blob: a5e6891f7bb6f366025e56dd701f9794d29ce67d [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"
140 "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
141 "Trying to fix it up, but a reboot is needed\n"
142 "Backtrace:\n",
143 current->comm, page, (int)(2*sizeof(unsigned long)),
144 (unsigned long)page->flags, page->mapping,
145 page_mapcount(page), page_count(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 dump_stack();
Hugh Dickins334795e2005-06-21 17:15:08 -0700147 page->flags &= ~(1 << PG_lru |
148 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 1 << PG_active |
151 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700152 1 << PG_reclaim |
153 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 1 << PG_swapcache |
Hugh Dickins689bceb2005-11-21 21:32:20 -0800155 1 << PG_writeback );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 set_page_count(page, 0);
157 reset_page_mapcount(page);
158 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700159 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160}
161
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162/*
163 * Higher-order pages are called "compound pages". They are structured thusly:
164 *
165 * The first PAGE_SIZE page is called the "head page".
166 *
167 * The remaining PAGE_SIZE pages are called "tail pages".
168 *
169 * All pages have PG_compound set. All pages have their ->private pointing at
170 * the head page (even the head page has this).
171 *
172 * The first tail page's ->mapping, if non-zero, holds the address of the
173 * compound page's put_page() function.
174 *
175 * The order of the allocation is stored in the first tail page's ->index
176 * This is only for debug at present. This usage means that zero-order pages
177 * may not be compound.
178 */
179static void prep_compound_page(struct page *page, unsigned long order)
180{
181 int i;
182 int nr_pages = 1 << order;
183
184 page[1].mapping = NULL;
185 page[1].index = order;
186 for (i = 0; i < nr_pages; i++) {
187 struct page *p = page + i;
188
189 SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700190 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 }
192}
193
194static void destroy_compound_page(struct page *page, unsigned long order)
195{
196 int i;
197 int nr_pages = 1 << order;
198
Nick Piggin224abf92006-01-06 00:11:11 -0800199 if (unlikely(page[1].index != order))
200 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
202 for (i = 0; i < nr_pages; i++) {
203 struct page *p = page + i;
204
Nick Piggin224abf92006-01-06 00:11:11 -0800205 if (unlikely(!PageCompound(p) |
206 (page_private(p) != (unsigned long)page)))
207 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 ClearPageCompound(p);
209 }
210}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
212/*
213 * function for dealing with page's order in buddy system.
214 * zone->lock is already acquired when we use these.
215 * So, we don't need atomic page->flags operations here.
216 */
217static inline unsigned long page_order(struct page *page) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700218 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219}
220
221static inline void set_page_order(struct page *page, int order) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700222 set_page_private(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 __SetPagePrivate(page);
224}
225
226static inline void rmv_page_order(struct page *page)
227{
228 __ClearPagePrivate(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700229 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230}
231
232/*
233 * Locate the struct page for both the matching buddy in our
234 * pair (buddy1) and the combined O(n+1) page they form (page).
235 *
236 * 1) Any buddy B1 will have an order O twin B2 which satisfies
237 * the following equation:
238 * B2 = B1 ^ (1 << O)
239 * For example, if the starting buddy (buddy2) is #8 its order
240 * 1 buddy is #10:
241 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
242 *
243 * 2) Any buddy B will have an order O+1 parent P which
244 * satisfies the following equation:
245 * P = B & ~(1 << O)
246 *
247 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
248 */
249static inline struct page *
250__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
251{
252 unsigned long buddy_idx = page_idx ^ (1 << order);
253
254 return page + (buddy_idx - page_idx);
255}
256
257static inline unsigned long
258__find_combined_index(unsigned long page_idx, unsigned int order)
259{
260 return (page_idx & ~(1 << order));
261}
262
263/*
264 * This function checks whether a page is free && is the buddy
265 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800266 * (a) the buddy is not in a hole &&
267 * (b) the buddy is free &&
268 * (c) the buddy is on the buddy system &&
269 * (d) a page and its buddy have the same order.
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700270 * for recording page's order, we use page_private(page) and PG_private.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 *
272 */
273static inline int page_is_buddy(struct page *page, int order)
274{
Nick Piggin13e74442006-01-06 00:10:58 -0800275#ifdef CONFIG_HOLES_IN_ZONE
276 if (!pfn_valid(page_to_pfn(page)))
277 return 0;
278#endif
279
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 if (PagePrivate(page) &&
281 (page_order(page) == order) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 page_count(page) == 0)
283 return 1;
284 return 0;
285}
286
287/*
288 * Freeing function for a buddy system allocator.
289 *
290 * The concept of a buddy system is to maintain direct-mapped table
291 * (containing bit values) for memory blocks of various "orders".
292 * The bottom level table contains the map for the smallest allocatable
293 * units of memory (here, pages), and each level above it describes
294 * pairs of units from the levels below, hence, "buddies".
295 * At a high level, all that happens here is marking the table entry
296 * at the bottom level available, and propagating the changes upward
297 * as necessary, plus some accounting needed to play nicely with other
298 * parts of the VM system.
299 * At each level, we keep a list of pages, which are heads of continuous
300 * free pages of length of (1 << order) and marked with PG_Private.Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700301 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 * So when we are allocating or freeing one, we can derive the state of the
303 * other. That is, if we allocate a small block, and both were
304 * free, the remainder of the region must be split into blocks.
305 * If a block is freed, and its buddy is also free, then this
306 * triggers coalescing into a block of larger size.
307 *
308 * -- wli
309 */
310
Nick Piggin48db57f2006-01-08 01:00:42 -0800311static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 struct zone *zone, unsigned int order)
313{
314 unsigned long page_idx;
315 int order_size = 1 << order;
316
Nick Piggin224abf92006-01-06 00:11:11 -0800317 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 destroy_compound_page(page, order);
319
320 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
321
322 BUG_ON(page_idx & (order_size - 1));
323 BUG_ON(bad_range(zone, page));
324
325 zone->free_pages += order_size;
326 while (order < MAX_ORDER-1) {
327 unsigned long combined_idx;
328 struct free_area *area;
329 struct page *buddy;
330
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 buddy = __page_find_buddy(page, page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 if (!page_is_buddy(buddy, order))
333 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800334
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 list_del(&buddy->lru);
336 area = zone->free_area + order;
337 area->nr_free--;
338 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800339 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 page = page + (combined_idx - page_idx);
341 page_idx = combined_idx;
342 order++;
343 }
344 set_page_order(page, order);
345 list_add(&page->lru, &zone->free_area[order].free_list);
346 zone->free_area[order].nr_free++;
347}
348
Nick Piggin224abf92006-01-06 00:11:11 -0800349static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350{
Nick Piggin92be2e32006-01-06 00:10:57 -0800351 if (unlikely(page_mapcount(page) |
352 (page->mapping != NULL) |
353 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 (page->flags & (
355 1 << PG_lru |
356 1 << PG_private |
357 1 << PG_locked |
358 1 << PG_active |
359 1 << PG_reclaim |
360 1 << PG_slab |
361 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700362 1 << PG_writeback |
Nick Piggin92be2e32006-01-06 00:10:57 -0800363 1 << PG_reserved ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800364 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700366 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800367 /*
368 * For now, we report if PG_reserved was found set, but do not
369 * clear it, and do not free the page. But we shall soon need
370 * to do more, for when the ZERO_PAGE count wraps negative.
371 */
372 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373}
374
375/*
376 * Frees a list of pages.
377 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700378 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 *
380 * If the zone was previously in an "all pages pinned" state then look to
381 * see if this freeing clears that state.
382 *
383 * And clear the zone's pages_scanned counter, to hold off the "all pages are
384 * pinned" detection logic.
385 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800386static void free_pages_bulk(struct zone *zone, int count,
387 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388{
Nick Pigginc54ad302006-01-06 00:10:56 -0800389 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 zone->all_unreclaimable = 0;
391 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800392 while (count--) {
393 struct page *page;
394
395 BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800397 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800399 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800401 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402}
403
Nick Piggin48db57f2006-01-08 01:00:42 -0800404static void free_one_page(struct zone *zone, struct page *page, int order)
405{
406 LIST_HEAD(list);
407 list_add(&page->lru, &list);
408 free_pages_bulk(zone, 1, &list, order);
409}
410
411static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412{
Nick Pigginc54ad302006-01-06 00:10:56 -0800413 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800415 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
417 arch_free_page(page, order);
Ingo Molnarde5097c2006-01-09 15:59:21 -0800418 if (!PageHighMem(page))
419 mutex_debug_check_no_locks_freed(page_address(page),
420 page_address(page+(1<<order)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422#ifndef CONFIG_MMU
Nick Piggin48db57f2006-01-08 01:00:42 -0800423 for (i = 1 ; i < (1 << order) ; ++i)
424 __put_page(page + i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425#endif
426
427 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800428 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800429 if (reserved)
430 return;
431
Nick Piggin48db57f2006-01-08 01:00:42 -0800432 kernel_map_pages(page, 1 << order, 0);
Nick Pigginc54ad302006-01-06 00:10:56 -0800433 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800434 __mod_page_state(pgfree, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800435 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800436 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437}
438
David Howellsa226f6c2006-01-06 00:11:08 -0800439/*
440 * permit the bootmem allocator to evade page validation on high-order frees
441 */
442void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
443{
444 if (order == 0) {
445 __ClearPageReserved(page);
446 set_page_count(page, 0);
447
448 free_hot_cold_page(page, 0);
449 } else {
450 LIST_HEAD(list);
451 int loop;
452
453 for (loop = 0; loop < BITS_PER_LONG; loop++) {
454 struct page *p = &page[loop];
455
456 if (loop + 16 < BITS_PER_LONG)
457 prefetchw(p + 16);
458 __ClearPageReserved(p);
459 set_page_count(p, 0);
460 }
461
462 arch_free_page(page, order);
463
464 mod_page_state(pgfree, 1 << order);
465
466 list_add(&page->lru, &list);
467 kernel_map_pages(page, 1 << order, 0);
468 free_pages_bulk(page_zone(page), 1, &list, order);
469 }
470}
471
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473/*
474 * The order of subdivision here is critical for the IO subsystem.
