blob: 97d6827c7d669529fb7e5607cad4ba838d1847bf [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
17#include <linux/config.h>
18#include <linux/stddef.h>
19#include <linux/mm.h>
20#include <linux/swap.h>
21#include <linux/interrupt.h>
22#include <linux/pagemap.h>
23#include <linux/bootmem.h>
24#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070025#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/module.h>
27#include <linux/suspend.h>
28#include <linux/pagevec.h>
29#include <linux/blkdev.h>
30#include <linux/slab.h>
31#include <linux/notifier.h>
32#include <linux/topology.h>
33#include <linux/sysctl.h>
34#include <linux/cpu.h>
35#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070036#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/nodemask.h>
38#include <linux/vmalloc.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080039#include <linux/mempolicy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#include <asm/tlbflush.h>
42#include "internal.h"
43
44/*
45 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
46 * initializer cleaner
47 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070048nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
Dean Nelson7223a932005-03-23 19:00:00 -070049EXPORT_SYMBOL(node_online_map);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070050nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
Dean Nelson7223a932005-03-23 19:00:00 -070051EXPORT_SYMBOL(node_possible_map);
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070052unsigned long totalram_pages __read_mostly;
53unsigned long totalhigh_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070054unsigned long totalreserve_pages __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055long nr_swap_pages;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080056int percpu_pagelist_fraction;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080058static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080059
Linus Torvalds1da177e2005-04-16 15:20:36 -070060/*
61 * results with 256, 32 in the lowmem_reserve sysctl:
62 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
63 * 1G machine -> (16M dma, 784M normal, 224M high)
64 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
65 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
66 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010067 *
68 * TBD: should special case ZONE_DMA32 machines here - in those we normally
69 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 */
Andi Kleena2f1b422005-11-05 17:25:53 +010071int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 256, 32 };
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
73EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
75/*
76 * Used by page_zone() to look up the address of the struct zone whose
77 * id is encoded in the upper bits of page->flags
78 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070079struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070080EXPORT_SYMBOL(zone_table);
81
Andi Kleena2f1b422005-11-05 17:25:53 +010082static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem" };
Linus Torvalds1da177e2005-04-16 15:20:36 -070083int min_free_kbytes = 1024;
84
85unsigned long __initdata nr_kernel_pages;
86unsigned long __initdata nr_all_pages;
87
Nick Piggin13e74442006-01-06 00:10:58 -080088#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -070089static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070090{
Dave Hansenbdc8cb92005-10-29 18:16:53 -070091 int ret = 0;
92 unsigned seq;
93 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -070094
Dave Hansenbdc8cb92005-10-29 18:16:53 -070095 do {
96 seq = zone_span_seqbegin(zone);
97 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
98 ret = 1;
99 else if (pfn < zone->zone_start_pfn)
100 ret = 1;
101 } while (zone_span_seqretry(zone, seq));
102
103 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700104}
105
106static int page_is_consistent(struct zone *zone, struct page *page)
107{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108#ifdef CONFIG_HOLES_IN_ZONE
109 if (!pfn_valid(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700110 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111#endif
112 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700113 return 0;
114
115 return 1;
116}
117/*
118 * Temporary debugging check for pages not lying within a given zone.
119 */
120static int bad_range(struct zone *zone, struct page *page)
121{
122 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700124 if (!page_is_consistent(zone, page))
125 return 1;
126
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 return 0;
128}
129
Nick Piggin13e74442006-01-06 00:10:58 -0800130#else
131static inline int bad_range(struct zone *zone, struct page *page)
132{
133 return 0;
134}
135#endif
136
Nick Piggin224abf92006-01-06 00:11:11 -0800137static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138{
Nick Piggin224abf92006-01-06 00:11:11 -0800139 printk(KERN_EMERG "Bad page state in process '%s'\n"
Hugh Dickins7365f3d2006-01-11 12:17:18 -0800140 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
141 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
142 KERN_EMERG "Backtrace:\n",
Nick Piggin224abf92006-01-06 00:11:11 -0800143 current->comm, page, (int)(2*sizeof(unsigned long)),
144 (unsigned long)page->flags, page->mapping,
145 page_mapcount(page), page_count(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 dump_stack();
Hugh Dickins334795e2005-06-21 17:15:08 -0700147 page->flags &= ~(1 << PG_lru |
148 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 1 << PG_active |
151 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700152 1 << PG_reclaim |
153 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 1 << PG_swapcache |
Nick Piggin676165a2006-04-10 11:21:48 +1000155 1 << PG_writeback |
156 1 << PG_buddy );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 set_page_count(page, 0);
158 reset_page_mapcount(page);
159 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700160 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161}
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163/*
164 * Higher-order pages are called "compound pages". They are structured thusly:
165 *
166 * The first PAGE_SIZE page is called the "head page".
167 *
168 * The remaining PAGE_SIZE pages are called "tail pages".
169 *
170 * All pages have PG_compound set. All pages have their ->private pointing at
171 * the head page (even the head page has this).
172 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800173 * The first tail page's ->lru.next holds the address of the compound page's
174 * put_page() function. Its ->lru.prev holds the order of allocation.
175 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800177
178static void free_compound_page(struct page *page)
179{
180 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
181}
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183static void prep_compound_page(struct page *page, unsigned long order)
184{
185 int i;
186 int nr_pages = 1 << order;
187
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800188 page[1].lru.next = (void *)free_compound_page; /* set dtor */
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800189 page[1].lru.prev = (void *)order;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 for (i = 0; i < nr_pages; i++) {
191 struct page *p = page + i;
192
Nick Piggin5e9dace2006-03-22 00:08:01 -0800193 __SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700194 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 }
196}
197
198static void destroy_compound_page(struct page *page, unsigned long order)
199{
200 int i;
201 int nr_pages = 1 << order;
202
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800203 if (unlikely((unsigned long)page[1].lru.prev != order))
Nick Piggin224abf92006-01-06 00:11:11 -0800204 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
206 for (i = 0; i < nr_pages; i++) {
207 struct page *p = page + i;
208
Nick Piggin224abf92006-01-06 00:11:11 -0800209 if (unlikely(!PageCompound(p) |
210 (page_private(p) != (unsigned long)page)))
211 bad_page(page);
Nick Piggin5e9dace2006-03-22 00:08:01 -0800212 __ClearPageCompound(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 }
214}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
Nick Piggin17cf4402006-03-22 00:08:41 -0800216static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
217{
218 int i;
219
220 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
Andrew Morton6626c5d2006-03-22 00:08:42 -0800221 /*
222 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
223 * and __GFP_HIGHMEM from hard or soft interrupt context.
224 */
225 BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800226 for (i = 0; i < (1 << order); i++)
227 clear_highpage(page + i);
228}
229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230/*
231 * function for dealing with page's order in buddy system.
232 * zone->lock is already acquired when we use these.
233 * So, we don't need atomic page->flags operations here.
234 */
235static inline unsigned long page_order(struct page *page) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700236 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237}
238
239static inline void set_page_order(struct page *page, int order) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700240 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000241 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242}
243
244static inline void rmv_page_order(struct page *page)
245{
Nick Piggin676165a2006-04-10 11:21:48 +1000246 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700247 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248}
249
250/*
251 * Locate the struct page for both the matching buddy in our
252 * pair (buddy1) and the combined O(n+1) page they form (page).
253 *
254 * 1) Any buddy B1 will have an order O twin B2 which satisfies
255 * the following equation:
256 * B2 = B1 ^ (1 << O)
257 * For example, if the starting buddy (buddy2) is #8 its order
258 * 1 buddy is #10:
259 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
260 *
261 * 2) Any buddy B will have an order O+1 parent P which
262 * satisfies the following equation:
263 * P = B & ~(1 << O)
264 *
265 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
266 */
267static inline struct page *
268__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
269{
270 unsigned long buddy_idx = page_idx ^ (1 << order);
271
272 return page + (buddy_idx - page_idx);
273}
274
275static inline unsigned long
276__find_combined_index(unsigned long page_idx, unsigned int order)
277{
278 return (page_idx & ~(1 << order));
279}
280
281/*
282 * This function checks whether a page is free && is the buddy
283 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800284 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000285 * (b) the buddy is in the buddy system &&
286 * (c) a page and its buddy have the same order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 *
Nick Piggin676165a2006-04-10 11:21:48 +1000288 * For recording whether a page is in the buddy system, we use PG_buddy.
289 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
290 *
291 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 */
293static inline int page_is_buddy(struct page *page, int order)
294{
Nick Piggin13e74442006-01-06 00:10:58 -0800295#ifdef CONFIG_HOLES_IN_ZONE
296 if (!pfn_valid(page_to_pfn(page)))
297 return 0;
298#endif
299
Nick Piggin676165a2006-04-10 11:21:48 +1000300 if (PageBuddy(page) && page_order(page) == order) {
301 BUG_ON(page_count(page) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000303 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 return 0;
305}
306
307/*
308 * Freeing function for a buddy system allocator.
309 *
310 * The concept of a buddy system is to maintain direct-mapped table
311 * (containing bit values) for memory blocks of various "orders".
312 * The bottom level table contains the map for the smallest allocatable
313 * units of memory (here, pages), and each level above it describes
314 * pairs of units from the levels below, hence, "buddies".
315 * At a high level, all that happens here is marking the table entry
316 * at the bottom level available, and propagating the changes upward
317 * as necessary, plus some accounting needed to play nicely with other
318 * parts of the VM system.
319 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000320 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700321 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 * So when we are allocating or freeing one, we can derive the state of the
323 * other. That is, if we allocate a small block, and both were
324 * free, the remainder of the region must be split into blocks.
325 * If a block is freed, and its buddy is also free, then this
326 * triggers coalescing into a block of larger size.
327 *
328 * -- wli
329 */
330
Nick Piggin48db57f2006-01-08 01:00:42 -0800331static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 struct zone *zone, unsigned int order)
333{
334 unsigned long page_idx;
335 int order_size = 1 << order;
336
Nick Piggin224abf92006-01-06 00:11:11 -0800337 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 destroy_compound_page(page, order);
339
340 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
341
342 BUG_ON(page_idx & (order_size - 1));
343 BUG_ON(bad_range(zone, page));
344
345 zone->free_pages += order_size;
346 while (order < MAX_ORDER-1) {
347 unsigned long combined_idx;
348 struct free_area *area;
349 struct page *buddy;
350
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 buddy = __page_find_buddy(page, page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 if (!page_is_buddy(buddy, order))
353 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800354
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 list_del(&buddy->lru);
356 area = zone->free_area + order;
357 area->nr_free--;
358 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800359 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 page = page + (combined_idx - page_idx);
361 page_idx = combined_idx;
362 order++;
363 }
364 set_page_order(page, order);
365 list_add(&page->lru, &zone->free_area[order].free_list);
366 zone->free_area[order].nr_free++;
367}
368
Nick Piggin224abf92006-01-06 00:11:11 -0800369static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370{
Nick Piggin92be2e332006-01-06 00:10:57 -0800371 if (unlikely(page_mapcount(page) |
372 (page->mapping != NULL) |
373 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 (page->flags & (
375 1 << PG_lru |
376 1 << PG_private |
377 1 << PG_locked |
378 1 << PG_active |
379 1 << PG_reclaim |
380 1 << PG_slab |
381 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700382 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000383 1 << PG_reserved |
384 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800385 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700387 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800388 /*
389 * For now, we report if PG_reserved was found set, but do not
390 * clear it, and do not free the page. But we shall soon need
391 * to do more, for when the ZERO_PAGE count wraps negative.
