages_high,
2273 zone->nr_active,
2274 zone->nr_inactive,
2275 zone->pages_scanned,
2276 zone->nr_scan_active, zone->nr_scan_inactive,
2277 zone->spanned_pages,
2278 zone->present_pages);
2279 seq_printf(m,
2280 "\n protection: (%lu",
2281 zone->lowmem_reserve[0]);
2282 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2283 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2284 seq_printf(m,
2285 ")"
2286 "\n pagesets");
Nick Piggin23316bc2006-01-08 01:00:41 -08002287 for_each_online_cpu(i) {
Nikita Danilov295ab932005-06-21 17:14:38 -07002288 struct per_cpu_pageset *pageset;
2289 int j;
2290
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002291 pageset = zone_pcp(zone, i);
Nikita Danilov295ab932005-06-21 17:14:38 -07002292 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2293 if (pageset->pcp[j].count)
2294 break;
2295 }
2296 if (j == ARRAY_SIZE(pageset->pcp))
2297 continue;
2298 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2299 seq_printf(m,
2300 "\n cpu: %i pcp: %i"
2301 "\n count: %i"
Nikita Danilov295ab932005-06-21 17:14:38 -07002302 "\n high: %i"
2303 "\n batch: %i",
2304 i, j,
2305 pageset->pcp[j].count,
Nikita Danilov295ab932005-06-21 17:14:38 -07002306 pageset->pcp[j].high,
2307 pageset->pcp[j].batch);
2308 }
2309#ifdef CONFIG_NUMA
2310 seq_printf(m,
2311 "\n numa_hit: %lu"
2312 "\n numa_miss: %lu"
2313 "\n numa_foreign: %lu"
2314 "\n interleave_hit: %lu"
2315 "\n local_node: %lu"
2316 "\n other_node: %lu",
2317 pageset->numa_hit,
2318 pageset->numa_miss,
2319 pageset->numa_foreign,
2320 pageset->interleave_hit,
2321 pageset->local_node,
2322 pageset->other_node);
2323#endif
2324 }
2325 seq_printf(m,
2326 "\n all_unreclaimable: %u"
2327 "\n prev_priority: %i"
2328 "\n temp_priority: %i"
2329 "\n start_pfn: %lu",
2330 zone->all_unreclaimable,
2331 zone->prev_priority,
2332 zone->temp_priority,
2333 zone->zone_start_pfn);
2334 spin_unlock_irqrestore(&zone->lock, flags);
2335 seq_putc(m, '\n');
2336 }
2337 return 0;
2338}
2339
2340struct seq_operations zoneinfo_op = {
2341 .start = frag_start, /* iterate over all zones. The same as in
2342 * fragmentation. */
2343 .next = frag_next,
2344 .stop = frag_stop,
2345 .show = zoneinfo_show,
2346};
2347
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348static char *vmstat_text[] = {
2349 "nr_dirty",
2350 "nr_writeback",
2351 "nr_unstable",
2352 "nr_page_table_pages",
2353 "nr_mapped",
2354 "nr_slab",
2355
2356 "pgpgin",
2357 "pgpgout",
2358 "pswpin",
2359 "pswpout",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360
Nick Piggin9328b8f2006-01-06 00:11:10 -08002361 "pgalloc_high",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 "pgalloc_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002363 "pgalloc_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 "pgalloc_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002365
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 "pgfree",
2367 "pgactivate",
2368 "pgdeactivate",
2369
2370 "pgfault",
2371 "pgmajfault",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002372
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 "pgrefill_high",
2374 "pgrefill_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002375 "pgrefill_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 "pgrefill_dma",
2377
2378 "pgsteal_high",
2379 "pgsteal_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002380 "pgsteal_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 "pgsteal_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002382
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 "pgscan_kswapd_high",
2384 "pgscan_kswapd_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002385 "pgscan_kswapd_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 "pgscan_kswapd_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002387
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 "pgscan_direct_high",
2389 "pgscan_direct_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002390 "pgscan_direct_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 "pgscan_direct_dma",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392
Nick Piggin9328b8f2006-01-06 00:11:10 -08002393 "pginodesteal",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 "slabs_scanned",
2395 "kswapd_steal",
2396 "kswapd_inodesteal",
2397 "pageoutrun",
2398 "allocstall",
2399
2400 "pgrotated",
KAMEZAWA Hiroyukiedfbe2b2005-05-01 08:58:37 -07002401 "nr_bounce",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402};
2403