475 * Please do not alter this order without good reasons and regression
476 * testing. Specifically, as large blocks of memory are subdivided,
477 * the order in which smaller blocks are delivered depends on the order
478 * they're subdivided in this function. This is the primary factor
479 * influencing the order in which pages are delivered to the IO
480 * subsystem according to empirical testing, and this is also justified
481 * by considering the behavior of a buddy system containing a single
482 * large block of memory acted on by a series of small allocations.
483 * This behavior is a critical factor in sglist merging's success.
484 *
485 * -- wli
486 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800487static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 int low, int high, struct free_area *area)
489{
490 unsigned long size = 1 << high;
491
492 while (high > low) {
493 area--;
494 high--;
495 size >>= 1;
496 BUG_ON(bad_range(zone, &page[size]));
497 list_add(&page[size].lru, &area->free_list);
498 area->nr_free++;
499 set_page_order(&page[size], high);
500 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501}
502
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503/*
504 * This page is about to be returned from the page allocator
505 */
Hugh Dickins689bceb2005-11-21 21:32:20 -0800506static int prep_new_page(struct page *page, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507{
Nick Piggin92be2e32006-01-06 00:10:57 -0800508 if (unlikely(page_mapcount(page) |
509 (page->mapping != NULL) |
510 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700511 (page->flags & (
512 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 1 << PG_private |
514 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 1 << PG_active |
516 1 << PG_dirty |
517 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700518 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700520 1 << PG_writeback |
Nick Piggin92be2e32006-01-06 00:10:57 -0800521 1 << PG_reserved ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800522 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
Hugh Dickins689bceb2005-11-21 21:32:20 -0800524 /*
525 * For now, we report if PG_reserved was found set, but do not
526 * clear it, and do not allocate the page: as a safety net.
527 */
528 if (PageReserved(page))
529 return 1;
530
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
532 1 << PG_referenced | 1 << PG_arch_1 |
533 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700534 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 set_page_refs(page, order);
536 kernel_map_pages(page, 1 << order, 1);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800537 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538}
539
540/*
541 * Do the hard work of removing an element from the buddy allocator.
542 * Call me with the zone->lock already held.
543 */
544static struct page *__rmqueue(struct zone *zone, unsigned int order)
545{
546 struct free_area * area;
547 unsigned int current_order;
548 struct page *page;
549
550 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
551 area = zone->free_area + current_order;
552 if (list_empty(&area->free_list))
553 continue;
554
555 page = list_entry(area->free_list.next, struct page, lru);
556 list_del(&page->lru);
557 rmv_page_order(page);
558 area->nr_free--;
559 zone->free_pages -= 1UL << order;
Nick Piggin085cc7d2006-01-06 00:11:01 -0800560 expand(zone, page, order, current_order, area);
561 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 }
563
564 return NULL;
565}
566
567/*
568 * Obtain a specified number of elements from the buddy allocator, all under
569 * a single hold of the lock, for efficiency. Add them to the supplied list.
570 * Returns the number of new pages which were placed at *list.
571 */
572static int rmqueue_bulk(struct zone *zone, unsigned int order,
573 unsigned long count, struct list_head *list)
574{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
Nick Pigginc54ad302006-01-06 00:10:56 -0800577 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 for (i = 0; i < count; ++i) {
Nick Piggin085cc7d2006-01-06 00:11:01 -0800579 struct page *page = __rmqueue(zone, order);
580 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 list_add_tail(&page->lru, list);
583 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800584 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800585 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586}
587
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700588#ifdef CONFIG_NUMA
589/* Called from the slab reaper to drain remote pagesets */
590void drain_remote_pages(void)
591{
592 struct zone *zone;
593 int i;
594 unsigned long flags;
595
596 local_irq_save(flags);
597 for_each_zone(zone) {
598 struct per_cpu_pageset *pset;
599
600 /* Do not drain local pagesets */
601 if (zone->zone_pgdat->node_id == numa_node_id())
602 continue;
603
Nick Piggin23316bc2006-01-08 01:00:41 -0800604 pset = zone_pcp(zone, smp_processor_id());
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700605 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
606 struct per_cpu_pages *pcp;
607
608 pcp = &pset->pcp[i];
Nick Piggin48db57f2006-01-08 01:00:42 -0800609 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
610 pcp->count = 0;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700611 }
612 }
613 local_irq_restore(flags);
614}
615#endif
616
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
618static void __drain_pages(unsigned int cpu)
619{
Nick Pigginc54ad302006-01-06 00:10:56 -0800620 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 struct zone *zone;
622 int i;
623
624 for_each_zone(zone) {
625 struct per_cpu_pageset *pset;
626
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700627 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
629 struct per_cpu_pages *pcp;
630
631 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800632 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800633 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
634 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800635 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 }
637 }
638}
639#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
640
641#ifdef CONFIG_PM
642
643void mark_free_pages(struct zone *zone)
644{
645 unsigned long zone_pfn, flags;
646 int order;
647 struct list_head *curr;
648
649 if (!zone->spanned_pages)
650 return;
651
652 spin_lock_irqsave(&zone->lock, flags);
653 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
654 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
655
656 for (order = MAX_ORDER - 1; order >= 0; --order)
657 list_for_each(curr, &zone->free_area[order].free_list) {
658 unsigned long start_pfn, i;
659
660 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
661
662 for (i=0; i < (1<<order); i++)
663 SetPageNosaveFree(pfn_to_page(start_pfn+i));
664 }
665 spin_unlock_irqrestore(&zone->lock, flags);
666}
667
668/*
669 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
670 */
671void drain_local_pages(void)
672{
673 unsigned long flags;
674
675 local_irq_save(flags);
676 __drain_pages(smp_processor_id());
677 local_irq_restore(flags);
678}
679#endif /* CONFIG_PM */
680
Nick Piggina74609f2006-01-06 00:11:20 -0800681static void zone_statistics(struct zonelist *zonelist, struct zone *z, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682{
683#ifdef CONFIG_NUMA
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 pg_data_t *pg = z->zone_pgdat;
685 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
686 struct per_cpu_pageset *p;
687
Nick Piggina74609f2006-01-06 00:11:20 -0800688 p = zone_pcp(z, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 if (pg == orig) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700690 p->numa_hit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 } else {
692 p->numa_miss++;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700693 zone_pcp(zonelist->zones[0], cpu)->numa_foreign++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 }
695 if (pg == NODE_DATA(numa_node_id()))
696 p->local_node++;
697 else
698 p->other_node++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699#endif
700}
701
702/*
703 * Free a 0-order page
704 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705static void fastcall free_hot_cold_page(struct page *page, int cold)
706{
707 struct zone *zone = page_zone(page);
708 struct per_cpu_pages *pcp;
709 unsigned long flags;
710
711 arch_free_page(page, 0);
712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 if (PageAnon(page))
714 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800715 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800716 return;
717
Hugh Dickins689bceb2005-11-21 21:32:20 -0800718 kernel_map_pages(page, 1, 0);
719
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700720 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800722 __inc_page_state(pgfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 list_add(&page->lru, &pcp->list);
724 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800725 if (pcp->count >= pcp->high) {
726 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
727 pcp->count -= pcp->batch;
728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 local_irq_restore(flags);
730 put_cpu();
731}
732
733void fastcall free_hot_page(struct page *page)
734{
735 free_hot_cold_page(page, 0);
736}
737
738void fastcall free_cold_page(struct page *page)
739{
740 free_hot_cold_page(page, 1);
741}
742
Al Virodd0fc662005-10-07 07:46:04 +0100743static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
745 int i;
746
747 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
748 for(i = 0; i < (1 << order); i++)
749 clear_highpage(page + i);
750}
751
752/*
753 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
754 * we cheat by calling it from here, in the order > 0 path. Saves a branch
755 * or two.
756 */
Nick Piggina74609f2006-01-06 00:11:20 -0800757static struct page *buffered_rmqueue(struct zonelist *zonelist,
758 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759{
760 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800761 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800763 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Hugh Dickins689bceb2005-11-21 21:32:20 -0800765again:
Nick Piggina74609f2006-01-06 00:11:20 -0800766 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800767 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 struct per_cpu_pages *pcp;
769
Nick Piggina74609f2006-01-06 00:11:20 -0800770 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800772 if (!pcp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 pcp->count += rmqueue_bulk(zone, 0,
774 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800775 if (unlikely(!pcp->count))
776 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 }
Nick Piggina74609f2006-01-06 00:11:20 -0800778 page = list_entry(pcp->list.next, struct page, lru);
779 list_del(&page->lru);
780 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800781 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 spin_lock_irqsave(&zone->lock, flags);
783 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800784 spin_unlock(&zone->lock);
785 if (!page)
786 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 }
788
Nick Piggina74609f2006-01-06 00:11:20 -0800789 __mod_page_state_zone(zone, pgalloc, 1 << order);
790 zone_statistics(zonelist, zone, cpu);
791 local_irq_restore(flags);
792 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
Nick Piggina74609f2006-01-06 00:11:20 -0800794 BUG_ON(bad_range(zone, page));
795 if (prep_new_page(page, order))
796 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
Nick Piggina74609f2006-01-06 00:11:20 -0800798 if (gfp_flags & __GFP_ZERO)
799 prep_zero_page(page, order, gfp_flags);
800
801 if (order && (gfp_flags & __GFP_COMP))
802 prep_compound_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800804
805failed:
806 local_irq_restore(flags);
807 put_cpu();
808 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809}
810
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800811#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800812#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
813#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
814#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
815#define ALLOC_HARDER 0x10 /* try to alloc harder */
816#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
817#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800818
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819/*
820 * Return 1 if free pages are above 'mark'. This takes into account the order
821 * of the allocation.
822 */
823int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800824 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825{
826 /* free_pages my go negative - that's OK */
827 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
828 int o;
829
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800830 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800832 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 min -= min / 4;
834
835 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
836 return 0;
837 for (o = 0; o < order; o++) {
838 /* At the next order, this order's pages become unavailable */
839 free_pages -= z->free_area[o].nr_free << o;
840
841 /* Require fewer higher order pages to be free */
842 min >>= 1;
843
844 if (free_pages <= min)
845 return 0;
846 }
847 return 1;
848}
849
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800850/*
851 * get_page_from_freeliest goes through the zonelist trying to allocate
852 * a page.
853 */
854static struct page *
855get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
856 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -0700857{
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800858 struct zone **z = zonelist->zones;
859 struct page *page = NULL;
860 int classzone_idx = zone_idx(*z);
861
862 /*
863 * Go through the zonelist once, looking for a zone with enough free.
864 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
865 */
866 do {
867 if ((alloc_flags & ALLOC_CPUSET) &&
868 !cpuset_zone_allowed(*z, gfp_mask))
869 continue;
870
871 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -0800872 unsigned long mark;
873 if (alloc_flags & ALLOC_WMARK_MIN)
874 mark = (*z)->pages_min;
875 else if (alloc_flags & ALLOC_WMARK_LOW)
876 mark = (*z)->pages_low;
877 else
878 mark = (*z)->pages_high;
879 if (!zone_watermark_ok(*z, order, mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800880 classzone_idx, alloc_flags))
881 continue;
882 }
883
Nick Piggina74609f2006-01-06 00:11:20 -0800884 page = buffered_rmqueue(zonelist, *z, order, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800885 if (page) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800886 break;
887 }
888 } while (*(++z) != NULL);
889 return page;
Martin Hicks753ee722005-06-21 17:14:41 -0700890}
891
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892/*
893 * This is the 'heart' of the zoned buddy allocator.
894 */
895struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +0100896__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 struct zonelist *zonelist)
898{
Al Viro260b2362005-10-21 03:22:44 -0400899 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800900 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 struct page *page;
902 struct reclaim_state reclaim_state;
903 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800905 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 int did_some_progress;
907
908 might_sleep_if(wait);
909
Jens Axboe6b1de912005-11-17 21:35:02 +0100910restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800911 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800913 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 /* Should this ever happen?? */
915 return NULL;
916 }
Jens Axboe6b1de912005-11-17 21:35:02 +0100917
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800918 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -0800919 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800920 if (page)
921 goto got_pg;
922
Jens Axboe6b1de912005-11-17 21:35:02 +0100923 do {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800924 wakeup_kswapd(*z, order);
Jens Axboe6b1de912005-11-17 21:35:02 +0100925 } while (*(++z));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800926
Paul Jackson9bf22292005-09-06 15:18:12 -0700927 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800928 * OK, we're below the kswapd watermark and have kicked background
929 * reclaim. Now things get more complex, so set up alloc_flags according
930 * to how we want to proceed.
931 *
932 * The caller may dip into page reserves a bit more if the caller
933 * cannot run direct reclaim, or if the caller has realtime scheduling
934 * policy.
Paul Jackson9bf22292005-09-06 15:18:12 -0700935 */
Nick Piggin31488902005-11-28 13:44:03 -0800936 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800937 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
938 alloc_flags |= ALLOC_HARDER;
939 if (gfp_mask & __GFP_HIGH)
940 alloc_flags |= ALLOC_HIGH;
Paul Jackson47f3a862006-01-06 00:10:32 -0800941 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
943 /*
944 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800945 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 *
947 * This is the last chance, in general, before the goto nopage.
948 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -0700949 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800951 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
952 if (page)
953 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
955 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -0700956
957 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
958 && !in_interrupt()) {
959 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -0800960nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -0700961 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800962 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -0800963 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800964 if (page)
965 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -0800966 if (gfp_mask & __GFP_NOFAIL) {
967 blk_congestion_wait(WRITE, HZ/50);
968 goto nofail_alloc;
969 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 }
971 goto nopage;
972 }
973
974 /* Atomic allocations - we can't balance anything */
975 if (!wait)
976 goto nopage;
977
978rebalance:
979 cond_resched();
980
981 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -0800982 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 p->flags |= PF_MEMALLOC;
984 reclaim_state.reclaimed_slab = 0;
985 p->reclaim_state = &reclaim_state;
986
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800987 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
989 p->reclaim_state = NULL;
990 p->flags &= ~PF_MEMALLOC;
991
992 cond_resched();
993
994 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800995 page = get_page_from_freelist(gfp_mask, order,
996 zonelist, alloc_flags);
997 if (page)
998 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1000 /*
1001 * Go through the zonelist yet one more time, keep
1002 * very high watermark here, this is only to catch
1003 * a parallel oom killing, we must fail if we're still
1004 * under heavy pressure.
1005 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001006 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001007 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001008 if (page)
1009 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010
Marcelo Tosatti79b9ce32005-07-07 17:56:04 -07001011 out_of_memory(gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 goto restart;
1013 }
1014
1015 /*
1016 * Don't let big-order allocations loop unless the caller explicitly
1017 * requests that. Wait for some write requests to complete then retry.
1018 *
1019 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1020 * <= 3, but that may not be true in other implementations.
1021 */
1022 do_retry = 0;
1023 if (!(gfp_mask & __GFP_NORETRY)) {
1024 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1025 do_retry = 1;
1026 if (gfp_mask & __GFP_NOFAIL)
1027 do_retry = 1;
1028 }
1029 if (do_retry) {
1030 blk_congestion_wait(WRITE, HZ/50);
1031 goto rebalance;
1032 }
1033
1034nopage:
1035 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1036 printk(KERN_WARNING "%s: page allocation failure."
1037 " order:%d, mode:0x%x\n",
1038 p->comm, order, gfp_mask);
1039 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001040 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 return page;
1044}
1045
1046EXPORT_SYMBOL(__alloc_pages);
1047
1048/*
1049 * Common helper functions.
1050 */
Al Virodd0fc662005-10-07 07:46:04 +01001051fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052{
1053 struct page * page;
1054 page = alloc_pages(gfp_mask, order);
1055 if (!page)
1056 return 0;
1057 return (unsigned long) page_address(page);
1058}
1059
1060EXPORT_SYMBOL(__get_free_pages);
1061
Al Virodd0fc662005-10-07 07:46:04 +01001062fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063{
1064 struct page * page;
1065
1066 /*
1067 * get_zeroed_page() returns a 32-bit address, which cannot represent
1068 * a highmem page
1069 */
Al Viro260b2362005-10-21 03:22:44 -04001070 BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071
1072 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1073 if (page)
1074 return (unsigned long) page_address(page);
1075 return 0;
1076}
1077
1078EXPORT_SYMBOL(get_zeroed_page);
1079
1080void __pagevec_free(struct pagevec *pvec)
1081{
1082 int i = pagevec_count(pvec);
1083
1084 while (--i >= 0)
1085 free_hot_cold_page(pvec->pages[i], pvec->cold);
1086}
1087
1088fastcall void __free_pages(struct page *page, unsigned int order)
1089{
Nick Pigginb5810032005-10-29 18:16:12 -07001090 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 if (order == 0)
1092 free_hot_page(page);
1093 else
1094 __free_pages_ok(page, order);
1095 }
1096}
1097
1098EXPORT_SYMBOL(__free_pages);
1099
1100fastcall void free_pages(unsigned long addr, unsigned int order)
1101{
1102 if (addr != 0) {
1103 BUG_ON(!virt_addr_valid((void *)addr));
1104 __free_pages(virt_to_page((void *)addr), order);
1105 }
1106}
1107
1108EXPORT_SYMBOL(free_pages);
1109
1110/*
1111 * Total amount of free (allocatable) RAM:
1112 */
1113unsigned int nr_free_pages(void)
1114{
1115 unsigned int sum = 0;
1116 struct zone *zone;
1117
1118 for_each_zone(zone)
1119 sum += zone->free_pages;
1120
1121 return sum;
1122}
1123
1124EXPORT_SYMBOL(nr_free_pages);
1125
1126#ifdef CONFIG_NUMA
1127unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1128{
1129 unsigned int i, sum = 0;
1130
1131 for (i = 0; i < MAX_NR_ZONES; i++)
1132 sum += pgdat->node_zones[i].free_pages;
1133
1134 return sum;
1135}
1136#endif
1137
1138static unsigned int nr_free_zone_pages(int offset)
1139{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001140 /* Just pick one node, since fallback list is circular */
1141 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 unsigned int sum = 0;
1143
Martin J. Blighe310fd42005-07-29 22:59:18 -07001144 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1145 struct zone **zonep = zonelist->zones;
1146 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
Martin J. Blighe310fd42005-07-29 22:59:18 -07001148 for (zone = *zonep++; zone; zone = *zonep++) {
1149 unsigned long size = zone->present_pages;
1150 unsigned long high = zone->pages_high;
1151 if (size > high)
1152 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 }
1154
1155 return sum;
1156}
1157
1158/*
1159 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1160 */
1161unsigned int nr_free_buffer_pages(void)
1162{
Al Viroaf4ca452005-10-21 02:55:38 -04001163 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164}
1165
1166/*
1167 * Amount of free RAM allocatable within all zones
1168 */
1169unsigned int nr_free_pagecache_pages(void)
1170{
Al Viroaf4ca452005-10-21 02:55:38 -04001171 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172}
1173
1174#ifdef CONFIG_HIGHMEM
1175unsigned int nr_free_highpages (void)
1176{
1177 pg_data_t *pgdat;
1178 unsigned int pages = 0;
1179
1180 for_each_pgdat(pgdat)
1181 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1182
1183 return pages;
1184}
1185#endif
1186
1187#ifdef CONFIG_NUMA
1188static void show_node(struct zone *zone)
1189{
1190 printk("Node %d ", zone->zone_pgdat->node_id);
1191}
1192#else
1193#define show_node(zone) do { } while (0)
1194#endif
1195
1196/*
1197 * Accumulate the page_state information across all CPUs.
1198 * The result is unavoidably approximate - it can change
1199 * during and after execution of this function.
1200 */
1201static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1202
1203atomic_t nr_pagecache = ATOMIC_INIT(0);
1204EXPORT_SYMBOL(nr_pagecache);
1205#ifdef CONFIG_SMP
1206DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1207#endif
1208
Nick Piggina86b1f52006-01-06 00:11:00 -08001209static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210{
1211 int cpu = 0;
1212
1213 memset(ret, 0, sizeof(*ret));
Andrew Morton84c20082006-01-08 01:00:28 -08001214 cpus_and(*cpumask, *cpumask, cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215
Martin Hicksc07e02d2005-09-03 15:55:11 -07001216 cpu = first_cpu(*cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 while (cpu < NR_CPUS) {
1218 unsigned long *in, *out, off;
1219
1220 in = (unsigned long *)&per_cpu(page_states, cpu);
1221
Martin Hicksc07e02d2005-09-03 15:55:11 -07001222 cpu = next_cpu(cpu, *cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223
1224 if (cpu < NR_CPUS)
1225 prefetch(&per_cpu(page_states, cpu));
1226
1227 out = (unsigned long *)ret;
1228 for (off = 0; off < nr; off++)
1229 *out++ += *in++;
1230 }
1231}
1232
Martin Hicksc07e02d2005-09-03 15:55:11 -07001233void get_page_state_node(struct page_state *ret, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234{
1235 int nr;
Martin Hicksc07e02d2005-09-03 15:55:11 -07001236 cpumask_t mask = node_to_cpumask(node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
1238 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1239 nr /= sizeof(unsigned long);
1240
Martin Hicksc07e02d2005-09-03 15:55:11 -07001241 __get_page_state(ret, nr+1, &mask);
1242}
1243
1244void get_page_state(struct page_state *ret)
1245{
1246 int nr;
1247 cpumask_t mask = CPU_MASK_ALL;
1248
1249 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1250 nr /= sizeof(unsigned long);
1251
1252 __get_page_state(ret, nr + 1, &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253}
1254
1255void get_full_page_state(struct page_state *ret)
1256{
Martin Hicksc07e02d2005-09-03 15:55:11 -07001257 cpumask_t mask = CPU_MASK_ALL;
1258
1259 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260}
1261
Nick Piggina74609f2006-01-06 00:11:20 -08001262unsigned long read_page_state_offset(unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263{
1264 unsigned long ret = 0;
1265 int cpu;
1266
Andrew Morton84c20082006-01-08 01:00:28 -08001267 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 unsigned long in;
1269
1270 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1271 ret += *((unsigned long *)in);
1272 }
1273 return ret;
1274}
1275
Nick Piggina74609f2006-01-06 00:11:20 -08001276void __mod_page_state_offset(unsigned long offset, unsigned long delta)
1277{
1278 void *ptr;
1279
1280 ptr = &__get_cpu_var(page_states);
1281 *(unsigned long *)(ptr + offset) += delta;
1282}
1283EXPORT_SYMBOL(__mod_page_state_offset);
1284
1285void mod_page_state_offset(unsigned long offset, unsigned long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286{
1287 unsigned long flags;
Nick Piggina74609f2006-01-06 00:11:20 -08001288 void *ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
1290 local_irq_save(flags);
1291 ptr = &__get_cpu_var(page_states);
Nick Piggina74609f2006-01-06 00:11:20 -08001292 *(unsigned long *)(ptr + offset) += delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 local_irq_restore(flags);
1294}
Nick Piggina74609f2006-01-06 00:11:20 -08001295EXPORT_SYMBOL(mod_page_state_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
1297void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1298 unsigned long *free, struct pglist_data *pgdat)
1299{
1300 struct zone *zones = pgdat->node_zones;
1301 int i;
1302
1303 *active = 0;
1304 *inactive = 0;
1305 *free = 0;
1306 for (i = 0; i < MAX_NR_ZONES; i++) {
1307 *active += zones[i].nr_active;
1308 *inactive += zones[i].nr_inactive;
1309 *free += zones[i].free_pages;
1310 }
1311}
1312
1313void get_zone_counts(unsigned long *active,
1314 unsigned long *inactive, unsigned long *free)
1315{
1316 struct pglist_data *pgdat;
1317
1318 *active = 0;
1319 *inactive = 0;
1320 *free = 0;
1321 for_each_pgdat(pgdat) {
1322 unsigned long l, m, n;
1323 __get_zone_counts(&l, &m, &n, pgdat);
1324 *active += l;
1325 *inactive += m;
1326 *free += n;
1327 }
1328}
1329
1330void si_meminfo(struct sysinfo *val)
1331{
1332 val->totalram = totalram_pages;
1333 val->sharedram = 0;
1334 val->freeram = nr_free_pages();
1335 val->bufferram = nr_blockdev_pages();
1336#ifdef CONFIG_HIGHMEM
1337 val->totalhigh = totalhigh_pages;
1338 val->freehigh = nr_free_highpages();
1339#else
1340 val->totalhigh = 0;
1341 val->freehigh = 0;
1342#endif
1343 val->mem_unit = PAGE_SIZE;
1344}
1345
1346EXPORT_SYMBOL(si_meminfo);
1347
1348#ifdef CONFIG_NUMA
1349void si_meminfo_node(struct sysinfo *val, int nid)
1350{
1351 pg_data_t *pgdat = NODE_DATA(nid);
1352
1353 val->totalram = pgdat->node_present_pages;
1354 val->freeram = nr_free_pages_pgdat(pgdat);
1355 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1356 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1357 val->mem_unit = PAGE_SIZE;
1358}
1359#endif
1360
1361#define K(x) ((x) << (PAGE_SHIFT-10))
1362
1363/*
1364 * Show free area list (used inside shift_scroll-lock stuff)
1365 * We also calculate the percentage fragmentation. We do this by counting the
1366 * memory on each free list with the exception of the first item on the list.
1367 */
1368void show_free_areas(void)
1369{
1370 struct page_state ps;
1371 int cpu, temperature;
1372 unsigned long active;
1373 unsigned long inactive;
1374 unsigned long free;
1375 struct zone *zone;
1376
1377 for_each_zone(zone) {
1378 show_node(zone);
1379 printk("%s per-cpu:", zone->name);
1380
Con Kolivasf3fe6512006-01-06 00:11:15 -08001381 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 printk(" empty\n");
1383 continue;
1384 } else
1385 printk("\n");
1386
Dave Jones6b482c62005-11-10 15:45:56 -05001387 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 struct per_cpu_pageset *pageset;
1389
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001390 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391
1392 for (temperature = 0; temperature < 2; temperature++)
Nick Piggin2d92c5c2006-01-06 00:10:59 -08001393 printk("cpu %d %s: high %d, batch %d used:%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 cpu,
1395 temperature ? "cold" : "hot",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 pageset->pcp[temperature].high,
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001397 pageset->pcp[temperature].batch,
1398 pageset->pcp[temperature].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 }
1400 }
1401
1402 get_page_state(&ps);
1403 get_zone_counts(&active, &inactive, &free);
1404
Denis Vlasenkoc0d62212005-06-21 17:15:14 -07001405 printk("Free pages: %11ukB (%ukB HighMem)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 K(nr_free_pages()),
1407 K(nr_free_highpages()));
1408
1409 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1410 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1411 active,
1412 inactive,
1413 ps.nr_dirty,
1414 ps.nr_writeback,
1415 ps.nr_unstable,
1416 nr_free_pages(),
1417 ps.nr_slab,
1418 ps.nr_mapped,
1419 ps.nr_page_table_pages);
1420
1421 for_each_zone(zone) {
1422 int i;
1423
1424 show_node(zone);
1425 printk("%s"
1426 " free:%lukB"
1427 " min:%lukB"
1428 " low:%lukB"
1429 " high:%lukB"
1430 " active:%lukB"
1431 " inactive:%lukB"
1432 " present:%lukB"
1433 " pages_scanned:%lu"
1434 " all_unreclaimable? %s"
1435 "\n",
1436 zone->name,
1437 K(zone->free_pages),
1438 K(zone->pages_min),
1439 K(zone->pages_low),
1440 K(zone->pages_high),
1441 K(zone->nr_active),
1442 K(zone->nr_inactive),
1443 K(zone->present_pages),
1444 zone->pages_scanned,
1445 (zone->all_unreclaimable ? "yes" : "no")
1446 );
1447 printk("lowmem_reserve[]:");
1448 for (i = 0; i < MAX_NR_ZONES; i++)
1449 printk(" %lu", zone->lowmem_reserve[i]);
1450 printk("\n");
1451 }
1452
1453 for_each_zone(zone) {
1454 unsigned long nr, flags, order, total = 0;
1455
1456 show_node(zone);
1457 printk("%s: ", zone->name);
Con Kolivasf3fe6512006-01-06 00:11:15 -08001458 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 printk("empty\n");
1460 continue;
1461 }
1462
1463 spin_lock_irqsave(&zone->lock, flags);
1464 for (order = 0; order < MAX_ORDER; order++) {
1465 nr = zone->free_area[order].nr_free;
1466 total += nr << order;
1467 printk("%lu*%lukB ", nr, K(1UL) << order);
1468 }
1469 spin_unlock_irqrestore(&zone->lock, flags);
1470 printk("= %lukB\n", K(total));
1471 }
1472
1473 show_swap_cache_info();
1474}
1475
1476/*
1477 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001478 *
1479 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 */
Christoph Lameter1a932052006-01-06 00:11:16 -08001481static int __init build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter070f8032006-01-06 00:11:19 -08001482 struct zonelist *zonelist, int nr_zones, int zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483{
Christoph Lameter1a932052006-01-06 00:11:16 -08001484 struct zone *zone;
1485
Christoph Lameter070f8032006-01-06 00:11:19 -08001486 BUG_ON(zone_type > ZONE_HIGHMEM);
Christoph Lameter02a68a52006-01-06 00:11:18 -08001487
1488 do {
Christoph Lameter070f8032006-01-06 00:11:19 -08001489 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001490 if (populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491#ifndef CONFIG_HIGHMEM
Christoph Lameter070f8032006-01-06 00:11:19 -08001492 BUG_ON(zone_type > ZONE_NORMAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493#endif
Christoph Lameter070f8032006-01-06 00:11:19 -08001494 zonelist->zones[nr_zones++] = zone;
1495 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 }
Christoph Lameter070f8032006-01-06 00:11:19 -08001497 zone_type--;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001498
Christoph Lameter070f8032006-01-06 00:11:19 -08001499 } while (zone_type >= 0);
1500 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501}
1502
Al Viro260b2362005-10-21 03:22:44 -04001503static inline int highest_zone(int zone_bits)
1504{
1505 int res = ZONE_NORMAL;
1506 if (zone_bits & (__force int)__GFP_HIGHMEM)
1507 res = ZONE_HIGHMEM;
Andi Kleena2f1b422005-11-05 17:25:53 +01001508 if (zone_bits & (__force int)__GFP_DMA32)
1509 res = ZONE_DMA32;
Al Viro260b2362005-10-21 03:22:44 -04001510 if (zone_bits & (__force int)__GFP_DMA)
1511 res = ZONE_DMA;
1512 return res;
1513}
1514
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515#ifdef CONFIG_NUMA
1516#define MAX_NODE_LOAD (num_online_nodes())
1517static int __initdata node_load[MAX_NUMNODES];
1518/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001519 * 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 -07001520 * @node: node whose fallback list we're appending
1521 * @used_node_mask: nodemask_t of already used nodes
1522 *
1523 * We use a number of factors to determine which is the next node that should
1524 * appear on a given node's fallback list. The node should not have appeared
1525 * already in @node's fallback list, and it should be the next closest node
1526 * according to the distance array (which contains arbitrary distance values
1527 * from each node to each node in the system), and should also prefer nodes
1528 * with no CPUs, since presumably they'll have very little allocation pressure
1529 * on them otherwise.
1530 * It returns -1 if no node is found.
1531 */
1532static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1533{
1534 int i, n, val;
1535 int min_val = INT_MAX;
1536 int best_node = -1;
1537
1538 for_each_online_node(i) {
1539 cpumask_t tmp;
1540
1541 /* Start from local node */
1542 n = (node+i) % num_online_nodes();
1543
1544 /* Don't want a node to appear more than once */
1545 if (node_isset(n, *used_node_mask))
1546 continue;
1547
1548 /* Use the local node if we haven't already */
1549 if (!node_isset(node, *used_node_mask)) {
1550 best_node = node;
1551 break;
1552 }
1553
1554 /* Use the distance array to find the distance */
1555 val = node_distance(node, n);
1556
1557 /* Give preference to headless and unused nodes */
1558 tmp = node_to_cpumask(n);
1559 if (!cpus_empty(tmp))
1560 val += PENALTY_FOR_NODE_WITH_CPUS;
1561
1562 /* Slight preference for less loaded node */
1563 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1564 val += node_load[n];
1565
1566 if (val < min_val) {
1567 min_val = val;
1568 best_node = n;
1569 }
1570 }
1571
1572 if (best_node >= 0)
1573 node_set(best_node, *used_node_mask);
1574
1575 return best_node;
1576}
1577
1578static void __init build_zonelists(pg_data_t *pgdat)
1579{
1580 int i, j, k, node, local_node;
1581 int prev_node, load;
1582 struct zonelist *zonelist;
1583 nodemask_t used_mask;
1584
1585 /* initialize zonelists */
1586 for (i = 0; i < GFP_ZONETYPES; i++) {
1587 zonelist = pgdat->node_zonelists + i;
1588 zonelist->zones[0] = NULL;
1589 }
1590
1591 /* NUMA-aware ordering of nodes */
1592 local_node = pgdat->node_id;
1593 load = num_online_nodes();
1594 prev_node = local_node;
1595 nodes_clear(used_mask);
1596 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1597 /*
1598 * We don't want to pressure a particular node.
1599 * So adding penalty to the first node in same
1600 * distance group to make it round-robin.
1601 */
1602 if (node_distance(local_node, node) !=
1603 node_distance(local_node, prev_node))
1604 node_load[node] += load;
1605 prev_node = node;
1606 load--;
1607 for (i = 0; i < GFP_ZONETYPES; i++) {
1608 zonelist = pgdat->node_zonelists + i;
1609 for (j = 0; zonelist->zones[j] != NULL; j++);
1610
Al Viro260b2362005-10-21 03:22:44 -04001611 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
1613 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1614 zonelist->zones[j] = NULL;
1615 }
1616 }
1617}
1618
1619#else /* CONFIG_NUMA */
1620
1621static void __init build_zonelists(pg_data_t *pgdat)
1622{
1623 int i, j, k, node, local_node;
1624
1625 local_node = pgdat->node_id;
1626 for (i = 0; i < GFP_ZONETYPES; i++) {
1627 struct zonelist *zonelist;
1628
1629 zonelist = pgdat->node_zonelists + i;
1630
1631 j = 0;
Al Viro260b2362005-10-21 03:22:44 -04001632 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 j = build_zonelists_node(pgdat, zonelist, j, k);
1634 /*
1635 * Now we build the zonelist so that it contains the zones
1636 * of all the other nodes.
1637 * We don't want to pressure a particular node, so when
1638 * building the zones for node N, we make sure that the
1639 * zones coming right after the local ones are those from
1640 * node N+1 (modulo N)
1641 */
1642 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1643 if (!node_online(node))
1644 continue;
1645 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1646 }
1647 for (node = 0; node < local_node; node++) {
1648 if (!node_online(node))
1649 continue;
1650 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1651 }
1652
1653 zonelist->zones[j] = NULL;
1654 }
1655}
1656
1657#endif /* CONFIG_NUMA */
1658
1659void __init build_all_zonelists(void)
1660{
1661 int i;
1662
1663 for_each_online_node(i)
1664 build_zonelists(NODE_DATA(i));
1665 printk("Built %i zonelists\n", num_online_nodes());
1666 cpuset_init_current_mems_allowed();
1667}
1668
1669/*
1670 * Helper functions to size the waitqueue hash table.
1671 * Essentially these want to choose hash table sizes sufficiently
1672 * large so that collisions trying to wait on pages are rare.
1673 * But in fact, the number of active page waitqueues on typical
1674 * systems is ridiculously low, less than 200. So this is even
1675 * conservative, even though it seems large.
1676 *
1677 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1678 * waitqueues, i.e. the size of the waitq table given the number of pages.
1679 */
1680#define PAGES_PER_WAITQUEUE 256
1681
1682static inline unsigned long wait_table_size(unsigned long pages)
1683{
1684 unsigned long size = 1;
1685
1686 pages /= PAGES_PER_WAITQUEUE;
1687
1688 while (size < pages)
1689 size <<= 1;
1690
1691 /*
1692 * Once we have dozens or even hundreds of threads sleeping
1693 * on IO we've got bigger problems than wait queue collision.
1694 * Limit the size of the wait table to a reasonable size.
1695 */
1696 size = min(size, 4096UL);
1697
1698 return max(size, 4UL);
1699}
1700
1701/*
1702 * This is an integer logarithm so that shifts can be used later
1703 * to extract the more random high bits from the multiplicative
1704 * hash function before the remainder is taken.
1705 */
1706static inline unsigned long wait_table_bits(unsigned long size)
1707{
1708 return ffz(~size);
1709}
1710
1711#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1712
1713static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1714 unsigned long *zones_size, unsigned long *zholes_size)
1715{
1716 unsigned long realtotalpages, totalpages = 0;
1717 int i;
1718
1719 for (i = 0; i < MAX_NR_ZONES; i++)
1720 totalpages += zones_size[i];
1721 pgdat->node_spanned_pages = totalpages;
1722
1723 realtotalpages = totalpages;
1724 if (zholes_size)
1725 for (i = 0; i < MAX_NR_ZONES; i++)
1726 realtotalpages -= zholes_size[i];
1727 pgdat->node_present_pages = realtotalpages;
1728 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1729}
1730
1731
1732/*
1733 * Initially all pages are reserved - free ones are freed
1734 * up by free_all_bootmem() once the early boot process is
1735 * done. Non-atomic initialization, single-pass.
1736 */
Dave Hansen3947be12005-10-29 18:16:54 -07001737void __devinit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 unsigned long start_pfn)
1739{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001741 unsigned long end_pfn = start_pfn + size;
1742 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001744 for (pfn = start_pfn; pfn < end_pfn; pfn++, page++) {
1745 if (!early_pfn_valid(pfn))
1746 continue;
1747 page = pfn_to_page(pfn);
1748 set_page_links(page, zone, nid, pfn);
Nick Pigginb5810032005-10-29 18:16:12 -07001749 set_page_count(page, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 reset_page_mapcount(page);
1751 SetPageReserved(page);
1752 INIT_LIST_HEAD(&page->lru);
1753#ifdef WANT_PAGE_VIRTUAL
1754 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1755 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001756 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 }
1759}
1760
1761void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1762 unsigned long size)
1763{
1764 int order;
1765 for (order = 0; order < MAX_ORDER ; order++) {
1766 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1767 zone->free_area[order].nr_free = 0;
1768 }
1769}
1770
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001771#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1772void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1773 unsigned long size)
1774{
1775 unsigned long snum = pfn_to_section_nr(pfn);
1776 unsigned long end = pfn_to_section_nr(pfn + size);
1777
1778 if (FLAGS_HAS_NODE)
1779 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1780 else
1781 for (; snum <= end; snum++)
1782 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1783}
1784
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785#ifndef __HAVE_ARCH_MEMMAP_INIT
1786#define memmap_init(size, nid, zone, start_pfn) \
1787 memmap_init_zone((size), (nid), (zone), (start_pfn))
1788#endif
1789
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001790static int __devinit zone_batchsize(struct zone *zone)
1791{
1792 int batch;
1793
1794 /*
1795 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001796 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001797 *
1798 * OK, so we don't know how big the cache is. So guess.
1799 */
1800 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001801 if (batch * PAGE_SIZE > 512 * 1024)
1802 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001803 batch /= 4; /* We effectively *= 4 below */
1804 if (batch < 1)
1805 batch = 1;
1806
1807 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001808 * Clamp the batch to a 2^n - 1 value. Having a power
1809 * of 2 value was found to be more likely to have
1810 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001811 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001812 * For example if 2 tasks are alternately allocating
1813 * batches of pages, one task can end up with a lot
1814 * of pages of one half of the possible page colors
1815 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001816 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001817 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001818
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001819 return batch;
1820}
1821
Christoph Lameter2caaad42005-06-21 17:15:00 -07001822inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1823{
1824 struct per_cpu_pages *pcp;
1825
Magnus Damm1c6fe942005-10-26 01:58:59 -07001826 memset(p, 0, sizeof(*p));
1827
Christoph Lameter2caaad42005-06-21 17:15:00 -07001828 pcp = &p->pcp[0]; /* hot */
1829 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001830 pcp->high = 6 * batch;
1831 pcp->batch = max(1UL, 1 * batch);
1832 INIT_LIST_HEAD(&pcp->list);
1833
1834 pcp = &p->pcp[1]; /* cold*/
1835 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001836 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001837 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001838 INIT_LIST_HEAD(&pcp->list);
1839}
1840
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001841/*
1842 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1843 * to the value high for the pageset p.
1844 */
1845
1846static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1847 unsigned long high)
1848{
1849 struct per_cpu_pages *pcp;
1850
1851 pcp = &p->pcp[0]; /* hot list */
1852 pcp->high = high;
1853 pcp->batch = max(1UL, high/4);
1854 if ((high/4) > (PAGE_SHIFT * 8))
1855 pcp->batch = PAGE_SHIFT * 8;
1856}
1857
1858
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001859#ifdef CONFIG_NUMA
1860/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001861 * Boot pageset table. One per cpu which is going to be used for all
1862 * zones and all nodes. The parameters will be set in such a way
1863 * that an item put on a list will immediately be handed over to
1864 * the buddy list. This is safe since pageset manipulation is done
1865 * with interrupts disabled.
1866 *
1867 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001868 *
1869 * The boot_pagesets must be kept even after bootup is complete for
1870 * unused processors and/or zones. They do play a role for bootstrapping
1871 * hotplugged processors.
1872 *
1873 * zoneinfo_show() and maybe other functions do
1874 * not check if the processor is online before following the pageset pointer.
1875 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001876 */
1877static struct per_cpu_pageset
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001878 boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001879
1880/*
1881 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001882 * per cpu pageset array in struct zone.
1883 */
1884static int __devinit process_zones(int cpu)
1885{
1886 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001887
1888 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001889
Nick Piggin23316bc2006-01-08 01:00:41 -08001890 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001891 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08001892 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001893 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001894
Nick Piggin23316bc2006-01-08 01:00:41 -08001895 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001896
1897 if (percpu_pagelist_fraction)
1898 setup_pagelist_highmark(zone_pcp(zone, cpu),
1899 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001900 }
1901
1902 return 0;
1903bad:
1904 for_each_zone(dzone) {
1905 if (dzone == zone)
1906 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08001907 kfree(zone_pcp(dzone, cpu));
1908 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001909 }
1910 return -ENOMEM;
1911}
1912
1913static inline void free_zone_pagesets(int cpu)
1914{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001915 struct zone *zone;
1916
1917 for_each_zone(zone) {
1918 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1919
1920 zone_pcp(zone, cpu) = NULL;
1921 kfree(pset);
1922 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001923}
1924
1925static int __devinit pageset_cpuup_callback(struct notifier_block *nfb,
1926 unsigned long action,
1927 void *hcpu)
1928{
1929 int cpu = (long)hcpu;
1930 int ret = NOTIFY_OK;
1931
1932 switch (action) {
1933 case CPU_UP_PREPARE:
1934 if (process_zones(cpu))
1935 ret = NOTIFY_BAD;
1936 break;
Andi Kleenb0d41692005-11-05 17:25:53 +01001937 case CPU_UP_CANCELED:
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001938 case CPU_DEAD:
1939 free_zone_pagesets(cpu);
1940 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001941 default:
1942 break;
1943 }
1944 return ret;
1945}
1946
1947static struct notifier_block pageset_notifier =
1948 { &pageset_cpuup_callback, NULL, 0 };
1949
Al Viro78d99552005-12-15 09:18:25 +00001950void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001951{
1952 int err;
1953
1954 /* Initialize per_cpu_pageset for cpu 0.
1955 * A cpuup callback will do this for every cpu
1956 * as it comes online
1957 */
1958 err = process_zones(smp_processor_id());
1959 BUG_ON(err);
1960 register_cpu_notifier(&pageset_notifier);
1961}
1962
1963#endif
1964
Dave Hansened8ece22005-10-29 18:16:50 -07001965static __devinit
1966void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
1967{
1968 int i;
1969 struct pglist_data *pgdat = zone->zone_pgdat;
1970
1971 /*
1972 * The per-page waitqueue mechanism uses hashed waitqueues
1973 * per zone.
1974 */
1975 zone->wait_table_size = wait_table_size(zone_size_pages);
1976 zone->wait_table_bits = wait_table_bits(zone->wait_table_size);
1977 zone->wait_table = (wait_queue_head_t *)
1978 alloc_bootmem_node(pgdat, zone->wait_table_size
1979 * sizeof(wait_queue_head_t));
1980
1981 for(i = 0; i < zone->wait_table_size; ++i)
1982 init_waitqueue_head(zone->wait_table + i);
1983}
1984
1985static __devinit void zone_pcp_init(struct zone *zone)
1986{
1987 int cpu;
1988 unsigned long batch = zone_batchsize(zone);
1989
1990 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1991#ifdef CONFIG_NUMA
1992 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08001993 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07001994 setup_pageset(&boot_pageset[cpu],0);
1995#else
1996 setup_pageset(zone_pcp(zone,cpu), batch);
1997#endif
1998 }
1999 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2000 zone->name, zone->present_pages, batch);
2001}
2002
2003static __devinit void init_currently_empty_zone(struct zone *zone,
2004 unsigned long zone_start_pfn, unsigned long size)
2005{
2006 struct pglist_data *pgdat = zone->zone_pgdat;
2007
2008 zone_wait_table_init(zone, size);
2009 pgdat->nr_zones = zone_idx(zone) + 1;
2010
2011 zone->zone_mem_map = pfn_to_page(zone_start_pfn);
2012 zone->zone_start_pfn = zone_start_pfn;
2013
2014 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2015
2016 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2017}
2018
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019/*
2020 * Set up the zone data structures:
2021 * - mark all pages reserved
2022 * - mark all memory queues empty
2023 * - clear the memory bitmaps
2024 */
2025static void __init free_area_init_core(struct pglist_data *pgdat,
2026 unsigned long *zones_size, unsigned long *zholes_size)
2027{
Dave Hansened8ece22005-10-29 18:16:50 -07002028 unsigned long j;
2029 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 unsigned long zone_start_pfn = pgdat->node_start_pfn;
2031
Dave Hansen208d54e2005-10-29 18:16:52 -07002032 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 pgdat->nr_zones = 0;
2034 init_waitqueue_head(&pgdat->kswapd_wait);
2035 pgdat->kswapd_max_order = 0;
2036
2037 for (j = 0; j < MAX_NR_ZONES; j++) {
2038 struct zone *zone = pgdat->node_zones + j;
2039 unsigned long size, realsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 realsize = size = zones_size[j];
2042 if (zholes_size)
2043 realsize -= zholes_size[j];
2044
Andi Kleena2f1b422005-11-05 17:25:53 +01002045 if (j < ZONE_HIGHMEM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 nr_kernel_pages += realsize;
2047 nr_all_pages += realsize;
2048
2049 zone->spanned_pages = size;
2050 zone->present_pages = realsize;
2051 zone->name = zone_names[j];
2052 spin_lock_init(&zone->lock);
2053 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002054 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 zone->zone_pgdat = pgdat;
2056 zone->free_pages = 0;
2057
2058 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2059
Dave Hansened8ece22005-10-29 18:16:50 -07002060 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 INIT_LIST_HEAD(&zone->active_list);
2062 INIT_LIST_HEAD(&zone->inactive_list);
2063 zone->nr_scan_active = 0;
2064 zone->nr_scan_inactive = 0;
2065 zone->nr_active = 0;
2066 zone->nr_inactive = 0;
Martin Hicks53e9a612005-09-03 15:54:51 -07002067 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 if (!size)
2069 continue;
2070
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002071 zonetable_add(zone, nid, j, zone_start_pfn, size);
Dave Hansened8ece22005-10-29 18:16:50 -07002072 init_currently_empty_zone(zone, zone_start_pfn, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 }
2075}
2076
2077static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2078{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 /* Skip empty nodes */
2080 if (!pgdat->node_spanned_pages)
2081 return;
2082
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002083#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 /* ia64 gets its own node_mem_map, before this, without bootmem */
2085 if (!pgdat->node_mem_map) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002086 unsigned long size;
2087 struct page *map;
2088
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002090 map = alloc_remap(pgdat->node_id, size);
2091 if (!map)
2092 map = alloc_bootmem_node(pgdat, size);
2093 pgdat->node_mem_map = map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002095#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 /*
2097 * With no DISCONTIG, the global mem_map is just set as node 0's
2098 */
2099 if (pgdat == NODE_DATA(0))
2100 mem_map = NODE_DATA(0)->node_mem_map;
2101#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002102#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103}
2104
2105void __init free_area_init_node(int nid, struct pglist_data *pgdat,
2106 unsigned long *zones_size, unsigned long node_start_pfn,
2107 unsigned long *zholes_size)
2108{
2109 pgdat->node_id = nid;
2110 pgdat->node_start_pfn = node_start_pfn;
2111 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2112
2113 alloc_node_mem_map(pgdat);
2114
2115 free_area_init_core(pgdat, zones_size, zholes_size);
2116}
2117
Dave Hansen93b75042005-06-23 00:07:47 -07002118#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119static bootmem_data_t contig_bootmem_data;
2120struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2121
2122EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002123#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
2125void __init free_area_init(unsigned long *zones_size)
2126{
Dave Hansen93b75042005-06-23 00:07:47 -07002127 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2129}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130
2131#ifdef CONFIG_PROC_FS
2132
2133#include <linux/seq_file.h>
2134
2135static void *frag_start(struct seq_file *m, loff_t *pos)
2136{
2137 pg_data_t *pgdat;
2138 loff_t node = *pos;
2139
2140 for (pgdat = pgdat_list; pgdat && node; pgdat = pgdat->pgdat_next)
2141 --node;
2142
2143 return pgdat;
2144}
2145
2146static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
2147{
2148 pg_data_t *pgdat = (pg_data_t *)arg;
2149
2150 (*pos)++;
2151 return pgdat->pgdat_next;
2152}
2153
2154static void frag_stop(struct seq_file *m, void *arg)
2155{
2156}
2157
2158/*
2159 * This walks the free areas for each zone.
2160 */
2161static int frag_show(struct seq_file *m, void *arg)
2162{
2163 pg_data_t *pgdat = (pg_data_t *)arg;
2164 struct zone *zone;
2165 struct zone *node_zones = pgdat->node_zones;
2166 unsigned long flags;
2167 int order;
2168
2169 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
Con Kolivasf3fe6512006-01-06 00:11:15 -08002170 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 continue;
2172
2173 spin_lock_irqsave(&zone->lock, flags);
2174 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2175 for (order = 0; order < MAX_ORDER; ++order)
2176 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2177 spin_unlock_irqrestore(&zone->lock, flags);
2178 seq_putc(m, '\n');
2179 }
2180 return 0;
2181}
2182
2183struct seq_operations fragmentation_op = {
2184 .start = frag_start,
2185 .next = frag_next,
2186 .stop = frag_stop,
2187 .show = frag_show,
2188};
2189
Nikita Danilov295ab932005-06-21 17:14:38 -07002190/*
2191 * Output information about zones in @pgdat.
2192 */
2193static int zoneinfo_show(struct seq_file *m, void *arg)
2194{
2195 pg_data_t *pgdat = arg;
2196 struct zone *zone;
2197 struct zone *node_zones = pgdat->node_zones;
2198 unsigned long flags;
2199
2200 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2201 int i;
2202
Con Kolivasf3fe6512006-01-06 00:11:15 -08002203 if (!populated_zone(zone))
Nikita Danilov295ab932005-06-21 17:14:38 -07002204 continue;
2205
2206 spin_lock_irqsave(&zone->lock, flags);
2207 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2208 seq_printf(m,
2209 "\n pages free %lu"
2210 "\n min %lu"
2211 "\n low %lu"
2212 "\n high %lu"
2213 "\n active %lu"
2214 "\n inactive %lu"
2215 "\n scanned %lu (a: %lu i: %lu)"
2216 "\n spanned %lu"
2217 "\n present %lu",
2218 zone->free_pages,
2219 zone->pages_min,
2220 zone->pages_low,
2221 zone->pages_high,
2222 zone->nr_active,
2223 zone->nr_inactive,
2224 zone->pages_scanned,
2225 zone->nr_scan_active, zone->nr_scan_inactive,
2226 zone->spanned_pages,
2227 zone->present_pages);
2228 seq_printf(m,
2229 "\n protection: (%lu",
2230 zone->lowmem_reserve[0]);
2231 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2232 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2233 seq_printf(m,
2234 ")"
2235 "\n pagesets");
Nick Piggin23316bc2006-01-08 01:00:41 -08002236 for_each_online_cpu(i) {
Nikita Danilov295ab932005-06-21 17:14:38 -07002237 struct per_cpu_pageset *pageset;
2238 int j;
2239
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002240 pageset = zone_pcp(zone, i);
Nikita Danilov295ab932005-06-21 17:14:38 -07002241 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2242 if (pageset->pcp[j].count)
2243 break;
2244 }
2245 if (j == ARRAY_SIZE(pageset->pcp))
2246 continue;
2247 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2248 seq_printf(m,
2249 "\n cpu: %i pcp: %i"
2250 "\n count: %i"
Nikita Danilov295ab932005-06-21 17:14:38 -07002251 "\n high: %i"
2252 "\n batch: %i",
2253 i, j,
2254 pageset->pcp[j].count,
Nikita Danilov295ab932005-06-21 17:14:38 -07002255 pageset->pcp[j].high,
2256 pageset->pcp[j].batch);
2257 }
2258#ifdef CONFIG_NUMA
2259 seq_printf(m,
2260 "\n numa_hit: %lu"
2261 "\n numa_miss: %lu"
2262 "\n numa_foreign: %lu"
2263 "\n interleave_hit: %lu"
2264 "\n local_node: %lu"
2265 "\n other_node: %lu",
2266 pageset->numa_hit,
2267 pageset->numa_miss,
2268 pageset->numa_foreign,
2269 pageset->interleave_hit,
2270 pageset->local_node,
2271 pageset->other_node);
2272#endif
2273 }
2274 seq_printf(m,
2275 "\n all_unreclaimable: %u"
2276 "\n prev_priority: %i"
2277 "\n temp_priority: %i"
2278 "\n start_pfn: %lu",
2279 zone->all_unreclaimable,
2280 zone->prev_priority,
2281 zone->temp_priority,
2282 zone->zone_start_pfn);
2283 spin_unlock_irqrestore(&zone->lock, flags);
2284 seq_putc(m, '\n');
2285 }
2286 return 0;
2287}
2288
2289struct seq_operations zoneinfo_op = {
2290 .start = frag_start, /* iterate over all zones. The same as in
2291 * fragmentation. */
2292 .next = frag_next,
2293 .stop = frag_stop,
2294 .show = zoneinfo_show,
2295};
2296
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297static char *vmstat_text[] = {
2298 "nr_dirty",
2299 "nr_writeback",
2300 "nr_unstable",
2301 "nr_page_table_pages",
2302 "nr_mapped",
2303 "nr_slab",
2304
2305 "pgpgin",
2306 "pgpgout",
2307 "pswpin",
2308 "pswpout",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309
Nick Piggin9328b8f2006-01-06 00:11:10 -08002310 "pgalloc_high",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 "pgalloc_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002312 "pgalloc_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 "pgalloc_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002314
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 "pgfree",
2316 "pgactivate",
2317 "pgdeactivate",
2318
2319 "pgfault",
2320 "pgmajfault",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002321
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 "pgrefill_high",
2323 "pgrefill_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002324 "pgrefill_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 "pgrefill_dma",
2326
2327 "pgsteal_high",
2328 "pgsteal_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002329 "pgsteal_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 "pgsteal_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002331
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 "pgscan_kswapd_high",
2333 "pgscan_kswapd_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002334 "pgscan_kswapd_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 "pgscan_kswapd_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002336
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 "pgscan_direct_high",
2338 "pgscan_direct_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002339 "pgscan_direct_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 "pgscan_direct_dma",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
Nick Piggin9328b8f2006-01-06 00:11:10 -08002342 "pginodesteal",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 "slabs_scanned",
2344 "kswapd_steal",
2345 "kswapd_inodesteal",
2346 "pageoutrun",
2347 "allocstall",
2348
2349 "pgrotated",
KAMEZAWA Hiroyukiedfbe2b2005-05-01 08:58:37 -07002350 "nr_bounce",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351};
2352
2353static void *vmstat_start(struct seq_file *m, loff_t *pos)
2354{
2355 struct page_state *ps;
2356
2357 if (*pos >= ARRAY_SIZE(vmstat_text))
2358 return NULL;
2359
2360 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2361 m->private = ps;
2362 if (!ps)
2363 return ERR_PTR(-ENOMEM);
2364 get_full_page_state(ps);
2365 ps->pgpgin /= 2; /* sectors -> kbytes */
2366 ps->pgpgout /= 2;
2367 return (unsigned long *)ps + *pos;
2368}
2369
2370static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2371{
2372 (*pos)++;
2373 if (*pos >= ARRAY_SIZE(vmstat_text))
2374 return NULL;
2375 return (unsigned long *)m->private + *pos;
2376}
2377
2378static int vmstat_show(struct seq_file *m, void *arg)
2379{
2380 unsigned long *l = arg;
2381 unsigned long off = l - (unsigned long *)m->private;
2382
2383 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2384 return 0;
2385}
2386
2387static void vmstat_stop(struct seq_file *m, void *arg)
2388{
2389 kfree(m->private);
2390 m->private = NULL;
2391}
2392
2393struct seq_operations vmstat_op = {
2394 .start = vmstat_start,
2395 .next = vmstat_next,
2396 .stop = vmstat_stop,
2397 .show = vmstat_show,
2398};
2399
2400#endif /* CONFIG_PROC_FS */
2401
2402#ifdef CONFIG_HOTPLUG_CPU
2403static int page_alloc_cpu_notify(struct notifier_block *self,
2404 unsigned long action, void *hcpu)
2405{
2406 int cpu = (unsigned long)hcpu;
2407 long *count;
2408 unsigned long *src, *dest;
2409
2410 if (action == CPU_DEAD) {
2411 int i;
2412
2413 /* Drain local pagecache count. */
2414 count = &per_cpu(nr_pagecache_local, cpu);
2415 atomic_add(*count, &nr_pagecache);
2416 *count = 0;
2417 local_irq_disable();
2418 __drain_pages(cpu);
2419
2420 /* Add dead cpu's page_states to our own. */
2421 dest = (unsigned long *)&__get_cpu_var(page_states);
2422 src = (unsigned long *)&per_cpu(page_states, cpu);
2423
2424 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2425 i++) {
2426 dest[i] += src[i];
2427 src[i] = 0;
2428 }
2429
2430 local_irq_enable();
2431 }
2432 return NOTIFY_OK;
2433}
2434#endif /* CONFIG_HOTPLUG_CPU */
2435
2436void __init page_alloc_init(void)
2437{
2438 hotcpu_notifier(page_alloc_cpu_notify, 0);
2439}
2440
2441/*
2442 * setup_per_zone_lowmem_reserve - called whenever
2443 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2444 * has a correct pages reserved value, so an adequate number of
2445 * pages are left in the zone after a successful __alloc_pages().