392 */
393 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394}
395
396/*
397 * Frees a list of pages.
398 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700399 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 *
401 * If the zone was previously in an "all pages pinned" state then look to
402 * see if this freeing clears that state.
403 *
404 * And clear the zone's pages_scanned counter, to hold off the "all pages are
405 * pinned" detection logic.
406 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800407static void free_pages_bulk(struct zone *zone, int count,
408 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409{
Nick Pigginc54ad302006-01-06 00:10:56 -0800410 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 zone->all_unreclaimable = 0;
412 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800413 while (count--) {
414 struct page *page;
415
416 BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800418 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800420 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800422 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423}
424
Nick Piggin48db57f2006-01-08 01:00:42 -0800425static void free_one_page(struct zone *zone, struct page *page, int order)
426{
427 LIST_HEAD(list);
428 list_add(&page->lru, &list);
429 free_pages_bulk(zone, 1, &list, order);
430}
431
432static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433{
Nick Pigginc54ad302006-01-06 00:10:56 -0800434 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800436 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
438 arch_free_page(page, order);
Ingo Molnarde5097c2006-01-09 15:59:21 -0800439 if (!PageHighMem(page))
440 mutex_debug_check_no_locks_freed(page_address(page),
David Woodhousea4fc7ab2006-01-11 14:41:26 +0000441 PAGE_SIZE<<order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800444 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800445 if (reserved)
446 return;
447
Nick Piggin48db57f2006-01-08 01:00:42 -0800448 kernel_map_pages(page, 1 << order, 0);
Nick Pigginc54ad302006-01-06 00:10:56 -0800449 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800450 __mod_page_state(pgfree, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800451 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800452 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453}
454
David Howellsa226f6c2006-01-06 00:11:08 -0800455/*
456 * permit the bootmem allocator to evade page validation on high-order frees
457 */
458void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
459{
460 if (order == 0) {
461 __ClearPageReserved(page);
462 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800463 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800464 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800465 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800466 int loop;
467
Nick Piggin545b1ea2006-03-22 00:08:07 -0800468 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800469 for (loop = 0; loop < BITS_PER_LONG; loop++) {
470 struct page *p = &page[loop];
471
Nick Piggin545b1ea2006-03-22 00:08:07 -0800472 if (loop + 1 < BITS_PER_LONG)
473 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800474 __ClearPageReserved(p);
475 set_page_count(p, 0);
476 }
477
Nick Piggin7835e982006-03-22 00:08:40 -0800478 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800479 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800480 }
481}
482
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
484/*
485 * The order of subdivision here is critical for the IO subsystem.
486 * Please do not alter this order without good reasons and regression
487 * testing. Specifically, as large blocks of memory are subdivided,
488 * the order in which smaller blocks are delivered depends on the order
489 * they're subdivided in this function. This is the primary factor
490 * influencing the order in which pages are delivered to the IO
491 * subsystem according to empirical testing, and this is also justified
492 * by considering the behavior of a buddy system containing a single
493 * large block of memory acted on by a series of small allocations.
494 * This behavior is a critical factor in sglist merging's success.
495 *
496 * -- wli
497 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800498static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 int low, int high, struct free_area *area)
500{
501 unsigned long size = 1 << high;
502
503 while (high > low) {
504 area--;
505 high--;
506 size >>= 1;
507 BUG_ON(bad_range(zone, &page[size]));
508 list_add(&page[size].lru, &area->free_list);
509 area->nr_free++;
510 set_page_order(&page[size], high);
511 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512}
513
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514/*
515 * This page is about to be returned from the page allocator
516 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800517static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518{
Nick Piggin92be2e332006-01-06 00:10:57 -0800519 if (unlikely(page_mapcount(page) |
520 (page->mapping != NULL) |
521 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700522 (page->flags & (
523 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 1 << PG_private |
525 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 1 << PG_active |
527 1 << PG_dirty |
528 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700529 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700531 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000532 1 << PG_reserved |
533 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800534 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
Hugh Dickins689bceb2005-11-21 21:32:20 -0800536 /*
537 * For now, we report if PG_reserved was found set, but do not
538 * clear it, and do not allocate the page: as a safety net.
539 */
540 if (PageReserved(page))
541 return 1;
542
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
544 1 << PG_referenced | 1 << PG_arch_1 |
545 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700546 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800547 set_page_refcounted(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800549
550 if (gfp_flags & __GFP_ZERO)
551 prep_zero_page(page, order, gfp_flags);
552
553 if (order && (gfp_flags & __GFP_COMP))
554 prep_compound_page(page, order);
555
Hugh Dickins689bceb2005-11-21 21:32:20 -0800556 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557}
558
559/*
560 * Do the hard work of removing an element from the buddy allocator.
561 * Call me with the zone->lock already held.
562 */
563static struct page *__rmqueue(struct zone *zone, unsigned int order)
564{
565 struct free_area * area;
566 unsigned int current_order;
567 struct page *page;
568
569 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
570 area = zone->free_area + current_order;
571 if (list_empty(&area->free_list))
572 continue;
573
574 page = list_entry(area->free_list.next, struct page, lru);
575 list_del(&page->lru);
576 rmv_page_order(page);
577 area->nr_free--;
578 zone->free_pages -= 1UL << order;
Nick Piggin085cc7d2006-01-06 00:11:01 -0800579 expand(zone, page, order, current_order, area);
580 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 }
582
583 return NULL;
584}
585
586/*
587 * Obtain a specified number of elements from the buddy allocator, all under
588 * a single hold of the lock, for efficiency. Add them to the supplied list.
589 * Returns the number of new pages which were placed at *list.
590 */
591static int rmqueue_bulk(struct zone *zone, unsigned int order,
592 unsigned long count, struct list_head *list)
593{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
Nick Pigginc54ad302006-01-06 00:10:56 -0800596 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 for (i = 0; i < count; ++i) {
Nick Piggin085cc7d2006-01-06 00:11:01 -0800598 struct page *page = __rmqueue(zone, order);
599 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 list_add_tail(&page->lru, list);
602 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800603 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800604 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605}
606
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700607#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800608/*
609 * Called from the slab reaper to drain pagesets on a particular node that
610 * belong to the currently executing processor.
Christoph Lameter879336c2006-03-22 00:09:08 -0800611 * Note that this function must be called with the thread pinned to
612 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800613 */
614void drain_node_pages(int nodeid)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700615{
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800616 int i, z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700617 unsigned long flags;
618
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800619 for (z = 0; z < MAX_NR_ZONES; z++) {
620 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700621 struct per_cpu_pageset *pset;
622
Nick Piggin23316bc2006-01-08 01:00:41 -0800623 pset = zone_pcp(zone, smp_processor_id());
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700624 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
625 struct per_cpu_pages *pcp;
626
627 pcp = &pset->pcp[i];
Christoph Lameter879336c2006-03-22 00:09:08 -0800628 if (pcp->count) {
629 local_irq_save(flags);
630 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
631 pcp->count = 0;
632 local_irq_restore(flags);
633 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700634 }
635 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700636}
637#endif
638
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
640static void __drain_pages(unsigned int cpu)
641{
Nick Pigginc54ad302006-01-06 00:10:56 -0800642 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 struct zone *zone;
644 int i;
645
646 for_each_zone(zone) {
647 struct per_cpu_pageset *pset;
648
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700649 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
651 struct per_cpu_pages *pcp;
652
653 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800654 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800655 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
656 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800657 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 }
659 }
660}
661#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
662
663#ifdef CONFIG_PM
664
665void mark_free_pages(struct zone *zone)
666{
667 unsigned long zone_pfn, flags;
668 int order;
669 struct list_head *curr;
670
671 if (!zone->spanned_pages)
672 return;
673
674 spin_lock_irqsave(&zone->lock, flags);
675 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
676 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
677
678 for (order = MAX_ORDER - 1; order >= 0; --order)
679 list_for_each(curr, &zone->free_area[order].free_list) {
680 unsigned long start_pfn, i;
681
682 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
683
684 for (i=0; i < (1<<order); i++)
685 SetPageNosaveFree(pfn_to_page(start_pfn+i));
686 }
687 spin_unlock_irqrestore(&zone->lock, flags);
688}
689
690/*
691 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
692 */
693void drain_local_pages(void)
694{
695 unsigned long flags;
696
697 local_irq_save(flags);
698 __drain_pages(smp_processor_id());
699 local_irq_restore(flags);
700}
701#endif /* CONFIG_PM */
702
Nick Piggina74609f2006-01-06 00:11:20 -0800703static void zone_statistics(struct zonelist *zonelist, struct zone *z, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704{
705#ifdef CONFIG_NUMA
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 pg_data_t *pg = z->zone_pgdat;
707 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
708 struct per_cpu_pageset *p;
709
Nick Piggina74609f2006-01-06 00:11:20 -0800710 p = zone_pcp(z, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 if (pg == orig) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700712 p->numa_hit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 } else {
714 p->numa_miss++;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700715 zone_pcp(zonelist->zones[0], cpu)->numa_foreign++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 }
717 if (pg == NODE_DATA(numa_node_id()))
718 p->local_node++;
719 else
720 p->other_node++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721#endif
722}
723
724/*
725 * Free a 0-order page
726 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727static void fastcall free_hot_cold_page(struct page *page, int cold)
728{
729 struct zone *zone = page_zone(page);
730 struct per_cpu_pages *pcp;
731 unsigned long flags;
732
733 arch_free_page(page, 0);
734
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 if (PageAnon(page))
736 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800737 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800738 return;
739
Hugh Dickins689bceb2005-11-21 21:32:20 -0800740 kernel_map_pages(page, 1, 0);
741
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700742 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800744 __inc_page_state(pgfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 list_add(&page->lru, &pcp->list);
746 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800747 if (pcp->count >= pcp->high) {
748 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
749 pcp->count -= pcp->batch;
750 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 local_irq_restore(flags);
752 put_cpu();
753}
754
755void fastcall free_hot_page(struct page *page)
756{
757 free_hot_cold_page(page, 0);
758}
759
760void fastcall free_cold_page(struct page *page)
761{
762 free_hot_cold_page(page, 1);
763}
764
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800765/*
766 * split_page takes a non-compound higher-order page, and splits it into
767 * n (1<<order) sub-pages: page[0..n]
768 * Each sub-page must be freed individually.