2404static void *vmstat_start(struct seq_file *m, loff_t *pos)
2405{
2406 struct page_state *ps;
2407
2408 if (*pos >= ARRAY_SIZE(vmstat_text))
2409 return NULL;
2410
2411 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2412 m->private = ps;
2413 if (!ps)
2414 return ERR_PTR(-ENOMEM);
2415 get_full_page_state(ps);
2416 ps->pgpgin /= 2; /* sectors -> kbytes */
2417 ps->pgpgout /= 2;
2418 return (unsigned long *)ps + *pos;
2419}
2420
2421static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2422{
2423 (*pos)++;
2424 if (*pos >= ARRAY_SIZE(vmstat_text))
2425 return NULL;
2426 return (unsigned long *)m->private + *pos;
2427}
2428
2429static int vmstat_show(struct seq_file *m, void *arg)
2430{
2431 unsigned long *l = arg;
2432 unsigned long off = l - (unsigned long *)m->private;
2433
2434 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2435 return 0;
2436}
2437
2438static void vmstat_stop(struct seq_file *m, void *arg)
2439{
2440 kfree(m->private);
2441 m->private = NULL;
2442}
2443
2444struct seq_operations vmstat_op = {
2445 .start = vmstat_start,
2446 .next = vmstat_next,
2447 .stop = vmstat_stop,
2448 .show = vmstat_show,
2449};
2450
2451#endif /* CONFIG_PROC_FS */
2452
2453#ifdef CONFIG_HOTPLUG_CPU
2454static int page_alloc_cpu_notify(struct notifier_block *self,
2455 unsigned long action, void *hcpu)
2456{
2457 int cpu = (unsigned long)hcpu;
2458 long *count;
2459 unsigned long *src, *dest;
2460
2461 if (action == CPU_DEAD) {
2462 int i;
2463
2464 /* Drain local pagecache count. */
2465 count = &per_cpu(nr_pagecache_local, cpu);
2466 atomic_add(*count, &nr_pagecache);
2467 *count = 0;
2468 local_irq_disable();
2469 __drain_pages(cpu);
2470
2471 /* Add dead cpu's page_states to our own. */
2472 dest = (unsigned long *)&__get_cpu_var(page_states);
2473 src = (unsigned long *)&per_cpu(page_states, cpu);
2474
2475 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2476 i++) {
2477 dest[i] += src[i];
2478 src[i] = 0;
2479 }
2480
2481 local_irq_enable();
2482 }
2483 return NOTIFY_OK;
2484}
2485#endif /* CONFIG_HOTPLUG_CPU */
2486
2487void __init page_alloc_init(void)
2488{
2489 hotcpu_notifier(page_alloc_cpu_notify, 0);
2490}
2491
2492/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002493 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2494 * or min_free_kbytes changes.
2495 */
2496static void calculate_totalreserve_pages(void)
2497{
2498 struct pglist_data *pgdat;
2499 unsigned long reserve_pages = 0;
2500 int i, j;
2501
2502 for_each_online_pgdat(pgdat) {
2503 for (i = 0; i < MAX_NR_ZONES; i++) {
2504 struct zone *zone = pgdat->node_zones + i;
2505 unsigned long max = 0;
2506
2507 /* Find valid and maximum lowmem_reserve in the zone */
2508 for (j = i; j < MAX_NR_ZONES; j++) {
2509 if (zone->lowmem_reserve[j] > max)
2510 max = zone->lowmem_reserve[j];
2511 }
2512
2513 /* we treat pages_high as reserved pages. */
2514 max += zone->pages_high;
2515
2516 if (max > zone->present_pages)
2517 max = zone->present_pages;
2518 reserve_pages += max;
2519 }
2520 }
2521 totalreserve_pages = reserve_pages;
2522}
2523
2524/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 * setup_per_zone_lowmem_reserve - called whenever
2526 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2527 * has a correct pages reserved value, so an adequate number of
2528 * pages are left in the zone after a successful __alloc_pages().
2529 */
2530static void setup_per_zone_lowmem_reserve(void)
2531{
2532 struct pglist_data *pgdat;
2533 int j, idx;
2534
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08002535 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 for (j = 0; j < MAX_NR_ZONES; j++) {
2537 struct zone *zone = pgdat->node_zones + j;
2538 unsigned long present_pages = zone->present_pages;
2539
2540 zone->lowmem_reserve[j] = 0;
2541
2542 for (idx = j-1; idx >= 0; idx--) {
2543 struct zone *lower_zone;
2544
2545 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2546 sysctl_lowmem_reserve_ratio[idx] = 1;
2547
2548 lower_zone = pgdat->node_zones + idx;
2549 lower_zone->lowmem_reserve[j] = present_pages /
2550 sysctl_lowmem_reserve_ratio[idx];
2551 present_pages += lower_zone->present_pages;
2552 }
2553 }
2554 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002555
2556 /* update totalreserve_pages */
2557 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558}
2559
2560/*
2561 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2562 * that the pages_{min,low,high} values for each zone are set correctly
2563 * with respect to min_free_kbytes.
