blob: 8c960b469593a9a6818fa4dd9c08a4b591916c86 [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);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070052struct pglist_data *pgdat_list __read_mostly;
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070053unsigned long totalram_pages __read_mostly;
54unsigned long totalhigh_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
David Howellsa226f6c2006-01-06 00:11:08 -080058static void fastcall free_hot_cold_page(struct page *page, int cold);
59
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 |
Hugh Dickins689bceb2005-11-21 21:32:20 -0800155 1 << PG_writeback );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 set_page_count(page, 0);
157 reset_page_mapcount(page);
158 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700159 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160}
161
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162/*
163 * Higher-order pages are called "compound pages". They are structured thusly:
164 *
165 * The first PAGE_SIZE page is called the "head page".
166 *
167 * The remaining PAGE_SIZE pages are called "tail pages".
168 *
169 * All pages have PG_compound set. All pages have their ->private pointing at
170 * the head page (even the head page has this).
171 *
172 * The first tail page's ->mapping, if non-zero, holds the address of the
173 * compound page's put_page() function.
174 *
175 * The order of the allocation is stored in the first tail page's ->index
176 * This is only for debug at present. This usage means that zero-order pages
177 * may not be compound.
178 */
179static void prep_compound_page(struct page *page, unsigned long order)
180{
181 int i;
182 int nr_pages = 1 << order;
183
184 page[1].mapping = NULL;
185 page[1].index = order;
186 for (i = 0; i < nr_pages; i++) {
187 struct page *p = page + i;
188
189 SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700190 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 }
192}
193
194static void destroy_compound_page(struct page *page, unsigned long order)
195{
196 int i;
197 int nr_pages = 1 << order;
198
Nick Piggin224abf92006-01-06 00:11:11 -0800199 if (unlikely(page[1].index != order))
200 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
202 for (i = 0; i < nr_pages; i++) {
203 struct page *p = page + i;
204
Nick Piggin224abf92006-01-06 00:11:11 -0800205 if (unlikely(!PageCompound(p) |
206 (page_private(p) != (unsigned long)page)))
207 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 ClearPageCompound(p);
209 }
210}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
212/*
213 * function for dealing with page's order in buddy system.
214 * zone->lock is already acquired when we use these.
215 * So, we don't need atomic page->flags operations here.
216 */
217static inline unsigned long page_order(struct page *page) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700218 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219}
220
221static inline void set_page_order(struct page *page, int order) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700222 set_page_private(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 __SetPagePrivate(page);
224}
225
226static inline void rmv_page_order(struct page *page)
227{
228 __ClearPagePrivate(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700229 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230}
231
232/*
233 * Locate the struct page for both the matching buddy in our
234 * pair (buddy1) and the combined O(n+1) page they form (page).
235 *
236 * 1) Any buddy B1 will have an order O twin B2 which satisfies
237 * the following equation:
238 * B2 = B1 ^ (1 << O)
239 * For example, if the starting buddy (buddy2) is #8 its order
240 * 1 buddy is #10:
241 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
242 *
243 * 2) Any buddy B will have an order O+1 parent P which
244 * satisfies the following equation:
245 * P = B & ~(1 << O)
246 *
247 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
248 */
249static inline struct page *
250__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
251{
252 unsigned long buddy_idx = page_idx ^ (1 << order);
253
254 return page + (buddy_idx - page_idx);
255}
256
257static inline unsigned long
258__find_combined_index(unsigned long page_idx, unsigned int order)
259{
260 return (page_idx & ~(1 << order));
261}
262
263/*
264 * This function checks whether a page is free && is the buddy
265 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800266 * (a) the buddy is not in a hole &&
267 * (b) the buddy is free &&
268 * (c) the buddy is on the buddy system &&
269 * (d) a page and its buddy have the same order.
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700270 * for recording page's order, we use page_private(page) and PG_private.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 *
272 */
273static inline int page_is_buddy(struct page *page, int order)
274{
Nick Piggin13e74442006-01-06 00:10:58 -0800275#ifdef CONFIG_HOLES_IN_ZONE
276 if (!pfn_valid(page_to_pfn(page)))
277 return 0;
278#endif
279
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 if (PagePrivate(page) &&
281 (page_order(page) == order) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 page_count(page) == 0)
283 return 1;
284 return 0;
285}
286
287/*
288 * Freeing function for a buddy system allocator.
289 *
290 * The concept of a buddy system is to maintain direct-mapped table
291 * (containing bit values) for memory blocks of various "orders".
292 * The bottom level table contains the map for the smallest allocatable
293 * units of memory (here, pages), and each level above it describes
294 * pairs of units from the levels below, hence, "buddies".
295 * At a high level, all that happens here is marking the table entry
296 * at the bottom level available, and propagating the changes upward
297 * as necessary, plus some accounting needed to play nicely with other
298 * parts of the VM system.
299 * At each level, we keep a list of pages, which are heads of continuous
300 * free pages of length of (1 << order) and marked with PG_Private.Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700301 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 * So when we are allocating or freeing one, we can derive the state of the
303 * other. That is, if we allocate a small block, and both were
304 * free, the remainder of the region must be split into blocks.
305 * If a block is freed, and its buddy is also free, then this
306 * triggers coalescing into a block of larger size.
307 *
308 * -- wli
309 */
310
Nick Piggin48db57f2006-01-08 01:00:42 -0800311static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 struct zone *zone, unsigned int order)
313{
314 unsigned long page_idx;
315 int order_size = 1 << order;
316
Nick Piggin224abf92006-01-06 00:11:11 -0800317 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 destroy_compound_page(page, order);
319
320 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
321
322 BUG_ON(page_idx & (order_size - 1));
323 BUG_ON(bad_range(zone, page));
324
325 zone->free_pages += order_size;
326 while (order < MAX_ORDER-1) {
327 unsigned long combined_idx;
328 struct free_area *area;
329 struct page *buddy;
330
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 buddy = __page_find_buddy(page, page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 if (!page_is_buddy(buddy, order))
333 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800334
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 list_del(&buddy->lru);
336 area = zone->free_area + order;
337 area->nr_free--;
338 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800339 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 page = page + (combined_idx - page_idx);
341 page_idx = combined_idx;
342 order++;
343 }
344 set_page_order(page, order);
345 list_add(&page->lru, &zone->free_area[order].free_list);
346 zone->free_area[order].nr_free++;
347}
348
Nick Piggin224abf92006-01-06 00:11:11 -0800349static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350{
Nick Piggin92be2e32006-01-06 00:10:57 -0800351 if (unlikely(page_mapcount(page) |
352 (page->mapping != NULL) |
353 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 (page->flags & (
355 1 << PG_lru |
356 1 << PG_private |
357 1 << PG_locked |
358 1 << PG_active |
359 1 << PG_reclaim |
360 1 << PG_slab |
361 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700362 1 << PG_writeback |
Nick Piggin92be2e32006-01-06 00:10:57 -0800363 1 << PG_reserved ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800364 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700366 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800367 /*
368 * For now, we report if PG_reserved was found set, but do not
369 * clear it, and do not free the page. But we shall soon need
370 * to do more, for when the ZERO_PAGE count wraps negative.
371 */
372 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373}
374
375/*
376 * Frees a list of pages.
377 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700378 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 *
380 * If the zone was previously in an "all pages pinned" state then look to
381 * see if this freeing clears that state.
382 *
383 * And clear the zone's pages_scanned counter, to hold off the "all pages are
384 * pinned" detection logic.
385 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800386static void free_pages_bulk(struct zone *zone, int count,
387 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388{
Nick Pigginc54ad302006-01-06 00:10:56 -0800389 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 zone->all_unreclaimable = 0;
391 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800392 while (count--) {
393 struct page *page;
394
395 BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800397 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800399 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800401 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402}
403
Nick Piggin48db57f2006-01-08 01:00:42 -0800404static void free_one_page(struct zone *zone, struct page *page, int order)
405{
406 LIST_HEAD(list);
407 list_add(&page->lru, &list);
408 free_pages_bulk(zone, 1, &list, order);
409}
410
411static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412{
Nick Pigginc54ad302006-01-06 00:10:56 -0800413 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800415 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
417 arch_free_page(page, order);
Ingo Molnarde5097c2006-01-09 15:59:21 -0800418 if (!PageHighMem(page))
419 mutex_debug_check_no_locks_freed(page_address(page),
David Woodhousea4fc7ab2006-01-11 14:41:26 +0000420 PAGE_SIZE<<order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422#ifndef CONFIG_MMU
Nick Piggin48db57f2006-01-08 01:00:42 -0800423 for (i = 1 ; i < (1 << order) ; ++i)
424 __put_page(page + i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425#endif
426
427 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800428 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800429 if (reserved)
430 return;
431
Nick Piggin48db57f2006-01-08 01:00:42 -0800432 kernel_map_pages(page, 1 << order, 0);
Nick Pigginc54ad302006-01-06 00:10:56 -0800433 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800434 __mod_page_state(pgfree, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800435 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800436 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437}
438
David Howellsa226f6c2006-01-06 00:11:08 -0800439/*
440 * permit the bootmem allocator to evade page validation on high-order frees
441 */
442void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
443{
444 if (order == 0) {
445 __ClearPageReserved(page);
446 set_page_count(page, 0);
447
448 free_hot_cold_page(page, 0);
449 } else {
450 LIST_HEAD(list);
451 int loop;
452
453 for (loop = 0; loop < BITS_PER_LONG; loop++) {
454 struct page *p = &page[loop];
455
456 if (loop + 16 < BITS_PER_LONG)
457 prefetchw(p + 16);
458 __ClearPageReserved(p);
459 set_page_count(p, 0);
460 }
461
462 arch_free_page(page, order);
463
464 mod_page_state(pgfree, 1 << order);
465
466 list_add(&page->lru, &list);
467 kernel_map_pages(page, 1 << order, 0);
468 free_pages_bulk(page_zone(page), 1, &list, order);
469 }
470}
471
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473/*
474 * The order of subdivision here is critical for the IO subsystem.