769 *
770 * Note: this is probably too low level an operation for use in drivers.
771 * Please consult with lkml before using this in your driver.
772 */
773void split_page(struct page *page, unsigned int order)
774{
775 int i;
776
777 BUG_ON(PageCompound(page));
778 BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -0800779 for (i = 1; i < (1 << order); i++)
780 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800781}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800782
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783/*
784 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
785 * we cheat by calling it from here, in the order > 0 path. Saves a branch
786 * or two.
787 */
Nick Piggina74609f2006-01-06 00:11:20 -0800788static struct page *buffered_rmqueue(struct zonelist *zonelist,
789 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790{
791 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800792 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800794 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Hugh Dickins689bceb2005-11-21 21:32:20 -0800796again:
Nick Piggina74609f2006-01-06 00:11:20 -0800797 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800798 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 struct per_cpu_pages *pcp;
800
Nick Piggina74609f2006-01-06 00:11:20 -0800801 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800803 if (!pcp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 pcp->count += rmqueue_bulk(zone, 0,
805 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800806 if (unlikely(!pcp->count))
807 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 }
Nick Piggina74609f2006-01-06 00:11:20 -0800809 page = list_entry(pcp->list.next, struct page, lru);
810 list_del(&page->lru);
811 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800812 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 spin_lock_irqsave(&zone->lock, flags);
814 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800815 spin_unlock(&zone->lock);
816 if (!page)
817 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 }
819
Nick Piggina74609f2006-01-06 00:11:20 -0800820 __mod_page_state_zone(zone, pgalloc, 1 << order);
821 zone_statistics(zonelist, zone, cpu);
822 local_irq_restore(flags);
823 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
Nick Piggina74609f2006-01-06 00:11:20 -0800825 BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -0800826 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -0800827 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800829
830failed:
831 local_irq_restore(flags);
832 put_cpu();
833 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834}
835
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800836#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800837#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
838#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
839#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
840#define ALLOC_HARDER 0x10 /* try to alloc harder */
841#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
842#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800843
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844/*
845 * Return 1 if free pages are above 'mark'. This takes into account the order
846 * of the allocation.
847 */
848int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800849 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850{
851 /* free_pages my go negative - that's OK */
852 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
853 int o;
854
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800855 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800857 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 min -= min / 4;
859
860 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
861 return 0;
862 for (o = 0; o < order; o++) {
863 /* At the next order, this order's pages become unavailable */
864 free_pages -= z->free_area[o].nr_free << o;
865
866 /* Require fewer higher order pages to be free */
867 min >>= 1;
868
869 if (free_pages <= min)
870 return 0;
871 }
872 return 1;
873}
874
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800875/*
876 * get_page_from_freeliest goes through the zonelist trying to allocate
877 * a page.
878 */
879static struct page *
880get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
881 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -0700882{
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800883 struct zone **z = zonelist->zones;
884 struct page *page = NULL;
885 int classzone_idx = zone_idx(*z);
886
887 /*
888 * Go through the zonelist once, looking for a zone with enough free.
889 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
890 */
891 do {
892 if ((alloc_flags & ALLOC_CPUSET) &&
893 !cpuset_zone_allowed(*z, gfp_mask))
894 continue;
895
896 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -0800897 unsigned long mark;
898 if (alloc_flags & ALLOC_WMARK_MIN)
899 mark = (*z)->pages_min;
900 else if (alloc_flags & ALLOC_WMARK_LOW)
901 mark = (*z)->pages_low;
902 else
903 mark = (*z)->pages_high;
904 if (!zone_watermark_ok(*z, order, mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800905 classzone_idx, alloc_flags))
Christoph Lameter9eeff232006-01-18 17:42:31 -0800906 if (!zone_reclaim_mode ||
907 !zone_reclaim(*z, gfp_mask, order))
908 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800909 }
910
Nick Piggina74609f2006-01-06 00:11:20 -0800911 page = buffered_rmqueue(zonelist, *z, order, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800912 if (page) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800913 break;
914 }
915 } while (*(++z) != NULL);
916 return page;
Martin Hicks753ee722005-06-21 17:14:41 -0700917}
918
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919/*
920 * This is the 'heart' of the zoned buddy allocator.
921 */
922struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +0100923__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 struct zonelist *zonelist)
925{
Al Viro260b2362005-10-21 03:22:44 -0400926 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800927 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 struct page *page;
929 struct reclaim_state reclaim_state;
930 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800932 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 int did_some_progress;
934
935 might_sleep_if(wait);
936
Jens Axboe6b1de912005-11-17 21:35:02 +0100937restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800938 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800940 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 /* Should this ever happen?? */
942 return NULL;
943 }
Jens Axboe6b1de912005-11-17 21:35:02 +0100944
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800945 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -0800946 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800947 if (page)
948 goto got_pg;
949
Jens Axboe6b1de912005-11-17 21:35:02 +0100950 do {
Christoph Lameter0b1303f2006-03-24 03:15:59 -0800951 if (cpuset_zone_allowed(*z, gfp_mask))
952 wakeup_kswapd(*z, order);
Jens Axboe6b1de912005-11-17 21:35:02 +0100953 } while (*(++z));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800954
Paul Jackson9bf22292005-09-06 15:18:12 -0700955 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800956 * OK, we're below the kswapd watermark and have kicked background
957 * reclaim. Now things get more complex, so set up alloc_flags according
958 * to how we want to proceed.
959 *
960 * The caller may dip into page reserves a bit more if the caller
961 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -0800962 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
963 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -0700964 */
Nick Piggin31488902005-11-28 13:44:03 -0800965 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800966 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
967 alloc_flags |= ALLOC_HARDER;
968 if (gfp_mask & __GFP_HIGH)
969 alloc_flags |= ALLOC_HIGH;
Paul Jackson47f3a862006-01-06 00:10:32 -0800970 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
972 /*
973 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800974 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 *
976 * This is the last chance, in general, before the goto nopage.
977 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -0700978 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800980 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
981 if (page)
982 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
984 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -0700985
986 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
987 && !in_interrupt()) {
988 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -0800989nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -0700990 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800991 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -0800992 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800993 if (page)
994 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -0800995 if (gfp_mask & __GFP_NOFAIL) {
996 blk_congestion_wait(WRITE, HZ/50);
997 goto nofail_alloc;
998 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 }
1000 goto nopage;
1001 }
1002
1003 /* Atomic allocations - we can't balance anything */
1004 if (!wait)
1005 goto nopage;
1006
1007rebalance:
1008 cond_resched();
1009
1010 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001011 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 p->flags |= PF_MEMALLOC;
1013 reclaim_state.reclaimed_slab = 0;
1014 p->reclaim_state = &reclaim_state;
1015
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001016 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
1018 p->reclaim_state = NULL;
1019 p->flags &= ~PF_MEMALLOC;
1020
1021 cond_resched();
1022
1023 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001024 page = get_page_from_freelist(gfp_mask, order,
1025 zonelist, alloc_flags);
1026 if (page)
1027 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1029 /*
1030 * Go through the zonelist yet one more time, keep
1031 * very high watermark here, this is only to catch
1032 * a parallel oom killing, we must fail if we're still
1033 * under heavy pressure.
1034 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001035 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001036 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001037 if (page)
1038 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001040 out_of_memory(zonelist, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 goto restart;
1042 }
1043
1044 /*
1045 * Don't let big-order allocations loop unless the caller explicitly
1046 * requests that. Wait for some write requests to complete then retry.
1047 *
1048 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1049 * <= 3, but that may not be true in other implementations.
1050 */
1051 do_retry = 0;
1052 if (!(gfp_mask & __GFP_NORETRY)) {
1053 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1054 do_retry = 1;
1055 if (gfp_mask & __GFP_NOFAIL)
1056 do_retry = 1;
1057 }
1058 if (do_retry) {
1059 blk_congestion_wait(WRITE, HZ/50);
1060 goto rebalance;
1061 }
1062
1063nopage:
1064 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1065 printk(KERN_WARNING "%s: page allocation failure."
1066 " order:%d, mode:0x%x\n",
1067 p->comm, order, gfp_mask);
1068 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001069 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 return page;
1073}
1074
1075EXPORT_SYMBOL(__alloc_pages);
1076
1077/*
1078 * Common helper functions.
1079 */
Al Virodd0fc662005-10-07 07:46:04 +01001080fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081{
1082 struct page * page;
1083 page = alloc_pages(gfp_mask, order);
1084 if (!page)
1085 return 0;
1086 return (unsigned long) page_address(page);
1087}
1088
1089EXPORT_SYMBOL(__get_free_pages);
1090
Al Virodd0fc662005-10-07 07:46:04 +01001091fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092{
1093 struct page * page;
1094
1095 /*
1096 * get_zeroed_page() returns a 32-bit address, which cannot represent
1097 * a highmem page
1098 */
Al Viro260b2362005-10-21 03:22:44 -04001099 BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
1101 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1102 if (page)
1103 return (unsigned long) page_address(page);
1104 return 0;
1105}
1106
1107EXPORT_SYMBOL(get_zeroed_page);
1108
1109void __pagevec_free(struct pagevec *pvec)
1110{
1111 int i = pagevec_count(pvec);
1112
1113 while (--i >= 0)
1114 free_hot_cold_page(pvec->pages[i], pvec->cold);
1115}
1116
1117fastcall void __free_pages(struct page *page, unsigned int order)
1118{
Nick Pigginb5810032005-10-29 18:16:12 -07001119 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 if (order == 0)
1121 free_hot_page(page);
1122 else
1123 __free_pages_ok(page, order);
1124 }
1125}
1126
1127EXPORT_SYMBOL(__free_pages);
1128
1129fastcall void free_pages(unsigned long addr, unsigned int order)
1130{
1131 if (addr != 0) {
1132 BUG_ON(!virt_addr_valid((void *)addr));
1133 __free_pages(virt_to_page((void *)addr), order);
1134 }
1135}
1136
1137EXPORT_SYMBOL(free_pages);
1138
1139/*
1140 * Total amount of free (allocatable) RAM:
1141 */
1142unsigned int nr_free_pages(void)
1143{
1144 unsigned int sum = 0;
1145 struct zone *zone;
1146
1147 for_each_zone(zone)
1148 sum += zone->free_pages;
1149
1150 return sum;
1151}
1152
1153EXPORT_SYMBOL(nr_free_pages);
1154
1155#ifdef CONFIG_NUMA
1156unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1157{
1158 unsigned int i, sum = 0;
1159
1160 for (i = 0; i < MAX_NR_ZONES; i++)
1161 sum += pgdat->node_zones[i].free_pages;
1162
1163 return sum;
1164}
1165#endif
1166
1167static unsigned int nr_free_zone_pages(int offset)
1168{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001169 /* Just pick one node, since fallback list is circular */
1170 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 unsigned int sum = 0;
1172
Martin J. Blighe310fd42005-07-29 22:59:18 -07001173 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1174 struct zone **zonep = zonelist->zones;
1175 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Martin J. Blighe310fd42005-07-29 22:59:18 -07001177 for (zone = *zonep++; zone; zone = *zonep++) {
1178 unsigned long size = zone->present_pages;
1179 unsigned long high = zone->pages_high;
1180 if (size > high)
1181 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 }
1183
1184 return sum;
1185}
1186
1187/*
1188 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1189 */
1190unsigned int nr_free_buffer_pages(void)
1191{
Al Viroaf4ca452005-10-21 02:55:38 -04001192 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193}
1194
1195/*
1196 * Amount of free RAM allocatable within all zones
1197 */
1198unsigned int nr_free_pagecache_pages(void)
1199{
Al Viroaf4ca452005-10-21 02:55:38 -04001200 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201}
1202
1203#ifdef CONFIG_HIGHMEM
1204unsigned int nr_free_highpages (void)
1205{
1206 pg_data_t *pgdat;
1207 unsigned int pages = 0;
1208
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08001209 for_each_online_pgdat(pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1211
1212 return pages;
1213}
1214#endif
1215
1216#ifdef CONFIG_NUMA
1217static void show_node(struct zone *zone)
1218{
1219 printk("Node %d ", zone->zone_pgdat->node_id);
1220}
1221#else
1222#define show_node(zone) do { } while (0)
1223#endif
1224
1225/*
1226 * Accumulate the page_state information across all CPUs.