2564 */
Dave Hansen3947be12005-10-29 18:16:54 -07002565void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566{
2567 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2568 unsigned long lowmem_pages = 0;
2569 struct zone *zone;
2570 unsigned long flags;
2571
2572 /* Calculate total number of !ZONE_HIGHMEM pages */
2573 for_each_zone(zone) {
2574 if (!is_highmem(zone))
2575 lowmem_pages += zone->present_pages;
2576 }
2577
2578 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07002579 u64 tmp;
2580
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07002582 tmp = (u64)pages_min * zone->present_pages;
2583 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 if (is_highmem(zone)) {
2585 /*
Nick Piggin669ed172005-11-13 16:06:45 -08002586 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2587 * need highmem pages, so cap pages_min to a small
2588 * value here.
2589 *
2590 * The (pages_high-pages_low) and (pages_low-pages_min)
2591 * deltas controls asynch page reclaim, and so should
2592 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 */
2594 int min_pages;
2595
2596 min_pages = zone->present_pages / 1024;
2597 if (min_pages < SWAP_CLUSTER_MAX)
2598 min_pages = SWAP_CLUSTER_MAX;
2599 if (min_pages > 128)
2600 min_pages = 128;
2601 zone->pages_min = min_pages;
2602 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08002603 /*
2604 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 * proportionate to the zone's size.
2606 */
Nick Piggin669ed172005-11-13 16:06:45 -08002607 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 }
2609
Andrew Mortonac924c62006-05-15 09:43:59 -07002610 zone->pages_low = zone->pages_min + (tmp >> 2);
2611 zone->pages_high = zone->pages_min + (tmp >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 spin_unlock_irqrestore(&zone->lru_lock, flags);
2613 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002614
2615 /* update totalreserve_pages */
2616 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617}
2618
2619/*
2620 * Initialise min_free_kbytes.
2621 *
2622 * For small machines we want it small (128k min). For large machines
2623 * we want it large (64MB max). But it is not linear, because network
2624 * bandwidth does not increase linearly with machine size. We use
2625 *
2626 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2627 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2628 *
2629 * which yields
2630 *
2631 * 16MB: 512k
2632 * 32MB: 724k
2633 * 64MB: 1024k
2634 * 128MB: 1448k
2635 * 256MB: 2048k
2636 * 512MB: 2896k
2637 * 1024MB: 4096k
2638 * 2048MB: 5792k
2639 * 4096MB: 8192k
2640 * 8192MB: 11584k
2641 * 16384MB: 16384k
2642 */
2643static int __init init_per_zone_pages_min(void)
2644{
2645 unsigned long lowmem_kbytes;
2646
2647 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2648
2649 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2650 if (min_free_kbytes < 128)
2651 min_free_kbytes = 128;
2652 if (min_free_kbytes > 65536)
2653 min_free_kbytes = 65536;
2654 setup_per_zone_pages_min();
2655 setup_per_zone_lowmem_reserve();
2656 return 0;
2657}
2658module_init(init_per_zone_pages_min)
2659
2660/*
2661 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2662 * that we can call two helper functions whenever min_free_kbytes
2663 * changes.
2664 */
2665int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2666 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2667{
2668 proc_dointvec(table, write, file, buffer, length, ppos);
2669 setup_per_zone_pages_min();
2670 return 0;
2671}
2672
2673/*
2674 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2675 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2676 * whenever sysctl_lowmem_reserve_ratio changes.
2677 *
2678 * The reserve ratio obviously has absolutely no relation with the
2679 * pages_min watermarks. The lowmem reserve ratio can only make sense
2680 * if in function of the boot time zone sizes.
2681 */
2682int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2683 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2684{
2685 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2686 setup_per_zone_lowmem_reserve();
2687 return 0;
2688}
2689
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002690/*
2691 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2692 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2693 * can have before it gets flushed back to buddy allocator.
2694 */
2695
2696int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2697 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2698{
2699 struct zone *zone;
2700 unsigned int cpu;
2701 int ret;
2702
2703 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2704 if (!write || (ret == -EINVAL))
2705 return ret;
2706 for_each_zone(zone) {
2707 for_each_online_cpu(cpu) {
2708 unsigned long high;
2709 high = zone->present_pages / percpu_pagelist_fraction;
2710 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2711 }
2712 }
2713 return 0;
2714}
2715
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716__initdata int hashdist = HASHDIST_DEFAULT;
2717
2718#ifdef CONFIG_NUMA
2719static int __init set_hashdist(char *str)
2720{
2721 if (!str)
2722 return 0;
2723 hashdist = simple_strtoul(str, &str, 0);
2724 return 1;
2725}
2726__setup("hashdist=", set_hashdist);
2727#endif
2728
2729/*
2730 * allocate a large system hash table from bootmem
2731 * - it is assumed that the hash table must contain an exact power-of-2
2732 * quantity of entries
2733 * - limit is the number of hash buckets, not the total allocation size
2734 */
2735void *__init alloc_large_system_hash(const char *tablename,
2736 unsigned long bucketsize,
2737 unsigned long numentries,
2738 int scale,
2739 int flags,
2740 unsigned int *_hash_shift,
2741 unsigned int *_hash_mask,
2742 unsigned long limit)
2743{
2744 unsigned long long max = limit;
2745 unsigned long log2qty, size;
2746 void *table = NULL;
2747
2748 /* allow the kernel cmdline to have a say */
2749 if (!numentries) {
2750 /* round applicable memory size up to nearest megabyte */
2751 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2752 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2753 numentries >>= 20 - PAGE_SHIFT;
2754 numentries <<= 20 - PAGE_SHIFT;
2755
2756 /* limit to 1 bucket per 2^scale bytes of low memory */
2757 if (scale > PAGE_SHIFT)
2758 numentries >>= (scale - PAGE_SHIFT);
2759 else
2760 numentries <<= (PAGE_SHIFT - scale);
2761 }
John Hawkes6e692ed2006-03-25 03:08:02 -08002762 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763
2764 /* limit allocation size to 1/16 total memory by default */
2765 if (max == 0) {
2766 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2767 do_div(max, bucketsize);
2768 }
2769
2770 if (numentries > max)
2771 numentries = max;
2772
2773 log2qty = long_log2(numentries);
2774
2775 do {
2776 size = bucketsize << log2qty;
2777 if (flags & HASH_EARLY)
2778 table = alloc_bootmem(size);
2779 else if (hashdist)
2780 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2781 else {
2782 unsigned long order;
2783 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2784 ;
2785 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2786 }
2787 } while (!table && size > PAGE_SIZE && --log2qty);
2788
2789 if (!table)
2790 panic("Failed to allocate %s hash table\n", tablename);
2791
2792 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2793 tablename,
2794 (1U << log2qty),
2795 long_log2(size) - PAGE_SHIFT,
2796 size);
2797
2798 if (_hash_shift)
2799 *_hash_shift = log2qty;
2800 if (_hash_mask)
2801 *_hash_mask = (1 << log2qty) - 1;
2802
2803 return table;
2804}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002805
2806#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
2807/*
2808 * pfn <-> page translation. out-of-line version.
2809 * (see asm-generic/memory_model.h)
2810 */
2811#if defined(CONFIG_FLATMEM)
2812struct page *pfn_to_page(unsigned long pfn)
2813{
2814 return mem_map + (pfn - ARCH_PFN_OFFSET);
2815}
2816unsigned long page_to_pfn(struct page *page)
2817{
2818 return (page - mem_map) + ARCH_PFN_OFFSET;
2819}
2820#elif defined(CONFIG_DISCONTIGMEM)
2821struct page *pfn_to_page(unsigned long pfn)
2822{
2823 int nid = arch_pfn_to_nid(pfn);
2824 return NODE_DATA(nid)->node_mem_map + arch_local_page_offset(pfn,nid);
2825}
2826unsigned long page_to_pfn(struct page *page)
2827{
KAMEZAWA Hiroyukia0140c12006-03-27 01:15:55 -08002828 struct pglist_data *pgdat = NODE_DATA(page_to_nid(page));
2829 return (page - pgdat->node_mem_map) + pgdat->node_start_pfn;
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002830}
2831#elif defined(CONFIG_SPARSEMEM)
2832struct page *pfn_to_page(unsigned long pfn)
2833{
2834 return __section_mem_map_addr(__pfn_to_section(pfn)) + pfn;
2835}
2836
2837unsigned long page_to_pfn(struct page *page)
2838{
2839 long section_id = page_to_section(page);
2840 return page - __section_mem_map_addr(__nr_to_section(section_id));
2841}
2842#endif /* CONFIG_FLATMEM/DISCONTIGMME/SPARSEMEM */
2843EXPORT_SYMBOL(pfn_to_page);
2844EXPORT_SYMBOL(page_to_pfn);
2845#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */