blob: df54e2fc8ee09760c67d2dd4e6bd496e07286124 [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 Piggin92be2e332006-01-06 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 Piggin92be2e332006-01-06 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 Piggin92be2e332006-01-06 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 Piggin92be2e332006-01-06 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))
Christoph Lameter9eeff232006-01-18 17:42:31 -0800881 if (!zone_reclaim_mode ||
882 !zone_reclaim(*z, gfp_mask, order))
883 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800884 }
885
Nick Piggina74609f2006-01-06 00:11:20 -0800886 page = buffered_rmqueue(zonelist, *z, order, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800887 if (page) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800888 break;
889 }
890 } while (*(++z) != NULL);
891 return page;
Martin Hicks753ee722005-06-21 17:14:41 -0700892}
893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894/*
895 * This is the 'heart' of the zoned buddy allocator.
896 */
897struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +0100898__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 struct zonelist *zonelist)
900{
Al Viro260b2362005-10-21 03:22:44 -0400901 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800902 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 struct page *page;
904 struct reclaim_state reclaim_state;
905 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800907 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 int did_some_progress;
909
910 might_sleep_if(wait);
911
Jens Axboe6b1de912005-11-17 21:35:02 +0100912restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800913 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800915 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 /* Should this ever happen?? */
917 return NULL;
918 }
Jens Axboe6b1de912005-11-17 21:35:02 +0100919
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800920 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -0800921 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800922 if (page)
923 goto got_pg;
924
Jens Axboe6b1de912005-11-17 21:35:02 +0100925 do {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800926 wakeup_kswapd(*z, order);
Jens Axboe6b1de912005-11-17 21:35:02 +0100927 } while (*(++z));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800928
Paul Jackson9bf22292005-09-06 15:18:12 -0700929 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800930 * OK, we're below the kswapd watermark and have kicked background
931 * reclaim. Now things get more complex, so set up alloc_flags according
932 * to how we want to proceed.
933 *
934 * The caller may dip into page reserves a bit more if the caller
935 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -0800936 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
937 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -0700938 */
Nick Piggin31488902005-11-28 13:44:03 -0800939 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800940 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
941 alloc_flags |= ALLOC_HARDER;
942 if (gfp_mask & __GFP_HIGH)
943 alloc_flags |= ALLOC_HIGH;
Paul Jackson47f3a862006-01-06 00:10:32 -0800944 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
946 /*
947 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800948 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 *
950 * This is the last chance, in general, before the goto nopage.
951 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -0700952 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800954 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
955 if (page)
956 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
958 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -0700959
960 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
961 && !in_interrupt()) {
962 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -0800963nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -0700964 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800965 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -0800966 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800967 if (page)
968 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -0800969 if (gfp_mask & __GFP_NOFAIL) {
970 blk_congestion_wait(WRITE, HZ/50);
971 goto nofail_alloc;
972 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 }
974 goto nopage;
975 }
976
977 /* Atomic allocations - we can't balance anything */
978 if (!wait)
979 goto nopage;
980
981rebalance:
982 cond_resched();
983
984 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -0800985 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 p->flags |= PF_MEMALLOC;
987 reclaim_state.reclaimed_slab = 0;
988 p->reclaim_state = &reclaim_state;
989
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800990 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
992 p->reclaim_state = NULL;
993 p->flags &= ~PF_MEMALLOC;
994
995 cond_resched();
996
997 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800998 page = get_page_from_freelist(gfp_mask, order,
999 zonelist, alloc_flags);
1000 if (page)
1001 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1003 /*
1004 * Go through the zonelist yet one more time, keep
1005 * very high watermark here, this is only to catch
1006 * a parallel oom killing, we must fail if we're still
1007 * under heavy pressure.
1008 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001009 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001010 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001011 if (page)
1012 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
Marcelo Tosatti79b9ce32005-07-07 17:56:04 -07001014 out_of_memory(gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 goto restart;
1016 }
1017
1018 /*
1019 * Don't let big-order allocations loop unless the caller explicitly
1020 * requests that. Wait for some write requests to complete then retry.
1021 *
1022 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1023 * <= 3, but that may not be true in other implementations.
1024 */
1025 do_retry = 0;
1026 if (!(gfp_mask & __GFP_NORETRY)) {
1027 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1028 do_retry = 1;
1029 if (gfp_mask & __GFP_NOFAIL)
1030 do_retry = 1;
1031 }
1032 if (do_retry) {
1033 blk_congestion_wait(WRITE, HZ/50);
1034 goto rebalance;
1035 }
1036
1037nopage:
1038 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1039 printk(KERN_WARNING "%s: page allocation failure."
1040 " order:%d, mode:0x%x\n",
1041 p->comm, order, gfp_mask);
1042 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001043 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 return page;
1047}
1048
1049EXPORT_SYMBOL(__alloc_pages);
1050
1051/*
1052 * Common helper functions.
1053 */
Al Virodd0fc662005-10-07 07:46:04 +01001054fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055{
1056 struct page * page;
1057 page = alloc_pages(gfp_mask, order);
1058 if (!page)
1059 return 0;
1060 return (unsigned long) page_address(page);
1061}
1062
1063EXPORT_SYMBOL(__get_free_pages);
1064
Al Virodd0fc662005-10-07 07:46:04 +01001065fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066{
1067 struct page * page;
1068
1069 /*
1070 * get_zeroed_page() returns a 32-bit address, which cannot represent
1071 * a highmem page
1072 */
Al Viro260b2362005-10-21 03:22:44 -04001073 BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
1075 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1076 if (page)
1077 return (unsigned long) page_address(page);
1078 return 0;
1079}
1080
1081EXPORT_SYMBOL(get_zeroed_page);
1082
1083void __pagevec_free(struct pagevec *pvec)
1084{
1085 int i = pagevec_count(pvec);
1086
1087 while (--i >= 0)
1088 free_hot_cold_page(pvec->pages[i], pvec->cold);
1089}
1090
1091fastcall void __free_pages(struct page *page, unsigned int order)
1092{
Nick Pigginb5810032005-10-29 18:16:12 -07001093 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 if (order == 0)
1095 free_hot_page(page);
1096 else
1097 __free_pages_ok(page, order);
1098 }
1099}
1100
1101EXPORT_SYMBOL(__free_pages);
1102
1103fastcall void free_pages(unsigned long addr, unsigned int order)
1104{
1105 if (addr != 0) {
1106 BUG_ON(!virt_addr_valid((void *)addr));
1107 __free_pages(virt_to_page((void *)addr), order);
1108 }
1109}
1110
1111EXPORT_SYMBOL(free_pages);
1112
1113/*
1114 * Total amount of free (allocatable) RAM:
1115 */
1116unsigned int nr_free_pages(void)
1117{
1118 unsigned int sum = 0;
1119 struct zone *zone;
1120
1121 for_each_zone(zone)
1122 sum += zone->free_pages;
1123
1124 return sum;
1125}
1126
1127EXPORT_SYMBOL(nr_free_pages);
1128
1129#ifdef CONFIG_NUMA
1130unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1131{
1132 unsigned int i, sum = 0;
1133
1134 for (i = 0; i < MAX_NR_ZONES; i++)
1135 sum += pgdat->node_zones[i].free_pages;
1136
1137 return sum;
1138}
1139#endif
1140
1141static unsigned int nr_free_zone_pages(int offset)
1142{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001143 /* Just pick one node, since fallback list is circular */
1144 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 unsigned int sum = 0;
1146
Martin J. Blighe310fd42005-07-29 22:59:18 -07001147 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1148 struct zone **zonep = zonelist->zones;
1149 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150
Martin J. Blighe310fd42005-07-29 22:59:18 -07001151 for (zone = *zonep++; zone; zone = *zonep++) {
1152 unsigned long size = zone->present_pages;
1153 unsigned long high = zone->pages_high;
1154 if (size > high)
1155 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 }
1157
1158 return sum;
1159}
1160
1161/*
1162 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1163 */
1164unsigned int nr_free_buffer_pages(void)
1165{
Al Viroaf4ca452005-10-21 02:55:38 -04001166 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167}
1168
1169/*
1170 * Amount of free RAM allocatable within all zones
1171 */
1172unsigned int nr_free_pagecache_pages(void)
1173{
Al Viroaf4ca452005-10-21 02:55:38 -04001174 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175}
1176
1177#ifdef CONFIG_HIGHMEM
1178unsigned int nr_free_highpages (void)
1179{
1180 pg_data_t *pgdat;
1181 unsigned int pages = 0;
1182
1183 for_each_pgdat(pgdat)
1184 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1185
1186 return pages;
1187}
1188#endif
1189
1190#ifdef CONFIG_NUMA
1191static void show_node(struct zone *zone)
1192{
1193 printk("Node %d ", zone->zone_pgdat->node_id);
1194}
1195#else
1196#define show_node(zone) do { } while (0)
1197#endif
1198
1199/*
1200 * Accumulate the page_state information across all CPUs.