1227 * The result is unavoidably approximate - it can change
1228 * during and after execution of this function.
1229 */
1230static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1231
1232atomic_t nr_pagecache = ATOMIC_INIT(0);
1233EXPORT_SYMBOL(nr_pagecache);
1234#ifdef CONFIG_SMP
1235DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1236#endif
1237
Nick Piggina86b1f52006-01-06 00:11:00 -08001238static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239{
Andrew Mortonb40607f2006-03-22 00:07:39 -08001240 unsigned cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001242 memset(ret, 0, nr * sizeof(unsigned long));
Andrew Morton84c20082006-01-08 01:00:28 -08001243 cpus_and(*cpumask, *cpumask, cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
Andrew Mortonb40607f2006-03-22 00:07:39 -08001245 for_each_cpu_mask(cpu, *cpumask) {
1246 unsigned long *in;
1247 unsigned long *out;
1248 unsigned off;
1249 unsigned next_cpu;
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001250
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 in = (unsigned long *)&per_cpu(page_states, cpu);
1252
Andrew Mortonb40607f2006-03-22 00:07:39 -08001253 next_cpu = next_cpu(cpu, *cpumask);
1254 if (likely(next_cpu < NR_CPUS))
1255 prefetch(&per_cpu(page_states, next_cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256
1257 out = (unsigned long *)ret;
1258 for (off = 0; off < nr; off++)
1259 *out++ += *in++;
1260 }
1261}
1262
Martin Hicksc07e02d2005-09-03 15:55:11 -07001263void get_page_state_node(struct page_state *ret, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264{
1265 int nr;
Martin Hicksc07e02d2005-09-03 15:55:11 -07001266 cpumask_t mask = node_to_cpumask(node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
1268 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1269 nr /= sizeof(unsigned long);
1270
Martin Hicksc07e02d2005-09-03 15:55:11 -07001271 __get_page_state(ret, nr+1, &mask);
1272}
1273
1274void get_page_state(struct page_state *ret)
1275{
1276 int nr;
1277 cpumask_t mask = CPU_MASK_ALL;
1278
1279 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1280 nr /= sizeof(unsigned long);
1281
1282 __get_page_state(ret, nr + 1, &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283}
1284
1285void get_full_page_state(struct page_state *ret)
1286{
Martin Hicksc07e02d2005-09-03 15:55:11 -07001287 cpumask_t mask = CPU_MASK_ALL;
1288
1289 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290}
1291
Nick Piggina74609f2006-01-06 00:11:20 -08001292unsigned long read_page_state_offset(unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293{
1294 unsigned long ret = 0;
1295 int cpu;
1296
Andrew Morton84c20082006-01-08 01:00:28 -08001297 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 unsigned long in;
1299
1300 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1301 ret += *((unsigned long *)in);
1302 }
1303 return ret;
1304}
1305
Nick Piggina74609f2006-01-06 00:11:20 -08001306void __mod_page_state_offset(unsigned long offset, unsigned long delta)
1307{
1308 void *ptr;
1309
1310 ptr = &__get_cpu_var(page_states);
1311 *(unsigned long *)(ptr + offset) += delta;
1312}
1313EXPORT_SYMBOL(__mod_page_state_offset);
1314
1315void mod_page_state_offset(unsigned long offset, unsigned long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316{
1317 unsigned long flags;
Nick Piggina74609f2006-01-06 00:11:20 -08001318 void *ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319
1320 local_irq_save(flags);
1321 ptr = &__get_cpu_var(page_states);
Nick Piggina74609f2006-01-06 00:11:20 -08001322 *(unsigned long *)(ptr + offset) += delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 local_irq_restore(flags);
1324}
Nick Piggina74609f2006-01-06 00:11:20 -08001325EXPORT_SYMBOL(mod_page_state_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326
1327void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1328 unsigned long *free, struct pglist_data *pgdat)
1329{
1330 struct zone *zones = pgdat->node_zones;
1331 int i;
1332
1333 *active = 0;
1334 *inactive = 0;
1335 *free = 0;
1336 for (i = 0; i < MAX_NR_ZONES; i++) {
1337 *active += zones[i].nr_active;
1338 *inactive += zones[i].nr_inactive;
1339 *free += zones[i].free_pages;
1340 }
1341}
1342
1343void get_zone_counts(unsigned long *active,
1344 unsigned long *inactive, unsigned long *free)
1345{
1346 struct pglist_data *pgdat;
1347
1348 *active = 0;
1349 *inactive = 0;
1350 *free = 0;
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08001351 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 unsigned long l, m, n;
1353 __get_zone_counts(&l, &m, &n, pgdat);
1354 *active += l;
1355 *inactive += m;
1356 *free += n;
1357 }
1358}
1359
1360void si_meminfo(struct sysinfo *val)
1361{
1362 val->totalram = totalram_pages;
1363 val->sharedram = 0;
1364 val->freeram = nr_free_pages();
1365 val->bufferram = nr_blockdev_pages();
1366#ifdef CONFIG_HIGHMEM
1367 val->totalhigh = totalhigh_pages;
1368 val->freehigh = nr_free_highpages();
1369#else
1370 val->totalhigh = 0;
1371 val->freehigh = 0;
1372#endif
1373 val->mem_unit = PAGE_SIZE;
1374}
1375
1376EXPORT_SYMBOL(si_meminfo);
1377
1378#ifdef CONFIG_NUMA
1379void si_meminfo_node(struct sysinfo *val, int nid)
1380{
1381 pg_data_t *pgdat = NODE_DATA(nid);
1382
1383 val->totalram = pgdat->node_present_pages;
1384 val->freeram = nr_free_pages_pgdat(pgdat);
1385 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1386 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1387 val->mem_unit = PAGE_SIZE;
1388}
1389#endif
1390
1391#define K(x) ((x) << (PAGE_SHIFT-10))
1392
1393/*
1394 * Show free area list (used inside shift_scroll-lock stuff)
1395 * We also calculate the percentage fragmentation. We do this by counting the
1396 * memory on each free list with the exception of the first item on the list.
1397 */
1398void show_free_areas(void)
1399{
1400 struct page_state ps;
1401 int cpu, temperature;
1402 unsigned long active;
1403 unsigned long inactive;
1404 unsigned long free;
1405 struct zone *zone;
1406
1407 for_each_zone(zone) {
1408 show_node(zone);
1409 printk("%s per-cpu:", zone->name);
1410
Con Kolivasf3fe6512006-01-06 00:11:15 -08001411 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 printk(" empty\n");
1413 continue;
1414 } else
1415 printk("\n");
1416
Dave Jones6b482c62005-11-10 15:45:56 -05001417 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 struct per_cpu_pageset *pageset;
1419
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001420 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
1422 for (temperature = 0; temperature < 2; temperature++)
Nick Piggin2d92c5c2006-01-06 00:10:59 -08001423 printk("cpu %d %s: high %d, batch %d used:%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 cpu,
1425 temperature ? "cold" : "hot",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 pageset->pcp[temperature].high,
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001427 pageset->pcp[temperature].batch,
1428 pageset->pcp[temperature].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 }
1430 }
1431
1432 get_page_state(&ps);
1433 get_zone_counts(&active, &inactive, &free);
1434
Denis Vlasenkoc0d62212005-06-21 17:15:14 -07001435 printk("Free pages: %11ukB (%ukB HighMem)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 K(nr_free_pages()),
1437 K(nr_free_highpages()));
1438
1439 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1440 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1441 active,
1442 inactive,
1443 ps.nr_dirty,
1444 ps.nr_writeback,
1445 ps.nr_unstable,
1446 nr_free_pages(),
1447 ps.nr_slab,
1448 ps.nr_mapped,
1449 ps.nr_page_table_pages);
1450
1451 for_each_zone(zone) {
1452 int i;
1453
1454 show_node(zone);
1455 printk("%s"
1456 " free:%lukB"
1457 " min:%lukB"
1458 " low:%lukB"
1459 " high:%lukB"
1460 " active:%lukB"
1461 " inactive:%lukB"
1462 " present:%lukB"
1463 " pages_scanned:%lu"
1464 " all_unreclaimable? %s"
1465 "\n",
1466 zone->name,
1467 K(zone->free_pages),
1468 K(zone->pages_min),
1469 K(zone->pages_low),
1470 K(zone->pages_high),
1471 K(zone->nr_active),
1472 K(zone->nr_inactive),
1473 K(zone->present_pages),
1474 zone->pages_scanned,
1475 (zone->all_unreclaimable ? "yes" : "no")
1476 );
1477 printk("lowmem_reserve[]:");
1478 for (i = 0; i < MAX_NR_ZONES; i++)
1479 printk(" %lu", zone->lowmem_reserve[i]);
1480 printk("\n");
1481 }
1482
1483 for_each_zone(zone) {
1484 unsigned long nr, flags, order, total = 0;
1485
1486 show_node(zone);
1487 printk("%s: ", zone->name);
Con Kolivasf3fe6512006-01-06 00:11:15 -08001488 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 printk("empty\n");
1490 continue;
1491 }
1492
1493 spin_lock_irqsave(&zone->lock, flags);
1494 for (order = 0; order < MAX_ORDER; order++) {
1495 nr = zone->free_area[order].nr_free;
1496 total += nr << order;
1497 printk("%lu*%lukB ", nr, K(1UL) << order);
1498 }
1499 spin_unlock_irqrestore(&zone->lock, flags);
1500 printk("= %lukB\n", K(total));
1501 }
1502
1503 show_swap_cache_info();
1504}
1505
1506/*
1507 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001508 *
1509 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 */
Christoph Lameter1a932052006-01-06 00:11:16 -08001511static int __init build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter070f8032006-01-06 00:11:19 -08001512 struct zonelist *zonelist, int nr_zones, int zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513{
Christoph Lameter1a932052006-01-06 00:11:16 -08001514 struct zone *zone;
1515
Christoph Lameter070f8032006-01-06 00:11:19 -08001516 BUG_ON(zone_type > ZONE_HIGHMEM);
Christoph Lameter02a68a52006-01-06 00:11:18 -08001517
1518 do {
Christoph Lameter070f8032006-01-06 00:11:19 -08001519 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001520 if (populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521#ifndef CONFIG_HIGHMEM
Christoph Lameter070f8032006-01-06 00:11:19 -08001522 BUG_ON(zone_type > ZONE_NORMAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523#endif
Christoph Lameter070f8032006-01-06 00:11:19 -08001524 zonelist->zones[nr_zones++] = zone;
1525 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 }
Christoph Lameter070f8032006-01-06 00:11:19 -08001527 zone_type--;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001528
Christoph Lameter070f8032006-01-06 00:11:19 -08001529 } while (zone_type >= 0);
1530 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531}
1532
Al Viro260b2362005-10-21 03:22:44 -04001533static inline int highest_zone(int zone_bits)
1534{
1535 int res = ZONE_NORMAL;
1536 if (zone_bits & (__force int)__GFP_HIGHMEM)
1537 res = ZONE_HIGHMEM;
Andi Kleena2f1b422005-11-05 17:25:53 +01001538 if (zone_bits & (__force int)__GFP_DMA32)
1539 res = ZONE_DMA32;
Al Viro260b2362005-10-21 03:22:44 -04001540 if (zone_bits & (__force int)__GFP_DMA)
1541 res = ZONE_DMA;
1542 return res;
1543}
1544
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545#ifdef CONFIG_NUMA
1546#define MAX_NODE_LOAD (num_online_nodes())
1547static int __initdata node_load[MAX_NUMNODES];
1548/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001549 * 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 -07001550 * @node: node whose fallback list we're appending
1551 * @used_node_mask: nodemask_t of already used nodes
1552 *
1553 * We use a number of factors to determine which is the next node that should
1554 * appear on a given node's fallback list. The node should not have appeared
1555 * already in @node's fallback list, and it should be the next closest node
1556 * according to the distance array (which contains arbitrary distance values
1557 * from each node to each node in the system), and should also prefer nodes
1558 * with no CPUs, since presumably they'll have very little allocation pressure
1559 * on them otherwise.
1560 * It returns -1 if no node is found.
1561 */
1562static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1563{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001564 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 int min_val = INT_MAX;
1566 int best_node = -1;
1567
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001568 /* Use the local node if we haven't already */
1569 if (!node_isset(node, *used_node_mask)) {
1570 node_set(node, *used_node_mask);
1571 return node;
1572 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001574 for_each_online_node(n) {
1575 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576
1577 /* Don't want a node to appear more than once */
1578 if (node_isset(n, *used_node_mask))
1579 continue;
1580
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 /* Use the distance array to find the distance */
1582 val = node_distance(node, n);
1583
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001584 /* Penalize nodes under us ("prefer the next node") */
1585 val += (n < node);
1586
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 /* Give preference to headless and unused nodes */
1588 tmp = node_to_cpumask(n);
1589 if (!cpus_empty(tmp))
1590 val += PENALTY_FOR_NODE_WITH_CPUS;
1591
1592 /* Slight preference for less loaded node */
1593 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1594 val += node_load[n];
1595
1596 if (val < min_val) {
1597 min_val = val;
1598 best_node = n;
1599 }
1600 }
1601
1602 if (best_node >= 0)
1603 node_set(best_node, *used_node_mask);
1604
1605 return best_node;
1606}
1607
1608static void __init build_zonelists(pg_data_t *pgdat)
1609{
1610 int i, j, k, node, local_node;
1611 int prev_node, load;
1612 struct zonelist *zonelist;
1613 nodemask_t used_mask;
1614
1615 /* initialize zonelists */
1616 for (i = 0; i < GFP_ZONETYPES; i++) {
1617 zonelist = pgdat->node_zonelists + i;
1618 zonelist->zones[0] = NULL;
1619 }
1620
1621 /* NUMA-aware ordering of nodes */
1622 local_node = pgdat->node_id;
1623 load = num_online_nodes();
1624 prev_node = local_node;
1625 nodes_clear(used_mask);
1626 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001627 int distance = node_distance(local_node, node);
1628
1629 /*
1630 * If another node is sufficiently far away then it is better
1631 * to reclaim pages in a zone before going off node.
1632 */
1633 if (distance > RECLAIM_DISTANCE)
1634 zone_reclaim_mode = 1;
1635
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 /*
1637 * We don't want to pressure a particular node.
1638 * So adding penalty to the first node in same
1639 * distance group to make it round-robin.
1640 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08001641
1642 if (distance != node_distance(local_node, prev_node))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 node_load[node] += load;
1644 prev_node = node;
1645 load--;
1646 for (i = 0; i < GFP_ZONETYPES; i++) {
1647 zonelist = pgdat->node_zonelists + i;
1648 for (j = 0; zonelist->zones[j] != NULL; j++);
1649
Al Viro260b2362005-10-21 03:22:44 -04001650 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651
1652 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1653 zonelist->zones[j] = NULL;
1654 }
1655 }
1656}
1657
1658#else /* CONFIG_NUMA */
1659
1660static void __init build_zonelists(pg_data_t *pgdat)
1661{
1662 int i, j, k, node, local_node;
1663
1664 local_node = pgdat->node_id;
1665 for (i = 0; i < GFP_ZONETYPES; i++) {
1666 struct zonelist *zonelist;
1667
1668 zonelist = pgdat->node_zonelists + i;
1669
1670 j = 0;
Al Viro260b2362005-10-21 03:22:44 -04001671 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 j = build_zonelists_node(pgdat, zonelist, j, k);
1673 /*
1674 * Now we build the zonelist so that it contains the zones
1675 * of all the other nodes.
1676 * We don't want to pressure a particular node, so when
1677 * building the zones for node N, we make sure that the
1678 * zones coming right after the local ones are those from
1679 * node N+1 (modulo N)
1680 */
1681 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1682 if (!node_online(node))
1683 continue;
1684 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1685 }
1686 for (node = 0; node < local_node; node++) {
1687 if (!node_online(node))
1688 continue;
1689 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1690 }
1691
1692 zonelist->zones[j] = NULL;
1693 }
1694}
1695
1696#endif /* CONFIG_NUMA */
1697
1698void __init build_all_zonelists(void)
1699{
1700 int i;
1701
1702 for_each_online_node(i)
1703 build_zonelists(NODE_DATA(i));
1704 printk("Built %i zonelists\n", num_online_nodes());
1705 cpuset_init_current_mems_allowed();
1706}
1707
1708/*
1709 * Helper functions to size the waitqueue hash table.
1710 * Essentially these want to choose hash table sizes sufficiently
1711 * large so that collisions trying to wait on pages are rare.
1712 * But in fact, the number of active page waitqueues on typical
1713 * systems is ridiculously low, less than 200. So this is even
1714 * conservative, even though it seems large.
1715 *
1716 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1717 * waitqueues, i.e. the size of the waitq table given the number of pages.
1718 */
1719#define PAGES_PER_WAITQUEUE 256
1720
1721static inline unsigned long wait_table_size(unsigned long pages)
1722{
1723 unsigned long size = 1;
1724
1725 pages /= PAGES_PER_WAITQUEUE;
1726
1727 while (size < pages)
1728 size <<= 1;
1729
1730 /*
1731 * Once we have dozens or even hundreds of threads sleeping
1732 * on IO we've got bigger problems than wait queue collision.
1733 * Limit the size of the wait table to a reasonable size.
1734 */
1735 size = min(size, 4096UL);
1736
1737 return max(size, 4UL);
1738}
1739
1740/*
1741 * This is an integer logarithm so that shifts can be used later
1742 * to extract the more random high bits from the multiplicative
1743 * hash function before the remainder is taken.
1744 */
1745static inline unsigned long wait_table_bits(unsigned long size)
1746{
1747 return ffz(~size);
1748}
1749
1750#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1751
1752static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1753 unsigned long *zones_size, unsigned long *zholes_size)
1754{
1755 unsigned long realtotalpages, totalpages = 0;
1756 int i;
1757
1758 for (i = 0; i < MAX_NR_ZONES; i++)
1759 totalpages += zones_size[i];
1760 pgdat->node_spanned_pages = totalpages;
1761
1762 realtotalpages = totalpages;
1763 if (zholes_size)
1764 for (i = 0; i < MAX_NR_ZONES; i++)
1765 realtotalpages -= zholes_size[i];
1766 pgdat->node_present_pages = realtotalpages;
1767 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1768}
1769
1770
1771/*
1772 * Initially all pages are reserved - free ones are freed
1773 * up by free_all_bootmem() once the early boot process is
1774 * done. Non-atomic initialization, single-pass.
1775 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001776void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 unsigned long start_pfn)
1778{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001780 unsigned long end_pfn = start_pfn + size;
1781 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001783 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001784 if (!early_pfn_valid(pfn))
1785 continue;
1786 page = pfn_to_page(pfn);
1787 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08001788 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 reset_page_mapcount(page);
1790 SetPageReserved(page);
1791 INIT_LIST_HEAD(&page->lru);
1792#ifdef WANT_PAGE_VIRTUAL
1793 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1794 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001795 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 }
1798}
1799
1800void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1801 unsigned long size)
1802{
1803 int order;
1804 for (order = 0; order < MAX_ORDER ; order++) {
1805 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1806 zone->free_area[order].nr_free = 0;
1807 }
1808}
1809
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001810#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1811void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1812 unsigned long size)
1813{
1814 unsigned long snum = pfn_to_section_nr(pfn);
1815 unsigned long end = pfn_to_section_nr(pfn + size);
1816
1817 if (FLAGS_HAS_NODE)
1818 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1819 else
1820 for (; snum <= end; snum++)
1821 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1822}
1823
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824#ifndef __HAVE_ARCH_MEMMAP_INIT
1825#define memmap_init(size, nid, zone, start_pfn) \
1826 memmap_init_zone((size), (nid), (zone), (start_pfn))
1827#endif
1828
Ashok Raj6292d9a2006-02-01 03:04:44 -08001829static int __cpuinit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001830{
1831 int batch;
1832
1833 /*
1834 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001835 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001836 *
1837 * OK, so we don't know how big the cache is. So guess.
1838 */
1839 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001840 if (batch * PAGE_SIZE > 512 * 1024)
1841 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001842 batch /= 4; /* We effectively *= 4 below */
1843 if (batch < 1)
1844 batch = 1;
1845
1846 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001847 * Clamp the batch to a 2^n - 1 value. Having a power
1848 * of 2 value was found to be more likely to have
1849 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001850 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001851 * For example if 2 tasks are alternately allocating
1852 * batches of pages, one task can end up with a lot
1853 * of pages of one half of the possible page colors
1854 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001855 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001856 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001857
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001858 return batch;
1859}
1860
Christoph Lameter2caaad42005-06-21 17:15:00 -07001861inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1862{
1863 struct per_cpu_pages *pcp;
1864
Magnus Damm1c6fe942005-10-26 01:58:59 -07001865 memset(p, 0, sizeof(*p));
1866
Christoph Lameter2caaad42005-06-21 17:15:00 -07001867 pcp = &p->pcp[0]; /* hot */
1868 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001869 pcp->high = 6 * batch;
1870 pcp->batch = max(1UL, 1 * batch);
1871 INIT_LIST_HEAD(&pcp->list);
1872
1873 pcp = &p->pcp[1]; /* cold*/
1874 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001875 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001876 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001877 INIT_LIST_HEAD(&pcp->list);
1878}
1879
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001880/*
1881 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1882 * to the value high for the pageset p.
1883 */
1884
1885static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1886 unsigned long high)
1887{
1888 struct per_cpu_pages *pcp;
1889
1890 pcp = &p->pcp[0]; /* hot list */
1891 pcp->high = high;
1892 pcp->batch = max(1UL, high/4);
1893 if ((high/4) > (PAGE_SHIFT * 8))
1894 pcp->batch = PAGE_SHIFT * 8;
1895}
1896
1897
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001898#ifdef CONFIG_NUMA
1899/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001900 * Boot pageset table. One per cpu which is going to be used for all
1901 * zones and all nodes. The parameters will be set in such a way
1902 * that an item put on a list will immediately be handed over to
1903 * the buddy list. This is safe since pageset manipulation is done
1904 * with interrupts disabled.
1905 *
1906 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001907 *
1908 * The boot_pagesets must be kept even after bootup is complete for
1909 * unused processors and/or zones. They do play a role for bootstrapping
1910 * hotplugged processors.
1911 *
1912 * zoneinfo_show() and maybe other functions do
1913 * not check if the processor is online before following the pageset pointer.
1914 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001915 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001916static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001917
1918/*
1919 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001920 * per cpu pageset array in struct zone.
1921 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08001922static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001923{
1924 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001925
1926 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001927
Nick Piggin23316bc2006-01-08 01:00:41 -08001928 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001929 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08001930 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001931 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001932
Nick Piggin23316bc2006-01-08 01:00:41 -08001933 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001934
1935 if (percpu_pagelist_fraction)
1936 setup_pagelist_highmark(zone_pcp(zone, cpu),
1937 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001938 }
1939
1940 return 0;
1941bad:
1942 for_each_zone(dzone) {
1943 if (dzone == zone)
1944 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08001945 kfree(zone_pcp(dzone, cpu));
1946 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001947 }
1948 return -ENOMEM;
1949}
1950
1951static inline void free_zone_pagesets(int cpu)
1952{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001953 struct zone *zone;
1954
1955 for_each_zone(zone) {
1956 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1957
1958 zone_pcp(zone, cpu) = NULL;
1959 kfree(pset);
1960 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001961}
1962
Ashok Raj6292d9a2006-02-01 03:04:44 -08001963static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001964 unsigned long action,
1965 void *hcpu)
1966{
1967 int cpu = (long)hcpu;
1968 int ret = NOTIFY_OK;
1969
1970 switch (action) {
1971 case CPU_UP_PREPARE:
1972 if (process_zones(cpu))
1973 ret = NOTIFY_BAD;
1974 break;
Andi Kleenb0d41692005-11-05 17:25:53 +01001975 case CPU_UP_CANCELED:
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001976 case CPU_DEAD:
1977 free_zone_pagesets(cpu);
1978 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001979 default:
1980 break;
1981 }
1982 return ret;
1983}
1984
1985static struct notifier_block pageset_notifier =
1986 { &pageset_cpuup_callback, NULL, 0 };
1987
Al Viro78d99552005-12-15 09:18:25 +00001988void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001989{
1990 int err;
1991
1992 /* Initialize per_cpu_pageset for cpu 0.
1993 * A cpuup callback will do this for every cpu
1994 * as it comes online
1995 */
1996 err = process_zones(smp_processor_id());
1997 BUG_ON(err);
1998 register_cpu_notifier(&pageset_notifier);
1999}
2000
2001#endif
2002
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002003static __meminit
Dave Hansened8ece22005-10-29 18:16:50 -07002004void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2005{
2006 int i;
2007 struct pglist_data *pgdat = zone->zone_pgdat;
2008
2009 /*
2010 * The per-page waitqueue mechanism uses hashed waitqueues
2011 * per zone.
2012 */
2013 zone->wait_table_size = wait_table_size(zone_size_pages);
2014 zone->wait_table_bits = wait_table_bits(zone->wait_table_size);
2015 zone->wait_table = (wait_queue_head_t *)
2016 alloc_bootmem_node(pgdat, zone->wait_table_size
2017 * sizeof(wait_queue_head_t));
2018
2019 for(i = 0; i < zone->wait_table_size; ++i)
2020 init_waitqueue_head(zone->wait_table + i);
2021}
2022
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002023static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07002024{
2025 int cpu;
2026 unsigned long batch = zone_batchsize(zone);
2027
2028 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2029#ifdef CONFIG_NUMA
2030 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002031 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002032 setup_pageset(&boot_pageset[cpu],0);
2033#else
2034 setup_pageset(zone_pcp(zone,cpu), batch);
2035#endif
2036 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08002037 if (zone->present_pages)
2038 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2039 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07002040}
2041
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002042static __meminit void init_currently_empty_zone(struct zone *zone,
Dave Hansened8ece22005-10-29 18:16:50 -07002043 unsigned long zone_start_pfn, unsigned long size)
2044{
2045 struct pglist_data *pgdat = zone->zone_pgdat;
2046
2047 zone_wait_table_init(zone, size);
2048 pgdat->nr_zones = zone_idx(zone) + 1;
2049
Dave Hansened8ece22005-10-29 18:16:50 -07002050 zone->zone_start_pfn = zone_start_pfn;
2051
2052 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2053
2054 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2055}
2056
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057/*
2058 * Set up the zone data structures:
2059 * - mark all pages reserved
2060 * - mark all memory queues empty
2061 * - clear the memory bitmaps
2062 */
2063static void __init free_area_init_core(struct pglist_data *pgdat,
2064 unsigned long *zones_size, unsigned long *zholes_size)
2065{
Dave Hansened8ece22005-10-29 18:16:50 -07002066 unsigned long j;
2067 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 unsigned long zone_start_pfn = pgdat->node_start_pfn;
2069
Dave Hansen208d54e2005-10-29 18:16:52 -07002070 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 pgdat->nr_zones = 0;
2072 init_waitqueue_head(&pgdat->kswapd_wait);
2073 pgdat->kswapd_max_order = 0;
2074
2075 for (j = 0; j < MAX_NR_ZONES; j++) {
2076 struct zone *zone = pgdat->node_zones + j;
2077 unsigned long size, realsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 realsize = size = zones_size[j];
2080 if (zholes_size)
2081 realsize -= zholes_size[j];
2082
Andi Kleena2f1b422005-11-05 17:25:53 +01002083 if (j < ZONE_HIGHMEM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 nr_kernel_pages += realsize;
2085 nr_all_pages += realsize;
2086
2087 zone->spanned_pages = size;
2088 zone->present_pages = realsize;
2089 zone->name = zone_names[j];
2090 spin_lock_init(&zone->lock);
2091 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002092 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 zone->zone_pgdat = pgdat;
2094 zone->free_pages = 0;
2095
2096 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2097
Dave Hansened8ece22005-10-29 18:16:50 -07002098 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 INIT_LIST_HEAD(&zone->active_list);
2100 INIT_LIST_HEAD(&zone->inactive_list);
2101 zone->nr_scan_active = 0;
2102 zone->nr_scan_inactive = 0;
2103 zone->nr_active = 0;
2104 zone->nr_inactive = 0;
Martin Hicks53e9a612005-09-03 15:54:51 -07002105 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 if (!size)
2107 continue;
2108
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002109 zonetable_add(zone, nid, j, zone_start_pfn, size);
Dave Hansened8ece22005-10-29 18:16:50 -07002110 init_currently_empty_zone(zone, zone_start_pfn, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 }
2113}
2114
2115static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2116{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 /* Skip empty nodes */
2118 if (!pgdat->node_spanned_pages)
2119 return;
2120
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002121#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 /* ia64 gets its own node_mem_map, before this, without bootmem */
2123 if (!pgdat->node_mem_map) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002124 unsigned long size;
2125 struct page *map;
2126
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002128 map = alloc_remap(pgdat->node_id, size);
2129 if (!map)
2130 map = alloc_bootmem_node(pgdat, size);
2131 pgdat->node_mem_map = map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002133#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 /*
2135 * With no DISCONTIG, the global mem_map is just set as node 0's
2136 */
2137 if (pgdat == NODE_DATA(0))
2138 mem_map = NODE_DATA(0)->node_mem_map;
2139#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002140#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141}
2142
2143void __init free_area_init_node(int nid, struct pglist_data *pgdat,
2144 unsigned long *zones_size, unsigned long node_start_pfn,
2145 unsigned long *zholes_size)
2146{
2147 pgdat->node_id = nid;
2148 pgdat->node_start_pfn = node_start_pfn;
2149 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2150
2151 alloc_node_mem_map(pgdat);
2152
2153 free_area_init_core(pgdat, zones_size, zholes_size);
2154}
2155
Dave Hansen93b75042005-06-23 00:07:47 -07002156#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157static bootmem_data_t contig_bootmem_data;
2158struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2159
2160EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002161#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162
2163void __init free_area_init(unsigned long *zones_size)
2164{
Dave Hansen93b75042005-06-23 00:07:47 -07002165 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2167}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168
2169#ifdef CONFIG_PROC_FS
2170
2171#include <linux/seq_file.h>
2172
2173static void *frag_start(struct seq_file *m, loff_t *pos)
2174{
2175 pg_data_t *pgdat;
2176 loff_t node = *pos;
KAMEZAWA Hiroyukiae0f15f2006-03-27 01:16:01 -08002177 for (pgdat = first_online_pgdat();
2178 pgdat && node;
2179 pgdat = next_online_pgdat(pgdat))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 --node;
2181
2182 return pgdat;
2183}
2184
2185static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
2186{
2187 pg_data_t *pgdat = (pg_data_t *)arg;
2188
2189 (*pos)++;
KAMEZAWA Hiroyukiae0f15f2006-03-27 01:16:01 -08002190 return next_online_pgdat(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191}
2192
2193static void frag_stop(struct seq_file *m, void *arg)
2194{
2195}
2196
2197/*
2198 * This walks the free areas for each zone.
2199 */
2200static int frag_show(struct seq_file *m, void *arg)
2201{
2202 pg_data_t *pgdat = (pg_data_t *)arg;
2203 struct zone *zone;
2204 struct zone *node_zones = pgdat->node_zones;
2205 unsigned long flags;
2206 int order;
2207
2208 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
Con Kolivasf3fe6512006-01-06 00:11:15 -08002209 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 continue;
2211
2212 spin_lock_irqsave(&zone->lock, flags);
2213 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2214 for (order = 0; order < MAX_ORDER; ++order)
2215 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2216 spin_unlock_irqrestore(&zone->lock, flags);
2217 seq_putc(m, '\n');
2218 }
2219 return 0;
2220}
2221
2222struct seq_operations fragmentation_op = {
2223 .start = frag_start,
2224 .next = frag_next,
2225 .stop = frag_stop,
2226 .show = frag_show,
2227};
2228
Nikita Danilov295ab932005-06-21 17:14:38 -07002229/*
2230 * Output information about zones in @pgdat.
2231 */
2232static int zoneinfo_show(struct seq_file *m, void *arg)
2233{
2234 pg_data_t *pgdat = arg;
2235 struct zone *zone;
2236 struct zone *node_zones = pgdat->node_zones;
2237 unsigned long flags;
2238
2239 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2240 int i;
2241
Con Kolivasf3fe6512006-01-06 00:11:15 -08002242 if (!populated_zone(zone))
Nikita Danilov295ab932005-06-21 17:14:38 -07002243 continue;
2244
2245 spin_lock_irqsave(&zone->lock, flags);
2246 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2247 seq_printf(m,
2248 "\n pages free %lu"
2249 "\n min %lu"
2250 "\n low %lu"
2251 "\n high %lu"
2252 "\n active %lu"
2253 "\n inactive %lu"
2254 "\n scanned %lu (a: %lu i: %lu)"
2255 "\n spanned %lu"
2256 "\n present %lu",
2257 zone->free_pages,
2258 zone->pages_min,
2259 zone->pages_low,
2260 zone->pages_high,
2261 zone->nr_active,
2262 zone->nr_inactive,
2263 zone->pages_scanned,
2264 zone->nr_scan_active, zone->nr_scan_inactive,
2265 zone->spanned_pages,
2266 zone->present_pages);
2267 seq_printf(m,
2268 "\n protection: (%lu",
2269 zone->lowmem_reserve[0]);
2270 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2271 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2272 seq_printf(m,
2273 ")"
2274 "\n pagesets");
Nick Piggin23316bc2006-01-08 01:00:41 -08002275 for_each_online_cpu(i) {
Nikita Danilov295ab932005-06-21 17:14:38 -07002276 struct per_cpu_pageset *pageset;
2277 int j;
2278
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002279 pageset = zone_pcp(zone, i);
Nikita Danilov295ab932005-06-21 17:14:38 -07002280 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2281 if (pageset->pcp[j].count)
2282 break;
2283 }
2284 if (j == ARRAY_SIZE(pageset->pcp))
2285 continue;
2286 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2287 seq_printf(m,
2288 "\n cpu: %i pcp: %i"
2289 "\n count: %i"
Nikita Danilov295ab932005-06-21 17:14:38 -07002290 "\n high: %i"
2291 "\n batch: %i",
2292 i, j,
2293 pageset->pcp[j].count,
Nikita Danilov295ab932005-06-21 17:14:38 -07002294 pageset->pcp[j].high,
2295 pageset->pcp[j].batch);
2296 }
2297#ifdef CONFIG_NUMA
2298 seq_printf(m,
2299 "\n numa_hit: %lu"
2300 "\n numa_miss: %lu"
2301 "\n numa_foreign: %lu"
2302 "\n interleave_hit: %lu"
2303 "\n local_node: %lu"
2304 "\n other_node: %lu",
2305 pageset->numa_hit,
2306 pageset->numa_miss,
2307 pageset->numa_foreign,
2308 pageset->interleave_hit,
2309 pageset->local_node,
2310 pageset->other_node);
2311#endif
2312 }
2313 seq_printf(m,
2314 "\n all_unreclaimable: %u"
2315 "\n prev_priority: %i"
2316 "\n temp_priority: %i"
2317 "\n start_pfn: %lu",
2318 zone->all_unreclaimable,
2319 zone->prev_priority,
2320 zone->temp_priority,
2321 zone->zone_start_pfn);
2322 spin_unlock_irqrestore(&zone->lock, flags);
2323 seq_putc(m, '\n');
2324 }
2325 return 0;
2326}
2327
2328struct seq_operations zoneinfo_op = {
2329 .start = frag_start, /* iterate over all zones. The same as in
2330 * fragmentation. */
2331 .next = frag_next,
2332 .stop = frag_stop,
2333 .show = zoneinfo_show,
2334};
2335
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336static char *vmstat_text[] = {
2337 "nr_dirty",
2338 "nr_writeback",
2339 "nr_unstable",
2340 "nr_page_table_pages",
2341 "nr_mapped",
2342 "nr_slab",
2343
2344 "pgpgin",
2345 "pgpgout",
2346 "pswpin",
2347 "pswpout",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348
Nick Piggin9328b8f2006-01-06 00:11:10 -08002349 "pgalloc_high",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 "pgalloc_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002351 "pgalloc_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 "pgalloc_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002353
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 "pgfree",
2355 "pgactivate",
2356 "pgdeactivate",
2357
2358 "pgfault",
2359 "pgmajfault",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002360
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 "pgrefill_high",
2362 "pgrefill_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002363 "pgrefill_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 "pgrefill_dma",
2365
2366 "pgsteal_high",
2367 "pgsteal_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002368 "pgsteal_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 "pgsteal_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002370
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 "pgscan_kswapd_high",
2372 "pgscan_kswapd_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002373 "pgscan_kswapd_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 "pgscan_kswapd_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002375
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 "pgscan_direct_high",
2377 "pgscan_direct_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002378 "pgscan_direct_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 "pgscan_direct_dma",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380
Nick Piggin9328b8f2006-01-06 00:11:10 -08002381 "pginodesteal",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 "slabs_scanned",
2383 "kswapd_steal",
2384 "kswapd_inodesteal",
2385 "pageoutrun",
2386 "allocstall",
2387
2388 "pgrotated",
KAMEZAWA Hiroyukiedfbe2b2005-05-01 08:58:37 -07002389 "nr_bounce",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390};
2391
2392static void *vmstat_start(struct seq_file *m, loff_t *pos)
2393{
2394 struct page_state *ps;
2395
2396 if (*pos >= ARRAY_SIZE(vmstat_text))
2397 return NULL;
2398
2399 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2400 m->private = ps;
2401 if (!ps)
2402 return ERR_PTR(-ENOMEM);
2403 get_full_page_state(ps);
2404 ps->pgpgin /= 2; /* sectors -> kbytes */
2405 ps->pgpgout /= 2;
2406 return (unsigned long *)ps + *pos;
2407}
2408
2409static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2410{
2411 (*pos)++;
2412 if (*pos >= ARRAY_SIZE(vmstat_text))
2413 return NULL;
2414 return (unsigned long *)m->private + *pos;
2415}
2416
2417static int vmstat_show(struct seq_file *m, void *arg)
2418{
2419 unsigned long *l = arg;
2420 unsigned long off = l - (unsigned long *)m->private;
2421
2422 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2423 return 0;
2424}
2425
2426static void vmstat_stop(struct seq_file *m, void *arg)
2427{
2428 kfree(m->private);
2429 m->private = NULL;
2430}
2431
2432struct seq_operations vmstat_op = {
2433 .start = vmstat_start,
2434 .next = vmstat_next,
2435 .stop = vmstat_stop,
2436 .show = vmstat_show,
2437};
2438
2439#endif /* CONFIG_PROC_FS */
2440
2441#ifdef CONFIG_HOTPLUG_CPU
2442static int page_alloc_cpu_notify(struct notifier_block *self,
2443 unsigned long action, void *hcpu)
2444{
2445 int cpu = (unsigned long)hcpu;
2446 long *count;
2447 unsigned long *src, *dest;
2448
2449 if (action == CPU_DEAD) {
2450 int i;
2451
2452 /* Drain local pagecache count. */
2453 count = &per_cpu(nr_pagecache_local, cpu);
2454 atomic_add(*count, &nr_pagecache);
2455 *count = 0;
2456 local_irq_disable();
2457 __drain_pages(cpu);
2458
2459 /* Add dead cpu's page_states to our own. */
2460 dest = (unsigned long *)&__get_cpu_var(page_states);
2461 src = (unsigned long *)&per_cpu(page_states, cpu);
2462
2463 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2464 i++) {
2465 dest[i] += src[i];
2466 src[i] = 0;
2467 }
2468
2469 local_irq_enable();
2470 }
2471 return NOTIFY_OK;
2472}
2473#endif /* CONFIG_HOTPLUG_CPU */
2474
2475void __init page_alloc_init(void)
2476{
2477 hotcpu_notifier(page_alloc_cpu_notify, 0);
2478}
2479
2480/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002481 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2482 * or min_free_kbytes changes.