475 * Please do not alter this order without good reasons and regression
476 * testing. Specifically, as large blocks of memory are subdivided,
477 * the order in which smaller blocks are delivered depends on the order
478 * they're subdivided in this function. This is the primary factor
479 * influencing the order in which pages are delivered to the IO
480 * subsystem according to empirical testing, and this is also justified
481 * by considering the behavior of a buddy system containing a single
482 * large block of memory acted on by a series of small allocations.
483 * This behavior is a critical factor in sglist merging's success.
484 *
485 * -- wli
486 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800487static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 int low, int high, struct free_area *area)
489{
490 unsigned long size = 1 << high;
491
492 while (high > low) {
493 area--;
494 high--;
495 size >>= 1;
496 BUG_ON(bad_range(zone, &page[size]));
497 list_add(&page[size].lru, &area->free_list);
498 area->nr_free++;
499 set_page_order(&page[size], high);
500 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501}
502
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503/*
504 * This page is about to be returned from the page allocator
505 */
Hugh Dickins689bceb2005-11-21 21:32:20 -0800506static int prep_new_page(struct page *page, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507{
Nick Piggin92be2e32006-01-06 00:10:57 -0800508 if (unlikely(page_mapcount(page) |
509 (page->mapping != NULL) |
510 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700511 (page->flags & (
512 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 1 << PG_private |
514 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 1 << PG_active |
516 1 << PG_dirty |
517 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700518 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700520 1 << PG_writeback |
Nick Piggin92be2e32006-01-06 00:10:57 -0800521 1 << PG_reserved ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800522 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
Hugh Dickins689bceb2005-11-21 21:32:20 -0800524 /*
525 * For now, we report if PG_reserved was found set, but do not
526 * clear it, and do not allocate the page: as a safety net.
527 */
528 if (PageReserved(page))
529 return 1;
530
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
532 1 << PG_referenced | 1 << PG_arch_1 |
533 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700534 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 set_page_refs(page, order);
536 kernel_map_pages(page, 1 << order, 1);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800537 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538}
539
540/*
541 * Do the hard work of removing an element from the buddy allocator.
542 * Call me with the zone->lock already held.
543 */
544static struct page *__rmqueue(struct zone *zone, unsigned int order)
545{
546 struct free_area * area;
547 unsigned int current_order;
548 struct page *page;
549
550 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
551 area = zone->free_area + current_order;
552 if (list_empty(&area->free_list))
553 continue;
554
555 page = list_entry(area->free_list.next, struct page, lru);
556 list_del(&page->lru);
557 rmv_page_order(page);
558 area->nr_free--;
559 zone->free_pages -= 1UL << order;
Nick Piggin085cc7d2006-01-06 00:11:01 -0800560 expand(zone, page, order, current_order, area);
561 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 }
563
564 return NULL;
565}
566
567/*
568 * Obtain a specified number of elements from the buddy allocator, all under
569 * a single hold of the lock, for efficiency. Add them to the supplied list.
570 * Returns the number of new pages which were placed at *list.
571 */
572static int rmqueue_bulk(struct zone *zone, unsigned int order,
573 unsigned long count, struct list_head *list)
574{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
Nick Pigginc54ad302006-01-06 00:10:56 -0800577 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 for (i = 0; i < count; ++i) {
Nick Piggin085cc7d2006-01-06 00:11:01 -0800579 struct page *page = __rmqueue(zone, order);
580 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 list_add_tail(&page->lru, list);
583 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800584 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800585 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586}
587
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700588#ifdef CONFIG_NUMA
589/* Called from the slab reaper to drain remote pagesets */
590void drain_remote_pages(void)
591{
592 struct zone *zone;
593 int i;
594 unsigned long flags;
595
596 local_irq_save(flags);
597 for_each_zone(zone) {
598 struct per_cpu_pageset *pset;
599
600 /* Do not drain local pagesets */
601 if (zone->zone_pgdat->node_id == numa_node_id())
602 continue;
603
Nick Piggin23316bc2006-01-08 01:00:41 -0800604 pset = zone_pcp(zone, smp_processor_id());
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700605 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
606 struct per_cpu_pages *pcp;
607
608 pcp = &pset->pcp[i];
Nick Piggin48db57f2006-01-08 01:00:42 -0800609 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
610 pcp->count = 0;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700611 }
612 }
613 local_irq_restore(flags);
614}
615#endif
616
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
618static void __drain_pages(unsigned int cpu)
619{
Nick Pigginc54ad302006-01-06 00:10:56 -0800620 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 struct zone *zone;
622 int i;
623
624 for_each_zone(zone) {
625 struct per_cpu_pageset *pset;
626
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700627 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
629 struct per_cpu_pages *pcp;
630
631 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800632 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800633 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
634 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800635 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 }
637 }
638}
639#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
640
641#ifdef CONFIG_PM
642
643void mark_free_pages(struct zone *zone)
644{
645 unsigned long zone_pfn, flags;
646 int order;
647 struct list_head *curr;
648
649 if (!zone->spanned_pages)
650 return;
651
652 spin_lock_irqsave(&zone->lock, flags);
653 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
654 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
655
656 for (order = MAX_ORDER - 1; order >= 0; --order)
657 list_for_each(curr, &zone->free_area[order].free_list) {
658 unsigned long start_pfn, i;
659
660 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
661
662 for (i=0; i < (1<<order); i++)
663 SetPageNosaveFree(pfn_to_page(start_pfn+i));
664 }
665 spin_unlock_irqrestore(&zone->lock, flags);
666}
667
668/*
669 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
670 */
671void drain_local_pages(void)
672{
673 unsigned long flags;
674
675 local_irq_save(flags);
676 __drain_pages(smp_processor_id());
677 local_irq_restore(flags);
678}
679#endif /* CONFIG_PM */
680
Nick Piggina74609f2006-01-06 00:11:20 -0800681static void zone_statistics(struct zonelist *zonelist, struct zone *z, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682{
683#ifdef CONFIG_NUMA
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 pg_data_t *pg = z->zone_pgdat;
685 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
686 struct per_cpu_pageset *p;
687
Nick Piggina74609f2006-01-06 00:11:20 -0800688 p = zone_pcp(z, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 if (pg == orig) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700690 p->numa_hit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 } else {
692 p->numa_miss++;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700693 zone_pcp(zonelist->zones[0], cpu)->numa_foreign++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 }
695 if (pg == NODE_DATA(numa_node_id()))
696 p->local_node++;
697 else
698 p->other_node++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699#endif
700}
701
702/*
703 * Free a 0-order page
704 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705static void fastcall free_hot_cold_page(struct page *page, int cold)
706{
707 struct zone *zone = page_zone(page);
708 struct per_cpu_pages *pcp;
709 unsigned long flags;
710
711 arch_free_page(page, 0);
712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 if (PageAnon(page))
714 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800715 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800716 return;
717
Hugh Dickins689bceb2005-11-21 21:32:20 -0800718 kernel_map_pages(page, 1, 0);
719
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700720 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800722 __inc_page_state(pgfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 list_add(&page->lru, &pcp->list);
724 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800725 if (pcp->count >= pcp->high) {
726 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
727 pcp->count -= pcp->batch;
728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 local_irq_restore(flags);
730 put_cpu();
731}
732
733void fastcall free_hot_page(struct page *page)
734{
735 free_hot_cold_page(page, 0);
736}
737
738void fastcall free_cold_page(struct page *page)
739{
740 free_hot_cold_page(page, 1);
741}
742
Al Virodd0fc662005-10-07 07:46:04 +0100743static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
745 int i;
746
747 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
748 for(i = 0; i < (1 << order); i++)
749 clear_highpage(page + i);
750}
751
752/*
753 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
754 * we cheat by calling it from here, in the order > 0 path. Saves a branch
755 * or two.
756 */
Nick Piggina74609f2006-01-06 00:11:20 -0800757static struct page *buffered_rmqueue(struct zonelist *zonelist,
758 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759{
760 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800761 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800763 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Hugh Dickins689bceb2005-11-21 21:32:20 -0800765again:
Nick Piggina74609f2006-01-06 00:11:20 -0800766 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800767 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 struct per_cpu_pages *pcp;
769
Nick Piggina74609f2006-01-06 00:11:20 -0800770 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800772 if (!pcp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 pcp->count += rmqueue_bulk(zone, 0,
774 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800775 if (unlikely(!pcp->count))
776 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 }
Nick Piggina74609f2006-01-06 00:11:20 -0800778 page = list_entry(pcp->list.next, struct page, lru);
779 list_del(&page->lru);
780 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800781 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 spin_lock_irqsave(&zone->lock, flags);
783 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800784 spin_unlock(&zone->lock);
785 if (!page)
786 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 }
788
Nick Piggina74609f2006-01-06 00:11:20 -0800789 __mod_page_state_zone(zone, pgalloc, 1 << order);
790 zone_statistics(zonelist, zone, cpu);
791 local_irq_restore(flags);
792 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
Nick Piggina74609f2006-01-06 00:11:20 -0800794 BUG_ON(bad_range(zone, page));
795 if (prep_new_page(page, order))
796 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
Nick Piggina74609f2006-01-06 00:11:20 -0800798 if (gfp_flags & __GFP_ZERO)
799 prep_zero_page(page, order, gfp_flags);
800
801 if (order && (gfp_flags & __GFP_COMP))
802 prep_compound_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800804
805failed:
806 local_irq_restore(flags);
807 put_cpu();
808 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809}
810
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800811#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800812#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
813#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
814#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
815#define ALLOC_HARDER 0x10 /* try to alloc harder */
816#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
817#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800818
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819/*
820 * Return 1 if free pages are above 'mark'. This takes into account the order
821 * of the allocation.