1201 * The result is unavoidably approximate - it can change
1202 * during and after execution of this function.
1203 */
1204static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1205
1206atomic_t nr_pagecache = ATOMIC_INIT(0);
1207EXPORT_SYMBOL(nr_pagecache);
1208#ifdef CONFIG_SMP
1209DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1210#endif
1211
Nick Piggina86b1f52006-01-06 00:11:00 -08001212static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213{
1214 int cpu = 0;
1215
1216 memset(ret, 0, sizeof(*ret));
Andrew Morton84c20082006-01-08 01:00:28 -08001217 cpus_and(*cpumask, *cpumask, cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218
Martin Hicksc07e02d2005-09-03 15:55:11 -07001219 cpu = first_cpu(*cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 while (cpu < NR_CPUS) {
1221 unsigned long *in, *out, off;
1222
1223 in = (unsigned long *)&per_cpu(page_states, cpu);
1224
Martin Hicksc07e02d2005-09-03 15:55:11 -07001225 cpu = next_cpu(cpu, *cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
1227 if (cpu < NR_CPUS)
1228 prefetch(&per_cpu(page_states, cpu));
1229
1230 out = (unsigned long *)ret;
1231 for (off = 0; off < nr; off++)
1232 *out++ += *in++;
1233 }
1234}
1235
Martin Hicksc07e02d2005-09-03 15:55:11 -07001236void get_page_state_node(struct page_state *ret, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237{
1238 int nr;
Martin Hicksc07e02d2005-09-03 15:55:11 -07001239 cpumask_t mask = node_to_cpumask(node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
1241 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1242 nr /= sizeof(unsigned long);
1243
Martin Hicksc07e02d2005-09-03 15:55:11 -07001244 __get_page_state(ret, nr+1, &mask);
1245}
1246
1247void get_page_state(struct page_state *ret)
1248{
1249 int nr;
1250 cpumask_t mask = CPU_MASK_ALL;
1251
1252 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1253 nr /= sizeof(unsigned long);
1254
1255 __get_page_state(ret, nr + 1, &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256}
1257
1258void get_full_page_state(struct page_state *ret)
1259{
Martin Hicksc07e02d2005-09-03 15:55:11 -07001260 cpumask_t mask = CPU_MASK_ALL;
1261
1262 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263}
1264
Nick Piggina74609f2006-01-06 00:11:20 -08001265unsigned long read_page_state_offset(unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266{
1267 unsigned long ret = 0;
1268 int cpu;
1269
Andrew Morton84c20082006-01-08 01:00:28 -08001270 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 unsigned long in;
1272
1273 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1274 ret += *((unsigned long *)in);
1275 }
1276 return ret;
1277}
1278
Nick Piggina74609f2006-01-06 00:11:20 -08001279void __mod_page_state_offset(unsigned long offset, unsigned long delta)
1280{
1281 void *ptr;
1282
1283 ptr = &__get_cpu_var(page_states);
1284 *(unsigned long *)(ptr + offset) += delta;
1285}
1286EXPORT_SYMBOL(__mod_page_state_offset);
1287
1288void mod_page_state_offset(unsigned long offset, unsigned long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289{
1290 unsigned long flags;
Nick Piggina74609f2006-01-06 00:11:20 -08001291 void *ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292
1293 local_irq_save(flags);
1294 ptr = &__get_cpu_var(page_states);
Nick Piggina74609f2006-01-06 00:11:20 -08001295 *(unsigned long *)(ptr + offset) += delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 local_irq_restore(flags);
1297}
Nick Piggina74609f2006-01-06 00:11:20 -08001298EXPORT_SYMBOL(mod_page_state_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299
1300void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1301 unsigned long *free, struct pglist_data *pgdat)
1302{
1303 struct zone *zones = pgdat->node_zones;
1304 int i;
1305
1306 *active = 0;
1307 *inactive = 0;
1308 *free = 0;
1309 for (i = 0; i < MAX_NR_ZONES; i++) {
1310 *active += zones[i].nr_active;
1311 *inactive += zones[i].nr_inactive;
1312 *free += zones[i].free_pages;
1313 }
1314}
1315
1316void get_zone_counts(unsigned long *active,
1317 unsigned long *inactive, unsigned long *free)
1318{
1319 struct pglist_data *pgdat;
1320
1321 *active = 0;
1322 *inactive = 0;
1323 *free = 0;
1324 for_each_pgdat(pgdat) {
1325 unsigned long l, m, n;
1326 __get_zone_counts(&l, &m, &n, pgdat);
1327 *active += l;
1328 *inactive += m;
1329 *free += n;
1330 }
1331}
1332
1333void si_meminfo(struct sysinfo *val)
1334{
1335 val->totalram = totalram_pages;
1336 val->sharedram = 0;
1337 val->freeram = nr_free_pages();
1338 val->bufferram = nr_blockdev_pages();
1339#ifdef CONFIG_HIGHMEM
1340 val->totalhigh = totalhigh_pages;
1341 val->freehigh = nr_free_highpages();
1342#else
1343 val->totalhigh = 0;
1344 val->freehigh = 0;
1345#endif
1346 val->mem_unit = PAGE_SIZE;
1347}
1348
1349EXPORT_SYMBOL(si_meminfo);
1350
1351#ifdef CONFIG_NUMA
1352void si_meminfo_node(struct sysinfo *val, int nid)
1353{
1354 pg_data_t *pgdat = NODE_DATA(nid);
1355
1356 val->totalram = pgdat->node_present_pages;
1357 val->freeram = nr_free_pages_pgdat(pgdat);
1358 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1359 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1360 val->mem_unit = PAGE_SIZE;
1361}
1362#endif
1363
1364#define K(x) ((x) << (PAGE_SHIFT-10))
1365
1366/*
1367 * Show free area list (used inside shift_scroll-lock stuff)
1368 * We also calculate the percentage fragmentation. We do this by counting the
1369 * memory on each free list with the exception of the first item on the list.