2446 */
2447static void setup_per_zone_lowmem_reserve(void)
2448{
2449 struct pglist_data *pgdat;
2450 int j, idx;
2451
2452 for_each_pgdat(pgdat) {
2453 for (j = 0; j < MAX_NR_ZONES; j++) {
2454 struct zone *zone = pgdat->node_zones + j;
2455 unsigned long present_pages = zone->present_pages;
2456
2457 zone->lowmem_reserve[j] = 0;
2458
2459 for (idx = j-1; idx >= 0; idx--) {
2460 struct zone *lower_zone;
2461
2462 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2463 sysctl_lowmem_reserve_ratio[idx] = 1;
2464
2465 lower_zone = pgdat->node_zones + idx;
2466 lower_zone->lowmem_reserve[j] = present_pages /
2467 sysctl_lowmem_reserve_ratio[idx];
2468 present_pages += lower_zone->present_pages;
2469 }
2470 }
2471 }
2472}
2473
2474/*
2475 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2476 * that the pages_{min,low,high} values for each zone are set correctly
2477 * with respect to min_free_kbytes.
2478 */
Dave Hansen3947be12005-10-29 18:16:54 -07002479void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480{
2481 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2482 unsigned long lowmem_pages = 0;
2483 struct zone *zone;
2484 unsigned long flags;
2485
2486 /* Calculate total number of !ZONE_HIGHMEM pages */
2487 for_each_zone(zone) {
2488 if (!is_highmem(zone))
2489 lowmem_pages += zone->present_pages;
2490 }
2491
2492 for_each_zone(zone) {
Nick Piggin669ed172005-11-13 16:06:45 -08002493 unsigned long tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 spin_lock_irqsave(&zone->lru_lock, flags);
Nick Piggin669ed172005-11-13 16:06:45 -08002495 tmp = (pages_min * zone->present_pages) / lowmem_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 if (is_highmem(zone)) {
2497 /*
Nick Piggin669ed172005-11-13 16:06:45 -08002498 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2499 * need highmem pages, so cap pages_min to a small
2500 * value here.
2501 *
2502 * The (pages_high-pages_low) and (pages_low-pages_min)
2503 * deltas controls asynch page reclaim, and so should
2504 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 */
2506 int min_pages;
2507
2508 min_pages = zone->present_pages / 1024;
2509 if (min_pages < SWAP_CLUSTER_MAX)
2510 min_pages = SWAP_CLUSTER_MAX;
2511 if (min_pages > 128)
2512 min_pages = 128;
2513 zone->pages_min = min_pages;
2514 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08002515 /*
2516 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 * proportionate to the zone's size.
2518 */
Nick Piggin669ed172005-11-13 16:06:45 -08002519 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 }
2521
Nick Piggin669ed172005-11-13 16:06:45 -08002522 zone->pages_low = zone->pages_min + tmp / 4;
2523 zone->pages_high = zone->pages_min + tmp / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 spin_unlock_irqrestore(&zone->lru_lock, flags);
2525 }
2526}
2527
2528/*
2529 * Initialise min_free_kbytes.
2530 *
2531 * For small machines we want it small (128k min). For large machines
2532 * we want it large (64MB max). But it is not linear, because network
2533 * bandwidth does not increase linearly with machine size. We use
2534 *
2535 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2536 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2537 *
2538 * which yields
2539 *
2540 * 16MB: 512k
2541 * 32MB: 724k
2542 * 64MB: 1024k
2543 * 128MB: 1448k
2544 * 256MB: 2048k
2545 * 512MB: 2896k
2546 * 1024MB: 4096k
2547 * 2048MB: 5792k
2548 * 4096MB: 8192k
2549 * 8192MB: 11584k
2550 * 16384MB: 16384k
2551 */
2552static int __init init_per_zone_pages_min(void)
2553{
2554 unsigned long lowmem_kbytes;
2555
2556 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2557
2558 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2559 if (min_free_kbytes < 128)
2560 min_free_kbytes = 128;
2561 if (min_free_kbytes > 65536)
2562 min_free_kbytes = 65536;
2563 setup_per_zone_pages_min();
2564 setup_per_zone_lowmem_reserve();
2565 return 0;
2566}
2567module_init(init_per_zone_pages_min)
2568
2569/*
2570 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2571 * that we can call two helper functions whenever min_free_kbytes
2572 * changes.
2573 */
2574int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2575 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2576{
2577 proc_dointvec(table, write, file, buffer, length, ppos);
2578 setup_per_zone_pages_min();
2579 return 0;
2580}
2581
2582/*
2583 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2584 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2585 * whenever sysctl_lowmem_reserve_ratio changes.
2586 *
2587 * The reserve ratio obviously has absolutely no relation with the
2588 * pages_min watermarks. The lowmem reserve ratio can only make sense
2589 * if in function of the boot time zone sizes.
2590 */
2591int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2592 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2593{
2594 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2595 setup_per_zone_lowmem_reserve();
2596 return 0;
2597}
2598
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002599/*
2600 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2601 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2602 * can have before it gets flushed back to buddy allocator.
2603 */
2604
2605int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2606 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2607{
2608 struct zone *zone;
2609 unsigned int cpu;
2610 int ret;
2611
2612 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2613 if (!write || (ret == -EINVAL))
2614 return ret;
2615 for_each_zone(zone) {
2616 for_each_online_cpu(cpu) {
2617 unsigned long high;
2618 high = zone->present_pages / percpu_pagelist_fraction;
2619 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2620 }
2621 }
2622 return 0;
2623}
2624
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625__initdata int hashdist = HASHDIST_DEFAULT;
2626
2627#ifdef CONFIG_NUMA
2628static int __init set_hashdist(char *str)
2629{
2630 if (!str)
2631 return 0;
2632 hashdist = simple_strtoul(str, &str, 0);
2633 return 1;
2634}
2635__setup("hashdist=", set_hashdist);
2636#endif
2637
2638/*
2639 * allocate a large system hash table from bootmem
2640 * - it is assumed that the hash table must contain an exact power-of-2
2641 * quantity of entries
2642 * - limit is the number of hash buckets, not the total allocation size
2643 */
2644void *__init alloc_large_system_hash(const char *tablename,
2645 unsigned long bucketsize,
2646 unsigned long numentries,
2647 int scale,
2648 int flags,
2649 unsigned int *_hash_shift,
2650 unsigned int *_hash_mask,
2651 unsigned long limit)
2652{
2653 unsigned long long max = limit;
2654 unsigned long log2qty, size;
2655 void *table = NULL;
2656
2657 /* allow the kernel cmdline to have a say */
2658 if (!numentries) {
2659 /* round applicable memory size up to nearest megabyte */
2660 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2661 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2662 numentries >>= 20 - PAGE_SHIFT;
2663 numentries <<= 20 - PAGE_SHIFT;
2664
2665 /* limit to 1 bucket per 2^scale bytes of low memory */
2666 if (scale > PAGE_SHIFT)
2667 numentries >>= (scale - PAGE_SHIFT);
2668 else
2669 numentries <<= (PAGE_SHIFT - scale);
2670 }
2671 /* rounded up to nearest power of 2 in size */
2672 numentries = 1UL << (long_log2(numentries) + 1);
2673
2674 /* limit allocation size to 1/16 total memory by default */
2675 if (max == 0) {
2676 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2677 do_div(max, bucketsize);
2678 }
2679
2680 if (numentries > max)
2681 numentries = max;
2682
2683 log2qty = long_log2(numentries);
2684
2685 do {
2686 size = bucketsize << log2qty;
2687 if (flags & HASH_EARLY)
2688 table = alloc_bootmem(size);
2689 else if (hashdist)
2690 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2691 else {
2692 unsigned long order;
2693 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2694 ;
2695 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2696 }
2697 } while (!table && size > PAGE_SIZE && --log2qty);
2698
2699 if (!table)
2700 panic("Failed to allocate %s hash table\n", tablename);
2701
2702 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2703 tablename,
2704 (1U << log2qty),
2705 long_log2(size) - PAGE_SHIFT,
2706 size);
2707
2708 if (_hash_shift)
2709 *_hash_shift = log2qty;
2710 if (_hash_mask)
2711 *_hash_mask = (1 << log2qty) - 1;
2712
2713 return table;
2714}