822 */
823int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800824 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825{
826 /* free_pages my go negative - that's OK */
827 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
828 int o;
829
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800830 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800832 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 min -= min / 4;
834
835 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
836 return 0;
837 for (o = 0; o < order; o++) {
838 /* At the next order, this order's pages become unavailable */
839 free_pages -= z->free_area[o].nr_free << o;
840
841 /* Require fewer higher order pages to be free */
842 min >>= 1;
843
844 if (free_pages <= min)
845 return 0;
846 }
847 return 1;
848}
849
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800850/*
851 * get_page_from_freeliest goes through the zonelist trying to allocate
852 * a page.
853 */
854static struct page *
855get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
856 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -0700857{
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800858 struct zone **z = zonelist->zones;
859 struct page *page = NULL;
860 int classzone_idx = zone_idx(*z);
861
862 /*
863 * Go through the zonelist once, looking for a zone with enough free.
864 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
865 */
866 do {
867 if ((alloc_flags & ALLOC_CPUSET) &&
868 !cpuset_zone_allowed(*z, gfp_mask))
869 continue;
870
871 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -0800872 unsigned long mark;
873 if (alloc_flags & ALLOC_WMARK_MIN)
874 mark = (*z)->pages_min;
875 else if (alloc_flags & ALLOC_WMARK_LOW)
876 mark = (*z)->pages_low;
877 else
878 mark = (*z)->pages_high;
879 if (!zone_watermark_ok(*z, order, mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800880 classzone_idx, alloc_flags))
881 continue;
882 }
883
Nick Piggina74609f2006-01-06 00:11:20 -0800884 page = buffered_rmqueue(zonelist, *z, order, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800885 if (page) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800886 break;
887 }
888 } while (*(++z) != NULL);
889 return page;
Martin Hicks753ee722005-06-21 17:14:41 -0700890}
891
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892/*
893 * This is the 'heart' of the zoned buddy allocator.
894 */
895struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +0100896__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 struct zonelist *zonelist)
898{
Al Viro260b2362005-10-21 03:22:44 -0400899 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800900 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 struct page *page;
902 struct reclaim_state reclaim_state;
903 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800905 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 int did_some_progress;
907
908 might_sleep_if(wait);
909
Jens Axboe6b1de912005-11-17 21:35:02 +0100910restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800911 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800913 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 /* Should this ever happen?? */
915 return NULL;
916 }
Jens Axboe6b1de912005-11-17 21:35:02 +0100917
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800918 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -0800919 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800920 if (page)
921 goto got_pg;
922
Jens Axboe6b1de912005-11-17 21:35:02 +0100923 do {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800924 wakeup_kswapd(*z, order);
Jens Axboe6b1de912005-11-17 21:35:02 +0100925 } while (*(++z));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800926
Paul Jackson9bf22292005-09-06 15:18:12 -0700927 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800928 * OK, we're below the kswapd watermark and have kicked background
929 * reclaim. Now things get more complex, so set up alloc_flags according
930 * to how we want to proceed.
931 *
932 * The caller may dip into page reserves a bit more if the caller
933 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -0800934 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
935 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -0700936 */
Nick Piggin31488902005-11-28 13:44:03 -0800937 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800938 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
939 alloc_flags |= ALLOC_HARDER;
940 if (gfp_mask & __GFP_HIGH)
941 alloc_flags |= ALLOC_HIGH;
Paul Jackson47f3a862006-01-06 00:10:32 -0800942 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
944 /*
945 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800946 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 *
948 * This is the last chance, in general, before the goto nopage.
949 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -0700950 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800952 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
953 if (page)
954 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
956 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -0700957
958 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
959 && !in_interrupt()) {
960 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -0800961nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -0700962 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800963 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -0800964 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800965 if (page)
966 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -0800967 if (gfp_mask & __GFP_NOFAIL) {
968 blk_congestion_wait(WRITE, HZ/50);
969 goto nofail_alloc;
970 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 }
972 goto nopage;
973 }
974
975 /* Atomic allocations - we can't balance anything */
976 if (!wait)
977 goto nopage;
978
979rebalance:
980 cond_resched();
981
982 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -0800983 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 p->flags |= PF_MEMALLOC;
985 reclaim_state.reclaimed_slab = 0;
986 p->reclaim_state = &reclaim_state;
987
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800988 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
990 p->reclaim_state = NULL;
991 p->flags &= ~PF_MEMALLOC;
992
993 cond_resched();
994
995 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800996 page = get_page_from_freelist(gfp_mask, order,
997 zonelist, alloc_flags);
998 if (page)
999 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1001 /*
1002 * Go through the zonelist yet one more time, keep
1003 * very high watermark here, this is only to catch
1004 * a parallel oom killing, we must fail if we're still
1005 * under heavy pressure.
1006 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001007 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001008 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001009 if (page)
1010 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
Marcelo Tosatti79b9ce32005-07-07 17:56:04 -07001012 out_of_memory(gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 goto restart;
1014 }
1015
1016 /*
1017 * Don't let big-order allocations loop unless the caller explicitly
1018 * requests that. Wait for some write requests to complete then retry.
1019 *
1020 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1021 * <= 3, but that may not be true in other implementations.
1022 */
1023 do_retry = 0;
1024 if (!(gfp_mask & __GFP_NORETRY)) {
1025 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1026 do_retry = 1;
1027 if (gfp_mask & __GFP_NOFAIL)
1028 do_retry = 1;
1029 }
1030 if (do_retry) {
1031 blk_congestion_wait(WRITE, HZ/50);
1032 goto rebalance;
1033 }
1034
1035nopage:
1036 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1037 printk(KERN_WARNING "%s: page allocation failure."
1038 " order:%d, mode:0x%x\n",
1039 p->comm, order, gfp_mask);
1040 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001041 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 return page;
1045}
1046
1047EXPORT_SYMBOL(__alloc_pages);
1048
1049/*
1050 * Common helper functions.
1051 */
Al Virodd0fc662005-10-07 07:46:04 +01001052fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053{
1054 struct page * page;
1055 page = alloc_pages(gfp_mask, order);
1056 if (!page)
1057 return 0;
1058 return (unsigned long) page_address(page);
1059}
1060
1061EXPORT_SYMBOL(__get_free_pages);
1062
Al Virodd0fc662005-10-07 07:46:04 +01001063fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064{
1065 struct page * page;
1066
1067 /*
1068 * get_zeroed_page() returns a 32-bit address, which cannot represent
1069 * a highmem page
1070 */
Al Viro260b2362005-10-21 03:22:44 -04001071 BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072
1073 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1074 if (page)
1075 return (unsigned long) page_address(page);
1076 return 0;
1077}
1078
1079EXPORT_SYMBOL(get_zeroed_page);
1080
1081void __pagevec_free(struct pagevec *pvec)
1082{
1083 int i = pagevec_count(pvec);
1084
1085 while (--i >= 0)
1086 free_hot_cold_page(pvec->pages[i], pvec->cold);
1087}
1088
1089fastcall void __free_pages(struct page *page, unsigned int order)
1090{
Nick Pigginb5810032005-10-29 18:16:12 -07001091 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 if (order == 0)
1093 free_hot_page(page);
1094 else
1095 __free_pages_ok(page, order);
1096 }
1097}
1098
1099EXPORT_SYMBOL(__free_pages);
1100
1101fastcall void free_pages(unsigned long addr, unsigned int order)
1102{
1103 if (addr != 0) {
1104 BUG_ON(!virt_addr_valid((void *)addr));
1105 __free_pages(virt_to_page((void *)addr), order);
1106 }
1107}
1108
1109EXPORT_SYMBOL(free_pages);
1110
1111/*
1112 * Total amount of free (allocatable) RAM:
1113 */
1114unsigned int nr_free_pages(void)
1115{
1116 unsigned int sum = 0;
1117 struct zone *zone;
1118
1119 for_each_zone(zone)
1120 sum += zone->free_pages;
1121
1122 return sum;
1123}
1124
1125EXPORT_SYMBOL(nr_free_pages);
1126
1127#ifdef CONFIG_NUMA
1128unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1129{
1130 unsigned int i, sum = 0;
1131
1132 for (i = 0; i < MAX_NR_ZONES; i++)
1133 sum += pgdat->node_zones[i].free_pages;
1134
1135 return sum;
1136}
1137#endif
1138
1139static unsigned int nr_free_zone_pages(int offset)
1140{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001141 /* Just pick one node, since fallback list is circular */
1142 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 unsigned int sum = 0;
1144
Martin J. Blighe310fd42005-07-29 22:59:18 -07001145 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1146 struct zone **zonep = zonelist->zones;
1147 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148
Martin J. Blighe310fd42005-07-29 22:59:18 -07001149 for (zone = *zonep++; zone; zone = *zonep++) {
1150 unsigned long size = zone->present_pages;
1151 unsigned long high = zone->pages_high;
1152 if (size > high)
1153 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 }
1155
1156 return sum;
1157}
1158
1159/*
1160 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1161 */
1162unsigned int nr_free_buffer_pages(void)
1163{
Al Viroaf4ca452005-10-21 02:55:38 -04001164 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165}
1166
1167/*
1168 * Amount of free RAM allocatable within all zones
1169 */
1170unsigned int nr_free_pagecache_pages(void)
1171{
Al Viroaf4ca452005-10-21 02:55:38 -04001172 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173}
1174
1175#ifdef CONFIG_HIGHMEM
1176unsigned int nr_free_highpages (void)
1177{
1178 pg_data_t *pgdat;
1179 unsigned int pages = 0;
1180
1181 for_each_pgdat(pgdat)
1182 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1183
1184 return pages;
1185}
1186#endif
1187
1188#ifdef CONFIG_NUMA
1189static void show_node(struct zone *zone)
1190{
1191 printk("Node %d ", zone->zone_pgdat->node_id);
1192}
1193#else
1194#define show_node(zone) do { } while (0)
1195#endif
1196
1197/*
1198 * Accumulate the page_state information across all CPUs.
1199 * The result is unavoidably approximate - it can change
1200 * during and after execution of this function.
1201 */
1202static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1203
1204atomic_t nr_pagecache = ATOMIC_INIT(0);
1205EXPORT_SYMBOL(nr_pagecache);
1206#ifdef CONFIG_SMP
1207DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1208#endif
1209
Nick Piggina86b1f52006-01-06 00:11:00 -08001210static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211{
1212 int cpu = 0;
1213
1214 memset(ret, 0, sizeof(*ret));
Andrew Morton84c20082006-01-08 01:00:28 -08001215 cpus_and(*cpumask, *cpumask, cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216
Martin Hicksc07e02d2005-09-03 15:55:11 -07001217 cpu = first_cpu(*cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 while (cpu < NR_CPUS) {
1219 unsigned long *in, *out, off;
1220
1221 in = (unsigned long *)&per_cpu(page_states, cpu);
1222
Martin Hicksc07e02d2005-09-03 15:55:11 -07001223 cpu = next_cpu(cpu, *cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224
1225 if (cpu < NR_CPUS)
1226 prefetch(&per_cpu(page_states, cpu));
1227
1228 out = (unsigned long *)ret;
1229 for (off = 0; off < nr; off++)
1230 *out++ += *in++;
1231 }
1232}
1233
Martin Hicksc07e02d2005-09-03 15:55:11 -07001234void get_page_state_node(struct page_state *ret, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235{
1236 int nr;
Martin Hicksc07e02d2005-09-03 15:55:11 -07001237 cpumask_t mask = node_to_cpumask(node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
1239 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1240 nr /= sizeof(unsigned long);
1241
Martin Hicksc07e02d2005-09-03 15:55:11 -07001242 __get_page_state(ret, nr+1, &mask);
1243}
1244
1245void get_page_state(struct page_state *ret)
1246{
1247 int nr;
1248 cpumask_t mask = CPU_MASK_ALL;
1249
1250 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1251 nr /= sizeof(unsigned long);
1252
1253 __get_page_state(ret, nr + 1, &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254}
1255
1256void get_full_page_state(struct page_state *ret)
1257{
Martin Hicksc07e02d2005-09-03 15:55:11 -07001258 cpumask_t mask = CPU_MASK_ALL;
1259
1260 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261}
1262
Nick Piggina74609f2006-01-06 00:11:20 -08001263unsigned long read_page_state_offset(unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264{
1265 unsigned long ret = 0;
1266 int cpu;
1267
Andrew Morton84c20082006-01-08 01:00:28 -08001268 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 unsigned long in;
1270
1271 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1272 ret += *((unsigned long *)in);
1273 }
1274 return ret;
1275}
1276
Nick Piggina74609f2006-01-06 00:11:20 -08001277void __mod_page_state_offset(unsigned long offset, unsigned long delta)
1278{
1279 void *ptr;
1280
1281 ptr = &__get_cpu_var(page_states);
1282 *(unsigned long *)(ptr + offset) += delta;
1283}
1284EXPORT_SYMBOL(__mod_page_state_offset);
1285
1286void mod_page_state_offset(unsigned long offset, unsigned long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287{
1288 unsigned long flags;
Nick Piggina74609f2006-01-06 00:11:20 -08001289 void *ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
1291 local_irq_save(flags);
1292 ptr = &__get_cpu_var(page_states);
Nick Piggina74609f2006-01-06 00:11:20 -08001293 *(unsigned long *)(ptr + offset) += delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 local_irq_restore(flags);
1295}
Nick Piggina74609f2006-01-06 00:11:20 -08001296EXPORT_SYMBOL(mod_page_state_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
1298void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1299 unsigned long *free, struct pglist_data *pgdat)
1300{
1301 struct zone *zones = pgdat->node_zones;
1302 int i;
1303
1304 *active = 0;
1305 *inactive = 0;
1306 *free = 0;
1307 for (i = 0; i < MAX_NR_ZONES; i++) {
1308 *active += zones[i].nr_active;
1309 *inactive += zones[i].nr_inactive;
1310 *free += zones[i].free_pages;
1311 }
1312}
1313
1314void get_zone_counts(unsigned long *active,
1315 unsigned long *inactive, unsigned long *free)
1316{
1317 struct pglist_data *pgdat;
1318
1319 *active = 0;
1320 *inactive = 0;
1321 *free = 0;
1322 for_each_pgdat(pgdat) {
1323 unsigned long l, m, n;
1324 __get_zone_counts(&l, &m, &n, pgdat);
1325 *active += l;
1326 *inactive += m;
1327 *free += n;
1328 }
1329}
1330
1331void si_meminfo(struct sysinfo *val)
1332{
1333 val->totalram = totalram_pages;
1334 val->sharedram = 0;
1335 val->freeram = nr_free_pages();
1336 val->bufferram = nr_blockdev_pages();
1337#ifdef CONFIG_HIGHMEM
1338 val->totalhigh = totalhigh_pages;
1339 val->freehigh = nr_free_highpages();
1340#else
1341 val->totalhigh = 0;
1342 val->freehigh = 0;
1343#endif
1344 val->mem_unit = PAGE_SIZE;
1345}
1346
1347EXPORT_SYMBOL(si_meminfo);
1348
1349#ifdef CONFIG_NUMA
1350void si_meminfo_node(struct sysinfo *val, int nid)
1351{
1352 pg_data_t *pgdat = NODE_DATA(nid);
1353
1354 val->totalram = pgdat->node_present_pages;
1355 val->freeram = nr_free_pages_pgdat(pgdat);
1356 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1357 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1358 val->mem_unit = PAGE_SIZE;
1359}
1360#endif
1361
1362#define K(x) ((x) << (PAGE_SHIFT-10))
1363
1364/*
1365 * Show free area list (used inside shift_scroll-lock stuff)
1366 * We also calculate the percentage fragmentation. We do this by counting the
1367 * memory on each free list with the exception of the first item on the list.
1368 */
1369void show_free_areas(void)
1370{
1371 struct page_state ps;
1372 int cpu, temperature;
1373 unsigned long active;
1374 unsigned long inactive;
1375 unsigned long free;
1376 struct zone *zone;
1377
1378 for_each_zone(zone) {
1379 show_node(zone);
1380 printk("%s per-cpu:", zone->name);
1381
Con Kolivasf3fe6512006-01-06 00:11:15 -08001382 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 printk(" empty\n");
1384 continue;
1385 } else
1386 printk("\n");
1387
Dave Jones6b482c62005-11-10 15:45:56 -05001388 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 struct per_cpu_pageset *pageset;
1390
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001391 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
1393 for (temperature = 0; temperature < 2; temperature++)
Nick Piggin2d92c5c2006-01-06 00:10:59 -08001394 printk("cpu %d %s: high %d, batch %d used:%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 cpu,
1396 temperature ? "cold" : "hot",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 pageset->pcp[temperature].high,
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001398 pageset->pcp[temperature].batch,
1399 pageset->pcp[temperature].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 }
1401 }
1402
1403 get_page_state(&ps);
1404 get_zone_counts(&active, &inactive, &free);
1405
Denis Vlasenkoc0d62212005-06-21 17:15:14 -07001406 printk("Free pages: %11ukB (%ukB HighMem)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 K(nr_free_pages()),
1408 K(nr_free_highpages()));
1409
1410 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1411 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1412 active,
1413 inactive,
1414 ps.nr_dirty,
1415 ps.nr_writeback,
1416 ps.nr_unstable,
1417 nr_free_pages(),
1418 ps.nr_slab,
1419 ps.nr_mapped,
1420 ps.nr_page_table_pages);
1421
1422 for_each_zone(zone) {
1423 int i;
1424
1425 show_node(zone);
1426 printk("%s"
1427 " free:%lukB"
1428 " min:%lukB"
1429 " low:%lukB"
1430 " high:%lukB"
1431 " active:%lukB"
1432 " inactive:%lukB"
1433 " present:%lukB"
1434 " pages_scanned:%lu"
1435 " all_unreclaimable? %s"
1436 "\n",
1437 zone->name,
1438 K(zone->free_pages),
1439 K(zone->pages_min),
1440 K(zone->pages_low),
1441 K(zone->pages_high),
1442 K(zone->nr_active),
1443 K(zone->nr_inactive),
1444 K(zone->present_pages),
1445 zone->pages_scanned,
1446 (zone->all_unreclaimable ? "yes" : "no")
1447 );
1448 printk("lowmem_reserve[]:");
1449 for (i = 0; i < MAX_NR_ZONES; i++)
1450 printk(" %lu", zone->lowmem_reserve[i]);
1451 printk("\n");
1452 }
1453
1454 for_each_zone(zone) {
1455 unsigned long nr, flags, order, total = 0;
1456
1457 show_node(zone);
1458 printk("%s: ", zone->name);
Con Kolivasf3fe6512006-01-06 00:11:15 -08001459 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 printk("empty\n");
1461 continue;
1462 }
1463
1464 spin_lock_irqsave(&zone->lock, flags);
1465 for (order = 0; order < MAX_ORDER; order++) {
1466 nr = zone->free_area[order].nr_free;
1467 total += nr << order;
1468 printk("%lu*%lukB ", nr, K(1UL) << order);
1469 }
1470 spin_unlock_irqrestore(&zone->lock, flags);
1471 printk("= %lukB\n", K(total));
1472 }
1473
1474 show_swap_cache_info();
1475}
1476
1477/*
1478 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001479 *
1480 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 */
Christoph Lameter1a932052006-01-06 00:11:16 -08001482static int __init build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter070f8032006-01-06 00:11:19 -08001483 struct zonelist *zonelist, int nr_zones, int zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484{
Christoph Lameter1a932052006-01-06 00:11:16 -08001485 struct zone *zone;
1486
Christoph Lameter070f8032006-01-06 00:11:19 -08001487 BUG_ON(zone_type > ZONE_HIGHMEM);
Christoph Lameter02a68a52006-01-06 00:11:18 -08001488
1489 do {
Christoph Lameter070f8032006-01-06 00:11:19 -08001490 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001491 if (populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492#ifndef CONFIG_HIGHMEM
Christoph Lameter070f8032006-01-06 00:11:19 -08001493 BUG_ON(zone_type > ZONE_NORMAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494#endif
Christoph Lameter070f8032006-01-06 00:11:19 -08001495 zonelist->zones[nr_zones++] = zone;
1496 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 }
Christoph Lameter070f8032006-01-06 00:11:19 -08001498 zone_type--;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001499
Christoph Lameter070f8032006-01-06 00:11:19 -08001500 } while (zone_type >= 0);
1501 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502}
1503
Al Viro260b2362005-10-21 03:22:44 -04001504static inline int highest_zone(int zone_bits)
1505{
1506 int res = ZONE_NORMAL;
1507 if (zone_bits & (__force int)__GFP_HIGHMEM)
1508 res = ZONE_HIGHMEM;
Andi Kleena2f1b422005-11-05 17:25:53 +01001509 if (zone_bits & (__force int)__GFP_DMA32)
1510 res = ZONE_DMA32;
Al Viro260b2362005-10-21 03:22:44 -04001511 if (zone_bits & (__force int)__GFP_DMA)
1512 res = ZONE_DMA;
1513 return res;
1514}
1515
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516#ifdef CONFIG_NUMA
1517#define MAX_NODE_LOAD (num_online_nodes())
1518static int __initdata node_load[MAX_NUMNODES];
1519/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001520 * 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 -07001521 * @node: node whose fallback list we're appending
1522 * @used_node_mask: nodemask_t of already used nodes
1523 *
1524 * We use a number of factors to determine which is the next node that should
1525 * appear on a given node's fallback list. The node should not have appeared
1526 * already in @node's fallback list, and it should be the next closest node
1527 * according to the distance array (which contains arbitrary distance values
1528 * from each node to each node in the system), and should also prefer nodes
1529 * with no CPUs, since presumably they'll have very little allocation pressure
1530 * on them otherwise.
1531 * It returns -1 if no node is found.
1532 */
1533static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1534{
1535 int i, n, val;
1536 int min_val = INT_MAX;
1537 int best_node = -1;
1538
1539 for_each_online_node(i) {
1540 cpumask_t tmp;
1541
1542 /* Start from local node */
1543 n = (node+i) % num_online_nodes();
1544
1545 /* Don't want a node to appear more than once */
1546 if (node_isset(n, *used_node_mask))
1547 continue;
1548
1549 /* Use the local node if we haven't already */
1550 if (!node_isset(node, *used_node_mask)) {
1551 best_node = node;
1552 break;
1553 }
1554
1555 /* Use the distance array to find the distance */
1556 val = node_distance(node, n);
1557
1558 /* Give preference to headless and unused nodes */
1559 tmp = node_to_cpumask(n);
1560 if (!cpus_empty(tmp))
1561 val += PENALTY_FOR_NODE_WITH_CPUS;
1562
1563 /* Slight preference for less loaded node */
1564 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1565 val += node_load[n];
1566
1567 if (val < min_val) {
1568 min_val = val;
1569 best_node = n;
1570 }
1571 }
1572
1573 if (best_node >= 0)
1574 node_set(best_node, *used_node_mask);
1575
1576 return best_node;
1577}
1578
1579static void __init build_zonelists(pg_data_t *pgdat)
1580{
1581 int i, j, k, node, local_node;
1582 int prev_node, load;
1583 struct zonelist *zonelist;
1584 nodemask_t used_mask;
1585
1586 /* initialize zonelists */
1587 for (i = 0; i < GFP_ZONETYPES; i++) {
1588 zonelist = pgdat->node_zonelists + i;
1589 zonelist->zones[0] = NULL;
1590 }
1591
1592 /* NUMA-aware ordering of nodes */
1593 local_node = pgdat->node_id;
1594 load = num_online_nodes();
1595 prev_node = local_node;
1596 nodes_clear(used_mask);
1597 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1598 /*
1599 * We don't want to pressure a particular node.
1600 * So adding penalty to the first node in same
1601 * distance group to make it round-robin.
1602 */
1603 if (node_distance(local_node, node) !=
1604 node_distance(local_node, prev_node))
1605 node_load[node] += load;
1606 prev_node = node;
1607 load--;
1608 for (i = 0; i < GFP_ZONETYPES; i++) {
1609 zonelist = pgdat->node_zonelists + i;
1610 for (j = 0; zonelist->zones[j] != NULL; j++);
1611
Al Viro260b2362005-10-21 03:22:44 -04001612 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613
1614 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1615 zonelist->zones[j] = NULL;
1616 }
1617 }
1618}
1619
1620#else /* CONFIG_NUMA */
1621
1622static void __init build_zonelists(pg_data_t *pgdat)
1623{
1624 int i, j, k, node, local_node;
1625
1626 local_node = pgdat->node_id;
1627 for (i = 0; i < GFP_ZONETYPES; i++) {
1628 struct zonelist *zonelist;
1629
1630 zonelist = pgdat->node_zonelists + i;
1631
1632 j = 0;
Al Viro260b2362005-10-21 03:22:44 -04001633 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 j = build_zonelists_node(pgdat, zonelist, j, k);
1635 /*
1636 * Now we build the zonelist so that it contains the zones
1637 * of all the other nodes.
1638 * We don't want to pressure a particular node, so when
1639 * building the zones for node N, we make sure that the
1640 * zones coming right after the local ones are those from
1641 * node N+1 (modulo N)
1642 */
1643 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1644 if (!node_online(node))
1645 continue;
1646 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1647 }
1648 for (node = 0; node < local_node; node++) {
1649 if (!node_online(node))
1650 continue;
1651 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1652 }
1653
1654 zonelist->zones[j] = NULL;
1655 }
1656}
1657
1658#endif /* CONFIG_NUMA */
1659
1660void __init build_all_zonelists(void)
1661{
1662 int i;
1663
1664 for_each_online_node(i)
1665 build_zonelists(NODE_DATA(i));
1666 printk("Built %i zonelists\n", num_online_nodes());
1667 cpuset_init_current_mems_allowed();
1668}
1669
1670/*
1671 * Helper functions to size the waitqueue hash table.
1672 * Essentially these want to choose hash table sizes sufficiently
1673 * large so that collisions trying to wait on pages are rare.
1674 * But in fact, the number of active page waitqueues on typical
1675 * systems is ridiculously low, less than 200. So this is even
1676 * conservative, even though it seems large.
1677 *
1678 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1679 * waitqueues, i.e. the size of the waitq table given the number of pages.
1680 */
1681#define PAGES_PER_WAITQUEUE 256
1682
1683static inline unsigned long wait_table_size(unsigned long pages)
1684{
1685 unsigned long size = 1;
1686
1687 pages /= PAGES_PER_WAITQUEUE;
1688
1689 while (size < pages)
1690 size <<= 1;
1691
1692 /*
1693 * Once we have dozens or even hundreds of threads sleeping
1694 * on IO we've got bigger problems than wait queue collision.
1695 * Limit the size of the wait table to a reasonable size.
1696 */
1697 size = min(size, 4096UL);
1698
1699 return max(size, 4UL);
1700}
1701
1702/*
1703 * This is an integer logarithm so that shifts can be used later
1704 * to extract the more random high bits from the multiplicative
1705 * hash function before the remainder is taken.
1706 */
1707static inline unsigned long wait_table_bits(unsigned long size)
1708{
1709 return ffz(~size);
1710}
1711
1712#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1713
1714static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1715 unsigned long *zones_size, unsigned long *zholes_size)
1716{
1717 unsigned long realtotalpages, totalpages = 0;
1718 int i;
1719
1720 for (i = 0; i < MAX_NR_ZONES; i++)
1721 totalpages += zones_size[i];
1722 pgdat->node_spanned_pages = totalpages;
1723
1724 realtotalpages = totalpages;
1725 if (zholes_size)
1726 for (i = 0; i < MAX_NR_ZONES; i++)
1727 realtotalpages -= zholes_size[i];
1728 pgdat->node_present_pages = realtotalpages;
1729 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1730}
1731
1732
1733/*
1734 * Initially all pages are reserved - free ones are freed
1735 * up by free_all_bootmem() once the early boot process is
1736 * done. Non-atomic initialization, single-pass.
1737 */
Dave Hansen3947be12005-10-29 18:16:54 -07001738void __devinit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 unsigned long start_pfn)
1740{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001742 unsigned long end_pfn = start_pfn + size;
1743 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001745 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001746 if (!early_pfn_valid(pfn))
1747 continue;
1748 page = pfn_to_page(pfn);
1749 set_page_links(page, zone, nid, pfn);
Nick Pigginb5810032005-10-29 18:16:12 -07001750 set_page_count(page, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 reset_page_mapcount(page);
1752 SetPageReserved(page);
1753 INIT_LIST_HEAD(&page->lru);
1754#ifdef WANT_PAGE_VIRTUAL
1755 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1756 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001757 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 }
1760}
1761
1762void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1763 unsigned long size)
1764{
1765 int order;
1766 for (order = 0; order < MAX_ORDER ; order++) {
1767 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1768 zone->free_area[order].nr_free = 0;
1769 }
1770}
1771
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001772#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1773void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1774 unsigned long size)
1775{
1776 unsigned long snum = pfn_to_section_nr(pfn);
1777 unsigned long end = pfn_to_section_nr(pfn + size);
1778
1779 if (FLAGS_HAS_NODE)
1780 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1781 else
1782 for (; snum <= end; snum++)
1783 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1784}
1785
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786#ifndef __HAVE_ARCH_MEMMAP_INIT
1787#define memmap_init(size, nid, zone, start_pfn) \
1788 memmap_init_zone((size), (nid), (zone), (start_pfn))
1789#endif
1790
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001791static int __devinit zone_batchsize(struct zone *zone)
1792{
1793 int batch;
1794
1795 /*
1796 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001797 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001798 *
1799 * OK, so we don't know how big the cache is. So guess.
1800 */
1801 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001802 if (batch * PAGE_SIZE > 512 * 1024)
1803 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001804 batch /= 4; /* We effectively *= 4 below */
1805 if (batch < 1)
1806 batch = 1;
1807
1808 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001809 * Clamp the batch to a 2^n - 1 value. Having a power
1810 * of 2 value was found to be more likely to have
1811 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001812 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001813 * For example if 2 tasks are alternately allocating
1814 * batches of pages, one task can end up with a lot
1815 * of pages of one half of the possible page colors
1816 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001817 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001818 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001819
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001820 return batch;
1821}
1822
Christoph Lameter2caaad42005-06-21 17:15:00 -07001823inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1824{
1825 struct per_cpu_pages *pcp;
1826
Magnus Damm1c6fe942005-10-26 01:58:59 -07001827 memset(p, 0, sizeof(*p));
1828
Christoph Lameter2caaad42005-06-21 17:15:00 -07001829 pcp = &p->pcp[0]; /* hot */
1830 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001831 pcp->high = 6 * batch;
1832 pcp->batch = max(1UL, 1 * batch);
1833 INIT_LIST_HEAD(&pcp->list);
1834
1835 pcp = &p->pcp[1]; /* cold*/
1836 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001837 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001838 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001839 INIT_LIST_HEAD(&pcp->list);
1840}
1841
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001842/*
1843 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1844 * to the value high for the pageset p.
1845 */
1846
1847static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1848 unsigned long high)
1849{
1850 struct per_cpu_pages *pcp;
1851
1852 pcp = &p->pcp[0]; /* hot list */
1853 pcp->high = high;
1854 pcp->batch = max(1UL, high/4);
1855 if ((high/4) > (PAGE_SHIFT * 8))
1856 pcp->batch = PAGE_SHIFT * 8;
1857}
1858
1859
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001860#ifdef CONFIG_NUMA
1861/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001862 * Boot pageset table. One per cpu which is going to be used for all
1863 * zones and all nodes. The parameters will be set in such a way
1864 * that an item put on a list will immediately be handed over to
1865 * the buddy list. This is safe since pageset manipulation is done
1866 * with interrupts disabled.
1867 *
1868 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001869 *
1870 * The boot_pagesets must be kept even after bootup is complete for
1871 * unused processors and/or zones. They do play a role for bootstrapping
1872 * hotplugged processors.
1873 *
1874 * zoneinfo_show() and maybe other functions do
1875 * not check if the processor is online before following the pageset pointer.
1876 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001877 */
1878static struct per_cpu_pageset
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001879 boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001880
1881/*
1882 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001883 * per cpu pageset array in struct zone.
1884 */
1885static int __devinit process_zones(int cpu)
1886{
1887 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001888
1889 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001890
Nick Piggin23316bc2006-01-08 01:00:41 -08001891 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001892 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08001893 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001894 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001895
Nick Piggin23316bc2006-01-08 01:00:41 -08001896 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001897
1898 if (percpu_pagelist_fraction)
1899 setup_pagelist_highmark(zone_pcp(zone, cpu),
1900 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001901 }
1902
1903 return 0;
1904bad:
1905 for_each_zone(dzone) {
1906 if (dzone == zone)
1907 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08001908 kfree(zone_pcp(dzone, cpu));
1909 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001910 }
1911 return -ENOMEM;
1912}
1913
1914static inline void free_zone_pagesets(int cpu)
1915{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001916 struct zone *zone;
1917
1918 for_each_zone(zone) {
1919 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1920
1921 zone_pcp(zone, cpu) = NULL;
1922 kfree(pset);
1923 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001924}
1925
1926static int __devinit pageset_cpuup_callback(struct notifier_block *nfb,
1927 unsigned long action,
1928 void *hcpu)
1929{
1930 int cpu = (long)hcpu;
1931 int ret = NOTIFY_OK;
1932
1933 switch (action) {
1934 case CPU_UP_PREPARE:
1935 if (process_zones(cpu))
1936 ret = NOTIFY_BAD;
1937 break;
Andi Kleenb0d41692005-11-05 17:25:53 +01001938 case CPU_UP_CANCELED:
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001939 case CPU_DEAD:
1940 free_zone_pagesets(cpu);
1941 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001942 default:
1943 break;
1944 }
1945 return ret;
1946}
1947
1948static struct notifier_block pageset_notifier =
1949 { &pageset_cpuup_callback, NULL, 0 };
1950
Al Viro78d99552005-12-15 09:18:25 +00001951void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001952{
1953 int err;
1954
1955 /* Initialize per_cpu_pageset for cpu 0.
1956 * A cpuup callback will do this for every cpu
1957 * as it comes online
1958 */
1959 err = process_zones(smp_processor_id());
1960 BUG_ON(err);
1961 register_cpu_notifier(&pageset_notifier);
1962}
1963
1964#endif
1965
Dave Hansened8ece22005-10-29 18:16:50 -07001966static __devinit
1967void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
1968{
1969 int i;
1970 struct pglist_data *pgdat = zone->zone_pgdat;
1971
1972 /*
1973 * The per-page waitqueue mechanism uses hashed waitqueues
1974 * per zone.
1975 */
1976 zone->wait_table_size = wait_table_size(zone_size_pages);
1977 zone->wait_table_bits = wait_table_bits(zone->wait_table_size);
1978 zone->wait_table = (wait_queue_head_t *)
1979 alloc_bootmem_node(pgdat, zone->wait_table_size
1980 * sizeof(wait_queue_head_t));
1981
1982 for(i = 0; i < zone->wait_table_size; ++i)
1983 init_waitqueue_head(zone->wait_table + i);
1984}
1985
1986static __devinit void zone_pcp_init(struct zone *zone)
1987{
1988 int cpu;
1989 unsigned long batch = zone_batchsize(zone);
1990
1991 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1992#ifdef CONFIG_NUMA
1993 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08001994 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07001995 setup_pageset(&boot_pageset[cpu],0);
1996#else
1997 setup_pageset(zone_pcp(zone,cpu), batch);
1998#endif
1999 }
2000 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2001 zone->name, zone->present_pages, batch);
2002}
2003
2004static __devinit void init_currently_empty_zone(struct zone *zone,
2005 unsigned long zone_start_pfn, unsigned long size)
2006{
2007 struct pglist_data *pgdat = zone->zone_pgdat;
2008
2009 zone_wait_table_init(zone, size);
2010 pgdat->nr_zones = zone_idx(zone) + 1;
2011
2012 zone->zone_mem_map = pfn_to_page(zone_start_pfn);
2013 zone->zone_start_pfn = zone_start_pfn;
2014
2015 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2016
2017 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2018}
2019
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020/*
2021 * Set up the zone data structures:
2022 * - mark all pages reserved
2023 * - mark all memory queues empty
2024 * - clear the memory bitmaps
2025 */
2026static void __init free_area_init_core(struct pglist_data *pgdat,
2027 unsigned long *zones_size, unsigned long *zholes_size)
2028{
Dave Hansened8ece22005-10-29 18:16:50 -07002029 unsigned long j;
2030 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 unsigned long zone_start_pfn = pgdat->node_start_pfn;
2032
Dave Hansen208d54e2005-10-29 18:16:52 -07002033 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 pgdat->nr_zones = 0;
2035 init_waitqueue_head(&pgdat->kswapd_wait);
2036 pgdat->kswapd_max_order = 0;
2037
2038 for (j = 0; j < MAX_NR_ZONES; j++) {
2039 struct zone *zone = pgdat->node_zones + j;
2040 unsigned long size, realsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 realsize = size = zones_size[j];
2043 if (zholes_size)
2044 realsize -= zholes_size[j];
2045
Andi Kleena2f1b422005-11-05 17:25:53 +01002046 if (j < ZONE_HIGHMEM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 nr_kernel_pages += realsize;
2048 nr_all_pages += realsize;
2049
2050 zone->spanned_pages = size;
2051 zone->present_pages = realsize;
2052 zone->name = zone_names[j];
2053 spin_lock_init(&zone->lock);
2054 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002055 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 zone->zone_pgdat = pgdat;
2057 zone->free_pages = 0;
2058
2059 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2060
Dave Hansened8ece22005-10-29 18:16:50 -07002061 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 INIT_LIST_HEAD(&zone->active_list);
2063 INIT_LIST_HEAD(&zone->inactive_list);
2064 zone->nr_scan_active = 0;
2065 zone->nr_scan_inactive = 0;
2066 zone->nr_active = 0;
2067 zone->nr_inactive = 0;
Martin Hicks53e9a612005-09-03 15:54:51 -07002068 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 if (!size)
2070 continue;
2071
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002072 zonetable_add(zone, nid, j, zone_start_pfn, size);
Dave Hansened8ece22005-10-29 18:16:50 -07002073 init_currently_empty_zone(zone, zone_start_pfn, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 }
2076}
2077
2078static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2079{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 /* Skip empty nodes */
2081 if (!pgdat->node_spanned_pages)
2082 return;
2083
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002084#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 /* ia64 gets its own node_mem_map, before this, without bootmem */
2086 if (!pgdat->node_mem_map) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002087 unsigned long size;
2088 struct page *map;
2089
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002091 map = alloc_remap(pgdat->node_id, size);
2092 if (!map)
2093 map = alloc_bootmem_node(pgdat, size);
2094 pgdat->node_mem_map = map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002096#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 /*
2098 * With no DISCONTIG, the global mem_map is just set as node 0's
2099 */
2100 if (pgdat == NODE_DATA(0))
2101 mem_map = NODE_DATA(0)->node_mem_map;
2102#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002103#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104}
2105
2106void __init free_area_init_node(int nid, struct pglist_data *pgdat,
2107 unsigned long *zones_size, unsigned long node_start_pfn,
2108 unsigned long *zholes_size)
2109{
2110 pgdat->node_id = nid;
2111 pgdat->node_start_pfn = node_start_pfn;
2112 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2113
2114 alloc_node_mem_map(pgdat);
2115
2116 free_area_init_core(pgdat, zones_size, zholes_size);
2117}
2118
Dave Hansen93b75042005-06-23 00:07:47 -07002119#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120static bootmem_data_t contig_bootmem_data;
2121struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2122
2123EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002124#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125
2126void __init free_area_init(unsigned long *zones_size)
2127{
Dave Hansen93b75042005-06-23 00:07:47 -07002128 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2130}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
2132#ifdef CONFIG_PROC_FS
2133
2134#include <linux/seq_file.h>
2135
2136static void *frag_start(struct seq_file *m, loff_t *pos)
2137{
2138 pg_data_t *pgdat;
2139 loff_t node = *pos;
2140
2141 for (pgdat = pgdat_list; pgdat && node; pgdat = pgdat->pgdat_next)
2142 --node;
2143
2144 return pgdat;
2145}
2146
2147static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
2148{
2149 pg_data_t *pgdat = (pg_data_t *)arg;
2150
2151 (*pos)++;
2152 return pgdat->pgdat_next;
2153}
2154
2155static void frag_stop(struct seq_file *m, void *arg)
2156{
2157}
2158
2159/*
2160 * This walks the free areas for each zone.
2161 */
2162static int frag_show(struct seq_file *m, void *arg)
2163{
2164 pg_data_t *pgdat = (pg_data_t *)arg;
2165 struct zone *zone;
2166 struct zone *node_zones = pgdat->node_zones;
2167 unsigned long flags;
2168 int order;
2169
2170 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
Con Kolivasf3fe6512006-01-06 00:11:15 -08002171 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 continue;
2173
2174 spin_lock_irqsave(&zone->lock, flags);
2175 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2176 for (order = 0; order < MAX_ORDER; ++order)
2177 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2178 spin_unlock_irqrestore(&zone->lock, flags);
2179 seq_putc(m, '\n');
2180 }
2181 return 0;
2182}
2183
2184struct seq_operations fragmentation_op = {
2185 .start = frag_start,
2186 .next = frag_next,
2187 .stop = frag_stop,
2188 .show = frag_show,
2189};
2190
Nikita Danilov295ab932005-06-21 17:14:38 -07002191/*
2192 * Output information about zones in @pgdat.
2193 */
2194static int zoneinfo_show(struct seq_file *m, void *arg)
2195{
2196 pg_data_t *pgdat = arg;
2197 struct zone *zone;
2198 struct zone *node_zones = pgdat->node_zones;
2199 unsigned long flags;
2200
2201 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2202 int i;
2203
Con Kolivasf3fe6512006-01-06 00:11:15 -08002204 if (!populated_zone(zone))
Nikita Danilov295ab932005-06-21 17:14:38 -07002205 continue;
2206
2207 spin_lock_irqsave(&zone->lock, flags);
2208 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2209 seq_printf(m,
2210 "\n pages free %lu"
2211 "\n min %lu"
2212 "\n low %lu"
2213 "\n high %lu"
2214 "\n active %lu"
2215 "\n inactive %lu"
2216 "\n scanned %lu (a: %lu i: %lu)"
2217 "\n spanned %lu"
2218 "\n present %lu",
2219 zone->free_pages,
2220 zone->pages_min,
2221 zone->pages_low,
2222 zone->pages_high,
2223 zone->nr_active,
2224 zone->nr_inactive,
2225 zone->pages_scanned,
2226 zone->nr_scan_active, zone->nr_scan_inactive,
2227 zone->spanned_pages,
2228 zone->present_pages);
2229 seq_printf(m,
2230 "\n protection: (%lu",
2231 zone->lowmem_reserve[0]);
2232 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2233 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2234 seq_printf(m,
2235 ")"
2236 "\n pagesets");
Nick Piggin23316bc2006-01-08 01:00:41 -08002237 for_each_online_cpu(i) {
Nikita Danilov295ab932005-06-21 17:14:38 -07002238 struct per_cpu_pageset *pageset;
2239 int j;
2240
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002241 pageset = zone_pcp(zone, i);
Nikita Danilov295ab932005-06-21 17:14:38 -07002242 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2243 if (pageset->pcp[j].count)
2244 break;
2245 }
2246 if (j == ARRAY_SIZE(pageset->pcp))
2247 continue;
2248 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2249 seq_printf(m,
2250 "\n cpu: %i pcp: %i"
2251 "\n count: %i"
Nikita Danilov295ab932005-06-21 17:14:38 -07002252 "\n high: %i"
2253 "\n batch: %i",
2254 i, j,
2255 pageset->pcp[j].count,
Nikita Danilov295ab932005-06-21 17:14:38 -07002256 pageset->pcp[j].high,
2257 pageset->pcp[j].batch);
2258 }
2259#ifdef CONFIG_NUMA
2260 seq_printf(m,
2261 "\n numa_hit: %lu"
2262 "\n numa_miss: %lu"
2263 "\n numa_foreign: %lu"
2264 "\n interleave_hit: %lu"
2265 "\n local_node: %lu"
2266 "\n other_node: %lu",
2267 pageset->numa_hit,
2268 pageset->numa_miss,
2269 pageset->numa_foreign,
2270 pageset->interleave_hit,
2271 pageset->local_node,
2272 pageset->other_node);
2273#endif
2274 }
2275 seq_printf(m,
2276 "\n all_unreclaimable: %u"
2277 "\n prev_priority: %i"
2278 "\n temp_priority: %i"
2279 "\n start_pfn: %lu",
2280 zone->all_unreclaimable,
2281 zone->prev_priority,
2282 zone->temp_priority,
2283 zone->zone_start_pfn);
2284 spin_unlock_irqrestore(&zone->lock, flags);
2285 seq_putc(m, '\n');
2286 }
2287 return 0;
2288}
2289
2290struct seq_operations zoneinfo_op = {
2291 .start = frag_start, /* iterate over all zones. The same as in
2292 * fragmentation. */
2293 .next = frag_next,
2294 .stop = frag_stop,
2295 .show = zoneinfo_show,
2296};
2297
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298static char *vmstat_text[] = {
2299 "nr_dirty",
2300 "nr_writeback",
2301 "nr_unstable",
2302 "nr_page_table_pages",
2303 "nr_mapped",
2304 "nr_slab",
2305
2306 "pgpgin",
2307 "pgpgout",
2308 "pswpin",
2309 "pswpout",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310
Nick Piggin9328b8f2006-01-06 00:11:10 -08002311 "pgalloc_high",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 "pgalloc_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002313 "pgalloc_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 "pgalloc_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002315
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 "pgfree",
2317 "pgactivate",
2318 "pgdeactivate",
2319
2320 "pgfault",
2321 "pgmajfault",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002322
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 "pgrefill_high",
2324 "pgrefill_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002325 "pgrefill_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 "pgrefill_dma",
2327
2328 "pgsteal_high",
2329 "pgsteal_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002330 "pgsteal_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 "pgsteal_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002332
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 "pgscan_kswapd_high",
2334 "pgscan_kswapd_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002335 "pgscan_kswapd_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 "pgscan_kswapd_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002337
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 "pgscan_direct_high",
2339 "pgscan_direct_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002340 "pgscan_direct_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 "pgscan_direct_dma",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342
Nick Piggin9328b8f2006-01-06 00:11:10 -08002343 "pginodesteal",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 "slabs_scanned",
2345 "kswapd_steal",
2346 "kswapd_inodesteal",
2347 "pageoutrun",
2348 "allocstall",
2349
2350 "pgrotated",
KAMEZAWA Hiroyukiedfbe2b2005-05-01 08:58:37 -07002351 "nr_bounce",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352};
2353
2354static void *vmstat_start(struct seq_file *m, loff_t *pos)
2355{
2356 struct page_state *ps;
2357
2358 if (*pos >= ARRAY_SIZE(vmstat_text))
2359 return NULL;
2360
2361 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2362 m->private = ps;
2363 if (!ps)
2364 return ERR_PTR(-ENOMEM);
2365 get_full_page_state(ps);
2366 ps->pgpgin /= 2; /* sectors -> kbytes */
2367 ps->pgpgout /= 2;
2368 return (unsigned long *)ps + *pos;
2369}
2370
2371static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2372{
2373 (*pos)++;
2374 if (*pos >= ARRAY_SIZE(vmstat_text))
2375 return NULL;
2376 return (unsigned long *)m->private + *pos;
2377}
2378
2379static int vmstat_show(struct seq_file *m, void *arg)
2380{
2381 unsigned long *l = arg;
2382 unsigned long off = l - (unsigned long *)m->private;
2383
2384 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2385 return 0;
2386}
2387
2388static void vmstat_stop(struct seq_file *m, void *arg)
2389{
2390 kfree(m->private);
2391 m->private = NULL;
2392}
2393
2394struct seq_operations vmstat_op = {
2395 .start = vmstat_start,
2396 .next = vmstat_next,
2397 .stop = vmstat_stop,
2398 .show = vmstat_show,
2399};
2400
2401#endif /* CONFIG_PROC_FS */
2402
2403#ifdef CONFIG_HOTPLUG_CPU
2404static int page_alloc_cpu_notify(struct notifier_block *self,
2405 unsigned long action, void *hcpu)
2406{
2407 int cpu = (unsigned long)hcpu;
2408 long *count;
2409 unsigned long *src, *dest;
2410
2411 if (action == CPU_DEAD) {
2412 int i;
2413
2414 /* Drain local pagecache count. */
2415 count = &per_cpu(nr_pagecache_local, cpu);
2416 atomic_add(*count, &nr_pagecache);
2417 *count = 0;
2418 local_irq_disable();
2419 __drain_pages(cpu);
2420
2421 /* Add dead cpu's page_states to our own. */
2422 dest = (unsigned long *)&__get_cpu_var(page_states);
2423 src = (unsigned long *)&per_cpu(page_states, cpu);
2424
2425 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2426 i++) {
2427 dest[i] += src[i];
2428 src[i] = 0;
2429 }
2430
2431 local_irq_enable();
2432 }
2433 return NOTIFY_OK;
2434}
2435#endif /* CONFIG_HOTPLUG_CPU */
2436
2437void __init page_alloc_init(void)
2438{
2439 hotcpu_notifier(page_alloc_cpu_notify, 0);
2440}
2441
2442/*
2443 * setup_per_zone_lowmem_reserve - called whenever
2444 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2445 * has a correct pages reserved value, so an adequate number of
2446 * pages are left in the zone after a successful __alloc_pages().
2447 */
2448static void setup_per_zone_lowmem_reserve(void)
2449{
2450 struct pglist_data *pgdat;
2451 int j, idx;
2452
2453 for_each_pgdat(pgdat) {
2454 for (j = 0; j < MAX_NR_ZONES; j++) {
2455 struct zone *zone = pgdat->node_zones + j;
2456 unsigned long present_pages = zone->present_pages;
2457
2458 zone->lowmem_reserve[j] = 0;
2459
2460 for (idx = j-1; idx >= 0; idx--) {
2461 struct zone *lower_zone;
2462
2463 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2464 sysctl_lowmem_reserve_ratio[idx] = 1;
2465
2466 lower_zone = pgdat->node_zones + idx;
2467 lower_zone->lowmem_reserve[j] = present_pages /
2468 sysctl_lowmem_reserve_ratio[idx];
2469 present_pages += lower_zone->present_pages;
2470 }
2471 }
2472 }
2473}
2474
2475/*
2476 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2477 * that the pages_{min,low,high} values for each zone are set correctly
2478 * with respect to min_free_kbytes.
2479 */
Dave Hansen3947be12005-10-29 18:16:54 -07002480void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481{
2482 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2483 unsigned long lowmem_pages = 0;
2484 struct zone *zone;
2485 unsigned long flags;
2486
2487 /* Calculate total number of !ZONE_HIGHMEM pages */
2488 for_each_zone(zone) {
2489 if (!is_highmem(zone))
2490 lowmem_pages += zone->present_pages;
2491 }
2492
2493 for_each_zone(zone) {
Nick Piggin669ed172005-11-13 16:06:45 -08002494 unsigned long tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 spin_lock_irqsave(&zone->lru_lock, flags);
Nick Piggin669ed172005-11-13 16:06:45 -08002496 tmp = (pages_min * zone->present_pages) / lowmem_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 if (is_highmem(zone)) {
2498 /*
Nick Piggin669ed172005-11-13 16:06:45 -08002499 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2500 * need highmem pages, so cap pages_min to a small
2501 * value here.
2502 *
2503 * The (pages_high-pages_low) and (pages_low-pages_min)
2504 * deltas controls asynch page reclaim, and so should
2505 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 */
2507 int min_pages;
2508
2509 min_pages = zone->present_pages / 1024;
2510 if (min_pages < SWAP_CLUSTER_MAX)
2511 min_pages = SWAP_CLUSTER_MAX;
2512 if (min_pages > 128)
2513 min_pages = 128;
2514 zone->pages_min = min_pages;
2515 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08002516 /*
2517 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 * proportionate to the zone's size.
2519 */
Nick Piggin669ed172005-11-13 16:06:45 -08002520 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 }
2522
Nick Piggin669ed172005-11-13 16:06:45 -08002523 zone->pages_low = zone->pages_min + tmp / 4;
2524 zone->pages_high = zone->pages_min + tmp / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 spin_unlock_irqrestore(&zone->lru_lock, flags);
2526 }
2527}
2528
2529/*
2530 * Initialise min_free_kbytes.
2531 *
2532 * For small machines we want it small (128k min). For large machines
2533 * we want it large (64MB max). But it is not linear, because network
2534 * bandwidth does not increase linearly with machine size. We use
2535 *
2536 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2537 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2538 *
2539 * which yields
2540 *
2541 * 16MB: 512k
2542 * 32MB: 724k
2543 * 64MB: 1024k
2544 * 128MB: 1448k
2545 * 256MB: 2048k
2546 * 512MB: 2896k
2547 * 1024MB: 4096k
2548 * 2048MB: 5792k
2549 * 4096MB: 8192k
2550 * 8192MB: 11584k
2551 * 16384MB: 16384k
2552 */
2553static int __init init_per_zone_pages_min(void)
2554{
2555 unsigned long lowmem_kbytes;
2556
2557 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2558
2559 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2560 if (min_free_kbytes < 128)
2561 min_free_kbytes = 128;
2562 if (min_free_kbytes > 65536)
2563 min_free_kbytes = 65536;
2564 setup_per_zone_pages_min();
2565 setup_per_zone_lowmem_reserve();
2566 return 0;
2567}
2568module_init(init_per_zone_pages_min)
2569
2570/*
2571 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2572 * that we can call two helper functions whenever min_free_kbytes
2573 * changes.
2574 */
2575int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2576 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2577{
2578 proc_dointvec(table, write, file, buffer, length, ppos);
2579 setup_per_zone_pages_min();
2580 return 0;
2581}
2582
2583/*
2584 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2585 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2586 * whenever sysctl_lowmem_reserve_ratio changes.
2587 *
2588 * The reserve ratio obviously has absolutely no relation with the
2589 * pages_min watermarks. The lowmem reserve ratio can only make sense
2590 * if in function of the boot time zone sizes.
2591 */
2592int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2593 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2594{
2595 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2596 setup_per_zone_lowmem_reserve();
2597 return 0;
2598}
2599
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002600/*
2601 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2602 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2603 * can have before it gets flushed back to buddy allocator.
2604 */
2605
2606int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2607 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2608{
2609 struct zone *zone;
2610 unsigned int cpu;
2611 int ret;
2612
2613 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2614 if (!write || (ret == -EINVAL))
2615 return ret;
2616 for_each_zone(zone) {
2617 for_each_online_cpu(cpu) {
2618 unsigned long high;
2619 high = zone->present_pages / percpu_pagelist_fraction;
2620 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2621 }
2622 }
2623 return 0;
2624}
2625
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626__initdata int hashdist = HASHDIST_DEFAULT;
2627
2628#ifdef CONFIG_NUMA
2629static int __init set_hashdist(char *str)
2630{
2631 if (!str)
2632 return 0;
2633 hashdist = simple_strtoul(str, &str, 0);
2634 return 1;
2635}
2636__setup("hashdist=", set_hashdist);
2637#endif
2638
2639/*
2640 * allocate a large system hash table from bootmem
2641 * - it is assumed that the hash table must contain an exact power-of-2
2642 * quantity of entries
2643 * - limit is the number of hash buckets, not the total allocation size
2644 */
2645void *__init alloc_large_system_hash(const char *tablename,
2646 unsigned long bucketsize,
2647 unsigned long numentries,
2648 int scale,
2649 int flags,
2650 unsigned int *_hash_shift,
2651 unsigned int *_hash_mask,
2652 unsigned long limit)
2653{
2654 unsigned long long max = limit;
2655 unsigned long log2qty, size;
2656 void *table = NULL;
2657
2658 /* allow the kernel cmdline to have a say */
2659 if (!numentries) {
2660 /* round applicable memory size up to nearest megabyte */
2661 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2662 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2663 numentries >>= 20 - PAGE_SHIFT;
2664 numentries <<= 20 - PAGE_SHIFT;
2665
2666 /* limit to 1 bucket per 2^scale bytes of low memory */
2667 if (scale > PAGE_SHIFT)
2668 numentries >>= (scale - PAGE_SHIFT);
2669 else
2670 numentries <<= (PAGE_SHIFT - scale);
2671 }
2672 /* rounded up to nearest power of 2 in size */
2673 numentries = 1UL << (long_log2(numentries) + 1);
2674
2675 /* limit allocation size to 1/16 total memory by default */
2676 if (max == 0) {
2677 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2678 do_div(max, bucketsize);
2679 }
2680
2681 if (numentries > max)
2682 numentries = max;
2683
2684 log2qty = long_log2(numentries);
2685
2686 do {
2687 size = bucketsize << log2qty;
2688 if (flags & HASH_EARLY)
2689 table = alloc_bootmem(size);
2690 else if (hashdist)
2691 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2692 else {
2693 unsigned long order;
2694 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2695 ;
2696 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2697 }
2698 } while (!table && size > PAGE_SIZE && --log2qty);
2699
2700 if (!table)
2701 panic("Failed to allocate %s hash table\n", tablename);
2702
2703 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2704 tablename,
2705 (1U << log2qty),
2706 long_log2(size) - PAGE_SHIFT,
2707 size);
2708
2709 if (_hash_shift)
2710 *_hash_shift = log2qty;
2711 if (_hash_mask)
2712 *_hash_mask = (1 << log2qty) - 1;
2713
2714 return table;
2715}