1370 */
1371void show_free_areas(void)
1372{
1373 struct page_state ps;
1374 int cpu, temperature;
1375 unsigned long active;
1376 unsigned long inactive;
1377 unsigned long free;
1378 struct zone *zone;
1379
1380 for_each_zone(zone) {
1381 show_node(zone);
1382 printk("%s per-cpu:", zone->name);
1383
Con Kolivasf3fe6512006-01-06 00:11:15 -08001384 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 printk(" empty\n");
1386 continue;
1387 } else
1388 printk("\n");
1389
Dave Jones6b482c62005-11-10 15:45:56 -05001390 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 struct per_cpu_pageset *pageset;
1392
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001393 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394
1395 for (temperature = 0; temperature < 2; temperature++)
Nick Piggin2d92c5c2006-01-06 00:10:59 -08001396 printk("cpu %d %s: high %d, batch %d used:%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 cpu,
1398 temperature ? "cold" : "hot",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 pageset->pcp[temperature].high,
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001400 pageset->pcp[temperature].batch,
1401 pageset->pcp[temperature].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 }
1403 }
1404
1405 get_page_state(&ps);
1406 get_zone_counts(&active, &inactive, &free);
1407
Denis Vlasenkoc0d62212005-06-21 17:15:14 -07001408 printk("Free pages: %11ukB (%ukB HighMem)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 K(nr_free_pages()),
1410 K(nr_free_highpages()));
1411
1412 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1413 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1414 active,
1415 inactive,
1416 ps.nr_dirty,
1417 ps.nr_writeback,
1418 ps.nr_unstable,
1419 nr_free_pages(),
1420 ps.nr_slab,
1421 ps.nr_mapped,
1422 ps.nr_page_table_pages);
1423
1424 for_each_zone(zone) {
1425 int i;
1426
1427 show_node(zone);
1428 printk("%s"
1429 " free:%lukB"
1430 " min:%lukB"
1431 " low:%lukB"
1432 " high:%lukB"
1433 " active:%lukB"
1434 " inactive:%lukB"
1435 " present:%lukB"
1436 " pages_scanned:%lu"
1437 " all_unreclaimable? %s"
1438 "\n",
1439 zone->name,
1440 K(zone->free_pages),
1441 K(zone->pages_min),
1442 K(zone->pages_low),
1443 K(zone->pages_high),
1444 K(zone->nr_active),
1445 K(zone->nr_inactive),
1446 K(zone->present_pages),
1447 zone->pages_scanned,
1448 (zone->all_unreclaimable ? "yes" : "no")
1449 );
1450 printk("lowmem_reserve[]:");
1451 for (i = 0; i < MAX_NR_ZONES; i++)
1452 printk(" %lu", zone->lowmem_reserve[i]);
1453 printk("\n");
1454 }
1455
1456 for_each_zone(zone) {
1457 unsigned long nr, flags, order, total = 0;
1458
1459 show_node(zone);
1460 printk("%s: ", zone->name);
Con Kolivasf3fe6512006-01-06 00:11:15 -08001461 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 printk("empty\n");
1463 continue;
1464 }
1465
1466 spin_lock_irqsave(&zone->lock, flags);
1467 for (order = 0; order < MAX_ORDER; order++) {
1468 nr = zone->free_area[order].nr_free;
1469 total += nr << order;
1470 printk("%lu*%lukB ", nr, K(1UL) << order);
1471 }
1472 spin_unlock_irqrestore(&zone->lock, flags);
1473 printk("= %lukB\n", K(total));
1474 }
1475
1476 show_swap_cache_info();
1477}
1478
1479/*
1480 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001481 *
1482 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 */
Christoph Lameter1a932052006-01-06 00:11:16 -08001484static int __init build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter070f8032006-01-06 00:11:19 -08001485 struct zonelist *zonelist, int nr_zones, int zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486{
Christoph Lameter1a932052006-01-06 00:11:16 -08001487 struct zone *zone;
1488
Christoph Lameter070f8032006-01-06 00:11:19 -08001489 BUG_ON(zone_type > ZONE_HIGHMEM);
Christoph Lameter02a68a52006-01-06 00:11:18 -08001490
1491 do {
Christoph Lameter070f8032006-01-06 00:11:19 -08001492 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001493 if (populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494#ifndef CONFIG_HIGHMEM
Christoph Lameter070f8032006-01-06 00:11:19 -08001495 BUG_ON(zone_type > ZONE_NORMAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496#endif
Christoph Lameter070f8032006-01-06 00:11:19 -08001497 zonelist->zones[nr_zones++] = zone;
1498 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 }
Christoph Lameter070f8032006-01-06 00:11:19 -08001500 zone_type--;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001501
Christoph Lameter070f8032006-01-06 00:11:19 -08001502 } while (zone_type >= 0);
1503 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504}
1505
Al Viro260b2362005-10-21 03:22:44 -04001506static inline int highest_zone(int zone_bits)
1507{
1508 int res = ZONE_NORMAL;
1509 if (zone_bits & (__force int)__GFP_HIGHMEM)
1510 res = ZONE_HIGHMEM;
Andi Kleena2f1b422005-11-05 17:25:53 +01001511 if (zone_bits & (__force int)__GFP_DMA32)
1512 res = ZONE_DMA32;
Al Viro260b2362005-10-21 03:22:44 -04001513 if (zone_bits & (__force int)__GFP_DMA)
1514 res = ZONE_DMA;
1515 return res;
1516}
1517
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518#ifdef CONFIG_NUMA
1519#define MAX_NODE_LOAD (num_online_nodes())
1520static int __initdata node_load[MAX_NUMNODES];
1521/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001522 * 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 -07001523 * @node: node whose fallback list we're appending
1524 * @used_node_mask: nodemask_t of already used nodes
1525 *
1526 * We use a number of factors to determine which is the next node that should
1527 * appear on a given node's fallback list. The node should not have appeared
1528 * already in @node's fallback list, and it should be the next closest node
1529 * according to the distance array (which contains arbitrary distance values
1530 * from each node to each node in the system), and should also prefer nodes
1531 * with no CPUs, since presumably they'll have very little allocation pressure
1532 * on them otherwise.
1533 * It returns -1 if no node is found.
1534 */
1535static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1536{
1537 int i, n, val;
1538 int min_val = INT_MAX;
1539 int best_node = -1;
1540
1541 for_each_online_node(i) {
1542 cpumask_t tmp;
1543
1544 /* Start from local node */
1545 n = (node+i) % num_online_nodes();
1546
1547 /* Don't want a node to appear more than once */
1548 if (node_isset(n, *used_node_mask))
1549 continue;
1550
1551 /* Use the local node if we haven't already */
1552 if (!node_isset(node, *used_node_mask)) {
1553 best_node = node;
1554 break;
1555 }
1556
1557 /* Use the distance array to find the distance */
1558 val = node_distance(node, n);
1559
1560 /* Give preference to headless and unused nodes */
1561 tmp = node_to_cpumask(n);
1562 if (!cpus_empty(tmp))
1563 val += PENALTY_FOR_NODE_WITH_CPUS;
1564
1565 /* Slight preference for less loaded node */
1566 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1567 val += node_load[n];
1568
1569 if (val < min_val) {
1570 min_val = val;
1571 best_node = n;
1572 }
1573 }
1574
1575 if (best_node >= 0)
1576 node_set(best_node, *used_node_mask);
1577
1578 return best_node;
1579}
1580
1581static void __init build_zonelists(pg_data_t *pgdat)
1582{
1583 int i, j, k, node, local_node;
1584 int prev_node, load;
1585 struct zonelist *zonelist;
1586 nodemask_t used_mask;
1587
1588 /* initialize zonelists */
1589 for (i = 0; i < GFP_ZONETYPES; i++) {
1590 zonelist = pgdat->node_zonelists + i;
1591 zonelist->zones[0] = NULL;
1592 }
1593
1594 /* NUMA-aware ordering of nodes */
1595 local_node = pgdat->node_id;
1596 load = num_online_nodes();
1597 prev_node = local_node;
1598 nodes_clear(used_mask);
1599 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001600 int distance = node_distance(local_node, node);
1601
1602 /*
1603 * If another node is sufficiently far away then it is better
1604 * to reclaim pages in a zone before going off node.
1605 */
1606 if (distance > RECLAIM_DISTANCE)
1607 zone_reclaim_mode = 1;
1608
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 /*
1610 * We don't want to pressure a particular node.
1611 * So adding penalty to the first node in same
1612 * distance group to make it round-robin.
1613 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08001614
1615 if (distance != node_distance(local_node, prev_node))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 node_load[node] += load;
1617 prev_node = node;
1618 load--;
1619 for (i = 0; i < GFP_ZONETYPES; i++) {
1620 zonelist = pgdat->node_zonelists + i;
1621 for (j = 0; zonelist->zones[j] != NULL; j++);
1622
Al Viro260b2362005-10-21 03:22:44 -04001623 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
1625 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1626 zonelist->zones[j] = NULL;
1627 }
1628 }
1629}
1630
1631#else /* CONFIG_NUMA */
1632
1633static void __init build_zonelists(pg_data_t *pgdat)
1634{
1635 int i, j, k, node, local_node;
1636
1637 local_node = pgdat->node_id;
1638 for (i = 0; i < GFP_ZONETYPES; i++) {
1639 struct zonelist *zonelist;
1640
1641 zonelist = pgdat->node_zonelists + i;
1642
1643 j = 0;
Al Viro260b2362005-10-21 03:22:44 -04001644 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 j = build_zonelists_node(pgdat, zonelist, j, k);
1646 /*
1647 * Now we build the zonelist so that it contains the zones
1648 * of all the other nodes.
1649 * We don't want to pressure a particular node, so when
1650 * building the zones for node N, we make sure that the
1651 * zones coming right after the local ones are those from
1652 * node N+1 (modulo N)
1653 */
1654 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1655 if (!node_online(node))
1656 continue;
1657 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1658 }
1659 for (node = 0; node < local_node; node++) {
1660 if (!node_online(node))
1661 continue;
1662 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1663 }
1664
1665 zonelist->zones[j] = NULL;
1666 }
1667}
1668
1669#endif /* CONFIG_NUMA */
1670
1671void __init build_all_zonelists(void)
1672{
1673 int i;
1674
1675 for_each_online_node(i)
1676 build_zonelists(NODE_DATA(i));
1677 printk("Built %i zonelists\n", num_online_nodes());
1678 cpuset_init_current_mems_allowed();
1679}
1680
1681/*
1682 * Helper functions to size the waitqueue hash table.
1683 * Essentially these want to choose hash table sizes sufficiently
1684 * large so that collisions trying to wait on pages are rare.
1685 * But in fact, the number of active page waitqueues on typical
1686 * systems is ridiculously low, less than 200. So this is even
1687 * conservative, even though it seems large.
1688 *
1689 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1690 * waitqueues, i.e. the size of the waitq table given the number of pages.
1691 */
1692#define PAGES_PER_WAITQUEUE 256
1693
1694static inline unsigned long wait_table_size(unsigned long pages)
1695{
1696 unsigned long size = 1;
1697
1698 pages /= PAGES_PER_WAITQUEUE;
1699
1700 while (size < pages)
1701 size <<= 1;
1702
1703 /*
1704 * Once we have dozens or even hundreds of threads sleeping
1705 * on IO we've got bigger problems than wait queue collision.
1706 * Limit the size of the wait table to a reasonable size.
1707 */
1708 size = min(size, 4096UL);
1709
1710 return max(size, 4UL);
1711}
1712
1713/*
1714 * This is an integer logarithm so that shifts can be used later
1715 * to extract the more random high bits from the multiplicative
1716 * hash function before the remainder is taken.
1717 */
1718static inline unsigned long wait_table_bits(unsigned long size)
1719{
1720 return ffz(~size);
1721}
1722
1723#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1724
1725static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1726 unsigned long *zones_size, unsigned long *zholes_size)
1727{
1728 unsigned long realtotalpages, totalpages = 0;
1729 int i;
1730
1731 for (i = 0; i < MAX_NR_ZONES; i++)
1732 totalpages += zones_size[i];
1733 pgdat->node_spanned_pages = totalpages;
1734
1735 realtotalpages = totalpages;
1736 if (zholes_size)
1737 for (i = 0; i < MAX_NR_ZONES; i++)
1738 realtotalpages -= zholes_size[i];
1739 pgdat->node_present_pages = realtotalpages;
1740 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1741}
1742
1743
1744/*
1745 * Initially all pages are reserved - free ones are freed
1746 * up by free_all_bootmem() once the early boot process is
1747 * done. Non-atomic initialization, single-pass.
1748 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001749void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 unsigned long start_pfn)
1751{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001753 unsigned long end_pfn = start_pfn + size;
1754 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001756 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001757 if (!early_pfn_valid(pfn))
1758 continue;
1759 page = pfn_to_page(pfn);
1760 set_page_links(page, zone, nid, pfn);
Nick Pigginb5810032005-10-29 18:16:12 -07001761 set_page_count(page, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 reset_page_mapcount(page);
1763 SetPageReserved(page);
1764 INIT_LIST_HEAD(&page->lru);
1765#ifdef WANT_PAGE_VIRTUAL
1766 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1767 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001768 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 }
1771}
1772
1773void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1774 unsigned long size)
1775{
1776 int order;
1777 for (order = 0; order < MAX_ORDER ; order++) {
1778 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1779 zone->free_area[order].nr_free = 0;
1780 }
1781}
1782
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001783#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1784void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1785 unsigned long size)
1786{
1787 unsigned long snum = pfn_to_section_nr(pfn);
1788 unsigned long end = pfn_to_section_nr(pfn + size);
1789
1790 if (FLAGS_HAS_NODE)
1791 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1792 else
1793 for (; snum <= end; snum++)
1794 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1795}
1796
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797#ifndef __HAVE_ARCH_MEMMAP_INIT
1798#define memmap_init(size, nid, zone, start_pfn) \
1799 memmap_init_zone((size), (nid), (zone), (start_pfn))
1800#endif
1801
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001802static int __meminit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001803{
1804 int batch;
1805
1806 /*
1807 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001808 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001809 *
1810 * OK, so we don't know how big the cache is. So guess.
1811 */
1812 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001813 if (batch * PAGE_SIZE > 512 * 1024)
1814 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001815 batch /= 4; /* We effectively *= 4 below */
1816 if (batch < 1)
1817 batch = 1;
1818
1819 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001820 * Clamp the batch to a 2^n - 1 value. Having a power
1821 * of 2 value was found to be more likely to have
1822 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001823 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001824 * For example if 2 tasks are alternately allocating
1825 * batches of pages, one task can end up with a lot
1826 * of pages of one half of the possible page colors
1827 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001828 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001829 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001830
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001831 return batch;
1832}
1833
Christoph Lameter2caaad42005-06-21 17:15:00 -07001834inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1835{
1836 struct per_cpu_pages *pcp;
1837
Magnus Damm1c6fe942005-10-26 01:58:59 -07001838 memset(p, 0, sizeof(*p));
1839
Christoph Lameter2caaad42005-06-21 17:15:00 -07001840 pcp = &p->pcp[0]; /* hot */
1841 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001842 pcp->high = 6 * batch;
1843 pcp->batch = max(1UL, 1 * batch);
1844 INIT_LIST_HEAD(&pcp->list);
1845
1846 pcp = &p->pcp[1]; /* cold*/
1847 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001848 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001849 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001850 INIT_LIST_HEAD(&pcp->list);
1851}
1852
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001853/*
1854 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1855 * to the value high for the pageset p.
1856 */
1857
1858static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1859 unsigned long high)
1860{
1861 struct per_cpu_pages *pcp;
1862
1863 pcp = &p->pcp[0]; /* hot list */
1864 pcp->high = high;
1865 pcp->batch = max(1UL, high/4);
1866 if ((high/4) > (PAGE_SHIFT * 8))
1867 pcp->batch = PAGE_SHIFT * 8;
1868}
1869
1870
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001871#ifdef CONFIG_NUMA
1872/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001873 * Boot pageset table. One per cpu which is going to be used for all
1874 * zones and all nodes. The parameters will be set in such a way
1875 * that an item put on a list will immediately be handed over to
1876 * the buddy list. This is safe since pageset manipulation is done
1877 * with interrupts disabled.
1878 *
1879 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001880 *
1881 * The boot_pagesets must be kept even after bootup is complete for
1882 * unused processors and/or zones. They do play a role for bootstrapping
1883 * hotplugged processors.
1884 *
1885 * zoneinfo_show() and maybe other functions do
1886 * not check if the processor is online before following the pageset pointer.
1887 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001888 */
1889static struct per_cpu_pageset
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001890 boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001891
1892/*
1893 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001894 * per cpu pageset array in struct zone.
1895 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001896static int __meminit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001897{
1898 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001899
1900 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001901
Nick Piggin23316bc2006-01-08 01:00:41 -08001902 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001903 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08001904 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001905 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001906
Nick Piggin23316bc2006-01-08 01:00:41 -08001907 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001908
1909 if (percpu_pagelist_fraction)
1910 setup_pagelist_highmark(zone_pcp(zone, cpu),
1911 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001912 }
1913
1914 return 0;
1915bad:
1916 for_each_zone(dzone) {
1917 if (dzone == zone)
1918 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08001919 kfree(zone_pcp(dzone, cpu));
1920 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001921 }
1922 return -ENOMEM;
1923}
1924
1925static inline void free_zone_pagesets(int cpu)
1926{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001927 struct zone *zone;
1928
1929 for_each_zone(zone) {
1930 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1931
1932 zone_pcp(zone, cpu) = NULL;
1933 kfree(pset);
1934 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001935}
1936
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001937static int __meminit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001938 unsigned long action,
1939 void *hcpu)
1940{
1941 int cpu = (long)hcpu;
1942 int ret = NOTIFY_OK;
1943
1944 switch (action) {
1945 case CPU_UP_PREPARE:
1946 if (process_zones(cpu))
1947 ret = NOTIFY_BAD;
1948 break;
Andi Kleenb0d41692005-11-05 17:25:53 +01001949 case CPU_UP_CANCELED:
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001950 case CPU_DEAD:
1951 free_zone_pagesets(cpu);
1952 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001953 default:
1954 break;
1955 }
1956 return ret;
1957}
1958
1959static struct notifier_block pageset_notifier =
1960 { &pageset_cpuup_callback, NULL, 0 };
1961
Al Viro78d99552005-12-15 09:18:25 +00001962void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001963{
1964 int err;
1965
1966 /* Initialize per_cpu_pageset for cpu 0.
1967 * A cpuup callback will do this for every cpu
1968 * as it comes online
1969 */
1970 err = process_zones(smp_processor_id());
1971 BUG_ON(err);
1972 register_cpu_notifier(&pageset_notifier);
1973}
1974
1975#endif
1976
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001977static __meminit
Dave Hansened8ece22005-10-29 18:16:50 -07001978void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
1979{
1980 int i;
1981 struct pglist_data *pgdat = zone->zone_pgdat;
1982
1983 /*
1984 * The per-page waitqueue mechanism uses hashed waitqueues
1985 * per zone.
1986 */
1987 zone->wait_table_size = wait_table_size(zone_size_pages);
1988 zone->wait_table_bits = wait_table_bits(zone->wait_table_size);
1989 zone->wait_table = (wait_queue_head_t *)
1990 alloc_bootmem_node(pgdat, zone->wait_table_size
1991 * sizeof(wait_queue_head_t));
1992
1993 for(i = 0; i < zone->wait_table_size; ++i)
1994 init_waitqueue_head(zone->wait_table + i);
1995}
1996
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001997static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07001998{
1999 int cpu;
2000 unsigned long batch = zone_batchsize(zone);
2001
2002 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2003#ifdef CONFIG_NUMA
2004 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002005 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002006 setup_pageset(&boot_pageset[cpu],0);
2007#else
2008 setup_pageset(zone_pcp(zone,cpu), batch);
2009#endif
2010 }
2011 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2012 zone->name, zone->present_pages, batch);
2013}
2014
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002015static __meminit void init_currently_empty_zone(struct zone *zone,
Dave Hansened8ece22005-10-29 18:16:50 -07002016 unsigned long zone_start_pfn, unsigned long size)
2017{
2018 struct pglist_data *pgdat = zone->zone_pgdat;
2019
2020 zone_wait_table_init(zone, size);
2021 pgdat->nr_zones = zone_idx(zone) + 1;
2022
2023 zone->zone_mem_map = pfn_to_page(zone_start_pfn);
2024 zone->zone_start_pfn = zone_start_pfn;
2025
2026 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2027
2028 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2029}
2030
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031/*
2032 * Set up the zone data structures:
2033 * - mark all pages reserved
2034 * - mark all memory queues empty
2035 * - clear the memory bitmaps
2036 */
2037static void __init free_area_init_core(struct pglist_data *pgdat,
2038 unsigned long *zones_size, unsigned long *zholes_size)
2039{
Dave Hansened8ece22005-10-29 18:16:50 -07002040 unsigned long j;
2041 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 unsigned long zone_start_pfn = pgdat->node_start_pfn;
2043
Dave Hansen208d54e2005-10-29 18:16:52 -07002044 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 pgdat->nr_zones = 0;
2046 init_waitqueue_head(&pgdat->kswapd_wait);
2047 pgdat->kswapd_max_order = 0;
2048
2049 for (j = 0; j < MAX_NR_ZONES; j++) {
2050 struct zone *zone = pgdat->node_zones + j;
2051 unsigned long size, realsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 realsize = size = zones_size[j];
2054 if (zholes_size)
2055 realsize -= zholes_size[j];
2056
Andi Kleena2f1b422005-11-05 17:25:53 +01002057 if (j < ZONE_HIGHMEM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 nr_kernel_pages += realsize;
2059 nr_all_pages += realsize;
2060
2061 zone->spanned_pages = size;
2062 zone->present_pages = realsize;
2063 zone->name = zone_names[j];
2064 spin_lock_init(&zone->lock);
2065 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002066 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 zone->zone_pgdat = pgdat;
2068 zone->free_pages = 0;
2069
2070 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2071
Dave Hansened8ece22005-10-29 18:16:50 -07002072 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 INIT_LIST_HEAD(&zone->active_list);
2074 INIT_LIST_HEAD(&zone->inactive_list);
2075 zone->nr_scan_active = 0;
2076 zone->nr_scan_inactive = 0;
2077 zone->nr_active = 0;
2078 zone->nr_inactive = 0;
Martin Hicks53e9a612005-09-03 15:54:51 -07002079 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 if (!size)
2081 continue;
2082
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002083 zonetable_add(zone, nid, j, zone_start_pfn, size);
Dave Hansened8ece22005-10-29 18:16:50 -07002084 init_currently_empty_zone(zone, zone_start_pfn, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 }
2087}
2088
2089static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2090{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 /* Skip empty nodes */
2092 if (!pgdat->node_spanned_pages)
2093 return;
2094
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002095#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 /* ia64 gets its own node_mem_map, before this, without bootmem */
2097 if (!pgdat->node_mem_map) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002098 unsigned long size;
2099 struct page *map;
2100
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002102 map = alloc_remap(pgdat->node_id, size);
2103 if (!map)
2104 map = alloc_bootmem_node(pgdat, size);
2105 pgdat->node_mem_map = map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002107#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 /*
2109 * With no DISCONTIG, the global mem_map is just set as node 0's
2110 */
2111 if (pgdat == NODE_DATA(0))
2112 mem_map = NODE_DATA(0)->node_mem_map;
2113#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002114#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115}
2116
2117void __init free_area_init_node(int nid, struct pglist_data *pgdat,
2118 unsigned long *zones_size, unsigned long node_start_pfn,
2119 unsigned long *zholes_size)
2120{
2121 pgdat->node_id = nid;
2122 pgdat->node_start_pfn = node_start_pfn;
2123 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2124
2125 alloc_node_mem_map(pgdat);
2126
2127 free_area_init_core(pgdat, zones_size, zholes_size);
2128}
2129
Dave Hansen93b75042005-06-23 00:07:47 -07002130#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131static bootmem_data_t contig_bootmem_data;
2132struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2133
2134EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002135#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136
2137void __init free_area_init(unsigned long *zones_size)
2138{
Dave Hansen93b75042005-06-23 00:07:47 -07002139 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2141}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142
2143#ifdef CONFIG_PROC_FS
2144
2145#include <linux/seq_file.h>
2146
2147static void *frag_start(struct seq_file *m, loff_t *pos)
2148{
2149 pg_data_t *pgdat;
2150 loff_t node = *pos;
2151
2152 for (pgdat = pgdat_list; pgdat && node; pgdat = pgdat->pgdat_next)
2153 --node;
2154
2155 return pgdat;
2156}
2157
2158static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
2159{
2160 pg_data_t *pgdat = (pg_data_t *)arg;
2161
2162 (*pos)++;
2163 return pgdat->pgdat_next;
2164}
2165
2166static void frag_stop(struct seq_file *m, void *arg)
2167{
2168}
2169
2170/*
2171 * This walks the free areas for each zone.
2172 */
2173static int frag_show(struct seq_file *m, void *arg)
2174{
2175 pg_data_t *pgdat = (pg_data_t *)arg;
2176 struct zone *zone;
2177 struct zone *node_zones = pgdat->node_zones;
2178 unsigned long flags;
2179 int order;
2180
2181 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
Con Kolivasf3fe6512006-01-06 00:11:15 -08002182 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 continue;
2184
2185 spin_lock_irqsave(&zone->lock, flags);
2186 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2187 for (order = 0; order < MAX_ORDER; ++order)
2188 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2189 spin_unlock_irqrestore(&zone->lock, flags);
2190 seq_putc(m, '\n');
2191 }
2192 return 0;
2193}
2194
2195struct seq_operations fragmentation_op = {
2196 .start = frag_start,
2197 .next = frag_next,
2198 .stop = frag_stop,
2199 .show = frag_show,
2200};
2201
Nikita Danilov295ab932005-06-21 17:14:38 -07002202/*
2203 * Output information about zones in @pgdat.
2204 */
2205static int zoneinfo_show(struct seq_file *m, void *arg)
2206{
2207 pg_data_t *pgdat = arg;
2208 struct zone *zone;
2209 struct zone *node_zones = pgdat->node_zones;
2210 unsigned long flags;
2211
2212 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2213 int i;
2214
Con Kolivasf3fe6512006-01-06 00:11:15 -08002215 if (!populated_zone(zone))
Nikita Danilov295ab932005-06-21 17:14:38 -07002216 continue;
2217
2218 spin_lock_irqsave(&zone->lock, flags);
2219 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2220 seq_printf(m,
2221 "\n pages free %lu"
2222 "\n min %lu"
2223 "\n low %lu"
2224 "\n high %lu"
2225 "\n active %lu"
2226 "\n inactive %lu"
2227 "\n scanned %lu (a: %lu i: %lu)"
2228 "\n spanned %lu"
2229 "\n present %lu",
2230 zone->free_pages,
2231 zone->pages_min,
2232 zone->pages_low,
2233 zone->pages_high,
2234 zone->nr_active,
2235 zone->nr_inactive,
2236 zone->pages_scanned,
2237 zone->nr_scan_active, zone->nr_scan_inactive,
2238 zone->spanned_pages,
2239 zone->present_pages);
2240 seq_printf(m,
2241 "\n protection: (%lu",
2242 zone->lowmem_reserve[0]);
2243 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2244 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2245 seq_printf(m,
2246 ")"
2247 "\n pagesets");
Nick Piggin23316bc2006-01-08 01:00:41 -08002248 for_each_online_cpu(i) {
Nikita Danilov295ab932005-06-21 17:14:38 -07002249 struct per_cpu_pageset *pageset;
2250 int j;
2251
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002252 pageset = zone_pcp(zone, i);
Nikita Danilov295ab932005-06-21 17:14:38 -07002253 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2254 if (pageset->pcp[j].count)
2255 break;
2256 }
2257 if (j == ARRAY_SIZE(pageset->pcp))
2258 continue;
2259 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2260 seq_printf(m,
2261 "\n cpu: %i pcp: %i"
2262 "\n count: %i"
Nikita Danilov295ab932005-06-21 17:14:38 -07002263 "\n high: %i"
2264 "\n batch: %i",
2265 i, j,
2266 pageset->pcp[j].count,
Nikita Danilov295ab932005-06-21 17:14:38 -07002267 pageset->pcp[j].high,
2268 pageset->pcp[j].batch);
2269 }
2270#ifdef CONFIG_NUMA
2271 seq_printf(m,
2272 "\n numa_hit: %lu"
2273 "\n numa_miss: %lu"
2274 "\n numa_foreign: %lu"
2275 "\n interleave_hit: %lu"
2276 "\n local_node: %lu"
2277 "\n other_node: %lu",
2278 pageset->numa_hit,
2279 pageset->numa_miss,
2280 pageset->numa_foreign,
2281 pageset->interleave_hit,
2282 pageset->local_node,
2283 pageset->other_node);
2284#endif
2285 }
2286 seq_printf(m,
2287 "\n all_unreclaimable: %u"
2288 "\n prev_priority: %i"
2289 "\n temp_priority: %i"
2290 "\n start_pfn: %lu",
2291 zone->all_unreclaimable,
2292 zone->prev_priority,
2293 zone->temp_priority,
2294 zone->zone_start_pfn);
2295 spin_unlock_irqrestore(&zone->lock, flags);
2296 seq_putc(m, '\n');
2297 }
2298 return 0;
2299}
2300
2301struct seq_operations zoneinfo_op = {
2302 .start = frag_start, /* iterate over all zones. The same as in
2303 * fragmentation. */
2304 .next = frag_next,
2305 .stop = frag_stop,
2306 .show = zoneinfo_show,
2307};
2308
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309static char *vmstat_text[] = {
2310 "nr_dirty",
2311 "nr_writeback",
2312 "nr_unstable",
2313 "nr_page_table_pages",
2314 "nr_mapped",
2315 "nr_slab",
2316
2317 "pgpgin",
2318 "pgpgout",
2319 "pswpin",
2320 "pswpout",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321
Nick Piggin9328b8f2006-01-06 00:11:10 -08002322 "pgalloc_high",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 "pgalloc_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002324 "pgalloc_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 "pgalloc_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002326
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 "pgfree",
2328 "pgactivate",
2329 "pgdeactivate",
2330
2331 "pgfault",
2332 "pgmajfault",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002333
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 "pgrefill_high",
2335 "pgrefill_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002336 "pgrefill_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 "pgrefill_dma",
2338
2339 "pgsteal_high",
2340 "pgsteal_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002341 "pgsteal_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 "pgsteal_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002343
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 "pgscan_kswapd_high",
2345 "pgscan_kswapd_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002346 "pgscan_kswapd_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 "pgscan_kswapd_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002348
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 "pgscan_direct_high",
2350 "pgscan_direct_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002351 "pgscan_direct_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 "pgscan_direct_dma",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353
Nick Piggin9328b8f2006-01-06 00:11:10 -08002354 "pginodesteal",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 "slabs_scanned",
2356 "kswapd_steal",
2357 "kswapd_inodesteal",
2358 "pageoutrun",
2359 "allocstall",
2360
2361 "pgrotated",
KAMEZAWA Hiroyukiedfbe2b2005-05-01 08:58:37 -07002362 "nr_bounce",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363};
2364
2365static void *vmstat_start(struct seq_file *m, loff_t *pos)
2366{
2367 struct page_state *ps;
2368
2369 if (*pos >= ARRAY_SIZE(vmstat_text))
2370 return NULL;
2371
2372 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2373 m->private = ps;
2374 if (!ps)
2375 return ERR_PTR(-ENOMEM);
2376 get_full_page_state(ps);
2377 ps->pgpgin /= 2; /* sectors -> kbytes */
2378 ps->pgpgout /= 2;
2379 return (unsigned long *)ps + *pos;
2380}
2381
2382static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2383{
2384 (*pos)++;
2385 if (*pos >= ARRAY_SIZE(vmstat_text))
2386 return NULL;
2387 return (unsigned long *)m->private + *pos;
2388}
2389
2390static int vmstat_show(struct seq_file *m, void *arg)
2391{
2392 unsigned long *l = arg;
2393 unsigned long off = l - (unsigned long *)m->private;
2394
2395 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2396 return 0;
2397}
2398
2399static void vmstat_stop(struct seq_file *m, void *arg)
2400{
2401 kfree(m->private);
2402 m->private = NULL;
2403}
2404
2405struct seq_operations vmstat_op = {
2406 .start = vmstat_start,
2407 .next = vmstat_next,
2408 .stop = vmstat_stop,
2409 .show = vmstat_show,
2410};
2411
2412#endif /* CONFIG_PROC_FS */
2413
2414#ifdef CONFIG_HOTPLUG_CPU
2415static int page_alloc_cpu_notify(struct notifier_block *self,
2416 unsigned long action, void *hcpu)
2417{
2418 int cpu = (unsigned long)hcpu;
2419 long *count;
2420 unsigned long *src, *dest;
2421
2422 if (action == CPU_DEAD) {
2423 int i;
2424
2425 /* Drain local pagecache count. */
2426 count = &per_cpu(nr_pagecache_local, cpu);
2427 atomic_add(*count, &nr_pagecache);
2428 *count = 0;
2429 local_irq_disable();
2430 __drain_pages(cpu);
2431
2432 /* Add dead cpu's page_states to our own. */
2433 dest = (unsigned long *)&__get_cpu_var(page_states);
2434 src = (unsigned long *)&per_cpu(page_states, cpu);
2435
2436 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2437 i++) {
2438 dest[i] += src[i];
2439 src[i] = 0;
2440 }
2441
2442 local_irq_enable();
2443 }
2444 return NOTIFY_OK;
2445}
2446#endif /* CONFIG_HOTPLUG_CPU */
2447
2448void __init page_alloc_init(void)
2449{
2450 hotcpu_notifier(page_alloc_cpu_notify, 0);
2451}
2452
2453/*
2454 * setup_per_zone_lowmem_reserve - called whenever
2455 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2456 * has a correct pages reserved value, so an adequate number of
2457 * pages are left in the zone after a successful __alloc_pages().
2458 */
2459static void setup_per_zone_lowmem_reserve(void)
2460{
2461 struct pglist_data *pgdat;
2462 int j, idx;
2463
2464 for_each_pgdat(pgdat) {
2465 for (j = 0; j < MAX_NR_ZONES; j++) {
2466 struct zone *zone = pgdat->node_zones + j;
2467 unsigned long present_pages = zone->present_pages;
2468
2469 zone->lowmem_reserve[j] = 0;
2470
2471 for (idx = j-1; idx >= 0; idx--) {
2472 struct zone *lower_zone;
2473
2474 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2475 sysctl_lowmem_reserve_ratio[idx] = 1;
2476
2477 lower_zone = pgdat->node_zones + idx;
2478 lower_zone->lowmem_reserve[j] = present_pages /
2479 sysctl_lowmem_reserve_ratio[idx];
2480 present_pages += lower_zone->present_pages;
2481 }
2482 }
2483 }
2484}
2485
2486/*
2487 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2488 * that the pages_{min,low,high} values for each zone are set correctly
2489 * with respect to min_free_kbytes.
2490 */
Dave Hansen3947be12005-10-29 18:16:54 -07002491void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492{
2493 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2494 unsigned long lowmem_pages = 0;
2495 struct zone *zone;
2496 unsigned long flags;
2497
2498 /* Calculate total number of !ZONE_HIGHMEM pages */
2499 for_each_zone(zone) {
2500 if (!is_highmem(zone))
2501 lowmem_pages += zone->present_pages;
2502 }
2503
2504 for_each_zone(zone) {
Nick Piggin669ed172005-11-13 16:06:45 -08002505 unsigned long tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 spin_lock_irqsave(&zone->lru_lock, flags);
Nick Piggin669ed172005-11-13 16:06:45 -08002507 tmp = (pages_min * zone->present_pages) / lowmem_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 if (is_highmem(zone)) {
2509 /*
Nick Piggin669ed172005-11-13 16:06:45 -08002510 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2511 * need highmem pages, so cap pages_min to a small
2512 * value here.
2513 *
2514 * The (pages_high-pages_low) and (pages_low-pages_min)
2515 * deltas controls asynch page reclaim, and so should
2516 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 */
2518 int min_pages;
2519
2520 min_pages = zone->present_pages / 1024;
2521 if (min_pages < SWAP_CLUSTER_MAX)
2522 min_pages = SWAP_CLUSTER_MAX;
2523 if (min_pages > 128)
2524 min_pages = 128;
2525 zone->pages_min = min_pages;
2526 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08002527 /*
2528 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 * proportionate to the zone's size.
2530 */
Nick Piggin669ed172005-11-13 16:06:45 -08002531 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 }
2533
Nick Piggin669ed172005-11-13 16:06:45 -08002534 zone->pages_low = zone->pages_min + tmp / 4;
2535 zone->pages_high = zone->pages_min + tmp / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 spin_unlock_irqrestore(&zone->lru_lock, flags);
2537 }
2538}
2539
2540/*
2541 * Initialise min_free_kbytes.
2542 *
2543 * For small machines we want it small (128k min). For large machines
2544 * we want it large (64MB max). But it is not linear, because network
2545 * bandwidth does not increase linearly with machine size. We use
2546 *
2547 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2548 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2549 *
2550 * which yields
2551 *
2552 * 16MB: 512k
2553 * 32MB: 724k
2554 * 64MB: 1024k
2555 * 128MB: 1448k
2556 * 256MB: 2048k
2557 * 512MB: 2896k
2558 * 1024MB: 4096k
2559 * 2048MB: 5792k
2560 * 4096MB: 8192k
2561 * 8192MB: 11584k
2562 * 16384MB: 16384k
2563 */
2564static int __init init_per_zone_pages_min(void)
2565{
2566 unsigned long lowmem_kbytes;
2567
2568 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2569
2570 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2571 if (min_free_kbytes < 128)
2572 min_free_kbytes = 128;
2573 if (min_free_kbytes > 65536)
2574 min_free_kbytes = 65536;
2575 setup_per_zone_pages_min();
2576 setup_per_zone_lowmem_reserve();
2577 return 0;
2578}
2579module_init(init_per_zone_pages_min)
2580
2581/*
2582 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2583 * that we can call two helper functions whenever min_free_kbytes
2584 * changes.
2585 */
2586int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2587 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2588{
2589 proc_dointvec(table, write, file, buffer, length, ppos);
2590 setup_per_zone_pages_min();
2591 return 0;
2592}
2593
2594/*
2595 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2596 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2597 * whenever sysctl_lowmem_reserve_ratio changes.
2598 *
2599 * The reserve ratio obviously has absolutely no relation with the
2600 * pages_min watermarks. The lowmem reserve ratio can only make sense
2601 * if in function of the boot time zone sizes.
2602 */
2603int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2604 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2605{
2606 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2607 setup_per_zone_lowmem_reserve();
2608 return 0;
2609}
2610
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002611/*
2612 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2613 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2614 * can have before it gets flushed back to buddy allocator.
2615 */
2616
2617int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2618 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2619{
2620 struct zone *zone;
2621 unsigned int cpu;
2622 int ret;
2623
2624 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2625 if (!write || (ret == -EINVAL))
2626 return ret;
2627 for_each_zone(zone) {
2628 for_each_online_cpu(cpu) {
2629 unsigned long high;
2630 high = zone->present_pages / percpu_pagelist_fraction;
2631 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2632 }
2633 }
2634 return 0;
2635}
2636
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637__initdata int hashdist = HASHDIST_DEFAULT;
2638
2639#ifdef CONFIG_NUMA
2640static int __init set_hashdist(char *str)
2641{
2642 if (!str)
2643 return 0;
2644 hashdist = simple_strtoul(str, &str, 0);
2645 return 1;
2646}
2647__setup("hashdist=", set_hashdist);
2648#endif
2649
2650/*
2651 * allocate a large system hash table from bootmem
2652 * - it is assumed that the hash table must contain an exact power-of-2
2653 * quantity of entries
2654 * - limit is the number of hash buckets, not the total allocation size
2655 */
2656void *__init alloc_large_system_hash(const char *tablename,
2657 unsigned long bucketsize,
2658 unsigned long numentries,
2659 int scale,
2660 int flags,
2661 unsigned int *_hash_shift,
2662 unsigned int *_hash_mask,
2663 unsigned long limit)
2664{
2665 unsigned long long max = limit;
2666 unsigned long log2qty, size;
2667 void *table = NULL;
2668
2669 /* allow the kernel cmdline to have a say */
2670 if (!numentries) {
2671 /* round applicable memory size up to nearest megabyte */
2672 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2673 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2674 numentries >>= 20 - PAGE_SHIFT;
2675 numentries <<= 20 - PAGE_SHIFT;
2676
2677 /* limit to 1 bucket per 2^scale bytes of low memory */
2678 if (scale > PAGE_SHIFT)
2679 numentries >>= (scale - PAGE_SHIFT);
2680 else
2681 numentries <<= (PAGE_SHIFT - scale);
2682 }
2683 /* rounded up to nearest power of 2 in size */
2684 numentries = 1UL << (long_log2(numentries) + 1);
2685
2686 /* limit allocation size to 1/16 total memory by default */
2687 if (max == 0) {
2688 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2689 do_div(max, bucketsize);
2690 }
2691
2692 if (numentries > max)
2693 numentries = max;
2694
2695 log2qty = long_log2(numentries);
2696
2697 do {
2698 size = bucketsize << log2qty;
2699 if (flags & HASH_EARLY)
2700 table = alloc_bootmem(size);
2701 else if (hashdist)
2702 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2703 else {
2704 unsigned long order;
2705 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2706 ;
2707 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2708 }
2709 } while (!table && size > PAGE_SIZE && --log2qty);
2710
2711 if (!table)
2712 panic("Failed to allocate %s hash table\n", tablename);
2713
2714 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2715 tablename,
2716 (1U << log2qty),
2717 long_log2(size) - PAGE_SHIFT,
2718 size);
2719
2720 if (_hash_shift)
2721 *_hash_shift = log2qty;
2722 if (_hash_mask)
2723 *_hash_mask = (1 << log2qty) - 1;
2724
2725 return table;
2726}