2483 */
2484static void calculate_totalreserve_pages(void)
2485{
2486 struct pglist_data *pgdat;
2487 unsigned long reserve_pages = 0;
2488 int i, j;
2489
2490 for_each_online_pgdat(pgdat) {
2491 for (i = 0; i < MAX_NR_ZONES; i++) {
2492 struct zone *zone = pgdat->node_zones + i;
2493 unsigned long max = 0;
2494
2495 /* Find valid and maximum lowmem_reserve in the zone */
2496 for (j = i; j < MAX_NR_ZONES; j++) {
2497 if (zone->lowmem_reserve[j] > max)
2498 max = zone->lowmem_reserve[j];
2499 }
2500
2501 /* we treat pages_high as reserved pages. */
2502 max += zone->pages_high;
2503
2504 if (max > zone->present_pages)
2505 max = zone->present_pages;
2506 reserve_pages += max;
2507 }
2508 }
2509 totalreserve_pages = reserve_pages;
2510}
2511
2512/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 * setup_per_zone_lowmem_reserve - called whenever
2514 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2515 * has a correct pages reserved value, so an adequate number of
2516 * pages are left in the zone after a successful __alloc_pages().
2517 */
2518static void setup_per_zone_lowmem_reserve(void)
2519{
2520 struct pglist_data *pgdat;
2521 int j, idx;
2522
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08002523 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 for (j = 0; j < MAX_NR_ZONES; j++) {
2525 struct zone *zone = pgdat->node_zones + j;
2526 unsigned long present_pages = zone->present_pages;
2527
2528 zone->lowmem_reserve[j] = 0;
2529
2530 for (idx = j-1; idx >= 0; idx--) {
2531 struct zone *lower_zone;
2532
2533 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2534 sysctl_lowmem_reserve_ratio[idx] = 1;
2535
2536 lower_zone = pgdat->node_zones + idx;
2537 lower_zone->lowmem_reserve[j] = present_pages /
2538 sysctl_lowmem_reserve_ratio[idx];
2539 present_pages += lower_zone->present_pages;
2540 }
2541 }
2542 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002543
2544 /* update totalreserve_pages */
2545 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546}
2547
2548/*
2549 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2550 * that the pages_{min,low,high} values for each zone are set correctly
2551 * with respect to min_free_kbytes.
2552 */
Dave Hansen3947be12005-10-29 18:16:54 -07002553void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554{
2555 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2556 unsigned long lowmem_pages = 0;
2557 struct zone *zone;
2558 unsigned long flags;
2559
2560 /* Calculate total number of !ZONE_HIGHMEM pages */
2561 for_each_zone(zone) {
2562 if (!is_highmem(zone))
2563 lowmem_pages += zone->present_pages;
2564 }
2565
2566 for_each_zone(zone) {
Nick Piggin669ed172005-11-13 16:06:45 -08002567 unsigned long tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 spin_lock_irqsave(&zone->lru_lock, flags);
Nick Piggin669ed172005-11-13 16:06:45 -08002569 tmp = (pages_min * zone->present_pages) / lowmem_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 if (is_highmem(zone)) {
2571 /*
Nick Piggin669ed172005-11-13 16:06:45 -08002572 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2573 * need highmem pages, so cap pages_min to a small
2574 * value here.
2575 *
2576 * The (pages_high-pages_low) and (pages_low-pages_min)
2577 * deltas controls asynch page reclaim, and so should
2578 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 */
2580 int min_pages;
2581
2582 min_pages = zone->present_pages / 1024;
2583 if (min_pages < SWAP_CLUSTER_MAX)
2584 min_pages = SWAP_CLUSTER_MAX;
2585 if (min_pages > 128)
2586 min_pages = 128;
2587 zone->pages_min = min_pages;
2588 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08002589 /*
2590 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 * proportionate to the zone's size.
2592 */
Nick Piggin669ed172005-11-13 16:06:45 -08002593 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 }
2595
Nick Piggin669ed172005-11-13 16:06:45 -08002596 zone->pages_low = zone->pages_min + tmp / 4;
2597 zone->pages_high = zone->pages_min + tmp / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 spin_unlock_irqrestore(&zone->lru_lock, flags);
2599 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07002600
2601 /* update totalreserve_pages */
2602 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603}
2604
2605/*
2606 * Initialise min_free_kbytes.
2607 *
2608 * For small machines we want it small (128k min). For large machines
2609 * we want it large (64MB max). But it is not linear, because network
2610 * bandwidth does not increase linearly with machine size. We use
2611 *
2612 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2613 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2614 *
2615 * which yields
2616 *
2617 * 16MB: 512k
2618 * 32MB: 724k
2619 * 64MB: 1024k
2620 * 128MB: 1448k
2621 * 256MB: 2048k
2622 * 512MB: 2896k
2623 * 1024MB: 4096k
2624 * 2048MB: 5792k
2625 * 4096MB: 8192k
2626 * 8192MB: 11584k
2627 * 16384MB: 16384k
2628 */
2629static int __init init_per_zone_pages_min(void)
2630{
2631 unsigned long lowmem_kbytes;
2632
2633 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2634
2635 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2636 if (min_free_kbytes < 128)
2637 min_free_kbytes = 128;
2638 if (min_free_kbytes > 65536)
2639 min_free_kbytes = 65536;
2640 setup_per_zone_pages_min();
2641 setup_per_zone_lowmem_reserve();
2642 return 0;
2643}
2644module_init(init_per_zone_pages_min)
2645
2646/*
2647 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2648 * that we can call two helper functions whenever min_free_kbytes
2649 * changes.
2650 */
2651int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2652 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2653{
2654 proc_dointvec(table, write, file, buffer, length, ppos);
2655 setup_per_zone_pages_min();
2656 return 0;
2657}
2658
2659/*
2660 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2661 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2662 * whenever sysctl_lowmem_reserve_ratio changes.
2663 *
2664 * The reserve ratio obviously has absolutely no relation with the
2665 * pages_min watermarks. The lowmem reserve ratio can only make sense
2666 * if in function of the boot time zone sizes.
2667 */
2668int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2669 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2670{
2671 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2672 setup_per_zone_lowmem_reserve();
2673 return 0;
2674}
2675
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002676/*
2677 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2678 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2679 * can have before it gets flushed back to buddy allocator.
2680 */
2681
2682int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2683 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2684{
2685 struct zone *zone;
2686 unsigned int cpu;
2687 int ret;
2688
2689 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2690 if (!write || (ret == -EINVAL))
2691 return ret;
2692 for_each_zone(zone) {
2693 for_each_online_cpu(cpu) {
2694 unsigned long high;
2695 high = zone->present_pages / percpu_pagelist_fraction;
2696 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2697 }
2698 }
2699 return 0;
2700}
2701
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702__initdata int hashdist = HASHDIST_DEFAULT;
2703
2704#ifdef CONFIG_NUMA
2705static int __init set_hashdist(char *str)
2706{
2707 if (!str)
2708 return 0;
2709 hashdist = simple_strtoul(str, &str, 0);
2710 return 1;
2711}
2712__setup("hashdist=", set_hashdist);
2713#endif
2714
2715/*
2716 * allocate a large system hash table from bootmem
2717 * - it is assumed that the hash table must contain an exact power-of-2
2718 * quantity of entries
2719 * - limit is the number of hash buckets, not the total allocation size
2720 */
2721void *__init alloc_large_system_hash(const char *tablename,
2722 unsigned long bucketsize,
2723 unsigned long numentries,
2724 int scale,
2725 int flags,
2726 unsigned int *_hash_shift,
2727 unsigned int *_hash_mask,
2728 unsigned long limit)
2729{
2730 unsigned long long max = limit;
2731 unsigned long log2qty, size;
2732 void *table = NULL;
2733
2734 /* allow the kernel cmdline to have a say */
2735 if (!numentries) {
2736 /* round applicable memory size up to nearest megabyte */
2737 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2738 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2739 numentries >>= 20 - PAGE_SHIFT;
2740 numentries <<= 20 - PAGE_SHIFT;
2741
2742 /* limit to 1 bucket per 2^scale bytes of low memory */
2743 if (scale > PAGE_SHIFT)
2744 numentries >>= (scale - PAGE_SHIFT);
2745 else
2746 numentries <<= (PAGE_SHIFT - scale);
2747 }
John Hawkes6e692ed2006-03-25 03:08:02 -08002748 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749
2750 /* limit allocation size to 1/16 total memory by default */
2751 if (max == 0) {
2752 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2753 do_div(max, bucketsize);
2754 }
2755
2756 if (numentries > max)
2757 numentries = max;
2758
2759 log2qty = long_log2(numentries);
2760
2761 do {
2762 size = bucketsize << log2qty;
2763 if (flags & HASH_EARLY)
2764 table = alloc_bootmem(size);
2765 else if (hashdist)
2766 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2767 else {
2768 unsigned long order;
2769 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2770 ;
2771 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2772 }
2773 } while (!table && size > PAGE_SIZE && --log2qty);
2774
2775 if (!table)
2776 panic("Failed to allocate %s hash table\n", tablename);
2777
2778 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2779 tablename,
2780 (1U << log2qty),
2781 long_log2(size) - PAGE_SHIFT,
2782 size);
2783
2784 if (_hash_shift)
2785 *_hash_shift = log2qty;
2786 if (_hash_mask)
2787 *_hash_mask = (1 << log2qty) - 1;
2788
2789 return table;
2790}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002791
2792#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
2793/*
2794 * pfn <-> page translation. out-of-line version.
2795 * (see asm-generic/memory_model.h)
2796 */
2797#if defined(CONFIG_FLATMEM)
2798struct page *pfn_to_page(unsigned long pfn)
2799{
2800 return mem_map + (pfn - ARCH_PFN_OFFSET);
2801}
2802unsigned long page_to_pfn(struct page *page)
2803{
2804 return (page - mem_map) + ARCH_PFN_OFFSET;
2805}
2806#elif defined(CONFIG_DISCONTIGMEM)
2807struct page *pfn_to_page(unsigned long pfn)
2808{
2809 int nid = arch_pfn_to_nid(pfn);
2810 return NODE_DATA(nid)->node_mem_map + arch_local_page_offset(pfn,nid);
2811}
2812unsigned long page_to_pfn(struct page *page)
2813{
KAMEZAWA Hiroyukia0140c12006-03-27 01:15:55 -08002814 struct pglist_data *pgdat = NODE_DATA(page_to_nid(page));
2815 return (page - pgdat->node_mem_map) + pgdat->node_start_pfn;
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08002816}
2817#elif defined(CONFIG_SPARSEMEM)
2818struct page *pfn_to_page(unsigned long pfn)
2819{
2820 return __section_mem_map_addr(__pfn_to_section(pfn)) + pfn;
2821}
2822
2823unsigned long page_to_pfn(struct page *page)
2824{
2825 long section_id = page_to_section(page);
2826 return page - __section_mem_map_addr(__nr_to_section(section_id));
2827}
2828#endif /* CONFIG_FLATMEM/DISCONTIGMME/SPARSEMEM */
2829EXPORT_SYMBOL(pfn_to_page);
2830EXPORT_SYMBOL(page_to_pfn);
2831#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */